The in vivo molecular structure of plant mitochondrial DNA (mtDNA) has been a long-standing source of intrigue and controversy. Recent deep sequencing analyses of mitochondrial genomes from numerous plant species have provided the opportunity to revisit this decades-old question from a novel perspective. Whole-genome sequencing approaches have yielded new lines of evidence that the ‘master circle’ is not the predominant form of plant mtDNA and have revealed striking structural variation both within and among species. Here, I review these recent studies, including the discovery that at least two independent angiosperm lineages have evolved multichromosomal mitochondrial genome structures. These findings raise fascinating questions regarding the mechanisms of plant mtDNA replication and inheritance.
Plant mitochondrial genomes are conventionally reported as circular structures, much like the circular chromosomes found in animal mitochondria and bacteria. This convention reflects the fact that plant mitochondrial DNAs (mtDNAs) can generally be mapped as circles and that they, at least in some cases, occur in a supercoiled state (Sparks & Dale, 1980; Palmer & Shields, 1984; Palmer, 1988; Oda et al., 1992). However, for at least two reasons, the in vivo structure of plant mtDNA is far more complex than a simple circular chromosome model would suggest. First, mtDNA in vascular plants usually contains repeated sequences that are recombinationally active and serve as a constant source of rearrangements (Lonsdale et al., 1984; Palmer & Shields, 1984). As a result, the plant mitochondrial genome can exist as a population of alternative structures, even within a single individual. Therefore, plant mtDNA is typically described as multipartite, interconverting between a ‘master circle’ conformation which contains the entire sequence content of the genome and a set of subgenomic circles generated by repeat-mediated recombination (Fig. 1a; Table 1). Second, it is well known that circular genome maps do not necessarily imply the existence of circular molecules, as alternative structures, such as head-to-tail concatamers and overlapping linear fragments (i.e. circularly permuted linear molecules), can also map as circles (Fig. 1b) (Bendich, 1993). Attempts to more directly observe the in vivo structure of plant mtDNA have generally failed to recover a large population of genome-sized circular molecules, instead finding that linear and branched structures predominate (Oldenburg & Bendich, 1996; Backert & Börner, 2000; Manchekar et al., 2006). These observations are consistent with work on a number of other eukaryotes, which has suggested that noncircular molecules may be the rule rather than the exception for mtDNA outside of animals (Bendich, 1993). Nevertheless, the notion of circular plant mitochondrial genomes has persisted, and the evidence for the predominance of linear mtDNA molecules in plants has been questioned as a potential artifact resulting from DNA breakage during the extraction process.
Table 1. Definition of terms
A DNA molecule that does not share any large repeats with the main portion of the genome, and therefore experiences little or no recombination with other genomic regions. Such sequences cannot be incorporated into a conventional master circle map.
Circularly permuted linear molecules
A set of linear molecules with overlapping ends, allowing for a circular map or assembly even if no such structure exists in vivo
A large linear DNA molecule containing multiple genome copies connected end-to-end. Such molecules can map or assemble as circles even if circular conformations do not exist in vivo.
The co-occurrence of two different mitochondrial haplotypes in the same individual as a result of either de novo mutation or biparental inheritance of mitochondrial DNA (mtDNA)
The conventional format in which plant mitochondrial genome maps or assemblies are reported, consisting of a single circular structure that includes the entire sequence content of the genome
In the context of this review, a multichromosomal genome is one that contains one or more autonomous chromosomes (see definition above). This is not synonymous with classic multipartite models of plant mtDNA in which subgenomic circles can be converted into a master circle by recombination across shared repeats.
Broadly speaking, multipartite genomes are any that can exist in multiple pieces (e.g. segmented viral genomes), but, in the context of this review, the term is used to describe the model of plant mtDNA structure in which repeat-mediated recombination results in the interconversion between a master chromosome and subgenomic molecules.
