Genome-wide A-to-I RNA editing in fungi independent of ADAR enzymes

1RNA editing is a post-transcriptional event that recodes hereditary information by changing the nucleotide sequence of RNA molecules. Although various RNA editing systems that target major types of cellular RNA have been identified in eukaryotes, only two types of RNA editing are known to cause changes in nuclear-encoded messenger RNA (mRNA). One is the cytidine-to-uridine (C-to-U) RNA editing that is rare in mammals, such as the well-characterized target of APOB mRNA in human intestine. The other is the adenosine-to-inosine (A-to-I) editing, the most prevalent type of RNA editing known in the animal kingdom.

2A-to-I RNA editing occurs when A residues are converted to I residues via an enzymatic deamination reaction. Whereas editing of mRNA is mediated by the adenosine deaminase acting on RNA (ADAR) family of enzymes that convert A to I in double-stranded RNA (dsRNA) substrates, the adenosine deaminase acting on tRNA (ADAT) enzymes act on transfer RNAs (tRNAs). Because I is subsequently recognized as guanosine (G) by the translation machinery, A-to-I substitution in coding regions (CDSs) of mRNA may lead to codon changes and alter functional properties of proteins. However, editing events resulting in protein sequence recoding are not common in animals. To date, the majority of studies on the functional consequences of recoding editing are related to genes important for animal nervous systems. A-to-I editing of ligand- and voltage-gated ion channels and neurotransmitter receptors in invertebrates and vertebrates are among the best-studied ADAR-mediated recoding events. More recently, A-to-I editing of K+ channels responding to temperature adaptation also has been reported in octopuses. Despite these findings, the biological functions of mRNA editing in animals remains poorly understood because the majority of A-to-I editing events occur in the noncoding regions.

3By RNA-seq analyses, a large number of A-to-I RNA editing events have been identified in the transcriptomes of humans and other animals. In general, RNA editing occurs more frequently in noncoding than in coding regions. In humans, more than 1.4 million A-to-I editing sites have been identified. The vast majority (∼97%) of them target repetitive sequences located within introns and 5′- or 3′-untranslated regions (UTRs). Only a small fraction of these editing sites (approximately 100) occur in the coding regions and result in amino acid changes. Similar observations have been reported in other animals. Although most editing sites do not affect the primary sequences of coding proteins, only approximately 50 recoding events in mice, 645 in Drosophila, and eight in Caenorhabditis elegans have been identified. One remarkable exception among animals that have been studied for A-to-I editing is squid, in which a total of 57,108 recoding RNA editing sites were identified in its nervous system, raising the question about the prevalence of recoding RNA editing events in other animal lineages.

4ADATs and A-to-I editing on tRNA are found in all domains of life. To date, however, ADARs and A-to-I editing of mRNA are only found in metazoans and is thought to be a metazoan innovation. Yeasts and filamentous fungi do not have ADAR orthologs in their genomes and are believed to lack A-to-I mRNA editing. However, during this study with a protein kinase named Puk1 (Perithecium unique kinase 1) (gene ID: FGRRES_01058, FG01058.1, or FGSG_01058), we found that the two tandem stop codons, UAG UAG, in its ORF were changed to UGG UGG by RNA editing in Fusarium graminearum, a causal agent of Fusarium head blight (FHB) of wheat and barley. F. graminearum is a haploid homothallic ascomycete that overwinters in infected plant debris as saprophytic hyphae and produces perithecia and ascospores in the spring. Ascospores are forcibly discharged and dispersed by wind as the primary inoculum for FHB. We then conducted RNA-seq analysis to confirm the editing events of PUK1 transcripts during sexual reproduction and identify genome-wide A-to-I editing events in F. graminearum. In addition, to analyze characteristics of mRNA editing sites in F. graminearum, we also identified A-to-I RNA editing events in Neurospora crassa and F. verticillioides. Furthermore, we attempted to experimentally characterize biological functions of A-to-I RNA editing in late stages of sexual reproduction and identify responsible deaminase genes in F. graminearum.

