Hemiselmis andersenii mtDNA was isolated and sequenced to ~10× coverage as described in Lane et al. 25. About 1,200 end sequences were screened for quality and vector contamination with Pregap4 and automatically assembled using gap4 version 4.10 in the Staden package 26. Complete automated assembly of a large intergenic space between trnS and cox2 was unsuccessful due to the highly repetitive nature of this region. In an attempt to manually resolve this area, short (~30 bp) unique sequences within the trnS and cox2 genes were used to probe the sequence database for reads that extended from these two loci into the repeat region. These sequences were extracted and manually aligned using MacClade version 4.08 27. Sequences at the ends of the new constructs were then selected and the process was repeated. However, due to the presence of multiple identical copies of a >500 bp repeat, the assembly of a single unambiguous contig was not possible. When all available sequence reads were considered, three robust contigs were produced, each ending with similar repetitive sequences consisting of a ~340 bp repeat unit. These three contigs were joined to circularize the map. The complete H. andersenii mtDNA has been submitted to GenBank under the following accession number: EU651892.

DNA secondary structure within the repeat region was predicted using mfold version 3.2 28 at a folding temperature of 37°C and the ionic conditions of 1.0 M [Na⁺] and 0.0 M [Mg⁺⁺].

We used pulsed-field gel electrophoresis (PFGE) to obtain an independent size estimate of the H. andersenii mitochondrial genome. Hemiselmis andersenii total DNA plugs were prepared as described in Lane et al. 29 and digested overnight with the restriction enzymes PstI or BglII (Fermentas, Hanover, MD, USA). Based on the genome sequence, these enzymes were predicted to cut the mtDNA only once or twice. Both untreated and enzyme-digested H. andersenii DNA plugs were run on a 1% agarose gel (1× TBE) in 0.5× TBE buffer at 14.0°C for 18 h at a voltage of 6.0 V/cm with a switch time between 1–25 s using a CHEF-DR III Pulsed-Field Gel Electrophoresis System (Bio-Rad Laboratories, Hercules, CA, USA). DNA on the pulsed-field gel was transferred to a nylon membrane. Southern hybridization using a ~700 bp coxI probe as in Lane and Archibald 30 revealed that undigested mitochondrial DNA molecules were trapped in the wells or found in the 'compression zone'. The PstI or BglII endonuclease treated DNA plugs revealed mitochondrial molecules in a discrete band below the 'compression zone'. The corresponding bands could not be visualized on the ethidium-bromide stained pulsed-field gel image because of nuclear and nucleomorph DNA smears in the background. In order to visualize mtDNA on the pulsed-field gel, an initial PFGE run was used to remove the linear nuclear and nucleomorph chromosomes from the PFGE plugs. These plugs, which still contained organellar DNA, were subsequently removed from the gel and digested with the restriction enzymes PstI and BglII. Digested plugs were then inserted into a fresh gel and electrophoresed under the conditions described above. The 5 Kbp and Lambda CHEF DNA Size Standard (Bio-Rad Laboratories, Hercules, CA, USA) were used to estimate the size of the enzymatically linearized H. andersenii mtDNA.

Annotation of the H. andersenii mtDNA and the GC content and skew analyses were performed in Artemis version 8 31. Gene identification was carried out using BLASTX and BLASTN. Small and large ribosomal rRNA subunit genes were identified by comparison to rRNA gene sequences in the mitochondrial genome of Rhodomonas salina. Transfer RNAs were identified using tRNAscan-SE version 1.21 32.

The extent to which the H. andersenii and R. salina mitochondrial genomes are rearranged to each other was estimated using GRIMM 33. Each genome was designated as a sequence of 63 units, which include a repeat region and 62 genes common between the two cryptophyte mtDNAs.

