Theoretical arguments suggest that mutation rates influence the proliferation and maintenance of RNA editing. We identified RNA editing sites in five species within the angiosperm genus Silene that exhibit highly divergent mitochondrial mutation rates. We found that mutational acceleration has been associated with rapid loss of mitochondrial editing sites. In contrast, we did not find a significant difference in the frequency of editing in chloroplast genes, which lack the mutation rate variation observed in the mitochondrial genome. As found in other angiosperms, the rate of substitution at RNA editing sites in Silene greatly exceeds the rate at synonymous sites, a pattern that has previously been interpreted as evidence for selection against RNA editing. Alternatively, we suggest that editing sites may experience higher rates of C-to-T mutation than other portions of the genome. Such a pattern could be caused by gene conversion with reverse-transcribed mRNA (i.e., retroprocessing). If so, the genomic distribution of RNA editing site losses in Silene suggests that such conversions must be occurring at a local scale such that only one or two editing sites are affected at a time. Because preferential substitution at editing sites appears to occur in angiosperms regardless of the mutation rate, we conclude that mitochondrial rate accelerations within Silene have “fast-forwarded” a preexisting pattern but have not fundamentally changed the evolutionary forces acting on RNA editing sites.

IN the organelle genomes of land plants, a variable but often large number of sites undergo C-to-U RNA editing in which a cytidine is converted to uridine by deamination (Yu and Schuster 1995; Giege and Brennicke 1999). A generally much smaller number of sites undergo “reverse” U-to-C editing (Steinhauser et al. 1999). RNA editing is believed to be essential for organelle gene function in plants. Editing sites are preferentially located in protein genes and, within them, at first and second codon positions (Gray 2003). Editing at these sites generally results in the restoration of phylogenetically conserved (and presumably functionally constrained) amino acids in mitochondrial and chloroplast protein sequences (Gray and Covello 1993; Mower 2005; Yura and Go 2008). Therefore, there are obvious selective pressures for plants to maintain RNA editing in the short term. In contrast, the origin and long-term maintenance of RNA editing pose an evolutionary puzzle, as it is unclear what if any benefit this seemingly cumbersome process confers over direct encoding of the edited sequence in the genomic DNA. This puzzle mirrors broader evolutionary questions about the origin, maintenance, and function of a number of major features of gene and genome architecture (Lynch 2007).

Various adaptive effects of RNA editing have been proposed, including a role in gene regulation (Hirose et al. 1999; Farajollahi and Maas 2010), maintenance of alternative functional protein isoforms (Gott 2003; Farajollahi and Maas 2010), generation of genetic variation (Tillich et al. 2006; Gommans et al. 2009), optimization of genomic GC content (Jobson and Qiu 2008), nuclear control of selfish organelle genes (Burt and Trivers 2006), and mutational buffering (Borner et al. 1997). In humans, there is evidence for divergent functional roles for products of the edited and unedited forms of apolipoprotein B (Powell et al. 1987), but there is little evidence for the aforementioned adaptive mechanisms in plant organelles. Moreover, replacement of edited C's with T's at the genomic level appears to occur readily across lineages with no obvious detrimental effect (Shields and Wolfe 1997; Mower 2008). Accordingly, neutral and nonadaptive models for the proliferation of RNA editing have also been proposed (Covello and Gray 1993; Fiebig et al. 2004; Lynch et al. 2006).

C-to-U RNA editing appears to have evolved in a recent common ancestor of land plants, but the frequency of editing varies dramatically across lineages and between genomes (Turmel et al. 2003; Salone et al. 2007; Grewe et al. 2009). Seed plants experience relatively frequent editing in mitochondrial genes with ∼400 sites per species (>1% of all coding sequence) in the few angiosperms examined so far (e.g., Giege and Brennicke 1999) and even higher rates inferred in gymnosperms (Lu et al. 1998; Chaw et al. 2008; Ran et al. 2010). The rate of editing in seed plant chloroplast genomes is over an order of magnitude lower (Tillich et al. 2006). Outside seed plants, the moss Physcomitrella patens has only 11 edited sites in the mitochondrial genome, and the liverwort Marchantia polymorpha appears to have lost RNA editing altogether (Rüdinger et al. 2008, 2009). In contrast, other bryophytes, lycophytes, and ferns exhibit frequent mitochondrial editing (Malek et al. 1996; Grewe et al. 2009; Li et al. 2009), while the only two nonseed plant chloroplast genomes examined (from one hornwort and one fern) also have high levels of editing (Kugita et al. 2003; Wolf et al. 2004).

