The dinoflagellates Durinskia baltica and Kryptoperidinium foliaceum retain functionally overlapping mitochondria from two evolutionarily distinct lineages

Background The dinoflagellates Durinskia baltica and Kryptoperidinium foliaceum are distinguished by the presence of a tertiary plastid derived from a diatom endosymbiont. The diatom is fully integrated with the host cell cycle and is so altered in structure as to be difficult to recognize it as a diatom, and yet it retains a number of features normally lost in tertiary and secondary endosymbionts, most notably mitochondria. The dinoflagellate host is also reported to retain mitochondrion-like structures, making these cells unique in retaining two evolutionarily distinct mitochondria. This redundancy raises the question of whether the organelles share any functions in common or have distributed functions between them. Results We show that both host and endosymbiont mitochondrial genomes encode genes for electron transport proteins. We have characterized cytochrome c oxidase 1 (cox1), cytochrome oxidase 2 (cox2), cytochrome oxidase 3 (cox3), cytochrome b (cob), and large subunit of ribosomal RNA (LSUrRNA) of endosymbiont mitochondrial ancestry, and cox1 and cob of host mitochondrial ancestry. We show that all genes are transcribed and that those ascribed to the host mitochondrial genome are extensively edited at the RNA level, as expected for a dinoflagellate mitochondrion-encoded gene. We also found evidence for extensive recombination in the host mitochondrial genes and that recombination products are also transcribed, as expected for a dinoflagellate. Conclusion Durinskia baltica and K. foliaceum retain two mitochondria from evolutionarily distinct lineages, and the functions of these organelles are at least partially overlapping, since both express genes for proteins in electron transport.


Background
The endosymbiotic origins of plastids and mitochondria share a number of characteristics in common, [1,2], but differ in the complexity of their evolutionary history following their origin and initial integration. Whereas mitochondria originated once and have apparently never been lost [3][4][5], plastids have spread between eukaryotic lineages several times in events referred to as secondary and tertiary endosymbioses. Generally these secondary and tertiary endosymbionts have degenerated so far that all that remains is a plastid with extra membranes [6], but in a few exceptional cases intermediate stages of reduction are known, and these may provide interesting glimpses into how complexity is lost.
One of the characters that is absent from nearly all known examples of secondary and tertiary endosymbionts is the mitochondrion. This contrasts with the fact that mitochondria have never been lost in any other eukaryotic lineage. Even in the most severely reduced, anaerobic parasites which lack oxidative phosphorylation, highly reduced organelles called mitosomes and hydrogenosomes are found [3][4][5]. Some of these have no direct role in energy metabolism, but iron-sulfur cluster biosynthesis is a common function [5,7,8]. These relict organelles suggest mitochondria are resistant to outright loss, raising questions about why mitochondria appear to be one of the more dispensable features of algae taken up during secondary and tertiary endosymbiosis events.
The single clear exception to this is found in a group of related dinoflagellates that harbour a diatom tertiary endosymbiont. This group contains several species (see [9] for a recent summary), and here we have examined two: Durinskia baltica [10] and Kryptoperidinium foliaceum [11,12]. Several of these genera (including Durinskia and Kryptoperidinium) have been shown to share a common pennate diatom endosymbiont, arguing that the endosymbiosis is stable through evolutionary time [13,14]. Interestingly, this may not hold for the whole group, since the endosymbiont of Peridinium quinquecorne is a centric diatom [15], suggesting that the integration may have spanned a long period of time and different transient endosymbionts were ultimately fixed in these two subgroups. Nevertheless, the endosymbionts of D. baltica and K. foliaceum are no longer transient in the short term, they have lost motility and cell wall and, although some chromatin condensation occurs during sexual reproduction in D. baltica, typical chromosomes are not found within the endosymbiont nucleus at any stage of its life cycle [16,17]. During the endosymbiotic nuclear division, neither a spindle apparatus nor any microtubules have been observed [18], and the amitotic division of this nucleus results in unequal daughter nuclei and significantly larger amount of DNA in the nucleus than that reported in other diatoms [19].
The endosymbiont has clearly been reduced in many ways, but some of its most interesting characteristics are what it has retained. This includes plastids surrounded by endoplasmic reticulum (ER) that is continuous with the outer membrane of the nucleus, a plasma membrane that separates it from the host cytoplasm, a multi-lobed, prominent nucleus with a genome, ribosomes, dictyosomes, and mitochondria [16,20].
It is the retention of mitochondria that makes this endosymbiont stand out, in particular since D. baltica and K. foliaceum host cells have also been reported to retain mitochondria. The loss of endosymbiont mitochondria in virtually all known examples of secondary and tertiary endosymbiotic events suggests retaining two mitochondria is either unnecessary or even deleterious. The loss of one of these organelles may be ongoing, but it is also possible that both compartments require mitochondrial function or that they have distributed essential functions between them. We have previously shown [9] that the K. foliaceum endosymbiont mitochondrion contains a genome and expresses genes for cytochrome c oxidase subunit 1 (cox1), cytochrome c oxidase subunit 3 (cox3), and cytochrome b (cob). However, no data are available from the host mitochondrion, and with the function of this organelle completely unknown we cannot address the possibility that the two organelles have overlapping or differentiated function.
In order to determine whether this unique mitochondrial redundancy extends to the functional level, we characterized seven mitochondrial genes of D. baltica: five from the endosymbiont (cox1, cox2, cox3, cob, and LSUrRNA) and two from the host (cox1 and cob), and confirmed that cox2 and LSUrRNA from the endosymbiont and cox1 from the host are also present in K. foliaceum. Most significantly, in D. baltica, cox1 and cob are present and expressed in both mitochondria, and those in the host are heavily edited, as expected for a functional dinoflagellate mitochondrial gene [21][22][23]. All available data therefore suggest that both the host and endosymbiont mitochondria are actively expressing genes functional in oxidative phosphorylation and energy production.