A DNA molecule representing a subset of the genome produced by recombination between direct repeats within a master chromosome
With recent advances in DNA sequencing technology, there has been a proliferation of plant mitochondrial genome projects (Mower et al., 2012b), and > 50 complete land plant mitochondrial genome sequences are currently available in GenBank. The increased availability of genomic data has made it possible to glean new insights into plant mtDNA structure. For example, in addition to producing raw nucleotide data, deep sequencing technologies estimate the variation in relative sequence abundance based on the depth of coverage across the genome. Paired-end sequencing also provides a means of identifying structural rearrangements by mapping the relative positions of short reads generated from each end of longer DNA fragments (Fig. 2). Of course, DNA sequencing is not a panacea. Because the assembly of a genome from short sequenced fragments is a form of mapping, it is subject to the classic ambiguities involved in inferring in vivo molecular structure from genome maps (Fig. 1b). Therefore, it is important to interpret the findings from genome sequencing efforts in a broader context that incorporates more direct observations of mtDNA structure. The goal of this review is to distill the structural implications from recent genome projects and discuss some of the intriguing evolutionary and functional questions raised by the observed structural diversity in plant mtDNA.
Sequencing and assembly methods for plant mitochondrial genomes
Recent plant mitochondrial genome sequencing projects have used a diverse set of approaches. Some studies have sequenced large mitochondrial clones that were screened from fosmid libraries (Grewe et al., 2009; Hecht et al., 2011; Liu et al., 2012), but most have used a conventional shotgun sequencing approach, in which genomic DNA is randomly sheared, and the resulting fragments are sequenced to produce a sufficient level of redundancy in genome coverage for complete assembly. Typically, this approach has been performed with purified mtDNA to reduce the total amount of sequencing required and to limit assembly complications associated with the contamination of nuclear and plastid sequences. In particular, the abundant sequences of mitochondrial origin in the nucleus (and vice versa) can make it difficult to computationally disentangle sequences from different genomic compartments when working with total cellular DNA (Lough et al., 2008; Goremykin et al., 2012). Nevertheless, recent studies have been able to successfully assemble complete mitochondrial genomes from total cellular DNA. For example, the Mimulus guttatus mitochondrial genome was completed by mining mtDNA reads from large-insert libraries generated as part of the effort to sequence the M. guttatus nuclear genome (Mower et al., 2012a), and the mitochondrial genomes of other angiosperms have been assembled from total cellular DNA sequences by taking advantage of the advent of high-throughput sequencing technologies such as 454, Illumina and SOLiD (Iorizzo et al., 2012; Kazakoff et al., 2012; Wang et al., 2012; Zhang et al., 2012). In general, plant mitochondrial genome sequencing efforts have been somewhat slow to adopt these newer technologies, presumably because their short reads pose assembly challenges, given the repetitive nature of plant mtDNA. However, recent studies have overcome these difficulties by using large-insert, paired-end libraries to provide additional structural information, a classic technique that has been adapted to current sequencing platforms (Sloan et al., 2012a). In addition, as discussed below, researchers have begun to employ assembly methods that are customized to the specific challenges of plant mitochondrial genomes.
Repetitive content complicates any genome assembly, but the repeats in plant mtDNA pose a more fundamental challenge because they are a source of constant rearrangements (Fig. 1a). Assembly programs are typically designed under the assumption that the target genome exists as a single sequence, but plant mtDNA consists of a population of alternative structures that result from recombining repeats (Maréchal & Brisson, 2010). The biological reality of these genomes may be better conceptualized as a network of sequences with alternative connections, rather than as a single unbroken string. Therefore, assembly methods that explicitly recognize the network- or graph-like nature of plant mitochondrial genomes have proven to be effective (Zhang et al., 2011; Iorizzo et al., 2012). This is increasingly important, as current deep sequencing technologies are not only capable of detecting common recombination events, but also of identifying and quantifying low-frequency rearrangements associated with short repeats. Recent studies have used read pair conflicts from paired-end sequencing to measure the recombinational activity of short repeats on a genome-wide basis in multiple angiosperms (Alverson et al., 2011a; Mower et al., 2012a; Sloan et al., 2012a). Such recombination events have been linked to the phenomenon of substoichiometric shifting, in which rare mitochondrial genome conformations rapidly rise to high frequency, with potentially important phenotypic consequences (Maréchal & Brisson, 2010; Woloszynska, 2010; Arrieta-Montiel & Mackenzie, 2011; Davila et al., 2011).