5The Puk1 protein kinase is conserved in filamentous fungi but lacks a distinct ortholog in the budding and fission yeasts. Two puk1 deletion mutants generated in our previous study of the F. graminearum kinome were confirmed by Southern blot analysis in this study. All the puk1 mutants were normal in growth, conidiation, and plant infection but were defective in late stages of sexual reproduction. Close examinations showed that perithecia formed by the mutant were normal in size and morphology but defective in ascospore release and cirrhus formation (Fig. 1A). Whereas normal ascospores were four-celled, puk1 ascospores were often single- or two-celled and spherical or fragmented (Fig. 1B). When assayed by qRT-PCR, PUK1 transcription was barely detectable in conidia and vegetative hyphae but significantly up-regulated in perithecia collected at 8 d post-fertilization (dpf) (Fig. 1C). These results suggest that PUK1 has stage-specific expression and functions during sexual reproduction.

6To identify Puk1-interacting proteins in F. graminearum by yeast two-hybrid assays, we generated the bait construct of PUK1 with its ORF amplified from the first-strand cDNA synthesized with RNA isolated from mating cultures of the wild-type strain PH-1 harvested at 8 dpf. To our surprise, sequencing analysis showed that all the Puk1 bait constructs contained the third predicted intron, which is in the ninth subdomain of the kinase domain (Fig. 1D). Interestingly, the sequence of this putative intron in the Puk1 bait construct differed from its genomic DNA at nucleotides 1831 and 1834 (Fig. 1D). At both positions, A is in the genome sequence but G is in the yeast two-hybrid bait construct.

7To verify this observation, we sequenced the PUK1 fragments amplified from genomic DNA and cDNA. Whereas A1831 and A1834 were in the PCR products amplified from genomic DNA, G1831 and G1834 were in cDNA fragments, indicating that the third intron of PUK1 is incorrectly predicted and these two nucleotides may be subjected to A-to-I editing in F. graminearum. Although it has not been reported in fungi, A-to-I RNA editing is a well-known phenomenon in animals.

8We then conducted RNA-seq analysis with RNA isolated from conidia, 24-h hyphae, and 8-dpf perithecia of PH-1. For each fungal tissue, strand-specific RNA-seq data were generated by Illumina sequencing with 150-bp paired-end reads for two independent biological replicates. For each tissue, approximately 70 million high quality reads were obtained (Supplemental Table 1).

9Only rare PUK1 transcripts were present in RNA-seq data of conidia and hyphae (fragments per kilobases of exons per million mapped reads [FPKM] <2) and none of them had G1831 or G1834 (Supplemental Fig. 1). For the PUK1 reads present in RNA-seq data of perithecia, >99% of them had either G1831 or G1834. Of these, 90% had both G1831 and G1834 (Fig. 1D). These results indicate that A-to-I RNA editing of PUK1 transcripts occurred at a high frequency in perithecia.

10To determine the effects of A-to-I editing on PUK1 expression and function, we generated the PUK1TGGTGG allele with the TA1831G TA1834G sequence changed to TG1831G TG1834G and the PUK1TAATAA allele with two tandem stop codons TAA TAA added behind the TA1831G TA1834G sequence. Both mutant alleles were transformed into the puk1 deletion mutant. In the resulting transformants, both PUK1TGGTGG and PUK1TAATAA, similar to the wild-type PUK1 allele, had relatively low expression levels in vegetative hyphae (Supplemental Fig. 2A). In comparison with PUK1WT, the expression level of PUK1TGGTGG was increased ∼1.7-fold, whereas PUK1TAATAA expression was slightly reduced (Supplemental Fig. 2A). These results indicate that the promoter of PUK1 but not the editing events plays a major role in its up-regulated expression during sexual reproduction. Because the editing sites TG1831G TG1834G are in the conserved protein kinase domains, it is not surprising that expression of PUK1TAATAA failed to complement the defects of the puk1 mutant in ascospore morphology and release (Supplemental Fig. 2B,C). Therefore, A-to-I editing is essential for PUK1 to be functional during sexual reproduction.

11We then performed a genome-wide identification of A-to-I RNA editing events in F. graminearum. In comparison with genomic sequences, 27,301 single-nucleotide variant (SNV) sites were identified in the combined RNA-seq data of two independent biological replicates of 8-dpf perithecia (Fig. 2A; Supplemental Table 2). Strikingly, 26,056 of these SNVs (>95%) correspond to the A-to-G transition, which is consistent with A-to-I editing. Assuming that non-A-to-G variants (1245) are false positives and the error rate for all replacement types is equal, the estimated false-discovery rate in our A-to-G identification is only 0.43%, which is 10-fold lower than the error rate of A-to-I editing analysis in humans and animals. These results showed that genome-wide A-to-I editing (A-to-G SNV) occurs in F. graminearum.