tRNAscan-SE version 1.21 32 identified a putative intron of ~20 bp in the anticodon loop of the H. andersenii trnK(uuu) gene. To determine whether this prediction was correct, we performed RT-PCR using Lysine-tRNA-specific primer pairs and H. andersenii total RNA provided by H. Khan. To eliminate DNA contamination, 1 μl of total RNA was incubated for 30 min with RQ1 RNase-Free Dnase (Promega, Madison, WI, USA). RT-PCR was performed using the QIAGEN one-step RT-PCR kit (QIAGEN, Valencia, CA, USA) and with control reactions in which the reverse-transcription process was skipped. The following two pairs of primers were used: 1) The forward primer 5'-GAAGGTTGCTCGAATGGAA-3' with the reverse primer 5'-GAAGGTATAGGAATTGAACCTATTC-3' 2) and the forward primer 5'-GCCCAGAAGGTTGCTC-3' with the reverse primer 5'-AAGAAGGTATAGGAATTGAACCTAT-3'. RT-PCR was performed with the reverse transcription step for 30 min at 50°C and the subsequent inactivation of reverse transcriptase and activation of HotStart Taq DNA polymerase for 15 min at 95°C, followed by 35 cycles at 94°C for 1 min, 47°C for 1 min, and 72°C for 1 min, and a final extension at 72°C for 10 min. The amplified PCR fragments were cloned into pCR4-TOPO vector in the TOPO TA cloning kit for sequencing (Invitrogen, Carlsbad, CA, USA). Between 5 and 10 bacterial colonies from each reaction were selected for sequencing on a Beckman Coulter CEQ8000 (Beckman Coulter Inc., Fullerton, California, USA).

From the 36 protein-coding genes found in the H. andersenii mtDNA, 25 were selected for phylogenetic analyses. Eleven genes (atp8, nad8, rps2, rps3, rps4, rps7, rps8, rps13, rpl5, rpl6, tatC) were excluded because their sequences were poorly conserved and/or were only present in a few taxonomic groups. H. andersenii protein sequences were aligned with their homologs from other mitochondrial genomes available from GenBank. Amino acid sequences were aligned using MacClade version 4.08 27 and ambiguously aligned sites were manually removed. In addition to individual protein analyses, a concatenated protein data set containing 25 proteins was analyzed. To include the maximum number of gene sequences, we combined 25 protein-coding gene sequences encoded in 18 mitochondrial genomes across diverse eukaryotic taxa. As most mitochondrial genomes do not possess all 25 protein-coding genes selected for analysis, as many as 12 protein gene sequences were missing per taxon. A maximum likelihood tree was produced using RAxML-VI-HPC version 2.2.3 34 with the PROTOMIXJTT model of sequence evolution and the automatic tree rearrangement setting, and from 100 distinct randomized maximum parsimony starting trees. Bootstrap analysis was based on 100 re-samplings.

The mitochondrial DNA of the cryptophyte Hemiselmis andersenii CCMP644 was sequenced, assembled and manually edited to produce a circular-mapping genome 60,553 bp in size (Figure 1). Genome assembly was complicated by the presence of a highly repetitive non-coding region of ~20 Kbp (see below); genome size was thus verified using pulsed-field gel electrophoresis (PFGE). Several observations suggest that the H. andersenii mtDNA exists primarily in a linear-branched form comprised of multiple genome units. In PFGE, the H. andersenii mtDNA remains in the well or migrates within the 'compression zone' (i.e., the unresolved portion of DNA near the top of the gel), which contains primarily linear nuclear and nucleomorph chromosomes larger than ~150 Kbp (data not shown). The lack of mtDNA below the 'compression zone' suggests that the H. andersenii mtDNA is not composed of linear monomers or dimers. Furthermore, when the H. andersenii mtDNA is partially digested with PstI, an enzyme predicted to cut the genome only once, it produces a discrete band of ~60 Kbp in size (data not shown) but not a band ~120 Kbp in size, which would correspond to a dimeric linear form of the genome. This result indicates that the H. andersenii mtDNA is not composed of circular concatemers or linear head-to-tail concatemers consisting of three or more genomic units. Therefore, we suggest that the H. andersenii mtDNA exists primarily as a branched linear molecule although monomeric circles may also exist. Further studies using transmission electron microscopy or the 'moving picture' technique 10 will be necessary to confirm this hypothesis.

