Marking RNA: m6A writers, readers, and functions in Arabidopsis.

N6-methyladenosine (m6A) emerges as an important modification in eukaryotic mRNAs. m6A has first been reported in 1974, and its functional significance in mammalian gene regulation and importance for proper development have been well established. An arsenal of writer, eraser, and reader proteins accomplish deposition, removal, and interpretation of the m6A mark, resulting in dynamic function. This led to the concept of an epitranscriptome, the compendium of RNA species with chemical modification of the nucleobases in the cell, in analogy to the epigenome. While m6A has long been known to also exist in plant mRNAs, proteins involved in m6A metabolism have only recently been detected by mutant analysis, homology search, and mRNA interactome capture in the reference plant Arabidopsis thaliana. Dysregulation of the m6A modification causes severe developmental abnormalities of leaves and roots and altered timing of reproductive development. Furthermore, m6A modification affects viral infection. Here, we discuss recent progress in identifying m6A sites transcriptome-wide, in identifying the molecular players involved in writing, removing, and reading the mark, and in assigning functions to this RNA modification in A. thaliana. We highlight similarities and differences to m6A modification in mammals and provide an outlook on important questions that remain to be addressed.

Introduction

Noncoding RNAs (ncRNAs) including ribosomal RNAs (rRNAs), transfer RNAs (tRNAs), spliceosomal small nuclear RNAs (snRNAs), and small nucleolar RNAs (snoRNAs) undergo extensive modification at the posttranscriptional level. Among more than hundred different nucleotide variants, methylation of adenosine at N6 (m6A) is a predominant type of internal covalent RNA modification (Fray and Simpson, 2015; Meyer and Jaffrey, 2017). In the early 1970s, m6A modifications have been also found in mRNAs, both in mammals and plants. Nevertheless, they have only recently gained attention with respect to their function in regulating gene expression. The m6A mark is deposited by dedicated proteins, the ‘writers’, is interpreted by ‘readers’, and can be removed by enzymatic activities, the ‘erasers’ (Meyer and Jaffrey, 2017; Balacco and Soller, 2019).

In higher plants, the detection of m6A in the transcriptome of Arabidopsis thaliana and the prominent developmental abnormalities observed in mutants with altered m6A levels has boosted the interest in this modification in the past few years. Exploiting the genetic resources and applying state-of-the-art high-throughput approaches in this model plant has led to novel insights into the m6A methylome, to the identification of molecular players in m6A metabolism (Table 1), and to insights into the function this modification plays in plant development and responses to environmental threats (Fray and Simpson, 2015; Burgess et al., 2016; Bhat et al., 2018; Kramer et al., 2018).

Table 1

Orthologues of mammalian m6A writers, readers, and erasers in Arabidopsis thaliana.

Arabidopsis	Orthologues in mammals	Phenotype of Arabidopsis loss-of-function mutants	References
Writer complex
MTA	METTL3	Defective embryogenesis, abnormal flower morphology in hypomorphic adult plants	Zhong et al. (2008), Bodi et al. (2012), Vespa et al. (2004)
MTB	METTL14		Ruzicka et al. (2017)
FIP37	WTAP	Defective embryogenesis, overproliferation of stem cells in shoot apical meristem in hypomorphic adult plants	Zhong et al. (2008), Shen et al. (2016)
VIRILIZER	VIRMA/KIAA1429	Aberrant formation of lateral roots and root cap, aberrant development of cotyledons	Ruzicka et al. (2017)
HAKAI	HAKAI/Casitas B-lineage lymphoma-transforming sequence-like protein 1 (CBLL-1)/Cbl proto-oncogene like 1	Aphenotypic	Ruzicka et al. (2017)
Sequence not detected	Flacc/ZC3H13		Balacco and Soller (2019)
Readers
ECT2	YTHDF1/2/3	Increased trichome branching, delayed leaf initiation	Arribas-Hernández et al. (2018), Scutenaire et al. (2018), Wei et al. (2018)
ECT3	YTHDF1/2/3	Increased trichome branching, delayed leaf initiation	Arribas-Hernández et al. (2018), Scutenaire et al. (2018), Wei et al. (2018)
ECT4	YTHDF1/2/3	Delayed leaf initiation	Arribas-Hernández et al. (2018)
Eraser
atALKBH9B	AlkB5	Impaired AMV infection	Martinez-Perez et al. (2017)
atALKBH10B	AlkB5	Late flowering, reduced growth rate of leaves	Duan et al. (2017)
Sequence not detected	FTO		Balacco and Soller (2019)
N6-mAMP deaminase
AtADAL/MAPDA	HsADAL	Slight reduction in root growth	Chen et al. (2018)