12We also conducted a similar analysis with RNA-seq data of conidia and hyphae. In total, only 68 and 112 A-to-G SNV sites were identified in conidia and hyphae, respectively (Supplemental Fig. 3; Supplemental Table 2). In comparison with 335 and 452 non-A-to-G SNV sites, no enrichment for A-to-G SNV sites was observed in RNA-seq data of conidia and hyphae, indicating that A-to-I editing specifically occurs during sexual reproduction.

13We then examined the RNA editing level for each A-to-I editing site as the percentage of reads with the A-to-G variant among all the reads covering that site. Similar to what have been reported in humans, rhesus macaque, and flies, the A-to-I editing level varied from 3% to 100% among different editing sites, with the median level of 14.8% in F. graminearum (Fig. 2B). Among all 26,056 A-to-G sites in perithecia, 75.9% had editing levels <30%, whereas only 5.8% had editing levels >60%. However, we noticed that genes with editing sites of higher editing levels were enriched for genes likely important for sexual reproduction (Fig. 2C). Among the 10,652 genes expressed in perithecia, 971 (9.1%) were specifically expressed, and 1256 (11.8%) were up-regulated at least twofold during sexual reproduction in comparison with conidia and hyphae. Approximately 47% of the genes with editing sites of >60% A-to-I editing levels were either specifically expressed or up-regulated at least twofold in perithecia compared to conidia and hyphae. In contrast, only 14% of the genes with editing sites of editing levels <30% had increased expression levels during sexual reproduction (Fig. 2C). These results suggest that A-to-I editing occurs more efficiently in genes that display stage-specific or unregulated expression in perithecia in F. graminearum. Nevertheless, the median FPKM values for genes with editing sites of <30%, 30%–60%, and >60% editing levels are 31.5, 25.5, and 27.5, respectively. Therefore, there is no general correlation between expression levels and editing levels in F. graminearum.

14In humans and Drosophila, A-to-I editing primarily occurs in introns and UTRs. When the distribution of 27,301 A-to-I editing sites was analyzed, to our surprise, 21,095 of them (70.5%) are in the CDSs in F. graminearum (Fig. 2D). Only 0.9%, 12.9%, and 15.7% of A-to-I editing sites were in intronic, UTR, and intergenic regions, respectively. Therefore, unlike in metazoans, A-to-I editing may preferentially occur in the CDSs in fungi. Nevertheless, A-to-I editing sites appear to occur more frequently in 3′-UTRs (8.6%) than in 5′-UTRs (4.3%) in F. graminearum, which is similar to observations in humans and Drosophila.

15Among the 21,095 A-to-I editing sites in CDSs, 78.9% (16,649) of them are missense editing events that result in amino acid substitutions (Fig. 3A). In contrast, only 18.7% are synonymous editing. In total, 4594 of 10,652 (43%) protein coding genes expressed in perithecium samples (more than 10 counts per million) have at least one missense editing event (Fig. 3B). Almost half of them (2128) harbor three or more recoding sites. Transcripts of 349 genes (8%) contain at least 10 recoding sites. Two genes with the most abundant recoding events are FGRRES_06089 and FGRRES_08133 that have 38 and 37 editing sites, respectively. FGRRES_08133 is an essential gene in F. graminearum. It encodes a 2395-aa protein kinase orthologous to the Tor1 and Tor2 of the budding yeast that are involved in regulating various biochemical and cellular processes, including transcription, translation, autophagy, and meiosis. FGRRES_06089 encodes a 3880-aa protein that is orthologous to Tra1, a subunit of the yeast SAGA and NuA4 histone acetyltransferase complexes that are important for gene regulation and DNA repair. Interestingly, Tra1, Tor1, and Tor2 all are members of the PIKK (phosphoinositide three-kinase-related kinase) family.

16Furthermore, the editing level of missense editing sites (median, 15.4%) is statistically significantly higher than that of synonymous editing sites (median, 12.8%) (Fig. 3C). Notably, 76.2% (12,694) of the missense editing sites resulted in changes to amino acid residues of different physicochemical properties, including 22% lysine (K) (basic) to glutamate (E) (acidic), 14% asparagine (N) (nonacidic) to aspartate (D) (acidic), and 11% serine (S) (polar) to G (nonpolar) changes (Fig. 3D). These observations suggest that A-to-I editing may be important for adaptation and diversification protein functions during sexual reproduction in F. graminearum.

