The Arabidopsis genome information and the GO annotations were retrieved from TAIR (The Arabidopsis Information Resource; release 9, ftp://ftp.arabidopsis.org/home/tair/) 11. The miRNA information was retrieved from miRBase (release 14, http://www.mirbase.org/cgi-bin/mirna_summary.pl?org=ath) 12. The MPSS and PARE data were retrieved from the MPSS plus database (http://mpss.udel.edu/at/) and the PARE database (http://mpss.udel.edu/at_pare/), respectively 1314.

We retrieved MPSS sequences from 17 different libraries with normalized expression data (TPM, transcripts per million). The editing ratio for each editing site was defined as the expression value of all edited reads divided by that of the total reads surrounding the editing site. The single-base sequencing error rate of MPSS was estimated to be ~5.00% 1516. Thus, the average single-base sequencing error rate for each error pattern (12 patterns in all) is ~0.42%. To reduce the interference by sequencing errors, only the sites with editing ratios more than 2% in either library were clustered by using Cluster 3.0 17. Although the cutoff is arbitrary, the higher percentage surely reflects the higher editing efficiency in planta, and the possibility that the editing site may be a feint one generated by sequencing errors can be greatly reduced. The clustering results were visualized by using Treeview 18.

WebLogo 19 (http://weblogo.berkeley.edu/logo.cgi) was used for sequence conservation analysis. GO::TermFinder 20 was used for GO term enrichment analysis. RNAfold 21 (http://rna.tbi.univie.ac.at/cgi-bin/RNAfold.cgi) was used for pre-miRNA secondary structure prediction. miRU 22 (http://bioinfo3.noble.org/miRNA/miRU.htm) was used for miRNA target prediction.

Fahlman and colleagues revealed ubiquitous RNA modifications in plant tRNAs and miRNAs 10. However, no research has been carried out to elucidate if RNA editing occurs in nuclear protein-coding transcripts in plants. Here, we focus on one-base conversion in nuclear protein-coding transcripts and pre-miRNAs in Arabidopsis. The MPSS sequences derived from polyadenylation (poly(A))-tailed transcripts were retrieved from the MPSS plus database 13, and the PARE sequences from the 5' ends of miRNA-mediated poly(A)-tailed mRNA decays were retrieved from the PARE database 14. As nearly all the protein-coding and miRNA genes are transcribed by RNA polymerase II, resulting in poly(A)-tailed transcripts 2324, the MPSS and PARE data are applicable for this study.

All the short reads were mapped to the pre-miRNAs and the mRNAs of all the protein-coding genes including mitochondrial and chloroplast genes in Arabidopsis. The perfectly matched sequences were removed and the remaining reads were utilized to search for one-base-conversion editing sites. In light of the technological sequencing errors of MPSS and PARE, the protein-coding transcripts and the pre-miRNAs were considered to be edited based on the following criteria as a measure of caution: For each protein-coding transcript, more than two candidate editing sites should be detected and each editing site must be supported by more than five distinct short reads. For each pre-miRNA, the editing site should be supported by more than two distinct short reads. It was estimated that the single-base sequencing error rates were ~5.00% (20-nucleotide (nt) signatures) or ~4.25% (17-nt ones) for MPSS sequencing platform 1516, and 1.30 ± 0.90% for PARE sequencing 25. That is, the average sequencing error rates of each error pattern (12 in all) are ~0.42% (20 nt) or ~0.35% (17 nt) for MPSS, and ~0.03—0.18% for PARE. To further assess the reliability of our prediction criteria, the ratio of edited signatures to total signatures including non-edited ones surrounding each editing site was calculated. The ratios range from 12.50% to 100%, and the average ratios are 21.75% for the protein-coding transcripts and 42.05% for the pre-miRNAs (Additional Files 1 and 2). It indicates that a sizable portion of the predicted editing sites are not feint ones generated by sequencing errors.

The result indicates that all 12 RNA editing patterns may exist in the nuclear transcripts, although the number of editing sites in a specific pattern varies widely (Fig. 1A). Previous reports demonstrated that C-to-U conversion was the dominant editing pattern of mitochondrial and plastid transcripts 23. Consistently, our study shows that C-to-U conversion is the exclusive editing pattern in mitochondrial transcripts (Fig. 1A). From another perspective, it reflects that our search criteria are quite reliable, especially in excluding false positive. However, C-to-U editing is not the dominant pattern in the nuclear transcripts analyzed. Instead, U-to-C, A-to-G, G-to-U, and A-to-C are the dominant ones in the nuclear protein-coding transcripts, and U-to-C and G-to-A in the pre-miRNAs (Fig. 1A). A-to-I editing (A-to-G, recognized by sequencing) mediated by ADAR has been extensively characterized in mammals (see reviews in 626), whereas no such editing has been recognized in plants. Our results show that A-to-I editing is likely to be existed in plant nuclear transcripts. However, the ADAR homolog has not been identified in plants yet. Hence, this study will inspire further research to understand the intriguing mechanisms of this peculiar RNA editing pattern in plant nuclear transcripts. Taken together, our preliminary observation (Data S1 and S2) is a valuable repository for further studies on RNA editing in plant nuclear transcripts.