The m6A methylome

An experimental hurdle for studying the m6A distribution transcriptome-wide is the fact that it cannot be identified by conventional cDNA sequencing; in contrast to C to U or A to I conversions through RNA editing, adenosine methylation at the N6 position still leads to T incorporation during reverse transcription, because the methyl group does not reside at the Watson–Crick base-pairing edge. Furthermore, the chemical features of adenosine and N6-methylated adenosine are very similar, precluding chemical modification strategies to detect m6A at single nucleotide resolution, as done, for example, for 5-methylcytosine (Squires et al., 2012). Rather, transcriptome-wide identification of m6A sites has been possible through RNA immunoprecipitation using antibodies specific for m6A, followed by RNA-seq of the co-precipitated RNAs, a method designated m6A-seq (Dominissini et al., 2012) or m6A-specific methylated RNA immunoprecipitation (MeRIP)-seq (Meyer et al., 2012), respectively. m6A sites are determined with a resolution of about 200 nucleotides. One m6A peak was found about every 2000 nucleotides in mammals. Thus, if m6A marks would be distributed equally, a transcript would have on average 1.7 peaks (Dominissini et al., 2012). This correlates well with early estimates of three m6A marks per transcript obtained by determining the proportion of m6A among all nucleotides in mouse cells (Perry et al., 1975).

More recently, a UV crosslinking step has been added to covalently bind the antibody to the m6A mark. Similar to photoactivatable ribonucleoside-enhanced crosslinking and immunoprecipitation crosslinking (PAR-CLIP) detecting in vivo RNA–protein interaction sites genome-wide in human embryonic kidney 293 HEK293 cells (Hafner et al., 2010), HeLa cells were fed with 4-thiouridine (4-SU) with the rationale that some 4-SU is incorporated close to m6A sites (Chen et al., 2015). 4SU-containing mRNA is then purified by oligo (dT) capture and subjected to precipitation with the m6A antibody before irradiation with 365 nm UV light in this photo crosslinking-assisted (PA)-m6A-seq (Chen et al., 2015). After limited digestion of the crosslinked RNA and proteinase K digestion of the antibody, libraries are constructed. Because the crosslinked 4-SU is read as C during reverse transcription, resulting mutations are indicative of a close-by m6A site, increasing the resolution to about 30 nucleotides compared to m6A-seq and MeRIP-seq.

Another approach was to modify the HITS-CLIP technique that uses 254 nm UV light to crosslink RNA and bound proteins in vivo (Ule et al., 2003). In m6A-CLIP, RNA is fragmented before incubation with the m6A antibody. UV irradiation leads to crosslinking of the antibody at the m6A site. After precipitation of the m6A containing oligonucleotide and proteinase K digestion of the antibody, the residual peptide at the crosslink site leads to mutations or truncations during reverse transcription that are diagnostic for the modification site (Ke et al., 2015).

A similar approach employing m6A precipitation and UV crosslinking subsequently generated the sequencing libraries according to the individual resolution nucleotide crosslinking and immunoprecipitation (iCLIP) protocol developed for mapping RNA–protein interactions transcriptome-wide with single nucleotide resolution in mammalian cells (König et al., 2010; Müller-McNicoll et al., 2016). Due to intramolecular circularization and relinearization of the cDNAs, the antibody-induced truncations at the m6A site correspond to the nucleotide immediately upstream of the read. This approach is known as methyl iCLIP (miCLIP) (Linder et al., 2015).

These global studies in human and mouse cells showed that the m6A distribution on mRNAs is highly selective, with a preference for the 3′ untranslated region 3'UTR, the coding sequence and the region around the stop codon (Dominissini et al., 2012; Meyer et al., 2012). Recent results suggest that increasing m6A levels mark the early region of the last exon rather than the stop codon per se (Ke et al., 2015). In addition, m6A is found in ncRNAs and RNA viruses (Brocard et al., 2017). m6A sites are enriched in the previously established consensus sequence RRA*CH (R = G/A, H = A/C/U, * = methylation) (Wei and Moss, 1977; Harper et al., 1990; Dominissini et al., 2012).

An additional refinement of m6A probing aimed at determining the m6A methylome separately for three subcellular fractions, the chromatin-associated nascent pre-mRNAs, nucleoplasmic, and cytoplasmic mRNAs in HeLa cells (Ke et al., 2017). These data substantiated previous findings that m6A is mainly found in exons but rarely in introns, as chromatin-associated nascent RNA harbors many unspliced introns in contrast to poly(A) RNA usually employed for mapping of m6A. The m6A landscape in these fractions showed an overlap of ~90%, suggesting that the overall m6A patterns do not change much once the pre-mRNA is released from the chromatin into the nucleoplasm and following the export to the cytoplasm (Ke et al., 2017).

Importantly, the m6A methylome dynamically changes during development and in response to external stimuli. SCARLET (site-specific cleavage and radioactive-labeling followed by ligation-assisted extraction and thin-layer chromatography) allows determining the precise location and modification status of specific m6A sites. It has been estimated that the methylation status for a particular m6A site can vary between 6% and 80% and that many consensus motifs are not modified under particular conditions (Liu et al., 2013b).

