Complexity of mitochondrial genomes

Today there is no remaining doubt that the organelle genomes of mitochondria (mt) and chloroplasts (cp) are reduced forms of the genomes of α-Proteobacteria and Cyanobacteria, respectively, which as endosymbionts gave rise to these organelles in the emerging eukaryotic cell more than 109 years ago (Dyall et al. 2004; Gray 1999; Gray et al. 1999). The mainstream of organelle genomic evolution has been the massive loss and transfer of genes to the eukaryotic nucleus, thus reducing the bacterial genome of some 4 mbp and 3,000 genes to an organelle genome of 150 kbp and 120 genes, as for example in land plant cpDNAs (Timmis et al. 2004).

Investigation of plant mitochondrial DNA (mtDNA) has continually revealed surprises that one would not have expected to see in the supposedly simple and reduced genome of a former endosymbiotic bacterium (Fig. 1). The mitochondrial genome (the chondriome) of plants contrasts with its generally simple, streamlined, circular and compact counterpart of about 16 kbp in most animals in nearly every aspect. Even the smallest as yet known chondriome of a land plant (embryophyte), that of the liverwort Marchantia polymorpha (Oda et al. 1992), is 186.6 kbp, more than ten times as large as that of humans; and chondriome sizes can exceed 2,000 kbp in flowering plants of the Cucurbitaceae family (Ward et al. 1981). Instead of the compact, gene-dense and simple circular mitochondrial genome of animals (metazoa), a complex pool of frequently recombining molecules, the stoichiometry of which is controlled by nuclear genes (Abdelnoor et al. 2003), typically constitute flowering plant (angiosperm) chondriomes.

Fig. 1
figure 1

Typical idiosyncrasies in an angiosperm mitochondrial genome. Recombination commonly fragments the circularly mapping complete genomic entity of 200–2,000 kbp (the “master-circle”) into multipartite structures with sub-genomic circles. Some mitochondrial genes are continuous, but many others carry introns. Some of the organelle group II introns are disrupted, resulting in trans-arrangements of genes nad1, nad2 and nad5, which require trans-splicing for maturation. Virtually all plant mitochondrial transcripts are subject to RNA editing, which convert cytidines to uridines to reconstitute conserved codon identities. The origin of some intergenic sequences can be identified, given their homologies with nuclear DNA(nucDNA) or cpDNA. The latter, when expressed, can give rise to functional transfer RNA (tRNA) in angiosperm mitochondria, replacing their native counterparts

Full size image

Deviations from the gene-dense, small-circle chondriome architecture characteristic of the animal (metazoan) lineage were very recently described in phylogenetically early-branching protists, Monosiga brevicollis and Amoebidium parsiticum (Burger et al. 2003a), including genes that were transferred to the nucleus before the emergence of the metazoan lineage. The exceptional discovery of introns in the mtDNA of the sea anemone Metridium senile (Beagley et al. 1996) was a first hint that peculiar chondriome features can exist in early branches of the animal lineage.

The mtDNA of the liverwort Marchantia polymorpha was the first embryophyte chondriome sequence to be completely determined, later followed by those of angiosperms—the widely investigated model species Arabidopsis thaliana with a chondriome of 367 kbp (Unseld et al. 1997), sugar beet, Beta vulgaris, with an mtDNA of 369 kbp (Kubo et al. 2000) and rice, Oryza sativa, with a mitochondrial genome of 491 kbp (Notsu et al. 2002). The most recently completed chondriome sequence of rapeseed (Brassica napus) has a size of “only” 222 kbp, which immediately illustrates the puzzling increase in mtDNA size in Arabidopsis, another member of the family Brassicaceae and thus a closely related species (Handa 2003). The completely determined land plant chondriome sequences are now ideally complemented by those of the charophyte alga Chaetosphaeridium globosum of the order Coleochaetales (Turmel et al. 2002a) and Chara vulgaris (Turmel et al. 2003), a member of the Charales. Together with the land plants, the charophyte algae are placed in the clade Streptophyta. Common features of the Chara and Marchantia mtDNA sequence have strongly confirmed the placement of the Charales as an extant phylogenetic sister group to the land plants.

The circular mtDNAs of Chaetosphaeridium and Chara are 56 kbp and 67 kbp, respectively, in size. These chondriome sequences clearly show that most idiosyncratic features of land plant chondriomes—size increase, high recombinational activity, gain of several introns and trans-splicing, RNA editing and foreign DNA insertions (Fig. 1)—are derived features acquired with or after the establishment of the land plant lineage. The transition to the land plant life form extended chondriome size by a factor of three in the liverwort Marchantia, without a gain of genes or apparent functionality, but mainly through the increase of spacer DNA between genes.

Although not the subject of this review, it must be kept in mind that some independent, peculiar trends of mitochondrial genome evolution do exist in the chlorophyte algal lineages, not closely related to land plants. Ancestral, circular and gene-rich chondriomes are found in the genera Prototheca (Wolff et al. 1994) and Nephroselmis (Turmel et al. 1999), whereas a circular, yet highly gene-reduced chondriome is found in Pedinomonas (Turmel et al. 1999; Wolff et al. 1994). Chlorogonium (Kroymann and Zetsche 1998) also has a circular chondriome but carries fragmented ribosomal RNA (rRNA) genes that are typical for mtDNA in the genus Chlamydomonas (Nedelcu 1997), from which both circular and linear chondriomes are known. Fragmented rRNA genes are also found in Scenedesmus mtDNA, which additionally exhibits a deviant genetic code (Nedelcu et al. 2000). A compact circular chondriome of 45 kbp also exists in Mesostigma, an early branching lineage in algal phylogeny; and in this alga two group II introns, so far without orthologues in other species, have attained a trans-splicing configuration (Turmel et al. 2002b).

