Introduction

Since its first discovery in a trypanosome (Benne et al. 1986), RNA editing, in numerous forms, has been demonstrated in a diversity of eukaryotic organisms (Benne 1996; Smith et al. 1997). RNA editing in plants takes the form of conversion of a C to a U or, more infrequently, a U to a C (Steinhauser et al. 1999; Groth-Malonek et al. 2005). Among plants, RNA editing has been observed extensively among mitochondrial genomes of all land plants except the complex thalloid liverworts (Steinhauser et al. 1999). Similarly, editing has been observed in the plastid genomes of all major lineages of land plants (Freyer et al. 1997), although at a much lower frequency that observed in mitochondrial genomes.

The particular form of RNA editing observed in plant plastid and mitochondrial genomes seems to have originated before or soon after the origin of land plants. Among bryophytes, the frequency of editing events at a particular locus is quite variable in both the mitochondrial (Steinhauser et al. 1999) and the plastid (Freyer et al. 1997) genomes. In particular, the hornwort Anthoceros has been shown to have especially high rates of RNA editing. Steinhauser et al. (1999) examined cDNAs of the mitochondrial gene nad5 from two species of Anthoceros and demonstrated that these sequences have more than twice as many edited sites than observed in any other bryophyte. Groth-Malonek et al. (2004) have since discovered a number of other land plants that also exhibit similarly high editing rates in this same gene. Furthermore, among hornworts edited sites include nearly as many reverse, U-to-C, conversions as the typical C-to-U type. Among land plant chloroplast genomes high rates of editing are also observed in chloroplast rbcL transcripts of Anthoceros (Yoshinga et al. 1996), with 7 of 20 edited sites of the U-to-C type.

Recently generated rbcL sequences for use in a phylogenetic analysis of hornworts (Duff et al. 2004) Suggested the presence of extensive RNA editing across most genera. Here we use a number of these sequences, combined with newly generated DNA and cDNA sequences, to assess the extent and pattern of RNA editing among a diverse assemblage of extant hornworts. We find that extensive editing of rbcL transcripts among all lineages of hornworts except one, Leiosporoceros, which is shown to lack editing sites entirely. This genus is well supported as a basal hornwort. when edited sites are included in phylogenetic analyses; however, we show that the edited sites provide the support for this topology. When edited sites are removed the position of Leiosporoceros cannot be resolved.

Materials and Methods

Nucleic Acid Preparation

Nucleic acids were extracted from either fresh or liquid nitrogen-frozen field-collected material. Total DNA was extracted using the DNeasy Plant Mini Kit (Qiagen). Total RNA was extracted using the RNeasy Plant Mini Kit (Qiagen) followed by treatment with RNase-free DNase I (Invitrogen) for 1 h. The Sensuscript cDNA kit (Qiagen) was used to synthesize first-strand cDNAs using the rbcLRH primer (Duff et al. 2004) for rbcL. PCR, using the forward and reverse primers (Duff et al. 2004), was then performed on the cDNA, the total DNA, and the DNAsed sample prior to cDNA synthesis as a control. Positive PCR products representing approximately 1350 bp of the rbcL gene were cloned and multiple (at least two) positive colonies were sequenced by methods previously described (Duff et al. 2004).

Prediction of Edited Sites

Prediction of RNA editing sites was accomplished by comparison of 7 newly generated sequences and 13 previously reported rbcL sequences for hornworts (GenBank accession numbers are shown in Fig. 1) with 37 additional sequences (outgroups) including green algae, liverworts, mosses, lycopytes, and ferns. Only variable sites in which thymines or cytosines nucleotides were nearly universally conserved among the outgroups were examined. Subsequently, the putative edited sites were compared to the expected amino acid sequence of the same data set. Each site was examined and only those sites which should result in changes to an amino acid were retained as putative edited sites. Lastly, any stop codon found in the amino acid alignment, but not predicted by the above protocol, was assumed to be the result of editing and its position in the amino acid sequence used to predict the site of editing in the DNA sequence.

Figure 1
figure 1

Position and characteristics of the 72 edited sites among 20 hornwort taxa. Boldface numbers indicate predicted sites based on amino acid and nucleotide sequence alignments. Asterisks indicate sites for which editing will result in the repair of a stop codon. Bold-face names indicate samples for which cDNAs were generated or previously available. Numbers to the left of the top rows indicate total numbers of edited sites for each taxon. Numbers at the bottom of each column represent the base position of the edited site. The phylogenetic tree shown is the result of parsimony analysis of all sites, and numbers on the tree indicate bootstrap support for nodes (1000 replicates).