Mutation rate variation is one proposed mechanism to explain phylogenetic variation in the occurrence and frequency of RNA editing. Plant mitochondria tend to have very slow point mutation rates as evidenced by their low rate of synonymous substitution (Wolfe et al. 1987; Palmer and Herbon 1988; Mower et al. 2007; Drouin et al. 2008). Lynch et al. (2006) have argued that high mutation rates intensify selection against RNA editing because of the increased probability of disrupting sequences necessary for editing site recognition [the mutational burden hypothesis (Lynch 2006)]. This led them to conclude that the low mutation rates in plant mitochondrial genomes have created a permissive environment in which RNA editing sites have been able to proliferate by nonadaptive processes. Recent studies have documented a handful of angiosperm lineages that have experienced dramatic accelerations in synonymous substitution rate [and presumably the underlying mutation rate (Cho et al. 2004; Parkinson et al. 2005; Mower et al. 2007; Sloan et al. 2008, 2009)]. In one extreme example, RNA editing sites appear (based on a sampling of relatively few genes) to have been almost completely eliminated in Pelargonium hortorum, which has experienced a series of two successive ∼10-fold increases in mitochondrial mutation rate (Parkinson et al. 2005). While these results are grossly consistent with the role of selection proposed by Lynch et al., the data have also been interpreted as evidence for potentially frequent conversion of reverse-transcribed mRNA back into the genome. Parkinson et al. (2005) suggest that such “mutagenic retroprocessing” could explain both the high overall rate of mutation (because of the very high error rates in reverse transcription) and the loss of RNA editing (because the edited sequence is being incorporated into the genome).

This existing body of work raises two questions that we address in the present study: (1) Do major losses of RNA editing sites always accompany mutation rate accelerations in angiosperms? And (2) if so, to what extent can this correlation be explained by selection (Shields and Wolfe 1997; Lynch et al. 2006; Mower 2008) as opposed to the mutational bias introduced by retroprocessing [RNA-mediated gene conversion (Parkinson et al. 2005; Mulligan et al. 2007)]?

The genus Silene (Caryophyllaceae) offers a particularly valuable system for investigating the consequences of mitochondrial mutation rate variation. Multiple species within the genus exhibit increases of >100-fold in synonymous substitution rate (Figure 1; Sloan et al. 2009). These changes appear to have occurred very recently (<5–10 million years ago), allowing for close phylogenetic comparisons between accelerated and nonaccelerated species (Mower et al. 2007; Sloan et al. 2009). The recent timing of these accelerations enables tests of molecular evolution based on synonymous site divergence, which has yet to approach saturation despite the extreme substitution rates. Interestingly, a previous study found little difference in the predicted frequency of mitochondrial RNA editing sites between accelerated and nonaccelerated Silene species (Mower et al. 2007). This analysis, however, was restricted to a small sample of five relatively lightly edited Silene mitochondrial genes.

Figure 1.—

Mitochondrial rate accelerations in Silene. Branch lengths correspond to the number of synonymous substitutions per site for mitochondrial DNA (A) and chloroplast DNA (B). Analyses were performed in PAML v4.1 (Yang 2007) on concatenated data sets (see Table 2), using a codon-based model of evolution and constrained topology as described previously (Sloan et al. 2009).

We performed cDNA sequencing to empirically identify RNA editing sites for all mitochondrial protein-coding genes in Silene latifolia, which exhibits a slow rate of mitochondrial substitution typical of most plants, and a close relative, S. noctiflora, which has experienced a mitochondrial rate acceleration of >100-fold. We also analyzed a smaller sample of genes for three additional species including S. conica, which has a mitochondrial substitution rate comparably high to that of S. noctiflora. We find that the mitochondrial mutational accelerations in Silene have been associated with substantial loss of RNA editing sites via C-to-T substitutions. We explore these results along with previously published data in the context of competing models to infer the potential role of both adaptive and nonadaptive forces in the evolution of RNA editing.


Plant material and RNA and DNA extraction:

Silene contains hundreds of predominantly herbaceous species, some of which are widely used in studies of ecology and evolution (Bernasconi et al. 2009). Seeds for each of five Silene species (S. latifolia, S. noctiflora, S. conica, S. vulgaris, and S. paradoxa) were grown in the greenhouse under a 16-hr–8-hr light–dark cycle. Fresh leaf tissue was collected from plants that were >10 weeks old for both RNA and DNA extraction. Total cellular RNA was purified using the RNeasy Mini Kit (QIAGEN, Valencia, CA). Contaminating genomic DNA was removed by digestion with RNase-free DNase I (QIAGEN) for 30 min at room temperature, and postreaction cleanup was performed with a second RNeasy spin column. Genomic DNA was extracted using a modified CTAB method (Doyle and Doyle 1987).