Further development of informatic tools which can accommodate the populations of alternative structures that comprise plant mtDNA is an important goal for the field of plant mitochondrial genomics. The recently developed FASTG format, which elaborates on the familiar FASTA sequence format to retain information relating to variations in genome sequence and structure within an assembly, may be particularly well suited to the capture of the structural complexity of plant mtDNA (The FASTG Format Specification Working Group, 2012).
Evaluation of the master circle model in the light of genomic data
The results from the sequencing and mapping of plant mtDNA have historically supported the existence of circular genomes, and thus have been at odds with electrophoretic and microscopy studies, which have found that genome-sized circles are rare or even undetectable (Bendich, 1993). However, more recent plant mitochondrial sequencing efforts have also identified examples that conflict with a conventional master circle model. For instance, shotgun sequencing data from the M. guttatus mitochondrial genome (Fig. 2) could not be fully reconciled with a model consisting solely of a master circle and recombination-derived subgenomic circles (Mower et al., 2012a). By extracting mtDNA reads from the Mimulus genome project, Mower et al. obtained unusually high physical coverage for a plant mitochondrial genome, resulting from deep Sanger sequencing (> 100×) of libraries with larger insert sizes (up to 35 kb). These data were used to quantify alternative genome conformations. Specifically, each pair of large repeats in the genome was associated with four alternative ‘environments’, defined by the four possible pairwise combinations of single-copy flanking sequences. By counting the number of read pairs (i.e. clones) consistent with each of these repeat environments, the authors rejected the prediction of the master circle model that pairs of repeat environments that co-occur on the same chromosome(s) should be equimolar. Instead, they found that these repeat environments varied up to two-fold in relative abundance, which could potentially be explained by a genome composed of overlapping linear molecules that are present in unequal frequencies.
A better understanding of mtDNA replication (and degradation) will be essential to the interpretation of the variation in the relative abundance of different sequences and structures within plant mitochondrial genomes. A central assumption of the Mimulus analysis is that the physical linkage between sequences on a chromosome should constrain them to exist at the same copy number. In bacteria, however, where the existence of circular chromosomes is well established, sequences can vary in relative abundance across the length of the chromosome. For example, in actively dividing cells, sequences in close proximity to the origin of replication can be highly overrepresented (Frangeul et al., 1999). Little is known about the process of DNA replication in plant mitochondria or whether specific genomic regions are responsible for the initiation of replication, but it has been proposed that these genomes rely on a recombination-dependent mode of replication (Maréchal & Brisson, 2010), with small repeat sequences that experience asymmetric recombination events potentially serving as origins of replication (Davila et al., 2011). The use of deep sequencing to estimate relative copy number across the genome may provide a valuable tool to test the hypothesis that specific mtDNA regions act as origins of replication (and exhibit corresponding increases in sequencing coverage).
The master circle model was originally devised in the context of relatively simple genome structures. For example, the early analysis of the Brassica campestris mitochondrial genome produced a circular map with only a single pair of large repeats, which could potentially recombine to produce two subgenomic circles (Palmer & Shields, 1984). This proposed tripartite model therefore showed some interesting parallels to the flip–flop model of plastid genome structure (Palmer, 1983), in which recombination between the pair of large inverted repeats interconverts between two alternative genome conformations – coincidentally, another case in which circular maps may not necessarily reflect the existence of circular molecules (Bendich, 2004; Maréchal & Brisson, 2010). Given the simple origins of the master circle model, the discovery of exceptionally complex plant mitochondrial genomes with large numbers of recombining repeats and alternative structural connections has posed additional difficulties for this model.
The first mitochondrial genomes from lycophytes (the group that includes clubmosses, spikemosses and quillworts and is sister to the rest of the vascular plants) have been sequenced recently, revealing an abundance of recombinational breakpoints that could not be reconciled with any simple circular map (Grewe et al., 2009; Hecht et al., 2011). The network of alternative connections in each of these genomes was reported in a form that is more reminiscent of a crisscrossing subway map than a conventional genome.