Full size image

Phylogenetic Reconstruction

The effects of editing sites on phylogenetic reconstruction were assessed on a subset of the samples that were used to predict editing sites, which included the 20 hornwort genomic sequences, 2 liverworts, 4 mosses, 4 lycophytes, and 2 green algae. Phylogenetic analyses were performed using PAUP* 4.0b8 (Swofford 2001). All maximum parsimony heuristic searches were performed using random stepwise taxon addition with TBR branch swapping, MulTrees, with chararacters unordered and given equal weight. Analyses were performed on data sets that included (1) the entire data set, (2) all data excluding 72 hornwort edited sites, (3) a character set including only the 72 edited sites, and (4) only hornwort cDNAs with the outgroups. Statistical reliability of inferred clades resulting from these analyses was assessed using 1000 bootstrap replications (Felsenstein 1985).

Results

Characterization and Distribution of RNA Edited Sites

Prior to the generation of new cDNA sequences, sequence alignment comparisons of hornworts with outgroups resulted in the prediction, based on methods described above, of 37 edited sites in addition to the 20 previously reported for Anthoceros. The addition of six new cDNAs (Leiosporoceros dussii, Phaeoceros carolinianus, Phaeoceros pearsonii, Phaeoceros fimbriatus, Nothoceros sp., and Megaceros flagillaris) confirmed the presence of 26 of the predicted 27 sites for these samples. Ten additional predicted sites are from samples for which no cDNA sequence has yet been obtained. Of these 10, 6 represent potential U-to-C conversions that would restore stop codons to a conserved amino acid residue.

In addition to confirming the majority of predicted edited sites, 15 new edited sites were revealed. With these 15, the total number of known and predicted edited sites, among these 20 hornwort samples, stands at 72. Given the discovery of a number (six) of third base edited sites found only by cDNA sequence analyses, it is very likely that our estimate of 72 total edited positions among hornwort rbcL sequences examined here is an underestimate. Figure 1 shows the distribution of edited sites among the 20 taxa and includes the total number of known or predicted edited sites for each sample. Nucleotide positions 74, 141, 214, 339, 363, 870, and 1059 were either differentially edited or edited in both directions (U-to-C and C-to-U) in the cDNAs we examined. Four of the seven represent third base positions at which C-to-U transitions would not result in a change in amino acid. Edited site 214 does reverse a stop codon so the lack of editing detected in both cDNAs of P. pearsonii likely reflects the capture of immature transcripts. In fact, the two transcripts obtained for this sample exhibited different numbers of edited sites (28 and 22). Figure 1 shows the combined data for the two transcripts resulting in a minimum of 30 edited sites.

The distribution of the 72 edited sites is not homogeneous across taxa. While RNA editing is shown to be pervasive among most hornwort lineages, a single sample, Leiosporoceros, had no predicted edited sites, which was confirmed by cDNA sequencing. Among the remaining hornworts, no sample exhibited more than half of the total edited sites. Aside from Leiosporoceros, the total edited sites range from 20 in Anthoceros formosae (Yoshinga et al. 1996) to 34 predicted sites in Phaecoeros coriaceus, P. fimbriatus, and Megaceros giganteus.

Of the 72 total predicted and observed edited sites, 61% (43 of 71) are the typical C-to-U type found in chloroplast genomes of nearly all land plants, while 39% (28 of 71) are of the “reverse” U-to-C type reported only rarely for land plants (Bock 2000). One site exhibits both forward and reverse editing, with multiple transcripts of M. flagellaris all confirming U-to-C editing resulting in an isoleucine residue in place of the conserved threonine found in other land plants. Although we do find more second position edited sites (57%; 41 of 72) than first (35%; 25 of 72), the frequency of first position sites is much higher than in other land plants. U-to-C editing accounts for the high number of edited sites observed in first positions. Seventeen of the 28 U-to-C edited sites are found in the first base position. Of the 17 U-to-C edited sites at first base positions, 11 are reverse stop codons converting either UAA or UGA to CAA or CGA, respectively. Of these 11, only 2 were previously observed in A. formosae (Yoshinaga et al. 1996). However, this repair of stop codons by U-to-C editing is not unusual in hornworts, having been observed 51 total times among all protein-coding genes of A. formosae (Kugita et al. 2003). Only six third position edited sites were observed, none of which would effect the amino acid composition of the polypeptide. In addition, several of these sites (339, 363, 870, and 1059) are apparently not edited in all lineages.