Reverse transcription, PCR, and sequencing:

cDNA was produced by reverse transcription of purified RNA primed with random hexamers, using M-MulV reverse transcriptase (New England Biolabs, Beverly, MA) in accordance with manufacturer's recommendations.

PCR primers were designed to amplify overlapping fragments of up to 700 bp for all mitochondrial protein genes on the basis of draft mitochondrial genome assemblies for S. latifolia and S. noctiflora (our unpublished data). We also designed additional primers that were conserved across Silene species for portions of seven mitochondrial genes (ccmFn, cob, nad2, nad5, nad6, nad7, and nad9) and five chloroplast genes (ndhB, psbL, rpoB, rpoC1, and rps2). Primer sequences were chosen to exclude predicted RNA editing sites to avoid enriching for unedited transcripts (Mower 2005; Mower and Palmer 2006). cDNA was amplified using standard PCR techniques. We also PCR amplified genomic DNA when the corresponding genomic sequence was not already available. PCR products were cycle sequenced directly on both strands as described previously (Barr et al. 2007). Primer sequences are provided in Supporting Information, File S1. Sequences generated in this study were deposited in GenBank under accessions HM099771HM099885.

Identification of RNA editing sites:

RNA editing sites were identified by comparing aligned genomic and cDNA sequences and verified by manual inspection of sequencing electropherograms. Partially edited sites were identified and scored only when peaks for both the edited and the unedited bases were substantially above background on both strands. In some genes, all edited sites exhibited a degree of incomplete editing (see File S2 and File S3), which could reflect true partial editing but may also result from preferential amplification of immature transcripts or contaminating genomic DNA (Mower and Palmer 2006). In such cases, only editing sites that clearly differed from the observed baseline level of incomplete editing were classified as partially edited. This conservative approach is likely to underestimate the number of partially edited sites in favor of unedited or completely edited classifications (Mower and Palmer 2006).


The mitochondrial coding sequences of all five Silene species as well as Beta vulgaris, Arabidopsis thaliana, and Oryza sativa were aligned using ClustalW and adjusted manually. Regions that could not be reliably aligned in frame were excluded from subsequent analyses. RNA editing data for outgroup species were taken from REDIdb (Picardi et al. 2007). RNA editing data for matR in O. sativa were not available, so another monocot, Triticum aestivum, was used in its place.

To determine whether C-to-T substitutions occur preferentially at RNA editing sites, we compared the frequency of C-to-T substitutions at both RNA editing sites and twofold synonymous (nonedited) sites. The latter were chosen as a basis for comparison because they should be under the same pattern of selective constraint as editing sites with respect to protein sequence. In each case, C-to-A and C-to-G substitutions result in amino acid replacement, but C-to-T substitutions are silent. We analyzed two pairs of species, S. latifolia–S. noctiflora and A. thaliana–B. vulgaris. In each case, we used an outgroup (B. vulgaris and O. sativa, respectively) and an unweighted parsimony criterion to infer the ancestral state for each site. C-to-T substitutions were then identified by comparing the sequence of each species against the inferred ancestral state. Beta and Arabidopsis were chosen as a slowly evolving species pair to contrast with the rapid divergence observed within Silene. As members of the Caryophyllales and rosids, respectively, Beta and Arabidopsis span one of the deepest splits in the eudicot phylogeny (Wikström et al. 2001). Therefore, they were expected to exhibit relatively substantial sequence divergence despite their slow substitution rates. Codons for which the amino acid was not conserved between the two ingroup species were excluded from this analysis. We used Fisher's exact test to assess the statistical significance of differences between frequencies of C-to-T substitutions at edited and synonymous sites.

To determine whether C-to-T substitutions at RNA editing sites are clustered within the S. noctiflora mitochondrial genome, we employed the method previously described by Roy and Gilbert (2005) for analyzing spatial patterns of intron loss. We calculated the probability distribution for the expected number of pairs of adjacent losses (i.e., C-to-T substitutions) given the number of lost and retained sites in each gene and the assumption that each loss occurs as an independent event. We compared the observed number of pairs of adjacent losses to this distribution to assess whether the lost editing sites were significantly clustered relative to a random expectation. For this analysis, we excluded 24 editing sites that were inferred to be lost in S. noctiflora by a change other than a C-to-T substitution.

