New Edges of RNA Adenosine Methylation Modifications

Recently an article published in Molecular Cell reveals the mechanism of a nuclear N⁶-methyladenosine (m⁶A) reader, the YTH domain-containing protein 1 (YTHDC1), in regulating pre-mRNA splicing [1]. Meanwhile, two additional articles published in Nature and Nature Chemical Biology report the first transcriptome-wide maps of N¹-methyladenosine (m¹A) at high resolution, suggesting a functional role for m¹A in translation regulation [2], [3].

m⁶A reader YTHDC1 in pre-mRNA alternative splicing

m⁶A is the most abundant endogenous mRNA modification, which is conserved across archea, bacteria, and eukaryotes [4]. Nonetheless, the importance of m⁶A in mammals had been underappreciated for about 40 years until the discovery of its reversibility by an m⁶A demethylase—fat mass and obesity-associated protein (FTO) [5] in 2011. Ever since, the widespread regulatory roles of m⁶A have been unraveled through the transcriptome-wide mapping of m⁶A modification [6], [7], the characterization of the second m⁶A demethylase AlkB homolog 5 (ALKBH5) [8] and three subunits of m⁶A methyltransferase complex (methyltransferase like 3, METTL3; METTL14; and Wilms tumor 1 associated protein, WTAP) [9], [10], and the functional studies of m⁶A readers YTH domain family protein 1 (YTHDF1) and YTHDF2 in humans, which regulates m⁶A methylated RNA stability [11] and translational efficiency [12], respectively. In addition, m⁶A in primary microRNAs can be recognized by another m⁶A reader, the heterogeneous nuclear ribonucleoproteins A2/B1 (HNRNPA2B1), which consequently recruits DiGeorge syndrome critical region 8 (DGCR8) and DROSHA complex and promotes the maturation of microRNAs [13], [14].

YTHDC1, as reflected by its name, contains the YTH domain that selectively binds to m⁶A [15]. Unlike the other two cytoplasmic m⁶A binding proteins YTHDF1 and YTHDF2, YTHDC1 is localized in YT bodies near the nuclear speckles [16], supporting its association with pre-mRNA splicing. Xiao and colleagues [1] identified several YTHDC1 partners including five trans-acting splicing factors (serine/arginine-rich splicing factors; SRSF1/3/9/7/10) by tandem-affinity purification following by mass spectrometric analysis, suggesting the potential regulatory role of YTHDC1 in pre-mRNA splicing. To test such possibility, they measured the alternative splicing (AS) events using RNA-seq data upon knockdown of YTHDC1 and its potential SRSF partners in HeLa cells, respectively. Their findings indicate that YTHDC1 and SRSF3 facilitate exon inclusion, while SRSF10 promotes exon skipping; however, silencing of other SRSF proteins (SRSF1, SRSF7, and SRSF9) has no significant effect on AS events. Photoactivatable ribonucleoside crosslinking and immunoprecipitation (PAR-CLIP) sequencing shows that the targeted regions of YTHDC1, SRSF3, and SRSF10 are enriched in the coding sequences (CDS) and the 3′ untranslated regions (UTR). Through analyzing the targeted exons, they further confirmed the opposite roles of YTHDC1/SRSF3 and SRSF10 in AS regulation. The change of AS events on the transcripts targeted by both YTHDC1 and SRSF3 in HeLa cells with YTHDC1 or SRSF3 silenced shows similar features with that in METTL3-silenced HeLa cells, suggesting that YTHDC1 and SRSF3 co-regulates AS events in an m⁶A-dependent manner.