In plants, m6A in mRNAs was first detected in maize (Nichols, 1979). Only much later was it also discovered in Arabidopsis poly(A) RNA using 2D thin-layer chromatography (Zhong et al., 2008; Bodi et al., 2012). It was found that m6A is not uniformly distributed along the RNAs but is enriched toward the end of the transcripts. More recently, m6A has been mapped transcriptome-wide in Arabidopsis using m6A-seq (Luo et al., 2014). Again, it is the most frequent internal modification of mRNAs, with 75% of transcripts harboring at least one m6A site. m6A sites were enriched around the stop codon and within 3′UTRs, as found in mammals. Additionally, m6A was enriched around the start codons in this study (Luo et al., 2014). Albeit, m6A sequencing of three different Arabidopsis organs, leaves, flowers, and roots found an enrichment predominantly near the stop codon and in the 3′UTR (Wan et al., 2015). Around 70% of the transcripts were modified by m6A with an average of 1.4 to 2 m6A sites per transcript. Above 75% of the fragments recovered by RNA immunoprecipitation (RIP) contained the consensus sequence RRA*CH described in mammals, with AAA*CU and AAA*CA being the most frequent motifs, consistent with previous findings (Luo et al., 2014). Most of the methylated mRNA displayed the typical topology found in Arabidopsis with one or two high peaks at the stop codon or in the 3′UTR and very low m6A signals in the coding regions (Wan et al., 2015).

A comparison among the three organs revealed that >80% of the m6A-modified transcripts were common between leaves, flowers, and roots. One third of the transcripts showed differential m6A methylation among the organs whereas only one fourth showed differences in steady-state levels. Moreover, transcripts with particularly high methylation in one organ encode unique functions for this organ, e.g. photosynthesis, carbohydrate, and nitrogen metabolism in leaves, RNA degradation pathways, DNA synthesis, and protein synthesis in flowers, as well as alkaloid biosynthesis and carbonate metabolism in roots (Wan et al., 2015). This suggests that m6A methylation makes an important contribution to organ differentiation.

A direct comparison of the m6A profile has been performed for two Arabidopsis ecotypes, Can-0 and Hen-16 (Luo et al., 2014). Despite a substantial conservation in the patterns between the ecotypes, a suite of the common m6A peaks showed ecotype-specific changes in intensity in addition to peaks found exclusively in one of the ecotypes (Luo et al., 2014). In addition, m6A was also identified in the genomes of alfalfa mosaic virus and cucumber mosaic virus (see below) (Martinez-Perez et al., 2017).

Writers

In mammals, a high molecular weight protein complex of ~1 MDa, the methylosome, is responsible for deposition of the m6A mark (Figure 1) (Bokar et al., 1994; Liu et al., 2013a). This complex comprises methyltransferase-like protein 3 (METTL3), the active adenosine methyltransferase that binds the cofactor S-adenosyl methionine (SAM) (Bokar et al., 1997). METTL3 interacts with methyltransferase-like 14 (METTL14). A crystal structure of the METTL3–METTL14 heterodimer loaded with SAM identified the residues D337, D395, N539, and E532 as being critical for the enzymatic activity of METTL3 (Wang et al., 2016). In metazoa, D395, N539, and E532 are not conserved in METTL14. This may correlate with the observation that human METTL14 does not methylate N6A in vitro but may rather act to structurally support METTL3 (Wang et al., 2016; Schöller et al., 2018).

Figure 1

Schematic overview of the m6A machinery in plants and mammals. m6A is deposited on RNAs by ‘writers’, removed by ‘erasers’, and interpreted by ‘readers’. m6A RNA methylation is involved in almost all steps of RNA metabolism including splicing, alternative polyadenylation, RNA export, RNA stability, translation, RNA structure, and miRNA regulation.

Additionally, the methylosome comprises a number of regulatory subunits. WILMS’ TUMOR 1-ASSOCIATING PROTEIN (WTAP) stabilizes the interaction between METTL3 and METTL14 (Liu et al., 2013a; Ping et al., 2014; Schwartz et al., 2014). Moreover, it localizes METTL3 and METTL14 to nuclear speckles, which serve as a reservoir for splicing factors (Ping et al., 2014). Virilizer-like m6A methyltransferase associated protein (VIRMA/KIAA1429), a mammalian homolog of Drosophila Virilizer involved in splicing of a sex-determination factor, was identified through its interaction with METTL3 (Schwartz et al., 2014). The VIRMA N-terminus binds to the WTAP−METTL3−METTL14 complex via WTAP in an RNA-independent manner in HeLa cells (Yue et al., 2018), but how VIRMA is required for m6A methylation is unknown.

Similarly, RNA-binding motif protein 15 (RBM15) interacts with METTL3 dependent on WTAP in HEK293T cells (Patil et al., 2016). RBM15 belongs to the split end protein (Spen) family with three RNA-recognition motifs at the N-terminus and a Spen paralogues and orthologues C-terminal (SPOC) domain that engages in protein–protein interaction and has been suggested to function as an adapter protein recruiting the m6A methylosome to specific regions within transcripts (Patil et al., 2016). Another auxiliary subunit is Fl(2)d-associated complex component (Flacc) in Drosophila, the counterpart of mammalian Zinc finger CCCH domain-containing protein 13 (ZC3H13), an animal-specific protein that bridges WTAP and RBM15 (Balacco and Soller, 2019; Knuckles et al., 2018).