Retroprocessing is expected to result in the loss of both introns and RNA editing sites. To determine whether observed intron losses were significantly associated with losses of flanking editing sites, we compared the observed number of losses to the following null distribution,Math(1)where P[n] is the probability of observing n lost flanking editing sites around a given intron, and r and l are the total number of retained and lost editing sites in the genome, respectively. The n + 1 term in Equation 1 represents the number of possible ways to obtain n flanking editing site losses. For example, there are three ways to lose two flanking sites (i.e., with the intron positioned to the left, to the right, or in between the two sites). The rest of the expression on the right side of Equation 1 represents the probability of a single one of those patterns occurring.


Mitochondrial RNA editing in Silene:

The mitochondrial genomes of S. latifolia and S. noctiflora contain 26 and 27 putatively functional protein-coding genes, respectively (Table 1). The sole difference in intact gene content between the two species is rps13, which is intact in S. noctiflora but contains an internal stop codon in S. latifolia. Comparison of mitochondrial cDNA and genomic sequences revealed a total of 287 C-to-U edited sites in S. latifolia but only 189 sites in the rapidly evolving S. noctiflora genome—a 34% difference (Table 1; File S2 and File S3). No other types of editing (e.g., U-to-C) were detected. The total number of editing sites in either Silene genome is lower than in any other angiosperm mitochondrial genome analyzed to date (Giege and Brennicke 1999; Notsu et al. 2002; Handa 2003; Mower and Palmer 2006; Alverson et al. 2010; Picardi et al. 2010). This pattern results from both a reduced number of mitochondrial genes in Silene and a lower average density of editing sites within each gene (Table 1).

View this table:

RNA editing sites in the mitochondrial protein genes of four eudicots

By contrast, the substantial difference in RNA editing content between S. latifolia and S. noctiflora is caused entirely by a change in the density of editing sites rather than gene loss. The difference in editing site density derives mostly from sites that have been lost in S. noctiflora rather than gained in S. latifolia. There are 111 sites that are edited in S. latifolia but not edited in S. noctiflora. Of these, 106 sites (95.5%) are edited in Beta and/or Arabidopsis, suggesting that editing was the ancestral state. In contrast, there are only 13 sites that were edited in S. noctiflora but not in S. latifolia, and only 3 of these (23.1%) were edited in Beta and/or Arabidopsis. By far the most common pattern of editing site loss was C-to-T substitution at the genomic level, which obviates the need for editing (Figure 2). Although the total number of editing sites is reduced in S. noctiflora, the proportional distribution among genes is quite similar to that in S. latifolia and other angiosperms, indicating a genome-wide loss of mitochondrial RNA editing sites (Figure 3, Table 1).

Figure 2.—

The distribution of nucleotide states at 106 positions where a C-to-U RNA editing site was inferred to be lost in S. noctiflora. Shaded bars indicate sites of amino acid conservation between S. latifolia and S. noctiflora, while solid bars indicate amino acid difference.

Figure 3.—

Loss of RNA editing sites in rapidly evolving mitochondrial genes of S. noctiflora. Each point represents a single gene. Genes falling below the 1:1 line have fewer editing sites in S. noctiflora than in S. latifolia.

On the basis of the divergence between S. noctiflora and S. latifolia, we found a significantly higher C-to-T substitution rate at editing sites (dRE) than at twofold synonymous sites (dS2) (dRE/dS2 = 3.05; P < 0.001). This pattern, however, was not unique to these two rapidly diverging Silene species. Comparison of two slowly evolving lineages (B. vulgaris and A. thaliana) also revealed a faster C-to-T substitution rate at RNA editing sites (dRE/dS2 = 3.55; P < 0.001), and a similar analysis of monocot–eudicot divergence previously showed a comparable (∼3.9-fold) excess of C-to-T substitutions at editing sites (Shields and Wolfe 1997). When we restricted our analysis to synonymous RNA editing sites (dSRE), we still found rate elevation, but it was less pronounced. With so few synonymous editing sites, the rate increase was not significant in either the Silene comparison (dSRE/dS2 = 2.29; P = 0.096) or the Beta–Arabidopsis comparison (dSRE/dS2 = 1.84; P = 0.27). A similar (∼2.4-fold) excess was previously found on the basis of monocot–eudicot divergence (Shields and Wolfe 1997).

We did not find any evidence for an increase in the frequency of incomplete editing associated with the history of mutational acceleration in the S. noctiflora mitochondrial genome. Both Silene species exhibited similar frequencies of partial editing (7.7 and 6.9% of editing sites in S. latifolia and S. noctiflora, respectively). As previously observed (Mower and Palmer 2006), partial editing was common at synonymous sites (File S2 and File S3). Synonymous editing sites were 49 times more likely to be partially edited than nonsynonymous sites in S. latifolia and 24 times more likely in S. noctiflora.