Phylogenetic Analyses

To test the effect of edited sites on hornwort phylogeny, we performed phylogenetic analyses on a data set that included 32 rbcL sequences including 20 hornworts and a subset of 12 of the initial outgroups from the original larger alignment used for predicting RNA editing sites. Maximum parsimony analysis resulted in a single shortest tree (see Fig. 1 for the hornwort topology). A single tree was also obtained after removing all 72 editing the data set (Table 1). These two analyses yielded similar topologies, with the only noteworthy difference being the position and support of Leiosporoceros. In the former, Leiosporoceros has moderate bootstrap support (82%) as sister to the hornworts, however, the latter analysis places Leiosporoceros in a clade with Anthoceros and Folioceros with weak support (52%). When only the 72 edited sites are used in the analysis, the support for the Leiosporoceros sister topology increases to 94% despite the small number of informative characters. Lastly, we performed an analysis which included only the seven cDNA sequences along with the 12 outgroup taxa. As expected, this resulted in similar values to the data set that included the genomic sequences excluding all edited sites.

Table 1 Comparison of phylogenetic analyses resulting from analyses of data that included or excluded RNA editing sites
Full size table

To test if the observed support levels for the edited sites included versus excluded phylogenies were affected by differences in phylogenetic signal, we performed two further analyses. Both of these analyses resulted in the same total number of phylogenetically informative sites as the initial analysis specifically excluding all edited sites. First, 58 of the 215 phylogenetically informative sites, both edited and nonedited, among the hornwort clade were randomly removed while maintaining all other characters and the analysis performed again. As a randomization analysis, this was done 1000 times and a consensus tree constructed from all resultant trees. This analysis resulted in 95% of the trees supporting Leiosporoceros as sister to hornworts. Second, 58 of the 157 nonedited phylogenetically informative sites among hornworts were randomly removed prior to analysis. As above, this was done 1000 times and a consensus of all trees examined, 97%, supported Leiosporoceros as sister to hornworts.

Discussion

RNA Editing Frequency

Extensive RNA editing has been demonstrated to be more than just a peculiarity of a single hornwort, Anthoceros formosae (Kugita et al. 2003). Rather, it is a general phenomenon across the entire hornwort clade, possibly excluding Leiosporoceros. Several lycopsids and ferns also exhibit elevated numbers of edited sites (Kugita et al. 2003; Wolf et al. 2004) in their chloroplast genomes, although none reach the levels attained in hornworts. Among the bryophytes, cpDNA editing is far less common, with only a few known edited sites in rbcL. However, a basal moss, Takakia, is predicted to have 25 edited sites based on the methods described here (Duff, unpublished data). Unlike the hornworts, though, all 25 of these sites are predicted to be the more typical C-to-U type, and only 4 of these sites are potentially homologous to sites found in hornworts. This suggests an independent origin and/or cause of frequent editing in Takakia. The presence of multiple independent lineages with severalfold higher rates of RNA editing among chloroplast genes is similar to that seen in the mitochondrial gene nad5 (Groth-Malonek et al. 2004), where several—a single moss, liverwort, hornwort, and fern—are all observed to have very high rates of RNA editing.

The Evolution and Maintenance of RNA Editing

Why does editing occur at such high levels in hornworts? This simple question belies the complexity of a more fundamental question: What is the nature of editing as a trait? Editing is detected at individual sites, but the number of sites is variable across taxa. We wish to distinguish between the evolution of mechanisms that might increase the propensity to edit across sites (editing propensity) and the evolution of editing at given sites (site corrections). Given this clarification, we discuss the maintenance of editing propensity and its ramifications on the pattern of site correction.

If present, high mutation rates alone could force the maintenance of large numbers of site corrections in hornworts. It is therefore possible that high rates of site correction are the result of selection at each locus in response to an increased probability of mutation across all loci (a genome-wide mutation rate increase), with no increased propensity for correction overall. There are, however, several lines of evidence that indicate that the editing propensity is actually driving change in the rate of site correction.

If high mutation rate in the hornworts has selected for individual site corrections and not for increased propensity, we would expect the higher substitution rates to be reflected in higher divergences among third base positions and among intron sequences which are rarely edited. In fact, we see no such increased number of substitutions in these sequences (see third positions in Table 1). In addition, intron sequences and the chloroplast rpoC and accD–rbcL spacer region show substitution rates similar to those in other land plant groups (Duff, unpublished data). Finally, arguing against direct selection for the maintenance of particular edited sites is the observation that, despite the high frequency of editing in nearly all lineages, edited sites show no less homoplasy than other sites that exhibit nucleotides substitutions. Indeed, although Megaceros flagillaris and Anthoceros formosae both have 20 or more edited sites, they share only 7 of these. Anthoceros punctatus and Notothylas have similar numbers of edited sites but only share six. Therefore, we conclude that the machinery of editing itself is altered, perhaps in its efficiency, such that the editing propensity is altered. This in turn affects the increase in site correction.

