m⁶A RNA methylation: from mechanisms to therapeutic potential

Abstract

RNA carries a diverse array of chemical modifications that play important roles in the regulation of gene expression. N ⁶‐methyladenosine (m⁶A), installed onto mRNA by the METTL3/METTL14 methyltransferase complex, is the most prevalent mRNA modification. m⁶A methylation regulates gene expression by influencing numerous aspects of mRNA metabolism, including pre‐mRNA processing, nuclear export, decay, and translation. The importance of m⁶A methylation as a mode of post‐transcriptional gene expression regulation is evident in the crucial roles m⁶A‐mediated gene regulation plays in numerous physiological and pathophysiological processes. Here, we review current knowledge on the mechanisms by which m⁶A exerts its functions and discuss recent advances that underscore the multifaceted role of m⁶A in the regulation of gene expression. We highlight advances in our understanding of the regulation of m⁶A deposition on mRNA and its context‐dependent effects on mRNA decay and translation, the role of m⁶A methylation of non‐coding chromosomal‐associated RNA species in regulating transcription, and the activities of the RNA demethylase FTO on diverse substrates. We also discuss emerging evidence for the therapeutic potential of targeting m⁶A regulators in disease.

Roles of the most prevalent RNA modification in post‐transcriptional regulation of gene expression, the latest understanding of its deposition and removal, and the possibility of its targeting in disease are discussed in this comprehensive review.

Introduction

m⁶A is a chemical derivative of adenosine in RNA that plays important, wide‐ranging roles in the regulation of gene expression. m⁶A modification of RNA occurs in most eukaryotes and is the most prevalent internal mRNA modification in mammals, where it is present at tens of thousands of sites across the transcriptome, at a frequency of 0.15–0.6% of all adenosines (Table 1) (Dominissini et al, 2012; Liu et al, 2014; Ke et al, 2015). The presence of m⁶A in mRNA was first identified in the 1970s, and the m⁶A mRNA methyltransferase complex was purified in the 1990s, with METTL3 being identified as a key component (Desrosiers et al, 1974; Lavi & Shatkin, 1975; Perry et al, 1975; Wei et al, 1975; Wei & Moss, 1977; Bokar et al, 1994, 1997). Later work demonstrated that METTL3 forms a heterodimer with METTL14 to convert A to m⁶A on mRNA (Liu et al, 2014; Ping et al, 2014; Wang et al, 2014b, 2016a, 2016b; Śledź & Jinek, 2016). METTL3 is the catalytically active methyltransferase, with METTL14 playing an essential structural role to facilitate catalysis. The larger methyltransferase holocomplex contains accessory units including WTAP, VIRMA, RBM15A, RBM15B, ZC3H13, and HAKAI, and is predicted to approach 1,000 kDa in size (Patil et al, 2016; Růžička et al, 2017; Guo et al, 2018; ; Wen et al, 2018; Yue et al, 2018). Orthologs of METTL3 and METTL14 have been identified in the sequenced genomes of many animals and plants, as well as some fungi (Bujnicki et al, 2002), and m⁶A is present in the mRNA of all human and mouse tissues studied so far (Xiao et al, 2019; Liu et al, 2020b). It should be noted that while the vast majority of m⁶A on mRNA is deposited by METTL3/METTL14, the U6 snRNA m⁶A methyltransferase METTL16 has also been reported to catalyze the addition of m⁶A on the structured 3′UTR of MAT2A transcripts (Pendleton et al, 2017; Doxtader et al, 2018; Mendel et al, 2018; Ruszkowska et al, 2018). The METTL3/METTL14 methyltransferase complex typically exhibits a nuclear localization and co‐transcriptionally deposits m⁶A methylation onto mRNA transcripts, as well as other RNA polymerase II (Pol II)‐transcribed RNAs, including long non‐coding RNAs (lncRNAs) and primary microRNAs (pri‐miRNAs) (Alarcón et al, 2015b; Ke et al, 2017; Slobodin et al, 2017; Knuckles et al, 2018). The methyltransferase deposits m⁶A within on a specific subset of cellular transcripts, primarily near stop codons/start of terminal exons and within unusually long internal exons (Dominissini et al, 2012; Meyer et al, 2012; Ke et al, 2015).

Table 1.

Statistics for m⁶A in mRNA.

Quantity	Value	References	Notes
Ratio of m6A to A in polyadenylated RNA	0.15–0.6% (LC‐MS/MS)	Liu et al (2014, 2018, 2020b), Wang et al (2014b), Lin et al (2017), Du et al (2018), Wei et al (2018)	m6A to A ratios in polyadenylated RNA are heavily influenced by the highly expressed genes in the transcriptome. In humans, a small fraction of all expressed genes (10–1,000 genes, depending on the tissue type) make up the majority of the mRNA pool (Ramsköld et al, 2009)
Percentage of transcripts in the transcriptome that contain m6A	~ 25 to ~ 60% (m6A‐seq)	Dominissini et al (2012, unpublished)	Dominissini et al identified m6A in transcripts of ~ 7,000 genes. The 25% figure is derived from dividing 7,000 from the estimated number of total human genes, ~ 20,000–25,000. However, this is likely an underestimate. First, all genes are not expressed in any given cell type. Second, m6A in relatively lowly expressed genes escapes detection due to low sequence coverage (increasing sequencing depth increases the number of m6A peaks identified). By restricting the analysis to a set of transcripts above an expression cutoff set to the lower quartile of expression for detected methylated transcripts, the percentage of transcripts containing at least one m6A peak is higher, ~ 60% (unpublished). It is important to note here that this percentage reflects the number of genes whose transcripts exhibit m6A, not the stoichiometry of m6A in the transcripts of any given gene
Average number of m6A per transcript	~ 3 m6A residues per average mRNA transcript (LC‐MS/MS), ~ 1.7 m6A‐seq peaks per transcript (m6A‐seq)	Perry et al (1975), Dominissini et al (2012)	It is important to consider that these are averages. Many transcripts are unmethylated and many have significantly more m6A peaks than the average
Percentage of DRACH motifs that are methylated	~ 5%	Dominissini et al (2012)	Dominissini et al estimate that a prevalence of ~ 1 m6A peak every 2,000 bp. The number of DRACH motifs found in 2,000 bp of sequence is theoretically (3/4)(2/4)(1/4)(1/4)(2/4)*2,000 = 23.4. Thus, the percentage is estimated at 1/23.4 ~ 5%. The actual percentage may be several‐fold higher or lower, given that m6A peaks may contain several m6A sites and the true frequency of DRACH motifs may diverge from the theoretical estimate