With its evolutionary plasticity in the land plant lineage, mtDNA stands in contrast to the other organelle genome in the plant cell, that of the chloroplast (Palmer 1990). The chloroplast genome (the plastome) with its circular structure encompassing a large and a small single copy region separated by a pair of inverted repeats (IR) is structurally conserved, even in the ancient algal lineages such as Mesostigma (Lemieux et al. 2000). Most gene clusters and orders are conserved and expansions and contractions of the IR are the most significant differentiating feature of the plastome, which was fixed in structure and appearance early in plant evolution. Extensive structural rearrangements, such as in subclover (Milligan et al. 1989) or in Trachelium (Cosner et al. 1997), appear to be rare in plants. However, unique structural changes in the plastome can very well convey significant phylogenetic information. A 30-kbp genomic cpDNA inversion documents the earliest dichotomy in vascular plant phylogeny between lycophytes and other tracheophytes (Raubeson and Jansen 1992); and the identification of two introns common in embryophytes and charophyte algae identified the streptophyte clade over 10 years ago (Manhart and Palmer 1990).

Hence, 500×106 years of land plant evolution rather rarely resulted in significant alterations to the chloroplast genome, but significant changes are found in the mtDNA, even on very narrow taxonomic scales. Whereas the plant mitochondrial genome itself is structurally dynamic in evolution, the sequence drift in coding regions is generally very low in plant mtDNA (Wolfe et al. 1987) and this can offer advantages for phylogenetic analysis of deeper branches in phylogeny. In the marchantiid subclass of liverworts, this slow sequence drift appears to be yet further retarded (Beckert et al. 1999), but the opposite trends exist in other lineages. Mitochondrial sequence evolution is strikingly accelerated in the angiosperm genera Pelargonium and Plantago (Palmer et al. 2000) and the fern genus Lygodium (Vangerow et al. 1999). Thus, contradictory trends exist in embryophyte mtDNAs: whereas gene sequences mostly evolve slowly, chondriome structures evolve quickly and gene orders are soon rearranged (Palmer and Herbon 1988). Genes are continuously transferred from the mtDNA of angiosperms to the nuclear genome in recent evolutionary times, while at the same time their chondriomes readily accept DNA from the nucleus or the chloroplast genome for incorporation.

The phylogenetic perspective

Several land plant groups circumscribed by conventional taxonomy are now well corroborated as monophyletic clades by the overwhelming majority of molecular phylogenies (Pryer et al. 2002; Qiu and Palmer 1999): the flowering plants (angiosperms), the seed plants (spermatophytes), the vascular plants (tracheophytes), the lycophytes, the liverworts, the mosses and the hornworts (Fig. 2). Other superior land plant clades suggested more recently are now likewise clearly defined and resolve the issues of “fern allies” (Kolukisaoglu et al. 1995; Raubeson and Jansen 1992). The “moniliformopses” (Kenrick and Crane 1997a; Pryer et al. 2001) include the true ferns, the horsetails (Equisetum) and the whisk ferns (Psilotales), but exclude the lycophytes (Fig. 2). Other groups as yet retain a less clear status, such as the gymnosperms. The anthophyte hypothesis assuming the proximity of Gnetopsida to angiosperms (Crepet 2000; Donoghue and Doyle 2000; Rydin et al. 2002) is not recovered by molecular analyses, which instead frequently identify the two sister groups Ginkgopsida–Cycadopsida and Coniferopsida–Gnetopsida in monophyletic gymnosperms (Bowe et al. 2000; Chaw et al. 1997, 2000; Goremykin et al. 1996; Malek et al. 1996). While the first sister group relation is recovered by most molecular phylogenies, the peculiar evolution of the latter two classes, long branch attraction problems and taxon sampling seem to affect molecular phylogenetic analyses in that case in a particularly strong manner (Magallon and Sanderson 2002; Rydin et al. 2002). Hence, it may as yet be more reasonable to consider the gymnosperm classes as unresolved at the base of seed plants as a whole (Pryer et al. 2002).

Fig. 2
figure 2

Several clades in the land plant phylogeny identified and/or confirmed by molecular data as monophyletic groups are: A angiosperms, S seed plants, M moniliformopses, E euphyllophytes, L lycophytes, T tracheophytes. The monophyly of gymnosperms (G) as a whole and that of a clade comprising two of its classes, Gnetopsida and Coniferopsida, is somewhat less well supported. Likewise, more phylogenetic resolution is needed for a clade comprising the eusporangiate ferns of the order Marattiales, the horsetails (Equisetales) and the leptosporangiate ferns vs a clade comprising the Ophioglossales and whisk ferns (Psilotales). The tree shown is topologically consistent, albeit not always statistically supported. Several phylogenetic analyses of organelle genes (unpublished observations) place hornworts (H) as sister group to the tracheophytes (node Z), mosses (Ms) as a sister group to the joint clade (Y) and confirm marchantiid (ML) and jungermanniid (JL) liverworts jointly as a sister group to all other embryophytes (node X)

Full size image

Most importantly, the deepest branches of the plant phylogenetic backbone are unresolved. The origin of land plants (embryophytes) supposedly dates to Ordovician times (Graham et al. 2000; Kenrick and Crane 1997b), but the notorious lack of macrofossils has so far not allowed a consensus on the first land plants (Wellman and Gray 2000). Molecular data make it convincingly clear that, among extant algae, the Charales share a most recent common ancestor with land plants (embryophytes) and that this joint group is sister to the Coleochaetales algae (Karol et al. 2001; Malek et al. 1996; Turmel et al. 2003). The division bryophytes, including the base-most land plant classes liverworts, mosses and hornworts, is most likely not a monophyletic group. The three earliest bifurcations of land plant evolution seem to separate the three bryophyte clades, possibly placing only one of them as sister group to the vascular plants. As discussed in the next section, molecular support for the idea of liverworts as the base-most group of land plants was found in the identification of three mitochondrial introns present in all land plant clades except the liverworts (Qiu et al. 1998). However, alternative views place hornworts as the base-most land plant clade, based both on molecular (Nickrent et al. 2000) and morphological insights (Renzaglia et al. 2000).