Results and discussion
Characterization of endosymbiont mitochondrial genes and transcripts PCR amplification using diatom-specific primers and total D. baltica DNA (or RNA, see below), resulted in fragments of the expected sizes for five genes: cox1, cox3, cob, cox2, and LSUrRNA. Sequencing multiple clones yielded a single copy of each gene. Only a short fragment of cox1 could be amplified from DNA, probably due to the presence of long type II introns such as those found in diatoms Phaeodactylum tricornutum and Thalassiosira pseudonana and the endosymbiont of K. foliaceum [9], so the remainder was recovered from RNA by RT-PCR. Diatom-derived genes for cox1, cox3 and cob are already known from K. foliaceum [9], so to complement the D. baltica data we also amplified the K. foliaceum cox2 and LSUrRNA.
The phylogenies of all five genes ( Fig. 1, 2, 3, 4, 5) generally resembled trees based on nuclear genes, with relatively strong support for the monophyly of alveolates and Protein maximum likelihood phylogeny of cytochrome c oxidase 1 (cox1) Figure 1 Protein maximum likelihood phylogeny of cytochrome c oxidase 1 (cox1). Numbers at nodes indicate bootstrap support for major nodes over 50% from ML (top) and distance (bottom). A dash (-) indicates support less than 50%. Major groups are labeled to the right, with diatoms (red) and dinoflagellates (purple) indicated by a box and D. baltica and K. foliaceum genes in black. sister relationship between dinoflagellates and apicomplexans [24,25]. We also noted that haptophytes and cryptophytes are sister groups with strong support in cob trees ( Fig. 4) and weakly so in cox1 and cox3 trees ( Fig. 1 and 3), as has been suggested in other analyses [26][27][28][29]. Most importantly, however, in all five phylogenies the distinction between the expected positions of host and endosymbiont-derived genes was unambiguous, and in all five trees the D. baltica gene amplified with diatom-specific primers branched within the diatom clade with strong support (Fig. 1, 2, 3, 4, 5). Moreover, D. baltica consistently grouped with K. foliaceum with strong support (with the exception of distance analysis of cox1, where the overlap between the two sequences is relatively short). This is in disagreement with the proposal that these dinoflagellates are products of separate endosymbiotic events [30], but consistent with the analyses of nuclear small-subunit rRNA genes from the hosts in these two dinoflagellates [14] and the hypothesis that these two species, and most likely their other close relatives, resulted from a single endosymbiotic event [13,15,31,32]. Together, D. baltica and K. foliaceum branched specifically with pennate dia-toms (i.e. Nitzschia or Cylindrotheca, Phaodactylum), which is also consistent with evidence that the endosymbiont is a descendent of a pennate diatom [9,14,33,34]. Overall, these trees strongly support the conclusion that cox1, cox3, cob, cox2, and LSUrRNA are all present in the mitochondrial genome of the endosymbiont.
To confirm that all five genes are actively expressed, each was also amplified from RNA using RT-PCR. All diatomderived cDNA sequences were identical to their corresponding genes, providing further evidence for these genes being from the endosymbiont mitochondrion, because the mitochondrial transcripts in dinoflagellates are extensively edited [22,35,36].