Next, the authors set out to validate the interaction of YTHDC1 with either SRSF3 or SRSF10. PAR-CLIP data show that the YTHDC1 target regions are located closer to the binding sites of SRSF3 than those of SRSF10. In vivo and in vitro co-immunoprecipitation assay verifies that YTHDC1 directly interacts with SRSF3 and SRSF10 through the N-terminal of YTHDC1 and C-terminal of SRSF3 or SRSF10. The different AS events affected by YTHDC1/SRSF3 and SRSF10 prompts them to speculate that SRSF3 and SRSF10 might competitively bind to YTHDC1. Indeed they confirm the hypothesis using competing pull-down assays. The authors then examine whether YTHDC1 regulates localization of SRSF3 and SRSF10. Immunostaining assays show that silencing of YTHDC1 reduces SRSF3 but increases SRSF10 in nuclear speckle. Interestingly, this phenomenon can be rescued by complementation of wild-type YTHDC1, but not YTHDC1 mutant without m⁶A binding ability, indicating that YTHDC1 regulates the subcellular localization of SRSF3 and SRSF10 in an m⁶A-dependent manner. Further RNA binding assay shows that YTHDC1 deficiency disrupts the RNA binding of SRSF3 but enhances that of SRSF10, which can be complemented by wild-type YTHDC1, but not an m⁶A-binding-defective variant. These results indicate that the impact of YTHDC1 on AS events relies on the presence of m⁶A and the binding ability of YTHDC1 to methylated RNA.

Clearly, the comprehensive analysis presented by Xiao et al. reveals that m⁶A reader YTHDC1 facilitates exon inclusion by recruiting RNA splicing factor SRSF3 but blocking SRSF10 for its access to the binding regions of its target mRNAs (Figure 1). Indeed, apart from YTHDC1, m⁶A reader HNRNPA2B1 [14] and indirect m⁶A reader HNRNPC [17] are both involved in RNA splicing. What roles do these proteins play in AS? Are there any other splicing factors regulated by m⁶A? Does YTHDC1 play other regulatory role apart from splicing? These questions warrant further investigations.

Figure 1.

A proposed model of pre-mRNA splicing regulated by YTHDC1

Under the conditions that m⁶A in pre-mRNA is recognized by YTHDC1, YTHDC1 recruits SRSF3 to promote exon inclusion; under the conditions that pre-mRNA does not contain m⁶A or pre-mRNA with m⁶A is not bound by YTHDC1, SRSF10 facilitates exon skipping. YTHDC1, YTH domain-containing protein 1; SRSF, serine/arginine-rich splicing factor.

The reversible and dynamic m¹A methylome in eukaryotic mRNA

m¹A, another RNA adenosine methylation modification, has been identified in total RNA [18], rRNA [19], and tRNA [20] for decades. m¹A modification contains a methyl group on N¹ (hydrogen bond receptor) to form the positive charge and disturbs Watson–Crick base pairs. Unlike m⁶A, m¹A can cause both reverse transcription stops and read-throughs accompanied by mismatches. m¹A has been shown to affect the structure and function of tRNA and rRNA [21], [22]. However, the presence and functions of m¹A in mRNA remain unknown.

In the two recently-released papers, Dominissini et al. [2] and Li et al. [3] reported two transcriptome-wide sequencing methods (termed m¹A-seq and m¹A-ID-seq, respectively) to map m¹A in mRNA at high resolution (Figure 2). Their work reveals that m¹A is the second reversible and dynamic modification in eukaryotic mRNA. They firstly enrich m¹A-containing mRNA fragments from human or mouse cell lines by m¹A-specific antibody immunoprecipitation, and then take advantage of m¹A property in reverse transcription to improve the sequencing resolution, albeit later on the two groups employ different approaches for locating m¹A sites (Figure 2). As m¹A modification can be converted to m⁶A in alkaline conditions (Dimroth rearrangement), Dominissini et al. treated a portion of precipitated m¹A-containing mRNA fragments with alkaline buffer to chemically rearrange m¹A to m⁶A prior to cDNA synthesis. By comparing mismatch rates between treated and untreated samples, they located m¹A position within m¹A peaks, in which mutation rates are high in the treated sample but low in the untreated sample. In this way, they can achieve m¹A sequencing peaks at the resolution of 5–15 nucleotides (conserved m¹A sites in rRNA can be mapped at the resolution of one nucleotide) [2] (Figure 2). Different from Dominissini et al., Li et al. used Escherichia coli AlkB protein to demethylate m¹A to regular adenosine and performed cDNA synthesis with AMV reverse transcriptase to maximally confer cDNA truncations near m¹A sites. In this way, they achieved the m¹A map at the resolution of 55 nucleotides by comparing the m¹A peak features between the untreated and treated samples [3] (Figure 2). In fact, both strategies, based on mutations or truncations, sacrifice the sequencing signal and lose some sequence information near the modified sites, which make it difficult to obtain single-base resolution m¹A maps of high quality.