Evolution of gene content: nuclear transfer and replacement

The small “core” set of 13 genes in animal mitochondria coding for protein subunits of respiratory chain complexes I, III, IV and V (atp6, atp9, cob, cox1, cox2, cox3, nad1, nad2, nad3, nad4, nad4L, nad5, nad6) is extended by extra genes for additional subunits of complex I (nad genes) and complex V (atp genes) and genes encoding ribosomal proteins (rpl, rps genes), subunits of complex II (sdh genes) and proteins involved in cytochrome c biogenesis (ccb genes) in plants. A comprehensive table of genes and introns in the completely sequenced land plant chondriomes and other supplementary information, intended to be continuously updated, has been assembled for the purpose of this review and is available on the internet at http://izmb.de/knoop/chondriomes. An overview of the protein-encoding genes in five completely sequenced chondriomes of plants is shown in Fig. 3. The liverwort Marchantia has the most gene-rich (and ancestral) embryophyte chondriome known so far, with the exception of nad7, which is degenerated to a pseudogene here. The unknown land plant ancestor may also have carried rpl14 on its chondriome, which still exists in the mtDNA of the alga Chara. The rice chondriome is the most gene-rich angiosperm chondriome so far known. The two most closely related plants, the Brassicaceae Arabidopsis thaliana and Brassica napus, differ with respect to the gene for rps14, which is a pseudogene in the Arabidopsis mtDNA but a functional gene in the mitochondrion of B. napus. At the same time, the Arabidopsis mtDNA is significantly larger (367 kbp vs 222 kbp), proving that intergenic DNA without obvious function can accumulate in short time intervals in a flowering plant chondriome.

Fig. 3
figure 3

Gene content in five completely sequenced streptophyte chondriomes. Figure design is adapted from Gray et al. 1999. Intron-encoded open reading frames (orfs) and tRNA genes, which are frequently lost and replaced by chloroplast counterparts in angiosperms, are not considered but documented in the supplementary table under http://www.izmb.de/knoop/chondriomes.htm. Genes encode subunits of respiratory chain complexes I (nad), II (sdh), III (cob), IV (cox) and V (atp), proteins of the small and large ribosomal subunit (rps, rpl) and proteins for cytochrome c biogenesis (ccb). The latter have also been referred to as yej, ccm or hel genes, respectively. The ccb6 (=yejR) gene is split into separate orfs in plant mitochondria (see text; Fig. 4). Designations atp4 and atp8 are new nomenclature proposals (Burger et al. 2003b; Gray et al. 1998; Heazlewood et al. 2003; Sabar et al. 2003) for genes that until recently were referred to as orf25 (or ymf39) and orfB, respectively, in the literature and in database entries. The mtt2 gene is the homologue of the secY-independent membrane transport protein, which was referred to as orfX, tatC or ymf16, respectively. Genes that are not present in the completely sequenced angiosperm chondriomes but known to be present in other angiosperm mtDNAs are shown in blue. Genes of the core set for which nuclear transfer in other angiosperms is demonstrated are given in red

Full size image

When a gene is missing in the chondriome of one plant species but present in another, one can reasonably assume that a copy has been transferred to the nucleus, gained function through appropriate regulatory elements and a proper amino-terminal targeting sequence and that its product is imported into mitochondria after synthesis in the cytosol (Adams and Palmer 2003). The as yet best documented example of such a “gene transfer in progress” in land plant evolution is the cox2 gene (Nugent and Palmer 1991). The different steps of gene-copying to the nucleus, the activation of the nuclear copy and the degeneration of its mitochondrial counterpart can be traced for the cox2 genes of leguminous plants (Adams et al. 1999; Daley et al. 2002). However, being a respiratory chain complex subunit belonging to the core set of genes, which is also encoded in mtDNA of animals, cox2 is clearly an exception. In fact, the atp8 gene of Allium (onion) is the only other example of a gene encoding a protein in the respiratory chain complexes I, III, IV or V that has been identified as missing from the mtDNA in a wide survey of angiosperms (Adams et al. 2002b). However, gene transfer events are significantly more often observed for the sdh genes encoding complex II subunits and those encoding ribosomal proteins (Adams et al. 2002b). A prominent example is the rps10 gene encoding ribosomal protein S10, where many gene transfers took place independently (Adams et al. 2000; Knoop et al. 1995). Most of the ribosomal protein genes were lost from plant chondriomes independently in a patchy manner; and only the rps2 and rps12 genes were lost a few times deep in angiosperm phylogeny. Gene transfer frequencies to the nucleus are highly skewed, with some taking place frequently and independently (e.g. rps7, lost at least 42 times among 280 sampled flowering plants) and others much more rarely, such as rps12 (six times; Adams and Palmer 2003). All ribosomal protein genes were lost from the chondriome in Lachnocaulon (bog buttons), but none in the basal-most angiosperms (Qiu et al. 2001; Zanis et al. 2002); and the latter observation may indicate that the set of mitochondrial genes is conserved in phylogeny through vascular plant evolution up to the root of angiosperms. Obviously though, gene transfer activity could be increased independently in the other major land plant clades (Fig. 2). Taxonomically extensive studies like those by Adams and colleagues on angiosperms (Adams et al. 2002b; Adams and Palmer 2003) are as yet not available for older land plant groups—the gymnosperms, the pteridophytes (i.e. lycophytes, moniliformopses) or the bryophytes. So far, the nad7 gene in the nucleus of Marchantia is the only example of a functional gene transfer from the mitochondrion documented in the early embryophytes (Kobayashi et al. 1997). The degeneration of nad7 to a mitochondrial pseudo-gene by a gain of stop codons is not restricted to Marchantia among the liverworts. Stop codons are gained independently at different positions in nad7 in other liverwort genera (unpublished observations). However, the nad7 gene is functional in moss mitochondria (Hashimoto and Sato 2001; Pruchner et al. 2001).

Mitochondrial re-targeting peptide sequences can be acquired in different ways: duplication of existing pre-sequences in the nucleus (e.g. mitochondrial heat-shock proteins), altered use of other coding sequences, exon shuffling, insertion into an intron for a mitochondrially targeted protein gene and alternative splicing or de novo acquisition of a targeting sequence at the insertion site (Adams et al. 2000, 2001b; Figueroa et al. 1999; Kadowaki et al. 1996; Sanchez et al. 1996; Sandoval et al. 2004; Wischmann and Schuster 1995). A further alternative is that no cleavable targeting peptide sequence may be required at all when the nuclear copy fortuitously contains targeting information, as in the case of rps10 (Adams et al. 2000; Kubo et al. 2003).

It must be noted that an observed gene loss from the chondriome does not necessarily implicate that the mitochondrial gene was functionally transferred to the nucleus. It was suggested that an extension of the rps19 gene gained after transfer to the nucleus could functionally replace the missing rps13 gene in Arabidopsis mitochondria (Sanchez et al. 1996). Instead, a diverged nuclear copy of rps13 of chloroplast origin was found to replace the function of its mitochondrial counterpart (Adams et al. 2002a). The rps8 gene, which is not identified in angiosperm mitochondria (Fig. 3), is functionally replaced by the adapted copy of a nuclear gene originally coding for rps15A in cytosolic ribosomes (Adams et al. 2002a).