17Interestingly, none of the A-to-I editing events resulted in changes from amino acid codons to stop codons (nonsense change), suggesting that the editing mechanism avoids generation of truncated proteins in F. graminearum. However, the stop codons of 503 genes were edited (Fig. 3E). Among these, 323 had the stop-loss events, in which UAG was changed to the UGG tryptophan codon, resulting in the addition of an extra stretch of amino acid residues to the C-terminal end of predicted protein sequences. Although the biological significance of stop-loss editing is not clear, transcripts of 11 transcriptional factor and five protein kinase genes known to be important or essential for sexual reproduction in F. graminearum are among the genes with stop-loss A-to-I editing (Supplemental Table 3). Moreover, the editing level of stop-loss events is significantly higher than that of other editing events (Fig. 3C), suggesting selection toward read-through of these proteins.

18For the other 180 editing events that occurred at stop codons, 175 had the UAA to UGA change and five had the UAA to UAG change. In general, UAA is the most efficient stop codon, whereas UGA is a relative leaky stop codon that sometimes causes read-through in bacteria, fungi, and mammals. Therefore, even for these edited transcripts with stop codons retained, A-to-I editing likely increases the read-through frequency.

19To search for genes with similar editing events to PUK1, we manually examined editing sites that resided in the predicted introns. Sixty-nine of them had the UAG to UGG change in exons that were erroneously annotated as intronic sequences by automated annotation based on our RNA-seq data (Supplemental Table 4). Therefore, PUK1 is not the only gene in F. graminearum with stop codons in the coding regions that are subjected to stage-specific RNA editing but avoided by automated annotation with incorrectly predicted introns.

20Among these 69 genes with PUK1-like editing events, 54 of them (78%) were specifically expressed or up-regulated in perithecia, including the rid (RIP defective) ortholog (FGRRES_08648) and genes encoding putative Dcp1-like mRNA-decapping enzyme (FGRRES_02091), Rho GTPase (FGRRES_01649), and Rho-GAP protein (FGRRES_08999). However, a majority of them (Supplemental Table 4) have no known functional protein domains, and their functions during sexual reproduction remain to be identified. Nevertheless, for two of them, FGRRES_10728 and FGRRES_01563, their orthologs in the budding yeast are important for meiosis. Although FGRRES_10728 is orthologous to yeast AMA1 that encodes an activator of meiotic anaphase promoting complex, FGRRES_01563 is an ortholog of yeast SPO22 that encodes a meiosis-specific protein essential for chromosome synapsis. It is likely that A-to-I RNA editing of these two genes during sexual reproduction is important for meiosis in F. graminearum.

21We then selected five of them, FGRRES_01649, FGRRES_ 10094, FGRRES_08389, FGRRES_12623_M, and FGRRES_14031_ M (Supplemental Table 4), that encode hypothetical proteins in F. graminearum for functional characterization and identified at least three knockout mutants for each gene. None of the deletion mutants of FGRRES_08389, FGRRES_12623_M, and FGRRES_ 14031_M had obvious defects in growth and sexual reproduction (Supplemental Fig. 4). Deletion of FGRRES_10094 had no effects on hyphal growth, perithecium development, and ascospore formation. However, the FGRRES_10094 deletion mutant was defective in ascospore release, and the ascus wall became dissolved in 12-dpf perithecia. When 12-dpf perithecia were cracked open, only aggregates of ascospores were observed in the mutant, but fascicles of asci were observed in the wild type (Supplemental Fig. 5). Yellowish ascospore cirrhi were rarely observed, and massive discharging of ascospores from perithecia was not observed in the FGRRES_10094 deletion mutant (Supplemental Fig. 5). Deletion of FGRRES_01649 that encodes a putative Rho GTPase resulted in defects in ascospore formation and release. Perithecia formed by the FGRRES_01649 mutant were normal in morphology but produced and released fewer mature ascospores (Supplemental Fig. 5). These results showed that both FGRRES_10094 and FGRRES_01649 play roles in the late stages of sexual development.