Characterization of host mitochondrial genes and transcripts
The presence and expression of cox and cob genes in the endosymbiont mitochondria suggests this organelle is engaged in electron transport. The pressing question is therefore the nature of the host organelle, but no data have been gathered from it for comparison. Dinoflagellate Protein maximum likelihood phylogeny of cytochrome oxidase 2 (cox2) Figure 2 Protein maximum likelihood phylogeny of cytochrome oxidase 2 (cox2). Numbers at nodes indicate bootstrap support for major nodes over 50% from ML (top) and distance (bottom). A dash (-) indicates support less than 50%. Major groups are labeled to the right, with diatoms (red) and dinoflagellates (purple) indicated by a box and D. baltica and K. foliaceum genes in black.

Kryptoperidinium foliaceum
Diatoms mitochondrial genomes only encode three protein-coding genes: cox1, cox3, and cob [23]. The LSU rRNA has been extensively fragmented and rearranged [37], and cox2 has been split and moved to the nucleus [38], so only cox1, cox3, and cob were sought in the mitochondrial genome. We used dinoflagellate-specific primers for all three genes in RT-PCR with total D. baltica RNA, and identified fragments of cox1 and cob. Using 3' RACE, we also recovered the 3' end of the D. baltica cob gene. We also recovered 6 additional copies of cox1 transcripts (Fig. 6), each of which contained inserts that differed in position, size, and sequences, and disrupted the reading frame. Inserts ranged from 81 to 453 bp. Two inserts at slightly different positions of RNA2 and RNA3 were similar in size and sequences, and both of these contained a 151 bp portion of the cob gene flanked by two small (about 20 bp) noncoding fragments (Fig. 6). Another insert contained a 75 bp fragment with over 90% identity to non-coding fragments from the dinoflagellates Alexandrium catenella (Genbank accession: AB265207) Gonyaulax polyedra (Genbank accession: AF142472). We did not sequence a DNA clone lacking an insert, but the mitochondrial genome of other dinoflagellates is known to contain many copies of each gene and many rearrangements [21,23], so the intact copy of the gene was most likely simply not sampled. Given the highly fragmented and divergent nature of dinoflagellate mitochondrial rRNAs [37], it is possible this 75 bp represents an as yet unidentified fragment of either the LSU or SSU.
PCR using genomic DNA from D. baltica resulted in a single cob gene fragment (12 identical clones were sequenced), and six different cox1 fragments (seven different clones were sequenced). As with cDNAs, each cox1 gene contained a unique insert (Fig. 6). Most inserts were unique, but many contained small imperfect repeats in common, and the positions of the inserts within cox1 were variable and inevitably a portion of cox1 was missing at Protein maximum likelihood phylogeny of cytochrome oxidase 3 (cox3) Figure 3 Protein maximum likelihood phylogeny of cytochrome oxidase 3 (cox3). Numbers at nodes indicate bootstrap support for major nodes over 50% from ML (top) and distance (bottom). A dash (-) indicates support less than 50%. Major groups are labeled to the right, with diatoms (red) and dinoflagellates (purple) indicated by a box and D. baltica and K. foliaceum genes in black.  (Fig. 6, DNA1 and RNA3) were found to be identical, with the exception of edited sites (see below).
A single transcript of the K. foliaceum cox1 was also recovered from sequencing 15 identical clones. This transcript also contained an insert, but without sequence similarity to anything known and not at a position in common with any D. baltica clone. We failed to identify cob, but considering its presence in D. baltica we feel it is unlikely to be absent in K. foliaceum.
The many variants of the D. baltica host cox1 gene and the insert in the K. foliaceum cox1 gene are both consistent with the nature of mitochondrial genomes in other dinoflagellates. This has been best described in Crypthecodinium cohni and Amphidinium carterae, where protein-coding genes are flanked by non-coding, repeat-rich sequences and that the context of a gene can vary in different copies due to homologous recombination [21,39], and in Oxyrrhis marina where protein coding genes are found in different genomic contexts, and are often fragmented [23].
Protein maximum likelihood phylogeny of cytochrome b (cob) Figure 4 Protein maximum likelihood phylogeny of cytochrome b (cob). Numbers at nodes indicate bootstrap support for major nodes over 50% from ML (top) and distance (bottom). A dash (-) indicates support less than 50%. Major groups are labeled to the right, with diatoms (red) and dinoflagellates (purple) indicated by a box and D. baltica and K. foliaceum genes in black.