In Arabidopsis, orthologues of several methylosome subunits have been identified and shown to interact with each other (Figure 1). Inactivation of METHYLTRANSFERASE A (MTA), the orthologue of METTL3, led to embryo-lethality and reduced m6A levels in transcripts of the arrested seeds (Zhong et al., 2008). Complementation of the mta mutant by MTA driven by the seed-specific ABSCISSIC ACID INSENSITIVE promoter allows overcoming the defect in embryo development. Reduced levels of MTA in the resulting hypomorphic adult plants lead to abnormal flower architecture and trichomes with a higher number of branches (Bodi et al., 2012).

MTA interacts in vitro and in vivo with A. thaliana FKBP12 INTERACTING PROTEIN 37 (FIP37), a homolog of WTAP (Zhong et al., 2008). Subsequently, WTAP was also found in the METTL3 complex in humans (Liu et al., 2013a; Ping et al., 2014). Fip37-4 mutants have only 10% of the m6A level and are embryo-lethal, similar to mta mutants. Again, complementation with FIP37 expressed from the embryo-specific LEAFY COTYLEDON promoter allowed to rescue adult plants (Shen et al., 2016). The very low FIP37 expression in these plants led to massive proliferation of the stem cells in the shoot apical meristem that delivers aerial parts throughout the lifespan of the plant. This correlated with loss of m6A marks in the mRNA for two stem cell regulators, WUSCHEL (WUS) and SHOOT MERISTEMLESS (STM), and an increase in the domain expressing WUS. Notably, the m6A peak at the WUS and STM stop codons was reduced in fip37-4 LEC1:FIP37, correlating with reduced RNA degradation. RIP showed that FIP37 interacts with both WUS and STM transcripts in the shoot apex, pointing to a scenario where FIP37 in vivo binding promotes distinct m6A modification with concomitant limitation of mRNA half-life. Elevated levels of FIP37 lead to a higher number of trichomes with supernumerary branches, similar to what was observed for mutants with reduced MTA levels (Vespa et al., 2004). This may indicate that the m6A level needs to be precisely balanced, or alternatively, an excess of FIP37 has a dominant negative effect, perhaps interfering with methylosome assembly or function.

A screen for regulators of Arabidopsis vascular development identified a protein with homology to VIRMA/KIAA1429 involved in m6A formation in mammals (Schwartz et al., 2014). In the Arabidopsis vir-1 mutant, m6A levels were reduced to ~10% and the mutant showed aberrant formation of lateral roots and root caps as well as aberrant cotyledon development (Ruzicka et al., 2017). A proteomics search for VIR-interacting proteins again identified FIP37 in cell suspension cultures. In addition, METHYLTRANSFERASE B (MTB) was recovered, which is an orthologue of human METTL14 (Ruzicka et al., 2017). As mentioned above, METTL14 lacks residues critical for enzymatic activity. Notably, these residues are conserved in Arabidopsis and other plants, suggesting that MTB may display enzymatic activity in some plants (Balacco and Soller, 2019). In an inducible MTB RNAi line m6A levels were reduced to 50% (Ruzicka et al., 2017).

An additional component of the Arabidopsis writer complex was HAKAI, the orthologue of an E3 ubiquitin ligase (Ruzicka et al., 2017). Although m6A levels are reduced to 35% in the hakai mutants, there are no obvious phenotypes. In mammals, Hakai interacts with WTAP, and knockdown in HeLa cells leads to reduction of m6A levels by 23% (Yue et al., 2018). Flacc homologs have not been identified in Arabidopsis so far (Balacco and Soller, 2019). The Arabidopsis homolog of RBM15 is FPA, which regulates flowering time by RNA-mediated chromatin silencing of the floral repressor FLOWERING LOCUS C (FLC) (Bäurle et al., 2007). FPA controls FLC transcription by mediating alternative polyadenylation of embedded noncoding antisense RNAs, which leads to downregulation of FLC transcription (Hornyik et al., 2010). However, a role of FPA in m6A RNA methylation has not yet been demonstrated.

In mammals, apart from the METTL3−METTL14 writer complex, which is the major methyltransferase enzyme acting on polyadenylated mRNA, METTL16 can also methylate mRNA as well as U6 snRNA and various lncRNAs in humans (Warda et al., 2017; Pendleton et al., 2017). Interestingly, METTL16 is not a paralogue of METTL3 and METTL14 but part of a different methyltransferase protein family.

Readers

The m6A mark is decoded by reader proteins to mediate the downstream effects on posttranscriptional regulation (Figure 1). Best understood are YTH (YT512-B homology) domain proteins, which were identified as m6A-binding proteins in RNA-affinity chromatography using methylated RNA substrates as baits in mammals (Dominissini et al., 2012). The crystal structures of the human YTH domain revealed that three tryptophan residues, Trp411, Trp465, and Trp470, are crucial for m6A binding, forming a buried hydrophobic aromatic cage where the 6-methylamino group is accommodated (Li et al., 2014; Luo and Tong, 2014; Theler et al., 2014; Xu et al., 2014; Zhu et al., 2014). This pocket is conserved in animals and plants and discriminates between m6A and non-methylated mRNA with an increase in affinity of 20–50-fold. However, the YTH domain alone has low affinity for mRNA and needs additional low complexity protein regions, which assist in mRNA binding.