22When the flanking sequences of all the A-to-I editing sites identified in this study were analyzed, the 5′ − 1 site preference is U (85.74%) > A (8.03%) > G (4.98%) > C (1.25%). Although the overall trend is similar to that of human ADAR (also known as ADAR1) and ADARB1 (also known as ADAR2) (U > A > C > G), the significant enrichment of U at the −1 position is unique to F. graminearum (Fig. 4A). The 3′ +1 site preference is A (39.16%) ≈ G (36.92%) > U (17.00%) > C (6.93%), which is different from that of human ADAR (G > C ≈ A > U) or ADARB1 (G > C > U ≈ A). In F. graminearum, A and G are enriched at the +1 position of edited A sites (Fig. 4A).

23When the percentage of edited triplets was estimated for all 16 possible triplets centered on edited A's (NAN), we found that almost 9% of UAG or UAA have the A-to-I editing events (Fig. 4B), making these two preferred triplet sequences for RNA editing in F. graminearum. In contrast, triplets such as GAC, AAC, GAU, and CAC are rarely edited (Fig. 4B). Although the editing preference of UAG triplets has been reported in studies with metazoan ADARs, the preference for UAA triplets appears to be unique to F. graminearum. Furthermore, we noticed that the preferred triplets tend to have higher editing levels (Fig. 4C). Indeed, the median editing level for the non-UAN triplets is only ∼10% but is >16% for the UAG and UAA triplets (Fig. 4C).

24In addition, we observed a weak base preference at the positions beyond the nearest neighboring nucleotides (Fig. 4A). Whereas the −2 and −3 positions are slightly enriched for U and G, respectively, the +2 position is enriched for G. An overrepresentation of A and G at both the +3 and +4 positions also observed (Fig. 4A). These findings suggest that A-to-I editing is influenced by more than the −1 and +1 nucleotides of the edited A's, particularly by the downstream nucleotides.

25Because RNA structure is known to affect editing specificity, we predicted the secondary structures of RNA sequences containing the A-to-I editing sites identified in this study. Interestingly, 62.4% of the editing sites were in hairpin loops (Fig. 4D). Furthermore, the percentage of editing sites in hairpin loops appears to increase for those with higher editing levels. For edited A sites with 60% or higher editing levels, >70% of them are in hairpin loops (Fig. 4D). These results suggest that A-to-I editing preferentially occurs to A's in hairpin loops in F. graminearum, which is different from preferred editing sites in RNA stems in animals.

26In addition, we found that the minimum free energy (MFE) of predicted hairpin loops with A-to-I editing sites is significantly lower than that of random controls (Fig. 4E). Furthermore, sites with 60% or higher editing levels tend to be in hairpin loops with lower MFEs than those with <30% editing levels (Fig. 4E). These results indicate that RNA editing selectively targets structurally stable hairpin loops, and the stability of hairpin loops affects editing efficiency in F. graminearum.

27The F. graminearum genome contains no predicted genes that encode proteins with both adenine deaminase domain and dsRNA binding domain, the hallmark of ADARs. In fact, none of the fungal species that have been sequenced have putative ADAR genes. Therefore, like yeasts and other filamentous fungi, F. graminearum lacks distinct ADAR orthologs and may involve other proteins for A-to-I editing.

28A-to-I editing in F. graminearum preferentially targets A's in hairpin loops, which is similar to the anticodon loop of tRNA targeted by ADATs, implying a potential evolutionary link between mRNA editing and ADATs in fungi. The F. graminearum genome has three predicted ADAT genes, FGRRES_16992, FGRRES_ 11590, and FGRRES_01444, that are orthologous to yeast TAD1, TAD2, and TAD3, respectively (Fig. 5A). Although FgTad1 contains only an adenosine-deaminase domain, both FgTad2 and FgTad3 have a cytidine/deoxycytidylate deaminase domain (Fig. 5A). Based on our RNA-seq data, none of these three ADAT genes were specifically expressed during sexual reproduction. The expression level of FgTAD1 was relatively low under all the conditions assayed, but FgTAD2 and FgTAD2 had higher expression levels in hyphae and perithecia than in conidia (Fig. 5B), suggesting that FgTAD2 and/or FgTAD3 may play a more important role in A-to-I editing in perithecia in F. graminearum.