Figure 2.

Schematic outline of m¹A-seq and m¹A-ID-seq

In m¹A-seq, mismatch rates caused by m¹A (untreated sample) and m⁶A (chemical rearrangement) were compared. In m¹A-ID-seq, cDNA truncations conferred by m¹A (untreated sample) were compared to full-length cDNA (demethylation to A). IP, immunoprecipitation; RT, reverse transcription.

The relative abundance of m¹A in mammalian mRNA is much lower (m¹A/A: 0.015%–0.054% in cell lines and up to 0.16% in tissues) than that of m⁶A (m⁶A/A: 0.4%–0.6%) [2]. m¹A-seq identified 7154 m¹A peaks covering 4151 coding and 63 non-coding genes in humans [2], whereas m¹A-ID-seq detected 901 m¹A peaks with high confidence in 600 human genes [3]. Both studies show that most of the identified transcripts contain only one m¹A peak. Unlike m⁶A peaks that are enriched in the last transcribed exon [6], [7], [23], [24], m¹A peaks are highly enriched within 5’ UTR and near start codons.

According to the estimation of Dominissini and colleagues [2], ∼20% genes contain a single m¹A. Through the deep analysis, they find that m¹A is associated with canonical and alternative translation initiation sites, as well as the first splice site. Therefore they presume that the first spicing reaction might guide m¹A deposition. m¹A prefers more structured regions with high GC content and low minimum free energy. It is of note that m¹A level and distribution pattern in mouse embryonic fibroblasts (MEFs) and mouse embryonic stem cells (mESCs) are comparable to those in human cell lines, suggesting an evolutionarily-conserved pattern of m¹A methylome. They also survey the influence of different stress conditions on m¹A, and find that the total level and peak number of m¹A can be reduced by glucose starvation but enhanced by heat shock, indicating the dynamic feature of m¹A under different physiological conditions. Given the close association of m¹A with the translation initiation sites, Dominissini and colleagues examine whether m¹A affects mRNA translation by using published ribosome profiling and proteomics data. Notably, m¹A-containing genes have higher translation efficiency and protein levels compared to non-m¹A-containing genes, implying that m¹A modification is correlated with elevated translation.

Meanwhile, Li and colleagues [3] studied the m¹A dynamics induced by H₂O₂ treatment and serum starvation. They propose that m¹A may reside in a prominent motif with a GA-rich consensus. Similar with the aforementioned Nature paper, they state that m¹A prefers structured sequences with high GC content. It is notable that ALKBH3 (human ortholog of E. coli AlkB) is found to be able to demethylate m¹A in human mRNA, indicating that m¹A is a reversible modification and may play an important regulatory role on mRNA.

Collectively, the two studies by Dominissini and his colleagues [2] and Li and his colleagues [3] provide the first map of transcriptome-wide m¹A methylome and suggest new roles for m¹A: this reversible modification is enriched around start codon, dynamically regulated by stress conditions, and correlated with elevated translation. Although the two m¹A-seq techniques discussed here provide m¹A maps with relatively-high resolution compared to m⁶A-seq method (at the resolution of ∼200 nucleotides), a big challenge is to develop single-base resolution methods for m⁶A and for m¹A as well. Another challenge is to uncover the broader biological functions of m⁶A and m¹A modifications. Future studies will focus on the identification and characterization of writer and reader proteins and functional roles of these two modifications. Given that m⁶A as an RNA structure switch affects RNA–protein interaction [17], the RNA structure changed by m¹A modification might also play certain functions. We expect more investigations to draw a more comprehensive picture of RNA modification story.

Competing interests

The authors declare that they have no competing interests.

Handled by Yi Xing