The effects of m⁶A on gene expression are wide‐ranging (Fig 1). m⁶A alters pre‐mRNA processing, promotes mRNA nuclear export, alters mRNA stability, increases translation efficiency, and facilitates non‐canonical translation initiation (Wang et al, 2014a, 2015; Liu et al, 2015, 2017; Meyer et al, 2015; Xiao et al, 2016; Barbieri et al, 2017; Roundtree et al, 2017; Slobodin et al, 2017; Kasowitz et al, 2018; Zhou et al, 2018, 2019). Some of these effects are mediated by m⁶A “reader” proteins that can selectively recognize m⁶A and exert a regulatory function on the m⁶A‐marked mRNA. m⁶A readers include members of the YTH family, which bind m⁶A with their eponymous YTH domains (Wang et al, 2014a, 2015; Xiao et al, 2016; Hsu et al, 2017; Roundtree et al, 2017; Shi et al, 2017). Deposition of m⁶A on RNA can also alter local RNA secondary structure. Here m⁶A can exert stabilizing and destabilizing effects, depending on its sequence context, but results so far indicate that destabilizing effects are more predominant transcriptome‐wide (Kierzek & Kierzek, 2003; Roost et al, 2015; Spitale et al, 2015; Sun et al, 2019b). The RNA‐binding proteins (RBPs) HNRNPC and HNRNPG are indirect “structural‐switch” readers, that can selectively bind at “structural switches”, sites at which m⁶A alters the accessibility of proximal RNA sequence to facilitate binding of RBPs (Liu et al, 2015, 2017; Zhou et al, 2019). Several other RBPs that preferentially bind m⁶A have also been identified, but beyond the YTH family and HNRNPC/G, how these proteins selectively bind m⁶A has not been conclusively elucidated (Alarcón et al, 2015a; Ae et al, 2017; Edupuganti et al, 2017; Wu et al, 2018a). More recently, it has furthermore been suggested that m⁶A may also regulate gene expression by disfavoring the binding of certain RBPs to mRNA, as has been reported for G3BP1 and LIN28A (Edupuganti et al, 2017; Sun et al, 2019b). The wide array of mRNA processing factors whose activities are regulated by m⁶A enable m⁶A to regulate the large number of diverse molecular processes involved in gene expression. In addition to m⁶A‐mediated regulation of gene expression at the mRNA level through the direct recruitment of reader proteins to the transcript, it has also been recently discovered that m⁶A modification of non‐coding chromosome‐associated RNA transcripts functions to modulate gene expression by regulating transcription of nearby mRNAs (Liu et al, 2020a). Thus, it is becoming increasingly clear that m⁶A has numerous effects on gene expression through diverse mechanisms.

Figure 1. m⁶A is a multifaceted regulator of gene expression.

m⁶A (red circle) regulates transcription, alternative splicing, alternative polyadenylation, nuclear export, cap‐dependent and cap‐independent translation, mRNA degradation, and mRNA stabilization. A diverse set of reader proteins that selectively bind m⁶A, either directly or indirectly, mediate these multifaceted effects on gene expression. Note that for clarity, nuclear processes are shown to occur after release of RNA from the polymerase, but some of the depicted nuclear processes may also occur co‐transcriptionally.

An important facet of m⁶A methylation as a regulatory system is its reversibility. m⁶A on mRNA can be removed by FTO and ALKBH5, m⁶A demethylases, posing an additional dimension of regulation for the m⁶A epitranscriptome (Jia et al, 2011; Zheng et al, 2013). Both are iron‐ and α‐ketoglutarate‐dependent dioxygenases that mediate oxidative demethylation of m⁶A. FTO has been demonstrated to demethylate m⁶A on mRNA, m⁶A_m on mRNA and snRNAs, and m¹A on tRNA (Mauer et al, 2017, 2019; Wei et al, 2018). Due to its wide array of substrates, the activities of FTO that are most relevant for its crucial in vivo functions have been the subject of debate, as discussed later in this review.

Given its wide‐ranging roles in the regulation of gene expression, it is unsurprising that m⁶A is required for numerous physiological and pathophysiological processes. Depletion of METTL3 homologs leads to defective meiosis in yeast, and developmental arrest in flies and plants (Zhong et al, 2008; Hongay & Orr‐Weaver, 2011; Agarwala et al, 2012; Bodi et al, 2012; Schwartz et al, 2014). Genetic knockout of either Mettl3 or Mettl14 is developmentally lethal in mice, with embryos failing to thrive at around E5.5 (Batista et al, 2014; Geula et al, 2015). Analysis of tissue‐specific knockout mouse models of Mettl3 and Mettl14 has revealed essential roles for m⁶A in brain development and function, cardiac homeostasis, immune system development and function, spermatogenesis, and skeletal function (Lin et al, 2017; Yoon et al, 2017; Li et al, 2017a; Rubio et al, 2018; Wang et al, 2018a, 2018c, 2019; Wu et al, 2018b; Dorn et al, 2019; Winkler et al, 2019; Xu et al, 2020). Moreover, knockout of either Mettl3 or Mettl14 severely blocks or delays differentiation in numerous stem cell or progenitor cell systems, including embryonic stem cells, embryonic neuronal stem cells (Yoon et al, 2017; Wang et al, 2018c), hematopoietic stem cells (Vu et al, 2017; Zhang et al, 2017a; Weng et al, 2018a; Cheng et al, 2019; Lee et al, 2019), naïve T cells (Li et al, 2017a), and bone marrow mesenchymal stem cells (Wu et al, 2018b). Additionally, m⁶A has been implicated in the pathogenesis of a variety of diseases, including in numerous cancer types and in type 2 diabetes (Batista, 2017; De Jesus et al, 2019; Ianniello et al, 2019; Yang et al, 2019b; Huang et al, 2020). Three independent reports implicate METTL3 or METTL14 as important regulators of AML, suggesting that targeting m⁶A may hold therapeutic potential in this setting (Barbieri et al, 2017; Vu et al, 2017; Weng et al, 2018a). Overall, m⁶A has emerged as a key regulator of numerous important biological processes in normal physiology and in disease.