The H. andersenii mitochondrial genome is comprised of a gene-rich region ~40 Kbp in size and a large (19,675 bp) intergenic region between trnS and cox2 with complex repeats (Figures 1 and 2). The intergenic region accounts for 32.5% of the entire genome and 83.5% of the total amount of non-coding DNA (23,549 bp). The overall GC content of the genome is 28.72%, slightly higher than that of the nucleomorph genome of this organism 25. Interestingly, a ~40 bp region near the start of the coding portion of the genome is very GC-rich (78.38%) and is followed by a 100% AT-containing region ~190 bp in size (Figure 3). This unusual stretch of sequence is about 70 bp from a palindromic sequence that is predicted to form a Type II stem-loop (Figures 2 and 3; see discussion below), and may be involved in regulating replication or transcription.

The H. andersenii mitochondrial genome encodes 66 genes with predicted functions and 8 hypothetical protein-coding genes, a total somewhat higher than the average for eukaryotes (40–50 genes) 1. Ten genes–orf167, orf71, rps13, rps11, nad3, rps2, tatC, 'trnK (uuu)', rps12, and rps7–overlap by up to 51 bp, emphasizing the extreme compactness of the coding portion of the genome. The genome encodes small and large rRNA subunit genes and 28 tRNAs, one of which may be a pseudogene (see discussion below). Of the 36 identifiable protein-coding genes, 14 encode ribosomal proteins, 21 are involved in oxidative phosphorylation, and one gene encodes a membrane translocase protein (Table 1).

Comparison of the mtDNA gene order in H. andersenii to other genomes reveals the presence of five gene clusters shared among distantly related protists: two ribosomal protein clusters (rps12-rps7-rps19-rps3-rpl16-rpl14-rpl5-rps14 and rps8-rpl6-rps13-rps11) and three NADH dehydrogenase clusters (nad4L-nad5; nad4-nad2; nad10-nad9). These gene clusters have been suggested to represent vestiges of bacterial operons 1224. Interestingly, all 74 genes in the H. andersenii mitochondrial genome are encoded on the same strand. While the evolution of such an arrangement seems improbable, absolute strand polarity has been observed in the mitochondrial genomes of diverse eukaryotes such as the amoeba Acanthamoeba castellanii (59 genes), the fungus Penicillium marneffei (47 genes), and the green alga Chlamydomonas eugametos (20 genes) 353637. In addition, strikingly similar mtDNA architectures–gene-dense regions, a single large repetitive intergenic region, and all genes encoded on one strand–are seen in diverse protists such as the stramenopile Thraustochytrium aureum (The Organelle Genome Megasequencing Program; http://megasun.bch.umontreal.ca/ogmp/) and the green alga Pedinomonas minor 38. Understanding the biological significance of such convergence at the level of genome architecture will require comparative molecular and biochemical studies of mitochondria in these organisms.

H. andersenii is only the second cryptophyte, after R. salina 24, for which a mitochondrial genome has been completely sequenced and annotated. Comparative analyses of the two genomes revealed a number of similarities. Both genomes feature a compact gene arrangement and a single large repeat region (Figure 1) 24, although the size of the large intergenic region in H. andersenii (~20 Kbp) is more than four times as large as that of R. salina (~4.7 Kbp). All of the 36 predicted protein-coding genes in the H. andersenii mitochondrial genome are present in the R. salina mtDNA. Four R. salina mitochondrion-encoded genes–rps1, atp4, tatA, and sdh4–are not found in H. andersenii, although two open reading frames, orf45 and orf91, in the H. andersenii mtDNA show marginal sequence similarity to the R. salina tatA and sdh4 genes, respectively. Additionally, while two group II introns are present in R. salina mtDNA, the H. andersenii mtDNA is devoid of introns (Table 2) 24.

With respect to conservation of gene order, 64.5% of the shared genes between the two cryptophyte mitochondrial genomes (40 out of 62 genes–36 protein-coding genes, 24 tRNA genes (see below), 2 rRNA genes) are present in thirteen syntenic blocks, each consisting of 2–7 genes. These include: 1) cox1-cob-nad11, 2) nad4L-nad5, 3) atp1-trnP(ugg), 4) rps8-rpl6-rps13-rps11, 5) trnC(gca)-atp6, 6) trnI(gau)-trnQ(uug)-trnR(gcg)-trnE(uuc)-trnW(cca)-nad10-nad9, 7) nad4-nad2, 8) trnR(ucu)-trnG(ucc), 9) trnM(cau)f-trnS(uga), 10) trnY(gua)-trnL(uag), 11) tatC-'trnK(uuu)' [H. andersenii] /trnS(gcu) [R. salina]-nad7, 12)cox3-rps12-rps7-rps19, and 13) rps3-rpl16-rpl14-rpl5-rps14. As noted earlier, some of the conserved gene clusters, such as nad4L-nad5, are found in distantly related eukaryotes and appear to be vestiges of bacterial operons. Analysis using GRIMM 33 suggests that the observed difference in gene order between the two cryptophyte mitochondrial genomes can be explained by at least 31 instances of genome reversal events.