The YTH domain proteins are classified in two clades, the nuclear YTHDC proteins and YTHDF proteins that are mainly in the cytoplasm. Humans have two YTHDC proteins (YTHDC1 and YTHDC2) and three YTHDF proteins (YTHDF1, YTHDF2, and YTHDF3). YTHDC1 interacts with m6A in nuclear RNA to regulate pre-mRNA splicing and polyadenylation (Xiao et al., 2016; Kasowitz et al., 2018), while YTHDF1, YTHDF2, and YTHDF3 interact with mature mRNAs in the cytoplasm to affect their stability. A systematic homology search unveiled that the Arabidopsis genome harbors 13 YTH domain proteins, which have arisen from multiple duplication events. Due to their conserved C-terminal region they were termed EVOLUTIONARILY CONSERVED C-TERMINAL REGION (ECT). ECT11 belong to the DF clade and the other two belong to the DC clade (Arribas-Hernández et al., 2018; Scutenaire et al., 2018).

ECT proteins were initially associated with calcium signaling (Ok et al., 2005) and were only recently validated as RNA-binding proteins through mRNA interactome capture studies (Marondedze et al., 2016; Reichel et al., 2016; Köster et al., 2017). ECT2 binds m6A-containing RNA in vitro and in vivo, and binding is abolished by mutation of tryptophane residues of the aromatic cage (Scutenaire et al., 2018; Wei et al., 2018). Mutants defective in ECT2 share a particular phenotype with mutants defective in writer proteins, namely increased branching of trichomes (Arribas-Hernández et al., 2018; Scutenaire et al., 2018; Wei et al., 2018). Trichome cells have undergone endoreduplication and thus have an increased DNA content, with the number of trichome branches directly correlating to ploidy levels (Vespa et al., 2004). In the ect2 trichomes, the DNA content is elevated, indicating that the ect2 trichomes underwent extra rounds of replication without cell division (Scutenaire et al., 2018). The trichome branching phenotype is complemented by wild-type ECT2 but not a mutated version with the three aromatic cage tryptophanes changed, indicating that ECT2 binding to m6A underlies the phenotype (Arribas-Hernández et al., 2018). Mutants deficient in both, ECT2 and ECT3, also display a delayed leaf initiation rate from the shoot apical meristem (Arribas-Hernández et al., 2018).

Wei and colleagues developed a formaldehyde crosslinking and immunoprecipitation (FA-CLIP) strategy to identify ECT2-RNA interaction sites transcriptome-wide (Wei et al., 2018). They found 3680 transcripts with ECT2 binding sites that were strongly enriched in the 3′UTRs and identified a plant-specific ECT2 binding motif, URUAY. The motif was found 30 to 150 nucleotides upstream of poly(A) sites. In the ect2 mutant, the majority of ECT2 binding targets were expressed at lower levels in contrast to non-targets. This has been taken as evidence that ECT2 promotes m6A-dependent stability in contrast to mammalian YTHDF2 that promotes RNA degradation, albeit direct measurements of RNA stability for individual transcripts remain to be done. It should be kept in mind that formaldehyde leads not only to nucleic acid–protein crosslinks but also to protein–protein crosslinks; therefore, FA-CLIP is likely to also recover indirect targets.

A further YTH domain protein in Arabidopsis is CLEAVAGE AND POLYADENYLATION SPECIFICITY FACTOR 30 (AtCPSF30), which functions as part of a larger complex in mRNA 3′-end formation (Hunt et al., 2012). Analyses of cpsf30 mutants indicated roles in oxidative stress responses (Zhang et al., 2008), plant immunity, and programmed cell death (Bruggeman et al., 2014), apart from altered mRNA 3′-end cleavage site choice (Thomas et al., 2012). Arabidopsis CPSF30 gives rise to two protein variants through alternative polyadenylation: a shorter form of ~28 kDa that harbors three zinc finger domains and is homologous to yeast and mammalian CPSF30 and a longer form of ~70 kDa that additionally has a YTH domain and is unique to plants (Delaney et al., 2006; Hunt et al., 2012). Whether CPSF30 indeed interacts with m6A and whether there is a link to 3′-end formation remains to be determined.

Apart from YTH domain proteins, other RNA binding proteins have been suggested to function as m6A readers in mammals. METTL3, for instance, can act as an m6A reader, independently of its function as a writer (Lin et al., 2016). It was shown to bind m6A in the vicinity of the stop codon and engage in circularization of the mRNA through interaction with the eukaryotic initiation factor 3h (eIF3h), thus promoting translation (Choe et al., 2018).

Moreover, HNRNPA2B1 binds m6A-bearing RNAs in vivo and in vitro and its biochemical footprint matches the m6A consensus motif (Alarcón et al., 2015a). The K-homology domain containing insulin-like growth factor 2 mRNA-binding protein (IGF2BP) binds to thousands of mRNA transcripts through recognizing the consensus RRA*CH sequence and promotes their stability (Huang et al., 2018). Recently, proline rich coiled-coil 2 A (Prrc2a) was identified as another reader that stabilizes a transcript involved in the specification of oligodendrocytes in an m6A-dependent manner (Wu et al., 2019). The plethora of reader proteins without the dedicated binding cage suggests that additional proteins may be identified to interpret the m6A mark, either directly or indirectly.