In this review, we will delve into recent evidence that underscores the multifaceted nature of the role of m⁶A in regulating gene expression. We highlight new advances in our understanding of m⁶A deposition on mRNA and its effects on mRNA decay and translation, the role of m⁶A on non‐coding chromosomal‐associated RNA species in regulating mRNA transcription, the activities of the RNA demethylase FTO on diverse substrates, and discuss the therapeutic potential of targeting m⁶A regulators.

Regulation of m⁶A deposition on mRNA

As mentioned above, a notable characteristic of the METTL3/METTL14 methyltransferase is the specificity with which it deposits m⁶A on the transcriptome. The consensus motif for METTL3/METTL14 is a commonly occurring DRACH (D = A, G or T; R = A or G; H = A, C or U) consensus sequence, but only a fraction of sequences that contain this consensus motif are methylated (Table 1). Further, m⁶A exhibits a marked regional bias in its distribution across the transcriptome. m⁶A can be found throughout the length of transcripts, but is strongly enriched in the vicinity of stop codons and in unusually long internal exons (Fig 2A) (Ke et al, 2017). According to one analysis, internal exons that are > 200 nt comprise only ~ 15% of all internal exons, but internal exons > 200 nt contain ~ 80% of all m⁶A sites present within internal exons (Ke et al, 2017). This cannot be explained by overrepresentation of methyltransferase motifs in long internal exons, because internal exons > 200 nt only contain ~ 30% of all motifs. Additionally, transcript isoforms that use proximal alternative polyadenylation sites are generally enriched for m⁶A over longer isoforms using distal alternative polyadenylation sites (Molinie et al, 2016). These observations imply regulation of m⁶A deposition by METTL3/METTL14 that extends beyond primary sequence constraints.

Figure 2. Specificity of the m⁶A epitranscriptome.

(A) Schematic representing the distribution of m⁶A in the mammalian transcriptome. A subset of transcripts contain one or more m⁶A sites, while another subset are not methylated. m⁶A is enriched in unusually long internal exons and near stop codons/start of last exons. (B) Deposition is regulated by intrinsic factors, such as the preference of the METTL3/METTL14 methyltransferase for specific RNA sequences. m⁶A deposition is also regulated by external factors; transcription factors, RNA‐binding proteins, RNA polymerase II, and the H3K36me3 histone modification have been reported to recruit the METTL3/METTL14 methyltransferase to mRNAs to promote methylation. m⁶A demethylases FTO and ALKBH5 can also tune m⁶A levels at a subset of sites following their initial deposition by METTL3/METTL14.

Determinants of m⁶A specificity are central to m⁶A‐mediated regulation of gene expression, as the restriction of m⁶A to a subset of cellular transcripts (estimated at 25–60%) allows for selective regulation of specific transcripts by m⁶A reader proteins (Table 1). For example, in mouse embryonic stem cells (mESCs), 80% of genes that regulate naïve pluripotency exhibit m⁶A methylation on their transcripts. This is thought to facilitate their clearance during the termination of pluripotency to enable embryonic stem cell differentiation (Geula et al, 2015). Further, the specific position at which an mRNA is m⁶A methylated may also be relevant for gene expression, since the nature of m⁶A‐mediated regulation appears to differ depending on the location of m⁶A within the transcript (Wang et al, 2014a, 2015; Meyer et al, 2015; Choi et al, 2016; Barbieri et al, 2017; Slobodin et al, 2017; Zhou et al, 2018; Mao et al, 2019).

Given the relevance of m⁶A specificity for gene expression regulation, several recent studies have focused on elucidating the molecular mechanisms that govern regulation of m⁶A deposition in various cellular contexts. Certain transcription factors can promote methylation of specific transcripts by recruiting the methyltransferase complex to their target loci (Barbieri et al, 2017; Bertero et al, 2018). In addition, RBPs, such as TARBP2, can recruit the methyltransferase complex to its bound transcripts (Fish et al, 2019) and slow transcription rates have been proposed to promote m⁶A deposition by enhancing co‐transcriptional recruitment of the methyltransferase complex to its RNA substrate (Slobodin et al, 2017). The histone H3K36me3 modification has also been proposed to recruit the methyltransferase complex to chromatin in order to promote methylation of nascent RNAs (Huang et al, 2019a). Collectively, these papers have shed light on how diverse biological inputs can alter m⁶A deposition and lead to differential downstream effects on gene expression and cellular phenotypes (Fig 2B). The wide variety of mechanisms that affect m⁶A methylation support the notion that m⁶A deposition is a highly regulated event that is interwoven with diverse cellular processes.

While progress has been made in elucidating new mechanisms regulating m⁶A deposition, fundamental questions regarding m⁶A deposition remain unanswered. Notably, the general rules that govern m⁶A deposition and generate the observed distribution of m⁶A on the transcriptome are still poorly understood.

First, the degree to which intrinsic vs. extrinsic factors determine m⁶A levels has been the subject of debate: to what extent are m⁶A levels determined by the intrinsic preference of the METTL3/METTL14 methyltransferase for certain nucleotide sequences vs. the contribution of extrinsic regulation of methyltransferase complex activity by external factors such as RBPs, transcription factors, RNA polymerase, or other yet to be discovered factors? This question is important due to its potential implications for m⁶A as a regulatory system. If intrinsic determinants are dominant, this would suggest that m⁶A deposition is not highly regulated, and would generally operate independently of other cellular processes. It would suggest that global changes in m⁶A methylation could occur if enzyme levels or other reaction variables change, but relative changes of m⁶A levels at different sites would be limited. Conversely, if extrinsic determinants dominate, this would indicate a more dynamic role for m⁶A, in which m⁶A levels may react to changes in the activity of external factors in different cellular states. It would moreover imply that m⁶A deposition is interwoven with other cellular processes, allowing for subsets of sites to gain or lose methylation in response to specific cellular events.