The R. salina mtDNA is characterized by a pair of ~1.5 Kbp inverted repeats that are joined by 112 bp of sequence 24. In contrast, repeats in the H. andersenii mitochondrial genome are not inverted, but are instead dispersed or arranged in tandem throughout the large non-coding region, with individual repeat units ranging from 22 to 336 bp and occurring up to 100 times (Figure 2). Given that R. salina and H. andersenii are distantly related to one another 29, the large repeat region presumably arose during or prior to the early diversification of cryptophytes. While there is no obvious sequence similarity between the two repeat regions, both contain multiple copies of palindromic sequences, which are predicted to form stable stem-loop DNA structures 24. In H. andersenii, two types of stem-loop structures were identified–I and II–using the DNA MFOLD program 28. The Type I structure has two slight variations, I-a and I-b, which occur 21 and 5 times, respectively (Figures 2 and 3). Type I-a and I-b structures have 22 and 20 base pairings in their stems, respectively, and occur adjacent to tandem repeats (Figures 2 and 3). One copy of the type II stem-loop structure is located within a ~300 bp segment that is devoid of any discernable repeat units, but close to the high and low GC regions noted earlier (Figures 2 and 3). As was suggested for R. salina by Hauth et al. 24, tandem repeats and multiple stem-loop structures in H. andersenii mtDNA might be involved in the regulation of transcription and replication, a hypothesis that needs to be tested further.

Hauth et al. 24 demonstrated that the repeat region of the R. salina mtDNA roughly coincides with a change in the direction of 'cumulative GC skew' [calculated as (G-C)/(G+C)] and suggested that the repeat corresponds to the origin of replication. We investigated the GC skew in the H. andersenii mitochondrial genome to see whether a similar pattern exists. Unlike R. salina, however, the H. andersenii GC skew does not change direction near the repeat region. Instead, in both the H. andersenii and R. salina mtDNA, observed GC skew patterns strongly correlate with transcriptional orientations, where the coding strand tends to be G-rich (data not shown). Therefore, the GC skew patterns of the two cryptophyte mitochondrial genomes do not seem to be the result of replication-associated mutational bias, but rather the non-random distribution of the protein coding genes, as has been observed in some other genomes 39. Nevertheless, based on the presence of other features such as stem-loop structures, it seems reasonable to assume that the repeat region in both cryptophyte mitochondrial genomes corresponds to the origin of replication.

The H. andersenii mtDNA encodes 28 tRNAs, 27 of which are predicted to form standard cloverleaf secondary structures. One tRNA gene, 'trnK(uuu)', shows atypical structure in the anticodon loop and the variable region, and is probably a pseudogene (Figure 4A). Allowing for wobble pairings and some base modifications, 26 tRNAs are the theoretical minimum required to cover all codons in bacteria. For some mitochondria, even smaller sets of tRNAs, as few as 22–23, are possible by adopting several additional strategies 40. The H. andersenii mitochondrial genome lacks only one tRNA gene, trnK(uuu), which is minimally required in order to recognize all 61 codons (Table 3). It is thus predicted that nuclear-encoded cytosolic Lys-tRNA is imported into H. andersenii mitochondria. Mitochondrial tRNA import has been demonstrated in apicomplexans and trypanosomatids where tRNA genes are completely missing in their mitochondrial genomes 41, as well as in ciliates and plants where mitochondrial genomes encode fewer than the 22–23 minimally required tRNA genes 42. Although most animals and some fungi do not import tRNAs into mitochondria 43, the fungus Saccharomyces cerevisiae has been shown to import one specific cytosolic tRNA even though its mitochondrial genome encodes the full complement of tRNAs 44. Analyses of the tRNA repertoire of mitochondrial genomes suggest that a number of other protist taxa across the eukaryotic tree also import one or more tRNAs into their mitochondria 4345. It is thus reasonable to assume that H. andersenii imports at least Lys-tRNA, although it is possible that tRNA editing makes up for the Lys-tRNA deficit by changing the identity of an existing tRNA, as has been shown in marsupials 46.