Analysis of other angiosperms has shown that editing sites are distributed in a clustered fashion along the length of mitochondrial genes (Mulligan et al. 2007). This pattern is also evident in Silene, but the observed losses of editing sites in S. noctiflora are not highly clustered (after taking into account the ancestrally clustered distribution of editing sites; Figure 4). We found only 21 adjacent pairs of lost editing sites specific to S. noctiflora (counting only sites that were lost by C-to-T substitution), which is not significantly different from the random expectation (P = 0.31) (Roy and Gilbert 2005). This test for clustering of lost editing sites was similarly nonsignificant when the data were analyzed as individual exons (P = 0.40) or as separately transcribed and trans-spliced fragments (P = 0.42) rather than as whole genes.

Figure 4.—

Distribution of lost and retained RNA editing sites in Silene noctiflora. Vertical lines indicate ancestral RNA editing sites that have been retained in S. noctiflora. Open triangles indicate ancestral editing sites that have been lost in S. noctiflora (but not in S. latifolia) by C-to-T substitution at the genomic level. Editing sites lost by other mechanisms and putative editing site gains are not shown. Solid and shaded triangles indicate the presence of cis- and trans-splicing introns, respectively. Position values exclude intron sequences.

However, there are two genes in Silene that exhibit clear examples of intron loss, and in each of these cases, losses are associated with multiple C-to-T substitutions at RNA editing sites in flanking exons. First, S. latifolia and S. noctiflora lack both cox2 introns as well as the five surrounding editing sites present in Beta (Figure 5A). Although the second cox2 intron is also absent in Beta, it is retained in other species within the Amaranthaceae (our unpublished data), suggesting that it has been lost independently in the Beta and Silene lineages. Therefore, additional RNA editing sites surrounding the second cox2 intron may have been lost in parallel between these lineages (Figure 5A). Overall, the cox2 introns appear to have been lost independently numerous times during angiosperm evolution (Figure S1) (Joly et al. 2001). In the second case of intron loss, S. noctiflora lacks the third nad7 intron along with two adjacent editing sites (Figure 5B). These two cases suggest that retroprocessing led to the simultaneous loss of introns and flanking editing sites. For both genes, the observed number of lost flanking sites exceeds the random expectation. This difference is significant for cox2 (P < 0.001) but not for nad7 (P = 0.23).

Figure 5.—

Loss of introns in cox2 (A) and nad7 (B) is associated with loss of adjacent RNA editing sites in Silene. Solid triangles and vertical lines indicate introns and editing sites, respectively. Shading covers regions surrounding lost introns in which RNA editing sites have also been lost by C-to-T substitution. A more detailed phylogenetic distribution of cox2 RNA editing sites and introns is provided in Figure S1. RNA editing data for outgroup species were obtained from REDIdb (Picardi et al. 2007). Position values exclude intron sequences.

S. conica, another species exhibiting dramatic mitochondrial rate acceleration, contains even fewer RNA editing sites (28) than S. noctiflora (34) in a sample of seven mitochondrial genes (Figure 6, Table 2). The number of shared losses between S. noctiflora and S. conica (13) was significantly higher than expected at random (P = 0.042; Fisher's exact test), perhaps reflecting a shared phylogenetic history between the two species and/or variable selection and mutation pressures across sites. On the other hand, S. noctiflora and S. conica also exhibited a large number of unique losses (7 and 10, respectively), indicating that the specific sites being lost were largely (if not entirely) independent between the two lineages.

Figure 6.—

Number of RNA editing sites in portions of selected (see Table 2) mitochondrial genes (solid bars) and chloroplast genes (shaded bars) in five Silene species. The current understanding of the phylogenetic relationships among these species is depicted at the bottom (Erixon and Oxelman 2008; Sloan et al. 2009; A. Rautenberg, D. B. Sloan, V. Aldén and B. Oxelman, unpublished data).

View this table:

Mitochondrial and chloroplast RNA editing in five Silene species

In contrast to their rapidly evolving congeners, S. vulgaris (51) and S. paradoxa (53) maintain a set of editing sites that is nearly identical to that of S. latifolia (53) for the same seven-gene sample. These three slowly evolving species share an identical core set of 51 editing sites. In addition, S. latifolia and S. paradoxa share a nad5 editing site that has been lost from S. vulgaris, while S. latifolia and S. paradoxa each have a unique site not found in the other two species. In comparison, Beta, Arabidopsis, and Oryza respectively maintain a total of 56, 62, and 67 editing sites in this same gene sample.