Sequencing of the mitochondrial genome from the angiosperm Silene vulgaris also found a high density of recombining repeats, associated with extreme structural variation within individuals (Sloan et al., 2012a). In addition, comparisons of mitochondrial genomes from different populations of this species revealed extensive structural reorganization, with essentially no conservation of synteny (Sloan et al., 2012b). The mtDNA diversity in S. vulgaris also extends to genomic content, with genomes from any pair of populations sharing only about one-half of their total sequence in common. More generally, angiosperms often exhibit intraspecific variation in mtDNA structure, and genomic rearrangements have proven to be informative for the reconstruction of mitochondrial genealogies (Darracq et al., 2010). Interestingly, the two most extreme examples of intraspecific diversity observed thus far are from gynodioecious species (Beta vulgaris and S. vulgaris), in which mitochondrial sequences are responsible for a high frequency of male sterility in natural populations (Darracq et al., 2011; Sloan et al., 2012b). This suggests that the elevated levels of nucleotide polymorphism previously observed in mtDNA from gynodioecious species (Touzet & Delph, 2009) are paralleled by similar increases in structural variation.
Although a minor fraction of each of the S. vulgaris mitochondrial genomes was found in small, autonomous chromosomes (discussed in detail below), most of the sequence could be assembled into a single chromosome by tracing a circular path through the observed network of alternative structural connections. It should be noted, however, that the existence of large circular structures in S. vulgaris mtDNA represented more of an assumption than a finding of this assembly process. Similar to observations in M. guttatus (Mower et al., 2012a), the depth of sequencing coverage across the S. vulgaris mitochondrial genome was often inconsistent with the idea of a large circular chromosome as the predominant form of mtDNA in this species. Therefore, some recent sequencing efforts have yielded results that are more in line with the long-standing conclusion from electrophoretic and microscopy studies that noncircular structures predominate in plant mitochondrial genomes.
In contrast with the structurally labile mitochondrial genomes in lycophytes and S. vulgaris, genome sequencing has also identified numerous angiosperm mtDNAs that are largely devoid of repetitive content. For > 20 yr, the unicircular map of the white mustard (Brassica hirta) mitochondrial genome stood out as an anomaly among angiosperms with its lack of large recombining repeats (Palmer & Herbon, 1987). However, the recently sequenced genomes of grape (Vitis vinifera), zucchini (Cucurbita pepo), mung bean (Vigna radiata), duckweed (Spirodela polyrhiza) and bamboo (Bambusa oldhamii) also lack repeats greater than 1 kb in length, suggesting that such repeats do not play any essential role in plant mitochondrial genome function (Goremykin et al., 2009; Alverson et al., 2010, 2011ab; Wang et al., 2012). Genomes without recombining repeats can be thought of as a special case of the master circle model, in which no subgenomic circles are created. Therefore, they represent an attractive basis for comparison to test predictions of this model and examine the effects of large recombining repeats on in vivo genome structure.
Multichromosomal mitochondrial genomes in angiosperms
The increased availability of plant mitochondrial genome sequences has also revealed fundamental differences in structural organization across species. In particular, recent studies have found that multichromosomal genomes have evolved in at least two independent angiosperm groups (Table 2). This was first discovered in cucumber (Cucumis sativus). Most of the 1.6-Mb mitochondrial genome in this species can be assembled into a single master circle-like map (Alverson et al., 2011a). However, genome sequencing also identified two small, circular-mapping chromosomes of 84 and 45 kb in size. These two chromosomes share a pair of 3.6-kb recombining repeats, and also exist in a combined conformation that maps as a 129-kb circle. However, they do not share any significant repeats with the main chromosome, making the multichromosomal organization of cucumber distinct from the typical multipartite structure of plant mitochondrial genomes (Table 1). The small chromosomes in cucumber are autonomous in the sense that they exhibit little or no evidence of recombination with the rest of the genome, and therefore cannot be incorporated into a single master chromosome map. They are also fundamentally different from the extrachromosomal plasmids that are often found in plant mitochondria. In particular, their nucleotide composition, sequence content and relative size (more than an order of magnitude larger than any circular plasmid found in plant mtDNA) all suggest a more direct relationship to the genome than observed with plasmids (Alverson et al., 2011a).
Table 2. Examples of multichromosomal plant mitochondrial genomes
Size of main
NA, not applicable.