29To determine their functions in mRNA editing, we attempted to generate ADAT knockout mutants by targeted gene replacement. Five Fgtad1 deletion mutants were identified in 10 hygromycin-resistant transformants that were screened, which is consistent with earlier publications of high gene replacement efficiency in F. graminearum. Nevertheless, we failed to identify putative Fgtad2 or Fgtad3 mutants after screening hundreds of transformants generated in at least three different transformations, suggesting that deletion of FgTAD2 or FgTAD2 may be lethal. For the five Fgtad1 mutants, none had any obvious defects in perithecium formation and ascosporogenesis or ascospore release (Fig. 5C,D). When assayed by RT-PCR and sequencing analysis of transcripts of the FgSSN3 gene (FGRRES_04484), A-to-I editing events still occurred in perithecia of the Fgtad1 mutants (Fig. 5E). These results suggest that FgTAD1 is dispensable for A-to-I editing during sexual reproduction.

30PUK1 is a protein kinase gene conserved in N. crassa and other Sordariomycetes. The predicted gene NCU03242 of N. crassa has a UAG stop codon in the kinase domain that is equivalent to the first of two tandem stop codons in the PUK1 ORF (Fig. 6A). Comparative analysis of published RNA-seq data revealed that the A-to-I editing event at this site (UAGUGG to UGGUGG) also occurs in the transcripts of NCU03242 in perithecia (Fig. 6A). Similar to PUK1, the transcription level of NCU03242 was relatively low, and A-to-I editing events were not observed in vegetative hyphae and other developmental stages in N. crassa.

31Interestingly, the PUK1 ortholog in F. verticillioides (FVEG_ 01191) has a stop codon that is equivalent to the second of two tandem stop codons in its ORF. Based on published RNA-seq data, A-to-I editing also occurs at this site and results in the UGGUAG to UGGUGG change in FVEG_ 01191 transcripts in perithecia but not in hyphae (Fig. 6B). These results indicate that stage-specific A-to-I editing also occurs in N. crassa and F. verticillioides during sexual reproduction.

32We then analyzed genome-wide A-to-I editing sites in F. verticillioides with published RNA-seq data (Supplemental Table 1; ). In F. verticillioides, a total of 7659 A-to-I editing sites were identified in RNA-seq data of perithecia with an estimated false discovery rate of 5.7% (Fig. 6C). Among them, 1685 (22%) are conserved in F. graminearum. No obvious enrichment for A-to-G variants was observed in the RNA-seq data of hyphae (Supplemental Fig. 6). Therefore, stage-specific A-to-I RNA editing occurs during sexual reproduction in F. verticillioides and possibly many other Sordariomycetes. In comparison with that of F. graminearum, the number of A-to-I editing sites identified in F. verticillioides is relatively small, which is due to the limitation of the RNA-seq data that are currently available and more stringent filters used during our analysis for the latter (see Supplemental Methods).

33Furthermore, for the other 69 F. graminearum genes with PUK1-like editing events, 47 have orthologs in F. verticillioides (Supplemental Table 4). Among them, 34 (72%) also had the same editing events that resulted in stop codon to amino acid codon changes in perithecia in F. verticillioides (Supplemental Table 4). Because F. graminearum and F. verticillioides are closely related to each other, it is likely that these orthologous genes with the same stage-specific editing events have similar functions during sexual development.

34According to cDNA clone sequencing and RNA-seq data, the gene model of PUK1 based on automated annotation is incorrect with the third intron. However, the corrected coding region of PUK1 contains two tandem stop codons UA1831G UA1834G in the kinase domain, which codes for a truncated, nonfunctional protein kinase. It appears that automated annotation introduces an intron to avoid these two stop codons that are changed to UG1831G UG1834G after RNA editing. Besides PUK1, we identified at least 69 other genes that have introns erroneously introduced by automated annotation to avoid in-frame stop codons that are subjected to A-to-I editing for the UAG to UGG change during sexual reproduction. Without RNA editing in hyphae or conidia, these genes are pseudogene-like because of the stop codons in their ORFs. Although the truncated proteins encoded by these pseudogene-like genes are likely nonfunctional in other stages, our findings suggest that they may have an important biological function during sexual reproduction. To our knowledge, this kind of A-to-I RNA editing events have not been functionally characterized or reported in humans, Drosophila, and other animals.