If a deleterious mutation exists at certain loci, selection will favor site correction at those loci as long as true reversions have not arisen. Selection in this case is for site correction and not propensity. The total number of edited sites in the genome would thus be determined by the equilibrium between creation of correction sites and loss of correction sites due to reversions. Editing propensity will affect this equilibrium, with taxa that have a high editing propensity having large numbers of edited sites. This does not mean that individuals with a high editing propensity will be selected for, but only that when individuals have a high editing propensity they will have more edited sites. The frequency of reverse mutations (Fig. 1) suggests that the loss of edited sites is still being selected for even after site correction. This implies an underlying mechanism that allows for rapid integration of new edited sites into the genome while at the same time incurring frequent reversals to the nonedited state.

Enhanced propensity for editing may initially arise for several reasons. First, it is possible that some unknown plieotropic effect of increased editing propensity may be beneficial. Second, each new deleterious mutation might renew the high-propensity genotype through successive rounds of selection for individual site corrections. This second explanation would require high mutation loads and would hence be most likely in bottlenecked or chronically inbred populations.

Once site correction is established, maintenance of a strong editing propensity warrants explanation. This is especially true since reversions may eventually replace site corrections, as it is reasonable to expect that “revertant” lineages may be more fit than “edited” lineages. In this case, increased editing propensity may be a correlated selective response to selection for increased editing efficiency. Although some progress has been made on determining the mechanisms underlying editing and propensity (Börner and Pääbo 1996; Maas and Rick 2000; Bock 2000; Chateigner-Boutin and Hanson 2003), information on the possible ties between editing propensity and efficiency is not yet available. An understanding of the relationship between editing propensity and efficiency may ultimately clarify whether or not correlated response to selection on editing efficiency maintains a high propensity for editing.

We summarize the evolution of editing in hornworts as follows: Given the presence of editing among other land plant lineages, the basic apparatus for RNA editing appears to be ancestral in the hornworts. The total number of edited sites in the genome should be determined by the equilibrium between creation of new sites and loss of sites due to reverse mutations. The large number of site corrections in nearly all hornwort lineages combined with a high degree of heterogeneity in which sites are edited indicates that the editing propensity is enhanced in hornworts.

Ramifications of Editing for PhylogeneticReconstruction in the Hornworts

Phylogenetic analyses utilizing DNA sequences frequently incorporate samples that are highly edited. Several studies have looked at the possible effects of these sequences on phylogenetic reconstruction (Bowe and dePamphilis 1996; Malek et al. 1996; Szmidt et al. 2001). These studies generally agree that as long as orthologous sequences are utilized and assumptions of vertical transmission and common ancestry are met, reliable reconstruction of phylogenetic events should be possible. However, it is clear that RNA editing can result in misinformative characters in phylogenetic analyses since taxa may differ significantly in editing rates (Groth-Malenok et al. 2005). Therefore, if editing rates vary significantly among groups, erroneous topologies may result. Exacerbating these problems are potential patterns of selective convergence that might be expected in some lineages with a history of high editing propensity. Lineages that have accumulated a large number of substitutions that are maintained by site correction will be under strong selection for true reversions across such sites if the efficiency of editing is reduced. As a result, those lineages appear to converge on the ancestral type. In short, if multiple lineages experience decreased editing efficiency, they could be falsely inferred to be a sister clade due to convergent selection toward the ancestral genotype.

An examination of Table 1 shows a clear example of sequence heterogeneity between hornworts and most other land plants. Much of the phylogenetic signal that is used to resolve the hornwort clade is derived from the 57 phylogenetically informative edited sites that constitute 27% of the total phylogenetic signal in the hornwort clade. These sites can be expected to be informative only within the hornwort clade, as the majority of these informative edited sites are found at nearly universally conserved locations among all potential outgroup taxa. Thus, these sites can be expected to provide little assistance in resolving hornwort relations to other land plants. In fact, a number of the sites will be shared only with other taxa that may also exhibit independently originated high rates of editing, thus contributing to high levels of homoplasy in phylogenetic studies.