Another possible mechanism to account for the missing tRNA is that the structurally abnormal 'trnK(uuu)' gene (Figure 4A) forms a functional Lys-tRNA to decode the codons AAA and AAG. Several cases of atypically-structured tRNAs are known from animal and ciliate mitochondria 4748. Interestingly, tRNAscan-SE 32 predicted the existence of a 20 bp intron within the H. andersenii 'trnK(uuu)', and we conducted further experiments to test whether this is indeed the case. RT-PCR experiments using primer sets specific for 'trnK(uuu)' indicated that the putative intron was not removed in the mature tRNA. This results is not unexpected, given that the 20-bp putative intron is too short to be a self-splicing group I or II intron, which are the only known types of introns reported in mitochondrial genomes 49. Sequencing of ~20 clones also did not reveal any evidence for RNA editing within the 'trnK(uuu)'. These results suggest that if 'trnK(uuu)' is indeed expressed to form a functional Lys-tRNA, it is predicted to have an unusually AU-rich stem in the codon loop and a long variable region, atypical for Lys-tRNA (Figure 4A). Long variable regions ranging from 11 to 23 nucleotides are generally restricted to tRNA-Leu, tRNA-Ser, and bacterial tRNA-Tyr 40. The D- and T-loops of the 'trnK(uuu)' sequence show sequence similarity to one of the two mitochondrion-encoded tRNA-Ser genes (Figure 4A and 4B), both of which have a long variable region. In addition, comparative analysis with the R. salina mtDNA revealed genomic position conservation between the H. andersenii trnS-like 'trnK(uuu)' gene and the trnS(gcu) gene of R. salina, flanked by the tatC and nad7 genes. The H. andersenii 'trnK(uuu)' and R. salina trnS(gcu) genes both overlap tatC by 51 bp and 22 bp, respectively. This strongly suggests that the H. andersenii 'trnK(uuu)' is indeed derived from an ancestral gene that encoded tRNA-Ser, explaining the origin of its long variable region. The overlap between the H. andersenii 'trnK(uuu)' and tatC suggests that 'trnK(uuu)' may play a role in processing the 3' end of the tatC gene transcript. This hypothesis could explain why the 'trnK(uuu)' gene still remains in the genome and retains conserved secondary structure in the stem loop and D- and T-loops, even if it does not form a functional tRNA. Comprehensive molecular and biochemical experimentation will be necessary to confirm or refute the existence of mitochondrial tRNA import in H. andersenii and the functionality of the unusual 'trnK(uuu)' gene.

When the H. andersenii tRNA genes were compared to those of R. salina, 24 homologous pairs of tRNAs were identified, leaving only four H. andersenii tRNA and three R. salina tRNA genes not unambiguously matched to each other. Each of the tRNA pairs possess identical anticodons except for the H. andersenii 'trnK(uuu)' and R. salina trnS(gcu) pair, despite their common derivation. The trnS(gcu) of H. andersenii, having sequence homology to the 'trnK(uuu)', probably originated from a recent gene duplication event. Of the three remaining H. andersenii tRNA genes that are unmatched in R. salina, two–trnL(gag) and trnG(gcc)–are redundant because trnL(uag) and trnG(ucc) can decode all of their respective four-codon families 40. These redundant copies might have been lost in an ancestor of R. salina after it diverged from H. andersenii. Lastly, the H. andersenii trnI(cau) is somewhat similar to the trnK(uuu) of the R. salina and only marginally resembles the R. salina trnI(cau) at the 3' end. It is possible that the H. andersenii trnI(cau) originated through recombination between ancestral trnI(cau) and trnK(uuu) genes, which would explain the lack of an obvious trnK(uuu) homolog in H. andersenii comparable to the R. salina trnK(uuu). Substantial sequence divergence among the three genes, however, makes it difficult to accurately trace the origin of the trnI(cau) and the loss of the original trnK(uuu) gene in H. andersenii. On the other hand, the unusual trnI(uau) gene reported from R. salina is not found in H. andersenii. It was suggested that the R. salina trnI(uau) is derived from trnF(uuc) through a recent gene duplication event 24. Overall, the two cryptophyte mitochondrial genomes use similar tRNA sets to recognize codons. However, unlike H. andersenii, which may need to import at least trnK(uuu) from cytosol, the R. salina mtDNA does possess the minimal required set for tRNA autonomy.