The most common form of functional gene replacement in the plant mitochondrion involves tRNA genes derived from the plastome. Frequent insertion of promiscuous cpDNA sequences into plant chondriomes is common in flowering plants. Not all of these transferred sequences are rendered useless as intergenic DNA in their new location—some tRNA encoding genes of chloroplast origin functionally replace the original native mitochondrial tRNA species in the plant mitochondrial genome (Marechal-Drouard et al. 1990). For example, the plastid genes trnH(gug), trnM(cau), trnN(guu) and trnW(cca) are present in all completely sequenced angiosperm chondriomes (Fig. 1), whereas the native mitochondrial genes are (still) present in the liverwort chondriome. However, a chloroplast-derived gene known to function in one plant species may be inactive in another, indicating the import of a tRNA encoded in the nucleus (Duchene and Marechal-Drouard 2001; Kumar et al. 1996).

The presence of ccb genes encoding proteins participating in cytochrome c biogenesis in embryophytes and Chara, but their absence in other green algae, must be explained by several independent losses (at least in Mesostigma, Chaetosphaeridium and an early chlorophyte ancestor). An alternative gain of ccb genes in the common ancestor of Chara and the embryophytes involving horizontal gene transfer may sound far-fetched but would evidently be a more parsimonious explanation. Actually, two recent reports claim horizontal gene transfer of mtDNA sequences into several angiosperm genera: atp1 into Amborella, rps2 into Actinidia and rps11 into Betula and Sanguinaria (Bergthorsson et al. 2003) and an angiosperm-like intron in nad1 into the gymnosperm Gnetum (Won and Renner 2003). If true, such events would naturally add a highly complicating factor to gain, loss and transfer scenarios. So far, the reported observations are restricted to individual mitochondrial loci in the suspected acceptor plants and a more extensive characterization of their chondriomes will be very helpful to further substantiate the conclusions.

Whereas the functional organelle-to-nucleus transfer of a gene can proceed via RNA and cDNA (Nugent and Palmer 1991), inter-genomic sequence transfer can also proceed via DNA and is not necessarily functional or successful. Even at pre-genomic times, it was possible to identify mtDNA, such as intron fragments, in the nuclear DNA of angiosperm species (Blanchard and Schmidt 1995; Knoop and Brennicke 1991, 1994). Sequencing the Arabidopsis thaliana nuclear genome revealed big stretches of mtDNA (Stupar et al. 2001) and it is not surprising that organelle sequences that are derived from DNA, not RNA, show up in the now-completed genome of rice. A surprisingly frequent random transfer of cpDNA to the nucleus that can even be traced within generation time is now experimentally documented in tobacco (Huang et al. 2003; Stegemann et al. 2003).

Plant mitochondrial introns

A recent proposal to designate organelle introns with the name of their host gene and a number indicating the respective upstream exon nucleotide of the homologous Marchantia polymorpha orf (Dombrovska and Qiu 2004) is adapted here by adding the classifications into group I and group II introns, respectively. It is important to distinguish these two different types of “organelle” introns (with group II introns increasingly being identified also in eubacteria; Dai and Zimmerly 2002), which are classified according to their entirely different secondary structures and which have different modes of splicing and invading new loci. Both types are frequently referred to as self-splicing—a somewhat misleading simplification because autocatalytic activity is documented only for a minority of fungal and bacterial species, but not for the plant organelle introns; and protein cofactors are known to be essential in vivo in any case. Group I introns can carry orfs encoding homing endonucleases of the LAGLIDADG type which take part in invading intron-less loci; and some group II introns can carry maturase-type orfs with reverse transcriptase sequence motifs. A new level of complexity to intron evolution and the modes of intron propagation comes from the discovery of group II introns with LAGLIDADG type orfs (Toor and Zimmerly 2002).

Table 1 attempts to summarize the knowledge on intron insertion positions in those plant mitochondrial genes for which there is at least some taxonomic sampling in non-seed plants other than the liverwort Marchantia alone, i.e. for moss, hornwort or pteridophyte species. All introns in the completely sequenced streptophyte chondriomes, together with information on sizes, similarity groupings and intron-borne orfs are compiled in the complementary table at http://izmb.de/knoop/chondriomes.htm.

Table 1 Variable intron occurrence as so far documented in mitochondrial genes for which there is at least some taxon sampling exceeding Marchantia, algae and angiosperms, i.e. for mosses, hornworts, lycophytes or moniliformopses: cox1 (Bowe et al. 2000; Sperwhitis et al. 1996), cox2 (Qiu et al. 1998; Sperwhitis et al. 1994), cox3 (Malek et al. 1996), nad1 (Dombrovska and Qiu 2004), nad2 (Beckert et al. 2001; Pruchner et al. 2002), nad5 (Beckert et al. 1999; Vangerow et al. 1999), nad4 and nad7 (Pruchner et al. 2001). Plus/minus signs Presence/absence of a given intron documented for at least one member of the respective clade, E, M endonuclease-type and maturase-type intron-borne orfs in group I and group II introns, respectively, T disruption into a trans-arrangement, brackets within-clade variability, question marks a lack of information for the clade or gene region, respectively, intron names in italics orthologues in non-streptophyte groups (fungi, chlorophytes)
Full size table

Comparing the chondriome sequences of the flowering plants with that of the liverwort Marchantia, one observation is immediately evident: of 25 known group II intron sites in angiosperm mitochondria and an equal number in Marchantia, only a single intron, nad2i709, is conserved in position (Table 1). This contrasts with the situation in chloroplast genomes, where most introns (a total of 17) are highly conserved among land plants of all groups and were obviously gained “in the water”, even before the separation of the Coleochaetales alga Chaetosphaeridium and the land plant lineage (Turmel et al. 2002a). Only four other plastome introns may have been gained after this separation “on land”.