Host mitochondrial transcripts are extensively edited
Dinoflagellate mitochondria possess a distinctive form of RNA editing. Editing sites typically involve A to G, T to C, and C to U changes at first and second positions, affecting about 2% of positions in cox1 and cob genes [22,35,40].
The presence of such editing would provide further evidence for the dinoflagellate mitochondrial location of the cox1 and cob genes identified here, so all D. baltica gene fragments were aligned to their respective transcripts and conserved editing sites examined. In total, 786 bp of cox1 were comparable, and 352 bp of cob were comparable, from which 11 and 7 edited sites were identified, respectively. The nature of these edits was similar to that of other dinoflagellates (Table 1) and, significantly, all but three of the editing sites were conserved in other dinoflagellate mitochondria [22,35,40]. Overall, the characteristics of this editing are consistent with these genes and cDNAs being located in the dinoflagellate host mitochondria.

Reduction and functional redundancy of mitochondria
Electron microscopy has shown that mitocondria exist in both the host and endosymbiont cytosolic compartments of D. baltica and K. foliaceum [11,12,16]. More recently diatom-derived genes for cox1, cox3, and cob have been shown to be expressed in K. foliaceum [9]. However, without comparable data from the host mitochondria it is impossible to determine whether the two organelles are functionally redundant, or have distributed functions between them. Here, we have shown that both organelles contain at least two genes with central functions in electron transport, cox1 and cob. Accordingly, these two spe-Schematic representation of cox1 gene and cDNA fragments characterised from Durinskia baltica  cies are unique among eukaryotes in having retained active, functional mitochondrial genomes from two distantly related eukaryotic lineages, the dinoflagellate host and the pennate diatom endosymbiont. Moreover, these organelles now appear to be at least partially functionally redundant, since both express genes for proteins in the electron transport chain.
Indeed, no characteristic of either host or endosymbiont mitochondrial genes or genomes has so far been shown to be significantly different from those of their dinoflagellate or diatom relatives, which points to the conclusion that neither organelle has been much affected by the presence of the other. It seems unlikely that the two mitochondria are retained due to functional differentiation, but their genetic redundancy may be related to spatial differentiation. If the membrane separating the host and the endosymbiont, which is thought to be derived from the diatom plasma membrane [20], were deficient in transporters to efficiently shuttle either substrates or products of mitochondrial reactions between the two compartments, then neither compartment could eliminate those functions without consequences. In other endosymbiotic events such difficulties would have been overcome or made irrelevant by the continued reduction of the endosymbiont and integration with the host. Whether the mitochondrion of D. baltica is a snapshot in the progression of events that will ultimately lead to its loss, or whether the process has been 'stuck' in some way is unknown. Simi-larly, although it is generally assumed that the endosymbiont will be reduced and the corresponding host feature retained, this does not need to be the case. With the already highly reduced and unusual nature of dinoflagellate mitochondrial genome [21,23,39], it is not unreasonable to hypothesize that the host organelle may be lost as easily as the relatively normal mitochondrion of the endosymbiont.

Conclusion
We have shown that two related dinoflagellates, D. baltica and K. foliaceum, retain redundant genes in their hostderived and endosymbiont-derived mitochondrial genomes, including several genes related to electron transport. Host-derived genes are edited at the RNA level and subject to extensive recombination, as is expected for dinoflagellate mitochondria. All genes characterized have been shown to be expressed at the mRNA level, suggesting the two organelles overlap in function, making these unique among eukaryotes in retaining two partially redundant mitochondria with different evolutionary origins.