Chloroplast RNA editing in Silene:

We sequenced both cDNA and genomic DNA for portions of the chloroplast genes ndhB, psbL, rpoB, rpoC1, and rps2. These loci represented all chloroplast genes that were both predicted to undergo RNA editing in Spinacia (Tsudzuki et al. 2001) and had genomic sequence available for at least one species of Silene to aid in primer design. In contrast to the mitochondrial pattern, S. noctiflora and S. conica did not exhibit higher synonymous substitution rates than their congeners for these chloroplast genes (Figure 1). S. latifolia, S. vulgaris, and S. paradoxa share an identical set of 14 chloroplast RNA editing sites across these five genes (Figure 6, Table 2). S. noctiflora and S. conica each lost one site by C-to-T substitution (in rpoB and psbL, respectively). The loss of 1 of 14 chloroplast editing sites (7%) represents a significantly lower rate of loss than observed in the mitochondrial genome for both S. noctiflora (35.8%; P = 0.040; Fisher's exact test) and S. conica (45.1%; P = 0.011; Fisher's exact test). All five Silene species lacked the RNA editing site in rpoC1 that is present in other eudicots.


We found that recent increases in mitochondrial synonymous substitution rates within the genus Silene have been associated with substantial reductions in the frequency of RNA editing, adding further evidence for a relationship between mutation rate and the maintenance of RNA editing. These findings provide insight into the forces that guide the evolution of RNA editing. For example, as noted previously (Lynch et al. 2006), the negative relationship between mutation rate and the frequency of editing runs directly counter to predictions arising from the hypothesis that editing acts as a mutational buffer (Borner et al. 1997; Horton and Landweber 2002). In addition, these findings raise central questions about the relative importance of adaptive and nonadaptive processes in the evolution of RNA editing, which we discuss below.

Elevated C-to-T substitution rates at RNA editing sites—the role of selection vs. retroprocessing:

More than a decade ago, Shields and Wolfe (1997) found that mitochondrial RNA editing sites undergo elevated rates of divergence among angiosperms. Mower (2008) subsequently showed that the high divergence rates are the result of frequent loss (rather than gain) of editing sites by C-to-T substitution. This loss appears to be driving a general decline in the frequency of mitochondrial RNA editing across the angiosperm phylogeny. This pattern has been interpreted as evidence for selection acting against RNA editing by preferentially favoring the fixation of C-to-T mutations at editing sites. This interpretation, however, rests on the assumption that RNA editing sites experience the same mutation rates as other sites in the genome (Shields and Wolfe 1997). Homologous recombination (gene conversion) with edited cDNA intermediates, i.e., retroprocessing, could violate this assumption by disproportionately affecting editing sites (Parkinson et al. 2005; Mulligan et al. 2007). There is only limited biochemical evidence of reverse transcriptase activity in angiosperm mitochondria (Moenne et al. 1996), but there are numerous convincing examples of the incorporation of reverse-transcribed mRNA back into the mitochondrial genome, reflected by either the simultaneous loss of introns and editing sites in the flanking exons or the loss of an entire suite of editing sites across much or all of a mitochondrial gene. Examples of the former include the loss of introns and neighboring editing sites from nad4 in both the Caryophyllales and Asterales (Geiss et al. 1994; Itchoda et al. 2002), two separate cases involving rps3 in conifers (Ran et al. 2010), and the additional cases involving cox2 and nad7 reported in this study. Examples of editing site losses across much or all of a gene have been reported in numerous and diverse seed plant lineages (Krishnasamy et al. 1994; Lu et al. 1998; Petersen et al. 2006; Lopez et al. 2007). Moreover, it has been suggested that the clustered distribution of RNA editing sites in angiosperm mitochondrial genes reflects the elimination of stretches of previously intervening editing sites by retroprocessing (Mulligan et al. 2007).

The finding of both clustered (Figure 5) and dispersed (Figure 4) losses of editing sites in Silene raises intriguing possibilities with respect to the relative importance of selection and retroprocessing as forces acting to purge RNA editing sites. At one extreme, both forces could be operative but with selection predominating, i.e., with retroprocessing occurring relatively rarely and only across large regions. At the other extreme, retroprocessing could be the predominant if not the sole force driving the loss of RNA editing, acting across a wide range of spatial scales. The latter scenario would require that gene conversion acts on relatively short stretches of nucleotides in plant mitochondrial genomes, because we observed numerous cases where editing sites were lost by C-to-T substitution with no change at nearby sites. So-called “microconversions” are known to occur between short stretches of homologous sequence in other genomes and organisms (Wheeler et al. 1990; Semple and Wolfe 1999; Palmer et al. 2003), while recent evidence indicates that microconversion (presumably DNA mediated) readily occurs in plant mitochondrial genomes (Hao and Palmer 2009; our unpublished data). Single-nucleotide substitutions at editing sites (independent of retroprocessing) and gene conversion with cDNA from partially edited transcripts would be expected to further reduce the clustering of lost editing sites.