The involvement of extrinsic determinants is supported by a large body of literature that reports differential methylation of significant numbers of m⁶A sites in different cellular contexts (Xiao et al, 2019; Liu et al, 2020b). Additionally, several mechanisms through which trans‐acting factors can regulate methylation have been characterized, as described above. However, other reports have concluded that intrinsic determinants are predominant and that levels may be largely “hard‐coded” in cis by the local sequence surrounding the m⁶A site (Schwartz et al, 2014; Garcia‐Campos et al, 2019). Part of this apparent discrepancy may be due to differences in m⁶A regulation in different organisms. Most studies characterizing the importance of extrinsic factors have been carried out in mammalian cells (Barbieri et al, 2017; Slobodin et al, 2017; Bertero et al, 2018; Fish et al, 2019). One study reporting that intrinsic determinants largely control m⁶A levels primarily examined the ability of local sequence features to predict m⁶A methylation levels in the budding yeast Saccharomyces  cerevisiae (Garcia‐Campos et al, 2019). The authors found that in mammalian systems, the ability to predict m⁶A methylation levels from local sequence features was diminished compared to the ability to predict methylation levels in S. cerevisiae. Part of this discrepancy may also arise from an assumption that local sequence variation only alters intrinsic determinants and not regulation by extrinsic factors, which may not necessarily hold true. Overall, we propose that intrinsic determinants and extrinsic determinants likely both play significant roles in shaping the landscape of m⁶A on the transcriptome and that their relative contributions likely vary based on the specific site and cellular context. Future work using systematic approaches to directly assess the contributions of intrinsic and extrinsic determinants of m⁶A will advance our understanding of the overall determinants and general characteristics of m⁶A regulation. Since batch effects and other variables can cause significant variation in m⁶A measurements, ensuring the use of optimal experimental designs and appropriate statistical models for differential methylation analysis may help bring more clarity to this issue (Zhang et al, 2019). Further development and application of robust and quantitative high‐throughput methods of m⁶A detection will also enable greater understanding of m⁶A regulation by enhancing the capability to detect changes in methylation stoichiometry under different conditions.

We should note that beyond regulation of m⁶A deposition itself, an additional layer of regulation may be mediated by the differential activities of diverse m⁶A reader proteins. While extrinsic determinants may facilitate m⁶A dynamics in response to cell differentiation and signals from stimulation or stress, m⁶A deposited via both intrinsic and extrinsic determinants could be recognized by readers differently under different cellular contexts. Thus, even if m⁶A levels at a particular site do not change, the functional effect of that m⁶A site may still vary according to the cell state. For example, m⁶A readers could compete for binding to m⁶A sites, in which case differing levels or ratios of readers in different cell types could lead to diverging gene expression outcomes. In this context, post‐translational modifications of readers also may tune their cellular localization, m⁶A binding affinity and interacting protein partners. All these variables could affect the functional outcome of mRNA m⁶A regulation (Fig 3).

Figure 3. Theoretical model for how changes in cellular state may impact m⁶A regulation.

m⁶A regulatory factors can act at multiple levels to enable differential gene expression regulation. Upon a change in cell state, m⁶A sites that are extensively regulated by extrinsic determinants (e.g., RNA‐binding proteins that recruit METTL3/METTL14 to the site) may vary in their methylation level if the levels or activities of these factors change. In contrast, m⁶A sites that are primarily controlled by intrinsic determinants may exhibit stable methylation. Additionally, levels or activities of m⁶A readers may vary upon a change in cell state. This may result in differential reader binding to m⁶A sites. Given the diverse activities of various m⁶A readers, the effect of a particular m⁶A site on mRNA metabolism may change even if methylation levels remain stable.

Another major unknown in the regulation of m⁶A deposition is the mechanism(s) that govern the specific enrichment of m⁶A near certain transcriptomic features. Despite recognition of the strong enrichment of m⁶A within unusually long exons and near stop codons since the first transcriptome‐wide m⁶A mapping studies, the mechanisms regulating m⁶A methylation reported to date do not explain how m⁶A is selectively deposited in this specific distribution. The m⁶A methyltransferase accessory factor VIRMA appears to promote methylation of many sites in mRNA 3′ untranslated regions (UTRs) and near stop codons, but the mechanistic basis for this effect is to date still unclear (Yue et al, 2018). H3K36me3 directly recruits METTL14 to gene bodies to promote m⁶A deposition, and it does not appear to confer enrichment of m⁶A near stop codons, as this enrichment is still observed when H3K36me3 levels are reduced (Huang et al, 2019a). The inability of currently known mechanisms to explain how m⁶A is specifically enriched in certain transcriptome regions suggests that major mechanisms regulating m⁶A deposition globally remain to be discovered. Most reported mechanisms of m⁶A deposition regulation involve models in which a trans‐acting factor recruits the METTL3/METTL14 methyltransferase to RNAs in order to promote methylation at proximal sites. However, mechanisms with different modes of action may need to be considered in order to explain m⁶A specificity at distinct transcriptomic features. Future work on the mechanisms that govern m⁶A specificity will advance the current understanding of m⁶A‐mediated gene regulation.

As METTL3/METTL14 is the predominant m⁶A methyltransferase on mRNA, we have mainly focused on METTL3/METTL14‐catalyzed m⁶A in this review. However, it should be noted that other methyltransferases that methylate the N6 position of adenosine on various RNA species have recently been discovered and characterized. PCIF1 (phosphorylated CTD interacting factor 1), an RNA polymerase II‐associated factor that contains a N ⁶‐methyladenosine methyltransferase domain, mediates the methylation of N ⁶, 2′‐O‐dimethyladenosine (m⁶A_m), which occurs near the mRNA cap (Akichika et al, 2019; Boulias et al, 2019; Sendinc et al, 2019; Sun et al, 2019a). 5′ ends of eukaryotic mRNAs carry a 7‐methylguanosine (m⁷G) cap linked to the rest of the mRNA by a triphosphate linkage and it is well known that the first nucleotide after the m⁷G cap can be methylated on the ribose sugar. If that nucleotide is adenosine, it can be further methylated to m⁶A_m by PCIF1. Additional characterized m⁶A methyltransferases include METTL16, ZCCHC4, METTL5, and METTL4. METTL16 deposits m⁶A methylation on U6 snRNA and on MAT2A mRNA (Pendleton et al, 2017; Doxtader et al, 2018; Mendel et al, 2018), ZCCHC4 and METTL5 are rRNA m⁶A methyltransferases (Ma et al, 2019; van Tran et al, 2019; Ignatova et al, 2020; Leismann et al, 2020), and METTL4 mediates internal m⁶A and m⁶A_m methylation of U2 snRNA (Chen et al, 2020; preprint: Goh et al, 2020).