Ramifications for the Interpretation of Leiosporoceros

The lack of editing of the rbcL sequence of Leiosporoceros is very surprising given the presence of editing in this gene among the other hornworts and other land plant groups. Furthermore, absence of editing is predicted in partial sequences of Leiosporoceros from the protein-coding gene rpoC and demonstrated in the entire atpB gene (Duff, unpublished). However, we believe it is highly unlikely that the chloroplast genome lacks the mechanism for editing completely. The ability to edit is present in this sample. Both C-to-U and U-to-C editing have been observed in the mitochondrial gene nad5 of Leiosporoceros (Duff, unpublished). Although present, the frequency of editing is more than fourfold lower than observed in representatives of the other hornwort genera. Hence, the apparatus for editing is apparently present in Leiosporoceros but the extent of editing may be highly reduced in both its mitochondrial and its chloroplast genomes.

Is this observed lack, or low level, of editing in Leiosporoceros an ancestral or a derived feature? If Leiosporoceros represents a sister lineage to all other hornworts, then we may place the putative evolutionary event(s) that led to the increased number of edited sites after the divergence of Leiosporoceros from the other hornworts. A lower propensity for editing in the mitochondrial genome and likely very rare editing in the chloroplast genome of this genus may simply reflect the ancestral editing efficiency as it functions in most mosses and leafy liverworts. Alternatively, if Leiosporoceros is nested within the hornwort clade, then the observed level of editing may have resulted from a reduction in the efficiency of its editing mechanism with a subsequent loss of edited sites, as reverse mutations will experience higher selection pressure.

To distinguish between these two alternative histories it is necessary to have a good handle on outgroup relationships, the extent of editing in these outgroups when protein-coding genes are being used for phylogenetic analyses. Unfortunately, the relationships of the major bryophyte lineages are still controversial (Palmer et al. 2004; Shaw and Renzaglia 2004). Likewise, the distribution of editing sites across land plant lineages, genes, and organelles has received little attention. Within the hornworts, prior phylogenetic analyses using rbcL including Leiosporoceros have placed this genus as a sister taxon to the hornworts (Duff et al. 2004), though the effects of RNA editing sites were not considered in the analysis. The phylogenetic analyses presented here do not definitively resolve the position of Leiosporoceros, as the inclusion of edited sites may lend false support to a Leiosporoceros basal topology. Furthermore, analyses of nuclear and mitochondrial SSU ribosomal DNA sequences, trnL-F chloroplast spacer/intron sequences (Dietmar, personal communication), and an intron found only in the mitochondrial nad5 gene and shared only among Leiosporoceros, Anthoceros, and Folioceros (Duff, unpublished) all support Leiosporoceros as part of an Anthoceros s. lat. clade. If this phylogeny is correct, the most parsimonious explanation for the lack of editing in Leiosporoceros would be the loss of common ancestral hornwort editing sites.

Unlike prior studies that examined the effects of using edited nucleotides in phylogenetic reconstruction (Bowe and dePamphilis 1996; Szmidt et al. 2001), the present study examines a data set that exhibits much higher levels of editing. In addition, editing rates vary greatly among taxa when Leiosporoceros and the outgroups are considered. Without knowing the likelihood of the losses and gains of editing sites and the presence of diverse rates of editing, the present data set cannot resolve the position of Leiosporoceros. Final resolution of the position of Leiosporoceros will require examination of additional protein-coding genes and rDNA genes that are not affected by editing. Once the position of Leiosporoceros can be definitively identified, we will be better able to assess if lower observed editing is ancestral or, possibly, a derived condition in this genus.

Comparative Approach to Studying the Evolution of RNA Editing

The presence of high rates of editing across a number of ancient lineages, along with a well-resolved phylogeny within the hornworts, presents an opportunity to elucidate the evolution of RNA editing in plants. It is currently unclear if the presence of RNA editing in both mitochondria and chloroplasts is owed to a single or a multiple origin early in land plant evolution. The absence of editing in both organelles of Marchantia, along with a consistent pattern of editing among other land plant organelles, has been suggested as evidence of a mechanistic connection and origin of editing in land plants (Bock 2000). Coordinated changes in editing patterns between organelles of divergent lineages of plants may provide further support for such a conjecture. Interestingly, hornworts exhibit not only an increase in overall numbers of edited sites but a disproportionate increase in the number of the rarer U-to-C conversions. Did this increase in U-to-C editing result from an evolutionary event separate from that which resulted in the increased propensity for editing in general, or are the two causally connected? A comparison to mitochondrial genes suggests a tight link not only between the evolution of increased editing but also between increased editing in general and the additional mechanism to cause U-to-C editing. Ongoing sequencing of both mitochondrial and chloroplast protein and ribosomal genes of hornworts will allow greater insights into the evolutionary forces at play in maintaining this enigmatic process.