The Marchantia chondriome carries group I introns, one in nad5 and six in the cox1 gene—none are present in the completely sequenced angiosperm chondriomes. Group I introns were unknown in angiosperm mitochondrial genes until the discovery of cox1i729 in Peperomia (Vaughn et al. 1995). This intron turned out to be a highly invasive species with a patchy distribution among angiosperms (Cho et al. 1998). Strikingly, the cox1 gene generally appears to be a prime target for group I intron invasion in algae and Marchantia (Table 1), with orthologous introns being observed in identical positions in the mtDNAs of fungi and chlorophyte algae (Ohta et al. 1993; Ohyama et al. 1993; Turmel et al. 2002b; Wolff et al. 1993). A horizontal inter-species transfer is thus very likely for most of them. The nad5 group I intron nad5i753 (Table 1) even has a positional homologue in the mtDNA of Monosiga brevicollis (Burger et al. 2003a), very likely representing a very early branch of the animal lineage. However, this intron is present (so far without exception) in all mosses and liverworts investigated (Beckert et al. 1999), indicating a highly stable vertical transmission once land plants began to evolve. It may have been lost in a common ancestor of hornworts and tracheophytes under the phylogenetic scenario shown in Fig. 2.

The single mitochondrial intron conserved in position in angiosperms and Marchantia, nad2i709 (Table 1), appeared as an attractive candidate for an intron potentially conserved in all land plant clades but surprisingly turned out to be absent in all mosses (Beckert et al. 2001). In contrast, moss nad2 genes (again, so far without any exception) exclusively carry the “angiosperm-type” intron nad2i156. Similar observations are made for the nad4 and nad7 genes (Table 1), where mosses likewise only feature angiosperm-type introns nad4i461, nad7i140 and nad7i209 (Pruchner et al. 2001). Group I intron nad5i753 found to be present both in all mosses and in all liverworts therefore turns out be exceptional; and it is actually so far the only documented example of an intron shared between mosses and liverworts. Recently extended investigations of the nad5 gene, which is rich in angiosperm introns, find this observation extended: angiosperm type introns nad5i230 and nad5i1455 are absent in liverworts but present in mosses, the latter being trans-disrupted in flowering plants (M. Groth-Malonek, D. Pruchter, F. Grewe, V. Knoop, submitted).

Only some mitochondrial group II introns are of narrower taxonomic distribution and confirm classic taxonomic groups: nad2i830 in lycophytes (Pruchner et al. 2002), nad1i348 in hornworts and nad1i258 in moniliformopses (Dombrovska and Qiu 2004). Others link two clades, which are, however, not suggested as sister groups in phylogenetic analyses (Fig. 2): nad5i1242 in lycophytes and moniliformopses (Vangerow et al. 1999), nad5i285 in mosses and hornworts and nad5i753 in mosses and liverworts. One deep gain and one deep loss along the backbone of plant phylogeny would need to be assumed for those. Introns that so far have been identified in single genera only are rare: nad5i391 in the lycophyte Huperzia (Vangerow et al. 1999) and intron nad5i881 in a hornwort (Beckert et al. 1999). The Huperzia intron nad5i391 is particularly interesting as it is closely related to the common “pteridophyte” (moniliformopses, lycophytes) intron nad5i1242 and is thus a first example of what may be a within-gene group II intron-copying mechanism.

None of the known angiosperm group II introns is a characteristic “monocot” or “dicot” intron but introns can be lost independently in genera of both angiosperm groups. Variability in intron occurrence among angiosperms that can most consistently be explained by independent losses is reported for cox2i373 and cox2i691 (Kudla et al. 2002; Qiu et al. 1998), nad4i976 and nad4i1399 (Geiss et al. 1994; Itchoda et al. 2002), nad7i676 (Pla et al. 1995), nad1i477 (Bakker et al. 2000) and rps3i74 (Kubo et al. 2000). An extensive study investigating the loss of cox2i373 documents many independent losses among angiosperms (Joly et al. 2001). Interestingly, this intron has a positional homologue in a strain of the yeast Schizosaccharomyces pombe (Schafer et al. 1998) which may, apart from independent losses, also hint at independent gains on wider taxonomic scales. The independent loss of group II introns is also evident for non-flowering vascular plants: introns cox2i373 and cox2i691 are lost in some pteridophytes and gymnosperms (Qiu et al. 1998), nad5i1242 is lost in Equisetum and Ophioglossum (Vangerow et al. 1999), nad1i394 in Ophioglossum, nad1i477 in Isoetes, Asplenium, Equisetum, among gymnosperms in Welwitschia and in a conifer clade and nad1i669 is lost in Isoetes and Asplenium (Dombrovska and Qiu 2004; Gugerli et al. 2001).

Taking the Chara chondriome as a reference, it appears that invasion of mitochondrial genes with introns largely took place during establishment of the embryophytes. Chara shares only two introns with Marchantia: nad3i140 (absent in angiosperms) and cox1i729—the “rampant invader” found in some angiosperm cox1 genes. Likewise, Chara shares two introns with angiosperms, absent in Marchantia: nad4i976 and rps3i74.

The earliest gains of new mitochondrial introns coinciding with the emergence of land plants after separation from the Chara lineage could be nad5i753 and nad2i709 (Fig. 2). After that, introns may have invaded the chondriomes in a differential manner after an early split of land plants into liverworts and non-liverwort plants. Certainly, intron transposition and differential loss scenarios are an alternative, although the less parsimonious explanation. Given that nine mitochondrial group II introns in Marchantia encode maturase orfs, mobility may have been significantly higher in early land plant evolution. It may have ceased somewhere in the evolution of the angiosperm lineage, possibly reflected in the fact that the maturase-like orf in nad1i728 is the only one of its kind in angiosperms. It is noteworthy that, while intron nad1i728 is retained in mosses, the maturase orf is lost from the intron in derived moss genera (Dombrovska and Qiu 2004).

There is no doubt that some land plant lineages will yield further surprises with respect to the occurrence of mitochondrial introns, possibly comparable with those of suddenly accelerated mitochondrial sequence evolution in single angiosperm genera (Adams et al. 1998). For example, very small mitochondrial group II introns have been found in the quillwort Isoetes (Malek and Knoop 1998) and new mitochondrial sequences of this genus reveal additional introns so far without orthologues in other species (unpublished observations). At least one group II intron is a strong candidate for independent gains over a large evolutionary distance: cox3i171, which is present in liverworts and the lycophyte Lycopodium (Hiesel et al. 1994b), and the introns nad4i976 and rps3i74 shared between Chara and some angiosperms.