Context‐dependent roles of m⁶A in mRNA decay and translation

m⁶A is known to regulate mRNA decay and translation through a variety of mechanisms. The mammalian YTHDF family of proteins, consisting of YTHDF1, YTHDF2, and YTHDF3, are cytosolic m⁶A readers that regulate m⁶A degradation and translation. YTHDF2, the first to have been characterized, selectively recognizes and shuttles m⁶A‐modified RNAs for degradation by multiple mechanisms (Wang et al, 2014a). YTHDF2 recruits the CCR4‐NOT deadenylase complex to promote deadenylation and degradation of m⁶A‐marked mRNAs and also recruits the RNase P/MRP complex to promote endoribonucleolytic cleavage of m⁶A‐marked mRNAs (Du et al, 2016; Park et al, 2019). Whereas YTHDF2 mediates mRNA decay, the two other cytosolic YTHDF proteins, YTHDF1 and YTHDF3, facilitate translation of their target methylated mRNAs. YTHDF1 promotes translation by recruiting the translation initiation factor eIF3 to bound mRNAs (Wang et al, 2015). YTHDF3 in turn interacts with YTHDF1 and is thought to shuttle its bound transcripts to YTHDF1 to promote their translation (Shi et al, 2017). In addition to this YTHDF1/3‐mediated regulation, other reported mechanisms of m⁶A‐mediated translation regulation include the direct binding of eIF3 to m⁶A in the 5′UTR of transcripts to promote cap‐independent translation initiation, METTL3‐mediated mRNA circularization to promote translation initiation, and the promotion of translation elongation by YTHDC2‐mediated disruption of RNA secondary structures (Meyer et al, 2015; Choe et al, 2018; Mao et al, 2019). Interestingly, YTHDC2 has also been implicated in mRNA degradation, as it strongly interacts with the 5′–3′ exonuclease XRN1, though a direct effect on mRNA decay has not yet been demonstrated (Hsu et al, 2017; Wojtas et al, 2017; Kretschmer et al, 2018).

An emerging theme in m⁶A‐mediated gene regulation is the apparent context‐dependent activity of the m⁶A reader proteins. The action of m⁶A readers may depend on the specific characteristics of the m⁶A sites on the mRNA upon which they act. It is a consensus in the field that m⁶A controls mRNA stability through YTHDF2‐mediated mRNA destabilization. Recently, IGF2BP1, IGF2BP2, and IGF2BP3 were reported to bind m⁶A‐methylated transcripts and stabilize them (Huang et al, 2018). Interestingly, while the distribution of YTHDF proteins approximates the distribution of m⁶A on transcripts, the distribution of IGF2BP1–3 binding sites is distinct. IGF2BP1–3 do not appear to be enriched at the stop codon; rather, they exhibit constant enrichment throughout the 3′ UTR. Further, there appears to be little overlap in the transcriptome‐wide binding sites of IGF2BP1–3 and YTHDF2. Mechanistically, IGF2BP1–3 appear to recognize m⁶A through their KH domains, although their exact mode of binding has yet to be elucidated. Altogether, these data suggest that specific characteristics of m⁶A sites may result in differential binding to IGF2BP1–3 and YTHDF2, and suggest that different m⁶A sites can either increase or decrease mRNA stability, depending on the activity of the specific reader proteins that bind them.

The context dependency of m⁶A reader proteins has also been observed in the role of m⁶A on translation. YTHDF1 binds selectively to m⁶A‐modified mRNA transcripts and promotes translation efficiency of target transcripts in HeLa and endometrial cancer cells (Wang et al, 2015; Liu et al, 2018). However, the extent of translation upregulation mediated by YTHDF1 appears to depend on the cellular context; the effect can be small, as is observed in HEK293T cells and on artificial reporters bearing m⁶A in a breast cancer cell line (Meyer et al, 2015; Slobodin et al, 2017). Data from a recent report investigating the role of YTHDF1 in learning and memory defects in mice furthermore suggest that YTHDF1‐mediated translation upregulation may be stimulation‐induced in some contexts (Shi et al, 2018). In this study, the authors found that hippocampus‐specific knockdown of YTHDF1 leads to defects in spatial memory and contextual fear memory. Interestingly, the effect of YTHDF1 on translation was mild under homeostatic conditions in cultured hippocampal neurons, but was observable following depolarization of hippocampal neurons with potassium chloride. In line with this proposed model is a recent report that characterizes the function of YTHDF1 in a dorsal root ganglion model of injury‐induced axon regeneration and finds that while the basal effect of YTHDF1 on translation is minor, YTHDF1 is required to promote protein synthesis during axon regeneration process (Weng et al, 2018b).

Furthermore, in addition to functional consequences due to differential binding of reader proteins, the effect of m⁶A on translation may also differ according to the position of the m⁶A site within the transcript. 5′ UTR m⁶A has been proposed to promote non‐canonical, cap‐independent translation initiation (Meyer et al, 2015; Zhou et al, 2018), while m⁶A within the coding sequence has been suggested to regulate translation elongation by the ribosome, though studies have reported mixed results on the directionally of the latter effect (Choi et al, 2016; Barbieri et al, 2017; Slobodin et al, 2017; Mao et al, 2019). In addition, m⁶A near the stop codon and in the 3′ UTR (as well as m⁶A in other mRNA regions to a lesser extent) can be bound by the reader proteins YTHDF1 and YTHDF3, which promote translation initiation (Wang et al, 2015; Shi et al, 2017). We note that these various mechanisms were reported in independent studies; more systematic studies focusing on the extent to which m⁶A sites in different transcript regions differ in their functions would greatly strengthen this paradigm.

A recent study on m⁶A‐quantitative trait loci (QTL) supports this notion of context dependency by revealing that the impact of m⁶A on translation appears quite heterogeneous and may depend on specific RBP binding to or near m⁶A sites (Zhang et al, 2020). In this study, m⁶A‐QTLs were identified and correlated with other molecular QTL data. Low correlations in effect sizes between m⁶A‐QTLs and all molecular QTLs analyzed, including ribosome‐binding QTLs, were observed. One possible explanation for this result is that the functional effects of m⁶A on these processes are heterogenous due to context dependency of diverse downstream mechanisms. Consistent with this model, the authors found that for m⁶A‐modified transcripts, the effects of m⁶A depletion on translation efficiency are heterogeneous, with similar numbers of up‐ and downregulated genes. However, further subgroup analysis of m⁶A sites according to nearby RBP occupancy reveals that certain RBP binding events are associated with upregulation or downregulation of translation efficiency. Consistently, the authors confirmed the previously reported translation‐promoting effect of YTHDF1, but in addition also identified proteins that appear to suppress translation of m⁶A‐associated transcripts. Collectively, these recent studies reveal the multifaceted and context‐dependent mechanisms by which m⁶A impacts mRNA decay and translation.