35In this study, a total of 26,056 A-to-I editing sites were identified in F. graminearum, which is the first report of genome-wide A-to-I editing of mRNAs outside the animal kingdom. Interestingly, >70% of the editing events occur in coding regions and more than two-thirds of them are missense editing events, resulting in amino acid changes in 4594 F. graminearum genes. Although a similar finding has been reported recently in squid, A-to-I editing mainly occurs in noncoding regions, including introns and UTRs, in other animals. Furthermore, >76% of the missense editing sites in F. graminearum resulted in changing to amino acids of different physicochemical properties, which is different from squid. Therefore, unlike in metazoans, A-to-I editing tends to diversify protein functions in F. graminearum. If RNA editing also occurs in other fungi, predicted proteins based on automated annotation for genes subjected to A-to-I editing may display stage-specific differences.

36Another unique feature of RNA editing in F. graminearum is that both the frequency and editing levels of nonsynonymous A-to-I editing sites are significantly higher than those of synonymous editing sites. Therefore, although RNA editing events are generally not advantageous in humans, A-to-I editing of mRNA may be subjected to selection for new traits or functions in fungi. We also noticed that F. graminearum has transcripts of more than 300 genes with the stop-loss editing events, including a number of transcription factor and kinase genes known to be important or essential for sexual reproduction. The large number of stop-loss editing events is somewhat unique to F. graminearum. Thus, it will be important to determine the biological functions of editing events resulting in the read-through of UAG stop codons.

37Last, but not the least striking feature of A-to-I editing in F. graminearum is its stage-specific occurrence during sexual reproduction. Unlike in perithecium samples, no real RNA editing events were identified in RNA-seq data of conidia and hyphae. In animals, A-to-I RNA editing has been identified in virtually all the tissues examined. Nevertheless, A-to-I editing is known to have some tissue or developmental stage preference and is more frequent in brains than in other tissues. Because ascospores (sexual spores) are forcibly discharged and dispersed by wind as the primary inoculum for this important pathogen, RNA editing during sexual reproduction may provide additional flexibility or variations in protein coding genes in this important pathogen. Approximately 47% of the genes with >60% editing levels are specifically expressed or up-regulated in perithecia, confirming the specific role of RNA editing during sexual reproduction in F. graminearum. At least three of these genes with stage-specific editing events, PUK1, FGRRES_ 10094, and FGRRES_01649, were functionally characterized to be important for ascospore formation and release, which is a critical step in the wheat head blight disease cycle. In N. crassa, deletion of STK-21 (NCU03242), the ortholog of PUK1, did not result in obvious defects in sexual reproduction. Whereas F. graminearum is homothallic, N. crassa is a heterothallic fungus. Park and colleagues only crossed the stk-21 mutant with a wild-type strain (stk-21 × STK-21WT), and it remains possible that the stk-21 × stk-21 mutant crosses may be defective in sexual reproduction.

38For PUK1, its orthologs in N. crassa and F. verticillioides also are subjected to A-to-I editing at the same sites during sexual reproduction. Interestingly, the two A's in the UAGUAG sequence have been changed to G's in the genomic sequence of PUK1 orthologs in some Sordariomycetes such as Magnaporthe oryzae and Claviceps paspali. In metazoans, it has been reported that the genomic G-to-A mutation may be corrected by A-to-I RNA editing, whereas the edited I (G) may be fixed into the genome sequences during evolution. For the 47 genes orthologous to F. graminearum genes with PUK1-like editing events, 34 of them have similar editing sites in F. verticillioides. The other 13 genes have these editing sites that may function to correct the G-to-A DNA mutations that occurred specifically in F. graminearum or be fixed in the F. verticillioides genome during evolution. Furthermore, genome-wide A-to-I RNA editing events also were identified in RNA-seq data of perithecia in F. verticillioides. Therefore, it is likely that stage-specific A-to-I editing is a common phenomenon in some filamentous ascomycetes during sexual reproduction. RNA editing may play an important role in regulating the functions of genes important for ascosporogenesis or forcible ascospore release.

39A-to-I RNA editing in animals is mediated by members of the ADAR family that contain a conserved C-terminal catalytic deaminase domain and a variable number of N-terminal dsRNA binding domains (dsRBDs). In humans, the deaminase activities of ADAR and ADARB1 but not ADARB2 (also known as ADAR3) have been established by in vitro assays. Unlike the other two that are expressed in most human tissues, ADARB2 is specifically expressed in the central nervous system and it lacks a dsRBD. To date, ADAR genes have been identified in nearly all metazoans from sponges to humans, but not in plants, yeasts, or filamentous fungi. Indeed, we failed to identify ADAR genes in F. graminearum, F. verticillioides, and N. crassa. The lack of distinct ADAR orthologs in filamentous fungi suggests that the enzymes and related molecular mechanisms responsible for the A-to-I editing are different between fungi and animals.