Given (1) that some mitochondrial intron positions are shared between plants and fungi, (2) that mycorrhizal fungi were already present on earth in Ordovician times when the evolution of embryophytes began (Redecker et al. 2000), (3) that not only vascular plants but also bryophytes can be partners in such symbioses (Read et al. 2000) and (4) that very diverse non-mycorrhizal fungi also inhabit aboveground organs of all plants as endophytes (Gamboa et al. 2003; Ligrone et al. 1993), it is conceivable that different fungi may have acted differentially as donors of different mitochondrial introns in the plant lineages.

Recombinational activity disrupting gene continuity

Mono-circular chondriomes, such as in Brassica hirta (Palmer and Herbon 1987) and Marchantia, represent the ancestral state of mitochondrial genome organization in the plant lineage, but are probably the exception rather than the rule in angiosperms. Active recombination via a repeated sequence motif producing two sub-genomic circles is evident for other Brassica species (Palmer and Herbon 1988; Palmer and Shields 1984); and the completed B. napus chondriome sequence has identified this sequence as a directly repeated motif of 2,427 bp (Handa 2003). This sequence repeat is not related to those found in Arabidopsis, where two pairs of repeated sequences produce a more complex population of sub-genomic molecules (Klein et al. 1994). Maize (Zea mays) is an important model for the significantly higher chondriome complexity through coexisting sub-genomic molecules found in other angiosperms (Andre and Walbot 1995; Small et al. 1989). Additionally, the plant mtDNA complexity is yet further enhanced in some plant species by the presence of plasmids, for example a linear 11-kbp DNA molecule with terminal IR sequences in B. napus (Handa et al. 2002), which does not share significant homology with the now completed main chondriome sequence (Handa 2003).

Whereas the active homologous recombination via larger sequence repeats produces coexisting sub-genomic molecules in a plant mitochondrion, recombinations via shorter sequence motifs appear to be the major force rearranging plant chondriomes on evolutionary time-scales. Notably, even a supposedly simple mono-circular mtDNA, such as in the liverwort Marchantia, may in fact be physically represented (also) by circularly permutated, linear molecules (Bendich 1993; Oldenburg and Bendich 2001). Although there is no evidence for active recombination leading to coexisting molecules of Marchantia mtDNA, it should be noted that duplicated sequence stretches, potential substrates of homologous recombination, are present in the liverwort chondriome (Ohyama et al. 1993). These include duplication of a cob gene intron sequence contributing to a size increase in the nad5-nad4 spacer region for which, however, homologous arrangements exist in jungermanniid liverworts (unpublished observations), indicating a recombination event on evolutionary time-scales.

Among angiosperms, a variation in mitochondrial genome arrangements is common in plants, even between isolates of the same species (Scotti et al. 2004; Ullrich et al. 1997). Mitochondrial dysfunctions in mutant lines of flowering plants, typified by cytoplasmic male sterility (Budar and Pelletier 2001), are frequently connected to such recombination events when they affect conserved orfs or their expression.

Comparing the Marchantia and Chara chondriomes, the conservation of some gene linkages is obvious (Turmel et al. 2003). The larger of these arrangements are the nad5-nad4-nad2- and the trnC-trnF-rps2-ccb2-ccb3-ccb6-trnQ gene orders and the nad7-rps10-rpl2-rps19-rps3-rpl16-[rpl14]-rpl5-rps14-(rps8)-rpl6-(rps13)-rps11 cluster, the latter of which includes the ancient bacterial rps10 operon, from which rps8 and rps13 are lost in Chara, whereas rpl14 is inserted at this position. Gene orders generally are not found conserved in angiosperm mitochondria, with the exception of the nad3-rps12 linkage and the (overlapping) rps3-rpl16 genes (Nakazono et al. 1995), preceded by rps19 in some species.

In some cases, the frequent recombinational activity not only affected intergenic regions but also disrupted the continuity of genes, either within introns or even in the coding region. On the gene expression level, one consequence of frequent recombination in land plant chondriomes is the necessity of messenger RNA (mRNA) maturation by trans-splicing in some plant mitochondrial genes. Three mitochondrial genes in angiosperms carry trans-splicing introns: nad1 (Chapdelaine and Bonen 1991; Wissinger et al. 1991), nad2 (Binder et al. 1992) and nad5 (Knoop et al. 1991; Pereira et al. 1991). In these cases, group II introns are disrupted in their continuous sequence and the 5′ half of the affected intron follows the upstream exon in one genomic location, while the 3′ half precedes the downstream exon in another genomic location. The disruptions of the introns in a large and variable domain IV of their secondary structures (Michel and Ferat 1995) do not affect maturation of the mRNAs, which have to be assembled from independent precursor transcripts. Therefore, this process was named trans-splicing. Among angiosperms, five trans-splicing events (nad1i394T, nad1i669T, nad2i542T, nad5i1455T, nad5i1477T) are highly conserved (Table 1). As shown for nad5i1477 in Oenothera, an intron can even be split twice (Knoop et al. 1997). None of these trans-splicing introns has an algal or liverwort homologue. Tracing the origins of trans-splicing introns in plant phylogeny (Fig. 2), it was possible to identify cis-arranged homologues and thus presumable ancestors in pteridophytes (Malek et al. 1997; Malek and Knoop 1998). In the case of nad5i1477T, a cis-arranged homologue was identified in the hornwort Anthoceros, demonstrating that gene invasion by ancestral introns could predate the bryophyte–vascular plant separation. Exploring the complete nad5 gene structure in mosses, we recently found that this intron, nad5i1477, is absent in mosses. However, mosses do show the presence of a cis-arranged counterpart for nad5i1455T (M. Groth-Malonek, D. Pruchter, F. Grewe, V. Knoop, submitted). Hence, serial intron entries beginning in the bryophyte clades for nad5i1455 and nad5i1477 are the likely evolutionary scenario (Fig. 2). As an alternative to evolution towards trans-splicing (in angiosperms), the introns can get lost secondarily in pteridophytes, as shown for nad1i669 and nad5i1477 (Dombrovska and Qiu 2004; Malek and Knoop 1998). An additional disruption splits intron nad1i728 in some angiosperm species (Chapdelaine and Bonen 1991; Conklin et al. 1991). This intron, which is omnipresent in the non-liverwort lineage (Qiu et al. 1998), has now been shown to evolve a trans-splicing status several times independently and differently among angiosperms (Qiu et al. 2004).