Further highlighting the complexity of these multifaceted regulatory mechanisms and underlining the need for future studies to fully unravel the molecular details, is the fact that two recent studies have proposed that YTHDF proteins function more similarly in the regulation of gene expression than previously described (Lasman et al, 2020; Zaccara & Jaffrey, 2020). Zaccara and Jaffrey report that in HeLa cells, YTHDF proteins all promote RNA decay, share common binding sites on mRNAs, and share similar protein interaction partners and subcellular localizations. They also report that YTHDF1 does not promote translation in HeLa cells. In the second study, Lasman et al examine the roles of Ythdf proteins in vivo and report a dosage‐dependent redundancy in their roles during early development in mice and compensation between the three readers in mESCs. Defects in mESC differentiation and mRNA decay are observed in the triple Ythdf knockout cells and not in the single knockouts. They did not observe global changes in translation in either the triple or single Ythdf knockouts in mESCs. Although it has previously been reported that the YTHDF proteins cooperatively regulate stability of m⁶A‐methylated transcripts (Shi et al, 2017), these findings add additional complexities to functions of the YTHDF proteins.

One discrepancy is the translation‐promoting role of YTHDF1. As described above, this function by YTHDF1 can be context‐dependent, has been shown to be stimulation‐induced, and is reflected by genetic experiments in knockout mice and m⁶A‐QTL studies. Genetic approaches using knockout mice of these Ythdf genes reveal very different phenotypes: Ythdf2 KO mice exhibit developmental defects which can be explained by altered transcript stability (Ivanova et al, 2017; Li et al, 2018a); Ythdf1 KO mice exhibit long‐term learning and memory defects and strong anti‐tumor immunity, all caused by altered translation of m⁶A methylated transcripts (Shi et al, 2018; Han et al, 2019). Different cellular contexts and the local sequence features of specific transcripts may thus alter the effect of YTHDF1 on translation. Overall, we note that evidence for distinct roles for the individual YTHDF proteins has been reported by several different groups in several different systems. Targeting of YTHDF1 to transcripts via a nuclease‐inactivated Cas13b‐YTHDF1 fusion protein promoted translation, while targeting YTHDF2 via an analogous fusion protein led to degradation (Rauch et al, 2018). m⁶A has also been shown to enhance phase separation of YTHDF m⁶A reader proteins and regulate their subcellular localization to phase‐separated RNP granules, such as stress granules (Ries et al, 2019). A recent study describing the role of YTHDF proteins in stress granule formation reports that in U‐2 OS osteosarcoma cells, YTHDF2 localizes to both stress granules and P‐bodies, while YTHDF1 and YTHDF3 localize to stress granules, but not P‐bodies (Fu & Zhuang, 2020). YTHDF1 and YTHDF3 but not YTHDF2 were found to promote stress granule formation. Identification of mechanisms that determine whether YTHDF proteins act in similar or distinct manners may help shed light on the contrasting observations in different cellular systems. Future exploration of the functions of the YTHDF proteins in different contexts will is needed to determine if YTHDF proteins generally display distinct or cooperative effects on gene expression.

Regulation of gene expression by modification of non‐coding RNA: m⁶A on carRNA regulates chromatin state and transcription

In the past, the primary focus within the field in defining the effect of m⁶A on gene expression has been placed on the mechanisms by which m⁶A on a given mRNA regulates expression of the transcript itself. However, recent evidence also points to an important role of m⁶A on non‐coding chromosome‐associated regulatory RNA (carRNA) for gene expression. Our recent work demonstrates that m⁶A on carRNA facilitates transcriptional downregulation of proximal genes by inducing the decay of carRNA transcripts that regulate the chromatin state of proximal loci (Liu et al, 2020a).

m⁶A is known to play crucial roles in the self‐renewal and differentiation of ESCs and early investigations mainly focused on the contribution of cytoplasmic m⁶A readers to this phenotype. In mESCs, transcripts encoding pluripotency factor transcripts tend to be m⁶A methylated and subjected to the YTHDF2‐mediated decay in the cytoplasm, which affects the turnover of these transcripts during differentiation (Batista et al, 2014; Geula et al, 2015). However, Ythdf2 knockout mice can survive to late embryonic developmental stages, while Mettl3 knockout results in early embryonic lethality (Geula et al, 2015; Ivanova et al, 2017; Li et al, 2018a), similar to the knockout of the nuclear m⁶A reader Ythdc1 (Kasowitz et al, 2018). These observations imply that m⁶A could play additional roles within the nucleus that also affect cell survival and differentiation.

m⁶A‐sequencing of chromatin‐associated RNA using a procedure that retains non‐polyadenylated transcripts revealed that METTL3 is responsible for m⁶A methylation of not only Pol II‐transcribed mRNAs and lincRNAs, but also promoter‐associated RNA (paRNA), enhancer RNA (eRNA), and RNA transcribed from transposable elements (repeats RNA), collectively termed chromosome‐associated regulatory RNAs (carRNAs). Approximately 15–30% of all carRNAs in ESCs were found to contain m⁶A. Depletion of METTL3 elevates the levels of these transcripts on chromatin. YTHDC1, which has previously been implicated in regulation of splicing, alternative polyadenylation, and export of mRNA, appears to mediate decay of some of these transcripts (Xiao et al, 2016; Roundtree et al, 2017; Kasowitz et al, 2018). Interestingly, depletion of METTL3 also leads to more open chromatin and a global upregulation of transcription. paRNA and eRNAs have been proposed to regulate transcription and indeed, genes with m⁶A‐methylated upstream paRNA or eRNA exhibited upregulated transcription after knockout of Mettl3. Moreover, site‐specific reduction of m⁶A methylation resulted in increased half‐life of carRNAs, upregulation of downstream gene transcription and elevated local H3K4me3 and H3K27ac levels, indicating that m⁶A on caRNAs alters transcription.