Erasers

Proteins implicated in removal of the m6A methyl group belong to the family of AlkB homologous proteins comprising nonheme Fe (II) α-ketoglutarate-dependent dioxygenases (Figure 1) (Jia et al., 2011; Zheng et al., 2013). In animals, the protein family includes fat mass and obesity-associated protein (FTO) and ALKBH5 that act as RNA m6A demethylases. Overexpression or depletion of FTO led to subtle changes in m6A (Jia et al., 2011). In the brain of knockout mice, adenosine methylation increased in some mRNAs important for neuronal signaling (Hess et al., 2013). FTO is also highly expressed in acute myeloid leukemia and inhibits normal hematopoiesis by decreasing m6A levels in specific transcripts, leading to their reduced stability (Li et al., 2017b). Recently, it has been shown that FTO also demethylates m6Am, the first nucleotide of the 5′ cap (Mauer et al., 2017). FTO exhibits nearly 100 times greater catalytic activity against m6Am compared to m6A, leading to the speculation that FTO mostly acts on m6Am and only on a few specific m6A sites. Apart from the controversy about the physiological target of FTO, concerns have also been raised about the reversibility of m6A methylation in general, since Darnell and colleagues showed that m6A is retained in mRNA exons until the mRNA is degraded (Darnell et al., 2018). FTO is conserved among eukaryotes but absent in Arabidopsis (Balacco and Soller, 2019).

In the Arabidopsis genome, 13 ALKHB family proteins are predicted with diverse subcellular localization (Mielecki et al., 2012). Of five proteins with homology to ALKBH5 (Duan et al., 2017), atALKBH9B demethylates m6A in single-stranded RNA in vitro (Martinez-Perez et al., 2017). As mentioned above, m6A marks were found in plant RNA viruses. In atalkbh9b mutants, the genomic alfalfa mosaic virus RNAs show increased levels of m6A. This correlates with impaired accumulation of the virus and reduced spreading in infected Arabidopsis plants, suggesting that atALKBH9B demethylates m6A in vivo (Martinez-Perez et al., 2017).

In contrast, Duan et al. (2017) did not observe changes in the m6A/A ratio in atalkbh9b and atalkbh9c mutants. Rather, atALKBH10B was shown to demethylate m6A marks in vitro and in vivo, and atalkbh10b mutants had increased m6A levels (Duan et al., 2017). Overexpression of atALKBH10B leads to early flowering whereas atalkbh10b mutants flower later than wild-type plants. The delayed transition to reproductive development was corrected by wild-type atALKBH10B but not by a catalytically dead variant, atALKBH10BH366A/E368A. In the mutant, the floral integrator flowering locus T (FT) and two transcriptional activators of FT expression, SQUAMOSA PROMOTER BINDING PROTEIN-LIKE 3 (SPL3) and SPL9 mRNA showed reduced levels (Duan et al., 2017). This correlated with higher m6A levels around the FT start and stop codons and within the SPL3 and SPL9 3′UTRs and faster degradation in the mutant. Direct in vivo binding of atALKBH10B to FT, SPL3, and SPL9 indicates that atALKBH10B demethylates floral activators to control their half-life and consequently, accumulation. Thus, m6A emerges as yet another factor to be added to the regulatory network of floral transition (Srikanth and Schmid, 2011; Johansson and Staiger, 2015).

Impact of m6A on RNA processing steps

Alternative splicing

Early on, m6A has been linked to splicing, as changes in alternative splicing have been observed upon knockdown of METTL3, FTO, or ALK5B (Ping et al., 2014; Bartosovic et al., 2017; Tang et al., 2018). However, the number of targets affected varied widely in different cells studied (Martinez and Gilbert, 2018). Recently, transient N6 methyladenosine transcriptome sequencing (TNT-seq) on BrdU labeled nascent RNA and quantitative TNT pulse-chase sequencing were developed to assess the impact of m6A on splicing kinetics with high temporal resolution in HEK293 cells (Louloupi et al., 2018). These experiments showed that a significant fraction of m6A sites are deposited early near splice site junctions in exons and promote fast splicing, whereas m6A sites in introns are associated with slower splicing kinetics and alternative splicing (Louloupi et al., 2018).

The effect of m6A on alternative splicing can be mediated by YTHDC1, which interacts with the splicing factor SRSF3 to increase its ability to bind RNA and promotes exon inclusion. In contrast, YTHDC1 and SRSF3 block RNA binding of SRSF10, a factor that stimulates exon exclusion (Xiao et al., 2016). Furthermore, METTL16 promotes the expression of human MAT2A encoding SAM synthetase when SAM levels are low by enhanced splicing of a retained intron, resulting from binding to its target site in the MAT2A 3′UTR (Pendleton et al., 2017).