40However, F. graminearum has three putative ADAT genes. In eukaryotic organisms, ADAT enzymes specifically catalyze the deamination of adenosines to inosines at or adjacent to the tRNA anticodon. They have a single adenosine deaminase domain that is closely related to that of ADARs but lack the dsRBD. Nevertheless, it appears that the dsRBD of ADARs is not essential for A-to-I editing although important for the deamination rate. When overexpressed in S. cerevisiae, the deaminase domain of human ADARB1 is capable of binding to dsRNA and displays A-to-I editing activities independent of its dsRBD. Given the sequence similarities and phyletic distributions, it has been proposed that the ADAR enzymes are a metazoan innovation that may have evolved from an ADAT ancestor via the addition of the dsRBD. Some of the ADAT enzymes may have the stage-specific activity to edit mRNAs in F. graminearum and other filamentous ascomycetes, although in vitro assays of ADATs with yeast extracts and dsRNA substrates did not reveal any activity on mRNA. However, none of the three ADAT genes in F. graminearum were specifically expressed during sexual reproduction. It is possible that fungi have ADAT-interacting proteins that are specifically expressed during sexual reproduction, and they form protein complexes with ADATs for A-to-I editing of mRNA. The other possibility is that the ADAT enzymes responsible for mRNA editing in F. graminearum are subjected to stage-specific phosphorylation or other posttranslational modifications for activation.

41In animals, the ADARs bind to any dsRNA without apparent sequence specificity but perform A-to-I editing specifically within certain dsRNA substrates. Although target recognition by metazoan ADARs and the mechanisms of substrate interaction are not well understood, it has been shown that the A's targeted for RNA-editing are embedded in an RNA stem. Furthermore, both nucleotides surrounding the edited adenosines and secondary structure elements affect the specificity and efficiency of RNA editing. In this study, we found that the 5′ − 1 preference in F. graminearum is U, and the 3′ + 1 preference is A or G, which is different from that of human ADAR and ADARB1. We also identified additional base preference beyond the nearest neighbor nucleotides in both the 5′ and 3′ positions, particularly, in the 3′ positions, suggesting that the sequence preferences of A-to-I editing is more specific in F. graminearum.

42Regarding structure selectivity, unlike targeting A's embedded in RNA stems in animals, A-to-I editing preferentially targets A's in hairpin loops and the stability of hairpin loops affect editing efficiency in F. graminearum. In animals, the neighbor preferences of ADARs are mainly dictated by the catalytic domain, whereas selectivity derives mainly from the dsRBDs domain. F. graminearum lacks ADAR orthologs, but it has three ADATs. Interestingly, the anticodon loop of tRNA targeted by ADATs is a hairpin loop. The similar sequence-preference but distinct structure-selectivity of editing between F. graminearum and animals, indicative of the potential evolutionary link between mRNA editing and ADATs in fungi. Therefore, the A-to-I editing events identified in fungi together with their stage-specific roles in sexual reproduction provide great resources to study the functions, regulatory mechanisms, and evolution of RNA editing.

43Genomic resequencing and strand-specific RNA-seq libraries were prepared with the wild-type strain PH-1 that was sequenced at the Broad Institute. Experimental methods, including sample collection, library construction, and sequencing, are described in detail in the Supplemental Methods. For each library, at least 30 million high quality reads were obtained.

44RNA-seq reads of F. graminearum were aligned to the complete genome of PH-1 available in Ensembl Fungi with program HISAT v 0.1.6-beta. RNA editing sites were identified by CLC Genomics Workbench 7.5 (CLC Bio), and a series of stringent filters were implemented to eliminate false positives as described in the Supplemental Methods. The methods used to analyze functional consequences and sequence and structure features of A-to-I editing sites were also summarized in the Supplemental Methods.

45Experimental methods for targeted deletion of the five genes with PUK1-like editing events (Supplemental Table 4) and the FgTAD1 gene, and generation of the PUK1TGGTGG and PUK1TAATAA constructs and transformants were described in the Supplemental Methods.

46Sequencing data from this study have been submitted to the NCBI Sequence Read Archive (SRA; http://www.ncbi.nlm.nih.gov/sra/) under accession numbers SRP062731 and SRP067538.