These results appear to be concordant with other studies reporting instances of RNA regulating chromatin accessibility and transcription. eRNAs can stimulate histone acetyltransferase activity of CBP/EP300, leading to increased H3K27ac levels and increased local chromatin accessibility (Bose et al, 2017). In addition, transcriptional activators such as YY1 can bind transcribed RNA and promote local transcription (Sigova et al, 2015). Indeed, YY1 is one of the top enriched TFs around m⁶A‐marked carRNAs, and ChIP‐seq of EP300 and YY1 in Mettl3 knockout and control mESCs revealed a global increase in binding for both proteins in Mettl3 knockout mESCs compared to controls. Several groups have previously shown that RNA transcripts can repel PRC2 and prevent its deposition on chromatin in order to maintain chromatin accessibility (Kaneko et al, 2014; Beltran et al, 2016; Wang et al, 2017). Intriguingly, ChIP‐seq of JARID2, a component of PRC2 complex, revealed a global decrease of JARID2 binding in Mettl3 knockout mESCs, which correlates with an increased abundance of eRNA and repeats transcripts. Overall, m⁶A methylation of carRNAs can control carRNA stability and downstream transcription, thus representing a distinct mechanism by which m⁶A impacts gene expression (Fig 4).

Figure 4. m⁶A on chromosome‐associated regulatory RNA regulates chromatin state and transcription.

m⁶A methylation of non‐coding eRNA, paRNA, and LINE element RNA promotes their decay, leading to reduction in active histone marks such as H3K27ac and H3K4me3 and downregulation of transcription (top). In Mettl3 KO cells, these non‐coding transcripts are stabilized resulting in increased deposition of H3K27ac and H3K4me3 and upregulated transcription of associated mRNAs (bottom).

Relating back to the discussion in the previous section, we suspect that effects of m⁶A on carRNA could also be context‐dependent. We characterized a carRNA decay pathway through YTHDC1 (Liu et al, 2020a), but other m⁶A readers may exist to compete with YTHDC1 and affect the fate of carRNA differently in other biological processes. In addition, one must also consider modulation of chromatin modifiers through mRNA m⁶A regulation. In specific cellular systems, such as certain cancer cells, chromatin modulation could be dominated through mRNA m⁶A regulation. The effects from both carRNA and mRNA should be examined when dissecting m⁶A regulatory pathways in the future.

Reversing mRNA methylation: the diverse substrates of FTO

An important aspect of m⁶A methylation is its reversibility, allowing for regulation of m⁶A levels following initial deposition. Two RNA m⁶A demethylases have been identified, FTO and ALKBH5. Both are iron‐ and α‐ketoglutarate‐dependent dioxygenases that mediate oxidative demethylation of m⁶A. ALKBH5 is a m⁶A demethylase and plays critical roles in spermatogenesis, immunity, and cancer progression, while FTO has been reported to demethylate m⁶A and m⁶A_m on mRNA and m¹A on tRNA (Wei et al, 2018). Here we discuss recent reports of diverse substrates for FTO as well as the debate regarding the relative functional importance of FTO activity on various substrates.

In 2011, it was reported that FTO demethylates m⁶A on polyadenylated RNA, which has been confirmed by several subsequent studies. Since these initial reports, other substrates have since been reported and FTO has been shown to demethylate m⁶A, m⁶A_m on mRNA, m⁶A_m on U2 snRNA, and m¹A on tRNA (Mauer et al, 2017, 2019; Wei et al, 2018). The substrate preferences for FTO appear to vary based on its subcellular localization—the demethylation activity of FTO on m⁶A is more pronounced in the nucleus, while its demethylation activity on m⁶A_m is more pronounced in the cytoplasm (Wei et al, 2018). FTO has been reported to play roles in several steps of gene expression, from pre‐mRNA processing steps, such as alternative splicing and alternative polyadenylation, to translation (Yu et al, 2018). Due to the diversity of substrates, the relative functional importance of its activity on each of its various substrates has been debated. Functional roles of m⁶A demethylation have been clearly demonstrated, as summarized in Wei et al (2018). Notably, in certain AML cells, FTO mediates up to 40% of demethylation of all cellular mRNA m⁶A (Wei et al, 2018).

Care must be taken when dissecting the contributions of the various activities of FTO; for example, both its demethylation activity on mRNA and snRNA could likely contribute to its role in the regulation of splicing. Comparison of the phenotypes of relevant mouse models may be informative for this question. FTO is critical for development: most knockout mice die at the embryo stage or within the first month of birth; those that survive tended to lose body weight and were smaller than control mice (Fischer et al, 2009, unpublished). However, knockout of Pcif1 or Mettl4, the methyltransferases that mediate cap mRNA and U2 snRNA m⁶A_m methylation, respectively, exhibit mild phenotypes and do not affect mouse viability and fertility (Kweon et al, 2019; Pandey et al, 2020). This suggests that m⁶A_m methylation on mRNA or U2 snRNA is not broadly required for embryonic development. In contrast, Mettl3 and Mettl14 knockout mice die at a very early embryo stage. These mice exhibit a more severe developmental phenotype than Fto KO mice, while Fto KO mice show much more severe developmental phenotypes than Pcif1 or Mettl4 KO mice. The minimal phenotypic impact of loss of m⁶A_m on mRNA and U2 snRNA suggests that the role of FTO in regulating levels of m⁶A_m may not account for the strong Fto KO phenotype during development. However, alternative possibilities cannot be excluded. It is possible that, while hypomethylation of m⁶A_m on these substrates results in mild phenotypes, hypermethylation of these substrates upon loss of FTO causes more severe defects. Moving forward, precise dissection of the functional importance of FTO's activity on individual substrates would shed light on the molecular basis of the severe FTO knockout phenotype. Both m⁶A and m⁶A_m demethylation by FTO may exert regulatory roles and it appears likely that the major development phenotypes associated with FTO loss derive from demethylation of a portion of m⁶A installed by METTL3. In contrast, m⁶A_m demethylation may fine‐tune specific stress responses, as indicated by the mild phenotype observed so far with knockout of corresponding methyltransferases.