The 100-nt sequences (Additional File 3: Data S3) surrounding the editing sites (50-nt sequences both upstream and downstream) with specific patterns of nuclear protein-coding genes were submitted to WebLogo 19 for sequence conservation analysis. Conserved elements were detected surrounding the editing sites with certain editing patterns, such as G-to-U and C-to-G. The conserved elements surrounding A-to-U editing sites are quite interesting that the nearer positions, relative to the editing sites, show higher occurring frequency of A (Fig. 1B). However, no obvious sequence conservation was observed surrounding the C-to-U editing sites in the nuclear transcripts, although short conserved elements were present in the mitochondrial transcripts (Fig. 1B and Additional File 4: Data S4). Previous research suggested that a particular cis-element surrounding the editing site was required for the recognition by PPR-associated editing enzyme in plants 3. Our result shows that besides C-to-U editing in mitochondrial and plastid transcripts, other editing with potential conserved cis-elements surrounding the editing sites may exist in nuclear transcripts. To better understand the mechanisms implicated in various RNA editing processes, delicate experiments are needed for cis-element identification, editing enzyme isolation, and editing site validation.

For each editing pattern, all the edited protein-coding transcripts compared with whole-genome protein-coding ones were subjected to GO term enrichment analysis. Interestingly, for nearly all the editing patterns, the functionalities of the edited genes are highly enriched in photosynthesis, light response, or energy metabolism (Fig. 1C and Additional File 5: Data S5). Although a number of mitochondrial and chloroplast transcripts have been reported to be edited in plants 23, it is surprising that the nuclear transcripts, encoding proteins involved in chloroplast- or mitochondrion-related functions, are more susceptible to RNA editing.

We utilized MPSS data from 17 different libraries to investigate the tissue-specific editing patterns. The expression data of each library was normalized to enable cross-library comparison. For each editing site, the ratio of the expression value of all the edited reads to that of the total reads was calculated which represents the editing efficiency. To reduce the interference by sequencing errors, only the editing sites with editing ratios more than 2% in each library were clustered. The clustering result shows that tissue-specific RNA editing, such as in agamous inflorescence, callus, and silique, has been observed in a portion of transcripts (Fig. 1D). The MPSS sequences are composed of 17-nt and 20-nt ones, so we analyzed the two portions separately and tissue-specific editing was still observed in both cases (Additional File 6: Fig. S1).

A number of transcripts were subjected to site-specific editing. For AT1G29930.1 and AT1G52400.1, both the C-to-U and the U-to-C editing are highly enriched surrounding the boundaries between the CDSs and the 3’ UTRs, which are also known as translation borders (Fig. 2A and B). Moreover, the C-to-U and the U-to-C editing sites come together, indicating that an amino-group, dissociated from C which further converts to U, could be integrated with the neighboring U that subsequently converts to C. For AT2G21660.1 and AT2G21660.2, A-to-G editing sites are also highly enriched surrounding the translation boundaries (Fig. 2C). The biological means of these site-specific editing events should be further investigated.

RNA editing resulted in generation of new start or stop codons has been reported in both humans and plants 2728 (also see reviews in 629). In this study, a systemic search was performed to identify novel start or stop codons generated by RNA editing in nuclear CDSs. In summary, new start codons are generated predominantly by C-to-U and G-to-U editing, and novel stop codons by G-to-U, A-to-U, and C-to-A editing (Table 1 and Additional File 7: Data S6). These types of editing may produce premature proteins or even new functional ones.

It was reported that certain elements within exons and introns of eukaryotic genes were essential for the splicing of their transcripts, and RNA editing has great potential to affect RNA splicing 6293031. Because all the MPSS and PARE reads were derived from poly(A)-tailed mature mRNAs 1314, we investigated the RNA editing within the 5’ first and the 3’ last three nucleotides of each exon, both of which will potentially affect RNA alternative splicing. Although only a small portion of nuclear transcripts were found to be edited at either ends of their exons (Additional File 1: Data S1), it suggested that alternative splicing converting pre-mRNAs to mRNAs might be influenced by RNA editing in Arabidopsis.

Previous research showed that various types of RNA editing occurred in plant tRNAs and mature miRNAs 10. However, the scene of RNA editing in pre-miRNAs, which may result in secondary structure transformation, has never been unveiled. We searched for potential editing sites in pre-miRNAs (Table 1 and Additional File 2: Data S2) and some interesting secondary structure transformations of edited pre-miRNAs were observed. All the pre-miRNAs of ath-miR854 family were found to be edited in several sites. Taking ath-MIR854c for example, the secondary structure has markedly changed after editing. Notably, a mini stem-loop structure near the main stem region, generating the mature miRNA through Dicer-like 1 (DCL1) cleavage, has disappeared after editing (Fig. 2D). The other three members of ath-miR854 family were also investigated, and the similar results were obtained (Additional File 8: Fig. S2). Thus, we postulate that the edited versions of ath-miR854 family members may be much more efficient for mature miRNA production, considering more accessible structures near the main stem regions for DCL1. Another intriguing observation is that the only pre-miRNA in clustering analysis, ath-MIR161 (MI0000193), is subjected to leaf-specific editing (Fig. 1D and Additional File 6: Fig. S1). Moreover, the mature miRNAs ath-miR161.1 and ath-miR161.2 target transcripts belonging to PPR family based on our prediction results generating by miRU. On the other hand, C-to-U editing in mitochondrial and plastid transcripts was reported to be mediated by PPR family proteins 23. Since a few reports has pointed to the involvement of RNA editing in the maturation of miRNAs in metazoans 3233, our preliminary observations deserve experimental exploration in plants.