On the other hand, it was suggested that m6A is not prominently required for splicing regulation. Comparing the m6A methylome separately for three subcellular fractions, the chromatin-associated nascent pre-mRNAs, nucleoplasmic pre-mRNAs, and cytoplasmic mRNAs in HeLa cells revealed that only ∼10% of m6As in chromatin-associated nascent pre-mRNAs are within 50 nucleotides of 5′ or 3′ splice sites. Furthermore, the vast majority of exons harboring m6A in wild-type mouse stem cells is spliced the same in cells lacking METTL3 (Ke et al., 2017). Clearly, changes in overall m6A levels can also have indirect effects on splicing. A dedicated role of an m6A site in alternative splicing can be inferred from the analysis of splicing reporters with varying methylation levels of a defined site.

In Arabidopsis, a connection between m6A and splicing has not yet been investigated. In vir-1 mutants defective in the homolog of the Drosophila splicing factor VIR, no prominent changes in alternative splicing were detected (Ruzicka et al., 2017).

Alternative polyadenylation

The enrichment of m6A sites at the beginning of the last exon and 3′UTR indicated an involvement in the choice of polyadenylation sites (Ke et al., 2015). VIRMA provides a link between the writer complex and the polyadenylation machinery by recruiting the catalytic core components METTL3/METTL14/WTAP to the 3′UTR and interacting with the polyadenylation cleavage factors CPSF5 and CPSF6 (Yue et al., 2018). Of more than 2800 transcripts having their 3′UTR shortened upon CPSF5 knockdown, 84% have increased m6A peak density in the 3′UTR and near the stop codon.

YTHDC1 also interacts with CPSF6, and loss of YTHDC1 in mouse oocytes influences alternative polyadenylation apart from the effect on splicing (Kasowitz et al., 2018). In mouse male germ cells, knockout of the eraser protein ALKBH5 also results in aberrant splicing and in the production of longer 3′UTRs (Tang et al., 2018).

In Arabidopsis, one of the two CPSF30 protein variants harbors a YTH domain and is involved in processing of 3′-ends, which are enriched in m6A marks. Therefore, a role of m6A in governing 3′-end formation has been also been proposed for plants (Chakrabarti and Hunt, 2015; Fray and Simpson, 2015; Burgess et al., 2016), but not yet experimentally validated.

mRNA export

Several reader and writer proteins have been connected to mRNA export by the TREX:NXF1 pathway (Lesbirel et al., 2018). Once an mRNA has matured, TREX recruits the export receptor NXF1, which guides the mRNA through the nuclear pore to the cytoplasm. It was shown that the m6A writer complex associates with the TREX complex, and simultaneous knockdown of WTAP and VIRMA blocked export of a specific group of methylated transcripts (Lesbirel et al., 2018).

Interestingly, knockdown of RBM15 leads to cytoplasmic depletion and nuclear accumulation of mRNA (Uranishi et al., 2009; Zolotukhin et al., 2009). As RBM15 binds to NXF1, it may act as an export co-adaptor aiding in NXF1 loading on the mRNA. Furthermore, knockdown of METTL3 negatively affects the nuclear export of specific transcripts of circadian clock genes, resulting in long period circadian rhythms (Fustin et al., 2013).

Moreover, YTHDC1 was shown to interact with the splicing factor and nuclear export adaptor protein SRSF3, facilitating RNA binding to both SRSF3 and NXF1 (Roundtree et al., 2017). Knockdown of YTHDC1 or SRSF3 blocked nuclear export of a common set of transcripts suggesting the two proteins act in the same pathway. Thus, YTHDC1 was proposed to selectively mediate the export of m6A-containing mRNAs (Roundtree et al., 2017). Additionally, ALKBH5-deficient mice show increased levels of m6A containing transcripts in the nucleus leading to impaired fertility (Zheng et al., 2013).

mRNA stability

In mouse embryonic stem cells, knockdown of METTL3 and METTL14 and the resulting lack of m6A RNA methylation led to a loss of self-renewal capability. For many transcripts, including transcripts encoding developmental regulators, m6A methylation was inversely correlated with mRNA stability and gene expression (Wang et al., 2014b). Another study in HeLa and mouse embryonic stem cells also found that, upon knockout of METTL3, the half-lives of thousands of mRNAs increased at least 2-fold, indicating that mRNAs harboring m6As have shorter half-lives (Ke et al., 2017).

Accordingly, YTHDF2 destabilizes m6A-containing RNA by interacting with the CCR4-NOT complex in processing (P) bodies, which leads to poly(A) tail shortening and consequently to mRNA degradation (Wang et al., 2014a; Du et al., 2016). In addition, YTHDC2 acts as an adaptor recruiting the cytoplasmic 5′−3′ exonuclease Xrn1 via its ankyrin repeats on mRNA, promoting rapid degradation (Wojtas et al., 2017; Kretschmer et al., 2018). The YTHDC2 helicase activity is also essential for the decay of specific mitotic mRNAs for meiosis progression in mammalian germlines (Wojtas et al., 2017; Jain et al., 2018).

Recent evidence shows that m6A may also increase mRNA stability. As described above, FTO overexpression in acute myeloid leukemia decreases the stability of specific transcripts upon a decrease in m6A levels (Li et al., 2017b). Moreover, in contrast to the destabilizing function of YTHDF2, IGF2BP binding to mRNAs in an m6A-dependent manner promotes their stability (Huang et al., 2018).