In line with a lack of conserved gene order in its chondriome, the alga Mesostigma features trans-splicing of group II introns, indicating an independent increase in recombinational activity in this early algal lineage. The complete chondriome sequence of this species allowed the identification of two trans-splicing introns in the nad3 gene, for which no orthologues have been reported so far (Turmel et al. 2002b).

In cpDNA of the streptophyte lineage, one example of a trans-splicing intron is known, which disrupts the continuity of the rps12 gene but, as with most genomic features in the plastome, the emergence of this trans-splicing arrangement has also preceded the split of Chaetosphaeridium from the land plant lineage (Turmel et al. 2002a). A second example of chloroplast trans-splicing, taxonomically rather restricted, affects two introns in the psaA gene of the chlorophyte alga Chlamydomonas reinhardtii (Choquet et al. 1988; Turmel et al. 1995).

Gene disruptions are not restricted to introns, but can also affect coding regions in plant mitochondria. The ccb genes encoding components of cytochrome c biogenesis are embedded in the gene order trnC-trnF-rps2-ccb2-ccb3-ccb6-trnQ conserved in Chara and Marchantia. Gene spacers are extended significantly in the liverwort in comparison to the alga (Fig. 4), whereas the cluster is completely disintegrated in angiosperms. More notably, the ccb6 orf itself is disrupted in embryophytes. The conservation of peptide motifs suggest that a first disruption in the middle of the ccb6 gene is shared by Marchantia and the flowering plants. Non-coding spacers of 234 bp and 2,016 bp are found in Marchantia and sugar beet, respectively, at the corresponding positions, whereas linkage of the two resulting parts is completely lost in other angiosperms. A second disruption of the downstream part resulted in two orfs overlapping by one nucleotide in Marchantia and another independent disruption of the upstream orf (termed ccmFN) is found in Brassicaceae chondriomes (Fig. 4).

Fig. 4
figure 4

Disintegration of the ccb gene cluster in embryophyte evolution. The gene order trnC-trnF-rps2-ccb2(yejV)-ccb3(yejU)-ccb6(yejR)-trnQ is conserved in the chondriomes of Chara vulgaris and Marchantia polymorpha, but not in angiosperms. Arrows indicate transcription from the opposite strand. Numbers above and below the lines indicate spacer sizes in Chara and Marchantia, respectively. The ccb6 gene is split differently into separate orfs in the embryophytes. The ellipse indicates a group II intron in angiosperms

Full size image

A second example of a discontinuity in a coding region, which additionally involves transfer to the nucleus, is the rpl2 gene. The situation is particularly complex, since transfer of rpl2 to the nucleus can occur in the whole or in parts after fission of the mitochondrial gene (Adams et al. 2001a). As an intermediate state, after transfer, the nucleus can provide the 3′ part, with the 5′ part coming from the mitochondrion.

RNA editing

Evolutionary and phylogenetic interest in plant mtDNA originated in part in the peculiar phenomenon of RNA editing (Hiesel et al. 1994a, 1994b). Plant mitochondrial mRNAs are subject to RNA editing, which site-specifically replaces specific cytidines by uridines to restore conserved codon identities (Covello and Gray 1989; Gualberto et al. 1989; Hiesel et al. 1989). Editing a protein orf is the rule in plant mitochondrial transcripts; and examples of a lack of editing in a mRNA, such as rps12 in rice, are rare (Notsu et al. 2002). Non-coding RNA leader and trailer sequences, pseudo-genes, introns and tRNAs are affected much more rarely (Binder et al. 1992, 1994). In the latter, the editing events sometimes reconstitute base-pairing and can be a prerequisite for correct processing of the affected tRNA (Marchfelder et al. 1996; Marechal-Drouard et al. 1996a, 1996b). In other instances, similar base mismatches in tRNA stem regions are not corrected by RNA editing and this remains without consequence for correct tRNA processing (Schock et al. 1998). Similar observations are made for base-pairing regions in group II introns (Carrillo et al. 2001; Carrillo and Bonen 1997).

In chloroplasts, the same type of RNA editing process, exchanging pyrimidines, is active, although at a significantly lower rate. Whereas totals of 441 and 491 such sites are reported for transcripts of the Arabidopsis and Oryza chondriomes, respectively (Giege and Brennicke 1999; Notsu et al. 2002), only some 30 sites are observed in angiosperm chloroplast transcripts (Tsudzuki et al. 2001). Given this difference in frequency between the organelles is a factor of at least ten, yet more striking are the differences between land plant groups. Initially assumed to be absent in bryophytes, RNA editing in the organelles has been identified in all classes of bryophytes (Freyer et al. 1997; Malek et al. 1996). However, a so far puzzling observation shows that RNA editing is present in the jungermanniid subclass (leafy, simple thallus) of liverworts to varying degrees, but is apparently lacking in the marchantiid subclass (complex thallus; Steinhauser et al. 1999). RNA editing is likewise not reported for algae. Very high frequencies of mtRNA editing are observed in the lycophyte Isoetes (Malek et al. 1996) and yet more surprising is RNA editing in hornworts and ferns. Only very rare exceptions of reversing the standard direction of C-to-U editing have been identified in angiosperms (Schuster et al. 1990), but “reverse” editing exchanging U-to-C is found frequently in hornworts (Steinhauser et al. 1999; Yoshinaga et al. 1996) and ferns (Vangerow et al. 1999). With the full plastome sequence of the hornwort Anthoceros formosae now available (Kugita et al. 2003a), a total of 509 C-to-U and 433 U-to-C edits have been identified in its chloroplast transcripts (Kugita et al. 2003b).