Culture conditions and nucleic acid extraction
Cultures of Durinskia baltica (Peridinium balticum) CS-38 were obtained from CSIRO Microalgae Supply Service (CSIRO Marine and Atmospheric Research Laboratories, Tasmania, Australia) and maintained in GSe medium at
We also used three pairs of dinoflagellate-specific primers to search for the host cob in K. foliaceum, which were based on the most conserved regions of this gene found in dinoflagellate mitochondria. Two pairs of these primers were tested successfully to amplify this gene from D. baltica (data not shown). However, no product was obtained with any of these primers from the total DNA or RNA extracted from K. foliaceum used in PCR and RT-PCR respectively.
All PCR and RT-PCR products were gel purified and cloned using pCR 2.1 TOPO Cloning kit (Invitrogen). In each case, several clones were sequenced on both strands using BigDye terminator chemistry. New sequences have been deposited into GenBank as accessions EF434607-EF434629.

Phylogenetic analyses
The conceptual translations of new cox1, cox2, cox3, and cob, and DNA sequences for LSUrRNA from D. baltica and K. foliaceum were aligned with homologues from public database using ClustalX 1.83.1 [41] under the default gap opening and gap extension penalties and the alignments edited manually. Phaeodactylum tricornutum homologues were kindly provided by Marie-Pierre Oudot-Le Secq from the P. tricornutum genome sequencing project [42]. Phylogenetic analyses were carried out including a diversity of eukaryotes to determine the overall position of new sequences, and subsequently restricted to homologues from chromalveolate taxa (dinoflagellates, apicomplex-ans, ciliates, heterokonts, haptophytes, and cryptomonads), since both the host and endosymbiont are thought to be members of this supergroup [43]. No LSUrRNA sequences for dinoflagellates or apicomplexans were included since these genes are fragmented, only partially described to date, and highly divergent, so ciliates alone represent alveolates. These alignments consisted of 52, 20, 20, and 26 amino acid sequences, and 17 DNA sequences with 378, 109, 251, 372, and 1304 unambiguously aligned sites for cox1, cox2, cox3, cob, and LSUrRNA, respectively. For cox1 and cob sequences, several alignment alternatives were attempted, which were independently followed by phylogenetic analyses. These alternative alignments differed only in the inclusion or exclusion of missing sites. In order to make use of all the recovered sequence data, the alignments of cox1 and cob that were analysed included some missing data and either the 5' or 3' ends of certain D. baltica or K. foliaceum genes. Phylogenies of cox1 and cob were also performed using excluding either host or endosymbiont genes, resulting in no significant differences (not shown). All alignments are available upon request.
Phylogenetic trees were inferred using maximum likelihood and distance methods. The proportion of invariable sites (i) and shape parameter alpha (α) with 8 variable rate categories were estimated from the data with PhyML 2.4.4 [44] under the Whelan and Goldman (WAG) model of substitution for cox1, cox2, cox3, and cob phylogenies, and under General Time-Reversible (GTR) model of substitution for LSUrRNA phylogeny with the frequency of amino acid or nucleotide usage calculated from the data. The i and α parameters estimated from the data were 0.000, 0.057, 0.004, 0.010, and 0.113, and 1.002, 1.297, 1.186, 1.402 and 1.002 for cox1, cox2, cox3, cob, and LSUr-RNA, respectively. For all five data sets 1,000 bootstrap replicates were analyzed using PhyML. For distance trees, distances were calculated using TREE-PUZZLE 5.2 [45] with 8 variable rate categories and invariable sites. The i and α parameters were estimated by TREE-PUZZLE to be 0.00, 0.05, 0.00, 0.00, and 0.10, and 0.94, 1.18, 1.03, 1.12, and 0.91 for cox1, cox2, cox3, cob, and LSUrRNA, respectively. Trees were constructed by weighted neighbor-joining using WEIGHBOR 1.0.1a [46]. Distance bootstrapping of 1,000 replicates was carried out using PUZZLEBOOT (shell script by A. Roger and M. Holder).