Silene vulgaris (KOV)
Silene vulgaris (MTV)
Silene vulgaris (S9L)
Silene vulgaris (SD2)
Even more striking examples of multichromosomal organization have been identified in mitochondrial genomes from the genus Silene (Caryophyllaceae). The bladder campion S. vulgaris contains multiple autonomous mitochondrial chromosomes, some of which map as circular structures that are as small as 5 kb (Sloan et al., 2012a). Sampling of multiple populations found that the mitochondrial genomes within this species contain anywhere from one to four of these small, autonomous chromosomes, which in total represent between 1% and 15% of the genome content (Sloan et al., 2012b). In other Silene species, the mitochondrial genome has undergone massive expansion (up to 11.3 Mb in size), and the entire sequence has been fragmented into dozens of circular-mapping chromosomes of < 200 kb each (Sloan et al., 2012a). Many of these chromosomes do not share any large repeats with the rest of the genome, and those that do contain repeats appear to experience relatively infrequent recombination. Analysis of paired-end read conflicts from deep sequencing datasets found a much lower frequency of alternative genome conformations than is typically associated with large repeats in plant mitochondrial genomes (Sloan et al., 2012a).
Although the existence of autonomous chromosomes in cucumber and Silene mtDNA has been inferred almost entirely on the basis of shotgun sequencing data, DNA gel blot analysis of the smallest autonomous chromosome in S. vulgaris confirmed the presence of supercoiled circular molecules of the expected size (Sloan et al., 2012b). Interestingly, however, most copies of this sequence were found in a variety of higher molecular weight forms, indicating that the chromosomes also exist in alternative conformations, most probably involving tandemly repeated copies (Fig. 3). The extent to which the structure of this autonomous chromosome in S. vulgaris is representative of other multichromosomal mitochondrial genomes in Silene and cucumber is not clear. It also remains to be seen whether the in vivo structure of the autonomous chromosomes mirrors that of plant mtDNA more generally. Therefore, the application of techniques, such as electron microscopy (EM), pulsed-field gel electrophoresis (PFGE) and DNA gel blot hybridization, to a broad sample of plant mtDNAs would provide a valuable picture of structural diversity.
Intriguingly, many of the autonomous mitochondrial chromosomes in cucumber and Silene do not contain any identifiable genes (Alverson et al., 2011a; Sloan et al., 2012a). Therefore, it is unclear what, if any, functional role they play. In S. vulgaris, most of the identified autonomous chromosomes were found in only one of four sequenced mitochondrial genomes, with the chromosomal content either absent or distributed across the rest of the genome in other populations (Sloan et al., 2012b). However, the smallest S. vulgaris chromosome was found in all four sequenced genomes. Given the remarkable lack of conservation of noncoding sequences in S. vulgaris mitochondrial genomes (Sloan et al., 2012b), the retention of this chromosome suggests that it is under functional constraint, or that it maintains some selfish replication advantage, as has been proposed for B chromosomes in many eukaryotic nuclear genomes (Burt & Trivers, 2006). The only identifiable sequence element in the chromosome is a small (182-bp), duplicated and divergent fragment of the 26S ribosomal RNA gene, and so it is possible that it is involved in the generation of alternative genome structures via rare recombination events (Arrieta-Montiel & Mackenzie, 2011). Given the high prevalence of cytoplasmic male sterility (CMS) in S. vulgaris, the possibility that its autonomous chromosomes and variable mitochondrial genome structure might contribute to CMS is intriguing. Notably, a 21-kb autonomous chromosome in the S. vulgaris S9L mitochondrial genome harbors a chimeric open reading frame (ORF) that is absent from other sequenced mitochondrial genomes in this species (Sloan et al., 2012b). Similar chimeric ORFs are known to cause CMS in other angiosperms.
The existence of multichromosomal mitochondrial genomes in angiosperms adds to the long list of unanswered questions regarding the mechanisms of replication and inheritance of plant mtDNA. In nuclear genomes, DNA replication and chromosomal segregation are strictly coordinated with the cell cycle, limiting opportunities for the proliferation of selfish genetic elements and ensuring that daughter cells receive a full complement of chromosomes. It is less clear how faithful inheritance occurs in the context of a multichromosomal cytoplasmic genome. Genome sequencing has provided evidence that chromosomes exist at different relative abundances, indicating that they may replicate independently from each other (Alverson et al., 2011a). Nevertheless, the dozens of chromosomes found in some Silene mitochondrial genomes are maintained over a fairly narrow range of relative copy numbers, suggesting some degree of coordination (Sloan et al., 2012a).