Foreign sequence insertions

The land plant chondriomes carry large stretches of intergenic sequences and most of this extra DNA has no significant homology to other sequences which would allow identifying their origin or potential function. In some cases however, the identity and origin of such sequences are obvious. The initial identification of a 12-kbp cpDNA sequence inserted into the maize chondriome (Stern and Lonsdale 1982) was followed by many other examples in many other species. For example, a systematic survey showed that coding sequences of the chloroplast rbcL gene (encoding ribulose-bisphosphate carboxylase) have been transferred independently several times (Cummings et al. 2003). The inter-organellar transfer of sequences is not restricted to cpDNA. Some “promiscuous” sequences were identified as degenerated retrotransposon sequences of all types that normally reside in the nucleus (Knoop et al. 1996). A total of 13.4% of sequences were nuclear and 6.3% were of plastid origin, respectively, in the 490.5-kbp rice mtDNA (Notsu et al. 2002). Hence, at least flowering plant mitochondria readily accept DNA from the chloroplast and the nuclear genome. Interestingly, a recent report demonstrated that the import of DNA into potato mitochondria is an active process (Koulintchenko et al. 2003).

No recognizable homology to sequences from the chloroplast or nuclear genome is present in the Marchantia chondriome. Thus, it is an open question when and why during evolution did plants accept foreign DNA inserted into their chondriomes. Clearly though, it is not the import of DNA alone that results in the extraordinary size increase of plant chondriomes. In other cases such as the Cucurbitaceae family, which shows a striking diversity of chondriome sizes, extensive sequence duplications are among the major reasons for size increase (Lilly and Havey 2001).

Conclusions and outlook

Problems in establishing a robust land plant phylogenetic tree for the base-most (bryophyte) groups are at least two-fold: first, to identify (molecular) markers that have conserved very ancient phylogenetic information among the majority of non-informative exchanges, to document the very short internodes presumably resulting from a very early separation of the bryophyte lineages and, second, to identify the long branch attraction artifacts as a result of the evolutionary distance to even the most closely related outgroup algae (Charales, Coleochaetales). Most single-gene and multigene analyses of organelle genes at low taxon sampling result in liverwort-basal topologies with hornworts sister to tracheophytes (Fig. 2), albeit mostly with low statistical support (unpublished observations). How do features of the mitochondrial genome evolution fit into such a view of land plant phylogeny?

Parsimonious explanations of intron occurrence are complicated in cases where frequent independent gains are obvious, such as for the “rampant invader” group I intron cox1i729. Horizontal transfer, possibly from fungal sources, is an obvious explanation for the occurrence of most group I introns in the basal genera. Chaetosphaeridium, Chara and Marchantia combined have a total of 26 mitochondrial group I intron locations in cob, cox1, cox2, rrnl and rrns. Whereas only four of these are shared among them (two between the algae, one each between Marchantia and one of the algae), a full 19 of them have orthologues that can be identified in the available mitochondrial sequences of fungi or chlorophyte algae.

In contrast, the occurrence of group II introns in land plants seems to be largely explained by vertical transmission and occasional loss. Strong support for the liverwort basal topology originally deduced from cox2i373, cox2i691 and nad1i728 absent in liverworts (Qiu et al. 1998) is further corroborated by nad2i156, nad4i461, nad7i140, nad7i209, nad5i230 and nad5i1455 in the non-liverwort branch. Notably, the topology is consistent with serial entries of the introns later developing a trans-splicing status in angiosperms: nad5i1455 at node Y, nad5i1477 at Z and nad2i542, nad1i394 and nad1i669 in the tracheophyte ancestor (Fig. 2). A sister group relation of hornworts and tracheophytes would be further supported by the gain of nad2i1282 and loss of nad5i753, the group I intron which is vertically inherited in mosses and liverworts. However, some intron losses on the backbone of land plant phylogeny need to be postulated in any topology and, in the topology shown in Fig. 2, this would be the loss of nad2i709 in the last common ancestor of mosses and nad2i156 in hornworts.

The suite of genes present in the common ancestor of embryophytes was most likely that of Marchantia, including functional nad7 and rpl14 genes (Fig. 1). Surprises for additional genes in mosses, the potential base-most non-liverwort clade (Fig. 2) are not to be expected. Given its presence in the chondriomes of Chara and Marchantia and its general absence in angiosperms, the rpl6 gene is an interesting candidate for a gene transfer early in the non-liverwort embryophytes (NLE) lineage.

No conservation of gene order is discernible in angiosperms, but the gene clusters conserved between Marchantia and Chara could well be conserved in the basal NLE groups. At least the nad5-nad4-nad2 gene arrangement is indeed conserved in both mosses and hornworts but is only partially retained in Isoetes (unpublished data).

A significant increase in chondriome size occurred early in embryophyte establishment. The total amount of intron sequences doubled in Marchantia, compared with Chara (56 kbp vs 28 kbp, respectively), but the contribution of 100 kbp of intergenic DNA to the non-coding sequences in the Marchantia chondriome was considerably larger. The absence of evolutionary pressure on these sequences may have obliterated the origins of early sequence acquisitions, whereas the source of the likely more recently transferred DNA in angiosperms is clearly recognizable as the chloroplast or nucleus.

There is only scarce information on other non-flowering plant chondriomes. The mtDNA of the fern Onoclea sensibilis and the horsetail Equisetum arvense are approximately 300 kbp and 200 kbp in size, respectively (Palmer et al. 1992), and thus well within the size range of Marchantia and the smaller of the known angiosperm mtDNAs. Partial sequence information of clones from lycophyte mtDNAs of Selaginella and Isoetes show cpDNA homologies in the chondriomes of both genera (unpublished observations), indicating that mechanisms to stably integrate cpDNA into plant chondriomes evolved in the common ancestor of the tracheophytes (Fig. 2).

The occurrence of RNA editing remains puzzling. The high ratio of U-to-C in addition to C-to-U editing could be a transitory phenomenon arising in the common ancestor of hornworts and tracheophytes, retained in lycophytes and moniliformopses and on its way out in the ancestor of seed plants. While at first glance a gain in RNA editing may have coincided with the emergence of land plants, its apparent absence in the marchantiid liverworts is problematic, given that recent phylogenetic analyses support the monophyly of liverworts as a whole (Fig. 2), including the conservation of introns in both liverwort subclasses (for a divergent phylogenetic assessment, see Capesius and Bopp 1997). For now, a loss of RNA editing in marchantiid or alternatively a later separate gain in jungermanniid liverworts needs to be postulated.

Chondriome sequences of mosses, hornworts, lycophytes, moniliformopses and gymnosperms may reveal further surprises but will in any case most likely continue to provide valuable information on plant phylogeny.