Cryptophytes are a well-established eukaryotic lineage, supported by both molecular and morphological features 20. However, their relationship to other eukaryotic groups, particularly those containing plastids of secondary endosymbiotic origin, has been the subject of considerable debate. The cryptophyte plastid is the product of a secondary endosymbiosis involving a red algal cell, the same process which accounts for plastid origins in haptophytes, dinoflagellates, and stramenopiles 50. Cavalier-Smith 50 suggested that plastids in these four algal lineages arose from a single secondary endosymbiosis in a common ancestor that these organisms shared, to the exclusion of other eukaryotic groups. However, this "chromalveolate" hypothesis is controversial 5152. Recent molecular studies have shown that the katablepharids, an enigmatic collection of plastid-less flagellates, are a sister group to cryptophytes 5354, and large-scale concatenated analyses of nuclear genes suggest that cryptophytes and haptophytes are also related 5556.

To gain insight into the phylogenetic relationship of the cryptophytes H. andersenii and R. salina to other eukaryotes, and more specifically, to test the hypothesis that cryptophytes and haptophytes are related to one another, phylogenetic analyses of mitochondrial protein sequences were performed (Figure 5). Unlike the cryptophyte plastid genome, in which several cases of LGT have recently been discovered 5758, individual analyses of 25 mitochondrial proteins did not reveal any obvious instances of LGT between prokaryotes and eukaryotes or within eukaryotes (data not shown). However, the possibility of ancient LGTs cannot be ruled out, as the backbones of individual protein phylogenies were generally very poorly supported.

As expected, a close relationship between the two cryptophytes H. andersenii and R. salina was well supported in the mitochondrial protein phylogenies, with twenty of twenty-five individual protein phylogenies showing this relationship. Five individual gene phylogenies–nad2, rpl14, rpl16, rps12, rps14–did not recover a H. andersenii-R. salina clade, although alternative topologies were not supported with >50% bootstrap support values. Additionally, single protein phylogenies were not, for the most part, able to resolve the relationship of cryptophytes to other eukaryotes. The position of cryptophytes was highly variable from protein to protein and the group did not regularly associate with other taxonomic clades with >50% bootstrap support values, except for in the cob and nad1 gene trees, where cryptophytes branch with haptophytes (81%) and jakobids (77%), respectively.

We subsequently analyzed a set of 25 concatenated proteins to assess the phylogenetic position of cryptophytes. In this analysis, the H. andersenii-R. salina clade received 100% bootstrap support (Figure 5). Other well-established eukaryotic groups including opisthokonts, rhodophytes, stramenopiles, and Viridiplantae, were also strongly recovered, but the relationships among major lineages were not. The jakobid Reclinomonas branched as the sister group to the Viridiplantae with moderate support (89% bootstrap support), and Malawimonas showed an affinity for these two groups in two of the three data sets, as was previously inferred from a concatenate of ten mitochondrial proteins 59. It is not clear whether the jabokid (and/or malawimonad)-Viridiplantae affinity is a phylogenetic artifact or reflects the true evolutionary history of mitochondrial genes. Though growing evidence supports a relationship between cryptophytes and haptophytes 555658, our extensive mitochondrial protein analyses did not reveal this relationship with reasonable bootstrap support, other than in a single protein gene tree (cob). In summary, while mitochondrial gene sequences are able to resolve some of the eukaryotic lineages determined using other markers, they are at present incapable of resolve the deepest branches of the eukaryotic tree using current phylogenetic methods and with the present level of taxon sampling.