Exactly how selection would act against RNA editing is not well established, but it would presumably involve one or both of the following selective pressures: (1) avoiding the deleterious consequences of failed editing (Lynch et al. 2006) and (2) eliminating the need for potentially costly site-specific editing machinery (Kotera et al. 2005; Hammani et al. 2009; Zehrmann et al. 2009). Both of these assume that editing is functionally important. Otherwise, there would be no cost to failed editing and no need to maintain editing machinery. Therefore, if selection has been acting to eliminate RNA editing, we would expect its effects to vary across sites depending upon the functional consequences of editing at those sites. In particular, we would predict that synonymous sites that undergo RNA editing should be relatively free from selection, because the editing process has no effect on the resulting protein sequence. Therefore, if selection is the predominant force in the preferential C-to-T substitution at RNA editing sites, we would predict this pattern to be absent at synonymous editing sites. In contrast, retroprocessing should not discriminate between synonymous and nonsynonymous edits in an mRNA transcript.

Consistent with previous studies (Shields and Wolfe 1997; Mower 2008), we found a trend suggesting that synonymous editing sites also experience preferential C-to-T substitutions, albeit at a lower rate than at nonsynonymous editing sites. At first glance, the rate difference between synonymous and nonsynonymous editing sites might seem inconsistent with a neutral model driven by retroprocessing. However, because synonymous editing sites are prone to both partial editing and frequent evolutionary reversion to unedited status—unaccompanied by C-to-T substitution—reverse transcription is less likely to capture the edited state at synonymous sites than at nonsynonymous sites. Indeed, using a five-state maximum-likelihood model that accounted for the transitions between edited and unedited cytidines within a phylogenetic lineage, Mower (2008) found that rates of C-to-T substitution at synonymous edited sites are at least as high as at nonsynonymous editing sites. Overall then, empirical RNA editing data at synonymous and nonsynonymous sites are consistent with a neutral model based on retroprocessing. Although a role of selection certainly cannot be dismissed, any argument invoking selection to account for the elevated C-to-T substitution rates at RNA editing sites must explain why such selection similarly affects both nonsynonymous and synonymous sites.

Of course, synonymous editing could have important effects on codon usage, mRNA stability, or regulatory sequence conservation (Chamary et al. 2006). The pattern of RNA editing at synonymous sites does not, however, support such adaptive mechanisms. There are relatively few synonymous editing sites in Silene and other angiosperms (Table 1) (Gray and Covello 1993), most of which are only partially edited (File S2 and File S3) (Mower and Palmer 2006). The extent of editing at these sites is much more likely to vary among individuals and tissue types than at nonsynonymous sites (Bentolila et al. 2008), and editing is frequently lost without the C-to-T substitution needed to conserve its effects (Shields and Wolfe 1997; Mower 2008). Overall, synonymous editing has all the hallmarks of a relatively neutral misfiring of the RNA editing machinery (Rüdinger et al. 2009).

The potential for selection on synonymous sites is also important because, as in previous studies, we have used divergence at (nonedited) synonymous sites as a baseline for comparison. The underlying assumption is that these sites are relatively free from selection and, therefore, provide a measure of the neutral substitution rate. If instead synonymous sites are subject to strong selection pressures based on mRNA function, it is possible that the apparent excess of substitutions at RNA editing sites in angiosperms is actually the result of negative selection acting on synonymous sites rather than positive selection or retroprocessing acting on RNA editing sites. In other words, the rate of C-to-T substitutions at RNA editing sites could better reflect the true neutral rate, because such changes are silent at the mRNA level (unlike substitutions at nonedited synonymous sites). Although the available evidence for selection on translational efficiency in plant mitochondrial DNA is limited (Sloan and Taylor 2010), this possibility should be considered as an alternative to the hypotheses analyzed in this study.