In S. vulgaris, there is evidence of occasional biparental inheritance of mtDNA (Bentley et al., 2010). The resulting heteroplasmy, combined with frequent fusion of mitochondria and the highly recombinogenic nature of plant mtDNA, creates the opportunity for an effectively sexual mode of inheritance (McCauley, 2013). This is consistent with the observation that mitochondrial polymorphisms are not in complete linkage disequilibrium in some Silene species, in contrast with what would be expected for a genome evolving under strict uniparental inheritance (Städler & Delph, 2002; McCauley & Ellis, 2008). The existence of autonomous mitochondrial chromosomes in this genus creates the potential for another mechanism for the mixing of maternally and paternally inherited mtDNA that does not require physical recombination (i.e. if some but not all of the chromosomes are transmitted via pollen). This could be particularly relevant for species such as S. noctiflora and S. conica with highly multichromosomal genomes, in which the relative abundance of alternative genome conformations suggests that rates of repeat-mediated recombination are greatly reduced (Sloan et al., 2012a).
It is also possible that the structure and content of plant mitochondrial genomes differ between meristematic and vegetative tissues. For example, it has been hypothesized that the master circle could represent the inherited form of the genome, whereas alternative conformations and subgenomic molecules accumulate in vegetative tissues (Woloszynska, 2010). In this sense, it is conceivable that the observed multichromosomal mitochondrial genomes in cucumber and Silene could be a ‘somatic’ phenomenon, resulting from the fact the mtDNA was extracted predominantly from vegetative tissue, and that the genome structure found in meristematic tissues is fundamentally different. Addressing this possibility will require a detailed analysis of mtDNA structure in meristematic tissue. In addition, there is evidence that mitochondrial genome structure can vary with environmental conditions. For example, in stressful environments, plants may down-regulate genes that normally suppress recombination between short repeats in the mitochondrial genome, and this may explain the extensive structural rearrangements and increased frequency of supercoiled DNA when plant tissue is brought into in vitro culture (Arrieta-Montiel & Mackenzie, 2011). Measurement of the variation in chromosome structure and relative abundance in multichromosomal mtDNAs sampled across populations and in a range of different tissue types, developmental stages and environmental conditions should help to unravel the mechanisms of inheritance in these unique genomes.
DNA sequencing has proven to be a valuable tool for the capture of diversity in plant mtDNA structure and has identified novel genome architectures, highlighted by the discovery of multichromosomal genomes in two independent angiosperm lineages. Given that the sampling of mitochondrial genomes is still sparse in angiosperms, and even more so in other plant groups, it is likely that the evolution of similar genome complexities will turn out to be much more widespread in plant mitochondria. Sequencing studies have also provided new support for earlier EM and PFGE analyses, which found that genome-sized circular molecules are not the predominant form of plant mtDNA. Not surprisingly, the advent of deep sequencing technologies has intensified (rather than obviated) the need for the application of conventional tools for structural analysis to a more diverse sample of plant mtDNAs.
Although there is compelling evidence that both circular and noncircular genomic structures exist in plant mitochondria, the relative abundance and importance of each remain a source of controversy. This is perhaps best illustrated by the fact that the (rare) circular plant mtDNA molecules detected in vivo have alternatively been dismissed as a functionally irrelevant byproduct of DNA damage and repair in metabolically active organelles, or hypothesized to be the true inherited form of the genome that is replicated in meristematic tissues and transmitted to the next generation. Therefore, the pressing challenges for this field are to elucidate the evolutionary forces responsible for the origin and maintenance of structural complexity in plant mitochondrial genomes, and to determine the functional consequences of this complexity for the expression and inheritance of plant mtDNA. General rules may emerge, but the remarkable variation observed, even among closely related angiosperms, suggests that these rules are subject to rapid evolutionary change.
I would like to thank Dave McCauley and two anonymous reviewers for helpful comments on an earlier version of the manuscript, and I acknowledge support from the National Science Foundation (NSF) (MCB-1022128) for research on plant mtDNA structure in Silene.