The role of m⁶A in human disease pathology

Studies that implicate m⁶A in the pathogenesis of a variety of diseases have emerged at a rapid pace in recent years. The role of m⁶A in disease has been most extensively studied in the context of cancer, and important roles for METTL3/METTL14 in a variety of cancer types have been reported. Analysis of somatic mutations in cancer genomes from TCGA exome sequencing data revealed METTL3 and METTL14 as potential tumor suppressor genes in bladder and uterine cancer, respectively (Zhao et al, 2019). Further, METTL14 loss‐of‐function hotspot mutations have been observed in endometrial cancer. Loss of METTL14 increases the proliferation and tumorigenicity of endometrial cancer cells by altering the mRNA stability and translation of AKT pathway regulators (Liu et al, 2018). m⁶A has also been reported to play tumor suppressor roles in renal cell carcinoma (Li et al, 2017b). Conversely, METTL3 and METTL14 have also been reported to act as oncogenes in other cancer types. METTL3 and METTL14 promote acute myeloid leukemia (AML) oncogenesis by increasing the mRNA stability and translation of m⁶A‐marked oncogenes such as MYC (Barbieri et al, 2017; Vu et al, 2017; Weng et al, 2018a). METTL3/METTL14 have also been found to play oncogenic roles in glioblastoma (Cui et al, 2017). In other cases, such as in hepatocellular carcinoma (HCC), the role of METTL3 and METTL14 is unclear due to contrasting findings regarding whether the m⁶A methyltransferase complex promotes or inhibits cancer pathogenesis (Ma et al, 2017; Chen et al, 2018). Due to the oncogenic roles of m⁶A in certain cancer types, METTL3/METTL14 have been suggested to be promising drug targets. Several biotech companies have stated a primary focus on developing small molecule inhibitors for METTL3/METTL14 for testing in oncological indications, with phase 1 trials planned for 2021–2022 (Cully, 2019).

Beyond cancer, METTL3/METTL14 have also been reported to play roles in a variety of other diseases, including heart failure, viral infection, and type 2 diabetes (T2D) (Gokhale & Horner, 2017; De Jesus et al, 2019; Yang et al, 2019b; Berulava et al, 2020). Specifically, m⁶A affects multiple aspects of viral infection, both through modification of viral RNA and its presence on host RNA. METTL3/METTL14‐mediated methylation of viral RNA exerts a positive, neutral, or negative effect on infection, depending on the virus (Williams et al, 2019). m⁶A on host RNA appears to suppress the anti‐viral interferon response, as METTL3 or METTL14 depletion leads to an increase in IFNβ expression (Rubio et al, 2018; Winkler et al, 2019). New evidence in the T2D field suggests that m⁶A dysregulation may be involved in T2D pathogenesis. A recent report (De Jesus et al, 2019) concluded that mRNA m⁶A marks could segregate samples by disease status significantly better than RNA sequencing data, suggesting a potential relationship between m⁶A methylation and type 2 diabetes. Depletion of METTL3 or METTL14 in a human beta‐cell line resulted in cell cycle arrest and impaired insulin secretin, and beta‐cell‐specific Mettl14 knockout mice exhibit early diabetes onset and mortality due to decreased beta‐cell proliferation and insulin degranulation. Investigation into the therapeutic potential of targeting regulators of m⁶A in T2D may hold promise going forward.

In addition to METTL3/METTL14, the FTO and ALKBH5 demethylases have also been implicated in disease. FTO is upregulated in a subset of AML and plays oncogenic roles (Li et al, 2017c; Su et al, 2018), including the gain of resistance to tyrosine kinase inhibitor therapy (Yan et al, 2018). A small molecule inhibitor that inhibits the demethylation activity of FTO has been shown to suppress proliferation and promote differentiation of human AML cells (Huang et al, 2019b). Knockdown of FTO sensitizes melanoma to anti‐PD‐1 treatment in mice, indicating that FTO may promote melanoma tumorigenesis and anti‐PD‐1 resistance (Yang et al, 2019a). Additionally, decreased FTO expression has been associated with mammalian heart failure, with FTO overexpression in failing mouse hearts reducing ischemia‐induced loss of cardiac function (Mathiyalagan et al, 2019; Berulava et al, 2020). Moreover, ALKBH5 appears to promote tumorigenesis in breast cancer and glioblastoma (Zhang et al, 2016, 2017b) in addition to affecting innate immune responses to viral infections (Zheng et al, 2017; Liu et al, 2019).

Similarly, YTHDF reader proteins have also been linked to disease processes. YTHDF2 is essential for AML initiation and leukemia stem cell development, and loss of YTHDF2 impairs AML cell survival and engraftment, while also promoting stem or primitive progenitor cell expansion and enhancing their reconstitution and myeloid differentiation potentials (Wang et al, 2018b; Paris et al, 2019). Suppression of YTHDF2 promotes hematopoietic stem cell ex vivo expansion more than eightfold (Li et al, 2018b), making it a potential target in bone marrow transplant treatment of leukemia or other mutation‐based hematologic diseases. YTHDF1, on the other hand, plays a critical role in regulating dendritic cell activity in the response to cancer immunotherapy, with knockout of Ythdf1 in mice enhancing anti‐tumor responses and response to PD‐L1 checkpoint blockade (Han et al, 2019). Thus, YTHDF1 may represent a potential therapeutic target to enhance anti‐tumor immune responses.

In summary, m⁶A writers, readers, and erasers have all been implicated in disease processes. These findings further highlight the importance of m⁶A as a gene regulatory system not only for normal physiology, but also in disease states. Overall, the involvement of m⁶A in the pathogenesis of a variety of human diseases suggests that therapeutic targeting of this pathway may be a viable treatment strategy.

Summary and outlook

It is well established that m⁶A‐mediated regulation of gene expression is essential in numerous biological processes, crucial for both normal physiology and in pathophysiological states. Recent advances have highlighted the diverse mechanisms through which m⁶A regulates gene expression, both on mRNA and non‐coding RNA species. Looking toward the future, important directions include elucidating fundamental mechanisms by which m⁶A regulates gene expression and exploring ways that m⁶A can be targeted for the treatment of disease. In addition, further unraveling the mechanisms that regulate m⁶A epitranscriptome specificity will enhance our understanding of the nature of m⁶A as a regulatory system, and shed light on how m⁶A deposition interacts with other cellular processes. Further characterization of the role of m⁶A on non‐coding chromosome‐associated RNA may shed light on the multifaceted role of m⁶A in gene regulation. We hypothesize that other nuclear m⁶A readers may differentially regulate caRNAs by mechanisms distinct from decay by YTHDC1, analogously to the diverse set of readers that bind m⁶A on mRNA. Finally, exploration of the therapeutic potential of targeting m⁶A writers, readers, and erasers in disease may reveal promising novel approaches for treating disease.

Conflict of interest

C.H. is a scientific founder and a member of the scientific advisory board of Accent Therapeutics, Inc.

The EMBO Journal (2021) 40: e105977.