An RNA-mediated gene conversion model for the loss of RNA editing sites:

We propose a model in which double-stranded breaks and gene conversion occur regularly in angiosperm mitochondrial genomes, perhaps playing a role in DNA repair. Occasional gene conversion with cDNA produced from edited mRNA (retroprocessing) would result in preferential C-to-T substitution at RNA editing sites. The length of DNA sequence typically affected by gene conversion would have to be short to explain why C-to-T substitutions at editing sites frequently occur without any change at nearby editing sites. Occasionally, however, gene conversion must affect much larger fragments, explaining the loss of large stretches of editing sites and intervening introns. Finally, in plant lineages with high mitochondrial mutation rates (e.g., S. noctiflora), an elevated rate of DNA damage and double-stranded breaks would accelerate the entire process.

While this model provides an attractive explanation of the data, the available evidence is still insufficient to fully support it or to exclude a role of selection in eliminating RNA editing sites in angiosperms. Separating the effects of selection and mutation bias is one of the most difficult and important challenges in the field of molecular evolution, and doing so will require additional studies that integrate phylogenetic, population genetic, and mechanistic data.

Reduced RNA editing content in Silene species with low substitution rates:

Although S. latifolia has a higher density of mitochondrial RNA editing sites than its rapidly evolving congeners, it maintains fewer sites than any other angiosperm analyzed to date (Table 1). The similarity between S. latifolia and other slowly evolving Silene species in the number and identity of mitochondrial RNA editing sites (see results and Table 2) suggests that the Silene ancestor likely had a smaller complement of RNA editing sites than the typical angiosperm. Like the larger reductions in editing observed in S. noctiflora and S. conica, this pattern may be at least partially related to changes in mitochondrial substitution rate. Although dwarfed by recent accelerations in some Silene species (Figure 1), the ancestral mitochondrial substitution rate in the genus appears to be significantly higher than that of related angiosperms. Comparison of all mitochondrial protein gene sequences reveals that the S. latifolia lineage has exhibited a 56% higher synonymous substitution rate than the B. vulgaris lineage (Figure S2), which itself has experienced a higher substitution rate than other fully sequenced angiosperm mitochondrial genomes (Alverson et al. 2010). Therefore, the relationship between RNA editing and evolutionary rates may extend to finer scales of rate variation (see also Alverson et al. 2010).

Mutagenic retroprocessing and elevated substitution rates:

Parkinson et al. (2005) suggested that, because of the nearly complete loss of RNA editing sites in the few mitochondrial genes sequenced in Pelargonium and the high error rate of reverse transcriptase, an increase in retroprocessing itself might be at least part of the cause of mitochondrial mutation rate acceleration in Pelargonium. On the basis of a previous analysis that found little change in the frequency of RNA editing despite the mutational acceleration in S. noctiflora, it seemed unlikely that this mechanism would apply to Silene (Mower et al. 2007). However, given the evidence from the present study that S. noctiflora has, in fact, lost a substantial fraction of its RNA editing sites, it is worth reconsidering this possibility. If mutagenic retroprocessing has been a major cause of high mitochondrial mutation rates in Silene, we would expect to see a disproportionate increase in C-to-T substitution rate for sites that are edited in the mRNA sequence. We did not, however, observe such an effect in S. noctiflora, as the increase in substitution rate at RNA editing sites appears roughly proportional to the rate increase at nonedited sites. Therefore, an increase in the frequency of retroprocessing is unlikely to be the primary cause of the high mitochondrial mutation rates in Silene. Whether mitochondrial rate accelerations in other angiosperms (e.g., Plantago and Pelargonium) are the result of mutagenic retroprocessing remains an open question that is subject to testable predictions (Parkinson et al. 2005).


We have found that increases in mitochondrial mutation rate in Silene are correlated with a rapid loss of RNA editing sites. However, rather than fundamentally altering the evolutionary forces that act on RNA editing, the high mutation rates in these lineages appear to have simply accelerated a preexisting pattern in angiosperm evolution, a pattern that might depend more on a neutral model involving RNA-mediated gene conversion than on selection. Stephen Jay Gould (Gould 1990) famously asked what would happen if we were able to rewind and replay the tape of evolution. The mutational acceleration in some Silene mitochondrial genomes may instead allow us to effectively fast-forward that tape and thereby provide a glimpse into the future of RNA editing in angiosperms.


We thank Jeff Mower for his insightful comments on an earlier version of our manuscript and Michael Hood for providing S. paradoxa and S. conica seeds. This study was supported by National Science Foundation grant DEB-0808452 (to D.B.S. and D.R.T.) and National Institutes of Health (NIH) grant R01-GM-70612 (to J.D.P.). A.J.A. was supported by an NIH Ruth L. Kirschstein National Research Service Award Postdoctoral Fellowship (1F32GM080079-01A1).


  • Received April 20, 2010.
  • Accepted May 7, 2010.


View Abstract