So far, stabilizing effects have been reported for the m6A mark in Arabidopsis. Many m6A-modified mRNAs in Arabidopsis have reduced abundance in the absence of this mark. The decrease in abundance is due to transcript destabilization caused by cleavage occurring 4 or 5 nucleotides directly upstream of unmodified m6A sites (Anderson et al., 2018). Furthermore, ECT2 has been proposed to promote m6A-dependent stability of binding targets (Wei et al., 2018). In contrast, the analysis of fip37 mutants indicated that m6A levels are inversely correlated with RNA levels in WUS and STM in the shoot apical meristem (Shen et al., 2016). Similarly, in the atalkbh10b mutant higher m6A levels in FT and its regulators SPL3 and SPL9 correlate with faster degradation (Duan et al., 2017).

Translation

YTH domain proteins also play important roles in regulating translation. In the 5′UTR, m6A can be bound by the translation factor eIF3 to recruit the 43S pre-initiation complex internally and initiate cap-independent translation (Meyer et al., 2015). In contrast, m6A located near stop codons or in the 3′UTR is recognized by YTHDF1, which then interacts with eIF3 and other ribosome-associated proteins to stimulate cap-dependent ribosome loading (Wang et al., 2015). YTHDF3 interacts with YTHDF1 and has a synergistic effect on promoting translation by recruiting ribosomal proteins (Li et al., 2017a; Shi et al., 2017). METTL3 in the 3′UTR was also shown to interact with eIF3h bound to the translation start site to promote closed-loop conformation, stimulating translation through enhanced ribosome recycling (Lin et al., 2016; Choe et al., 2018).

m6A is also involved in translational control in response to heat shock. Upon heat stress, YTHDF2 relocates to the nucleus where it binds to m6A sites in the 5′UTR of stress-induced transcripts including HSP70, thereby preventing FTO from demethylation and promoting translation (Zhou et al., 2015). One study also reported relocation of Arabidopsis ECT2 to stress granules upon heat stress, suggesting that m6A might also play a role in plant stress response (Scutenaire et al., 2018).

In mammals, m6A can also have a negative effect on translation, as FTO promotes translation of mRNAs involved in neural development (Yu et al., 2018). Together, these findings suggest that m6A can affect translation via multiple mechanisms depending on cell type, developmental stage and cellular context.

m6A effects on RNA structure

As mentioned above, methylation of N6 does not affect the base pairing properties of adenine. Instead, m6A leads to reduced stability of RNA duplexes and thus altered RNA secondary structure (Kierzek and Kierzek, 2003). m6A in unpaired loop positions base-stacks stronger than the unmodified base, stabilizing the single-stranded regions (Roost et al., 2015). To determine the impact of m6A on RNA secondary structure globally, Roost and coworkers intersected data on in vivo RNA secondary structure determined for a human cell line with m6A peaks obtained by Me-RIP-seq. This revealed a tendency to single-stranded structure of the nucleotides adjacent to the m6A (Roost et al., 2015). Thus, in addition to direct interpretation of m6A by dedicated reader proteins, m6A can affect the accessibility for RNA-binding proteins indirectly by altering RNA secondary structure (Liu et al., 2015; Liu et al., 2017).

m6A in microRNAs

In mammals, METTL3 methylates pri-miRNAs to mark them for recognition by DGCR8 of the microprocessor complex. This modification promotes pri-miRNA processing (Alarcón et al., 2015b; Knuckles et al., 2017; Michlewski and Caceres, 2019). Human HNRNPA2B1 binds to the m6A mark in a subset of pri-miRNAs and interacts with DGCR8, which then recruits DROSHA for processing (Alarcón et al., 2015a). While a suite of RNA-binding proteins including hnRNP-like proteins have also been shown to affect pri-miRNA processing in Arabidopsis at the posttranscriptional level, it is not known whether this involves m6A (Dong et al., 2008; Ren et al., 2012; Ben Chaabane et al., 2013; Köster et al., 2014).

Demethylation also affects miRNA expression, as knockdown of FTO leads to aberrant miRNA steady-state levels (Berulava et al., 2015). m6A marks within 3′UTRs have been generally associated with the presence of miRNA binding sites, where about 2/3 of mRNAs containing an m6A site within their 3′UTR also have at least one miRNA binding site (Meyer et al., 2012; Chen et al., 2015)

Interestingly, miRNAs themselves positively regulate m6A deposition on mRNAs via a sequence-pairing mechanism (Chen et al., 2015). Manipulation of miRNA expression or sequences alters m6A levels through modulating binding of METTL3 to mRNAs containing miRNA targeting sites. In Arabidopsis, only ~ 1% of the 1000 most significant m6A peaks are in regions potentially targeted by miRNAs, suggesting that m6A is less likely to directly affect miRNA binding sites (Luo et al., 2014).

m6A in ncRNAs

In addition to mRNAs, m6A also occurs in a range of ncRNAs including U6 snRNA and lncRNAs. For instance, the lncRNA X-inactive specific transcript (XIST), which mediates silencing of the X-chromosome during female development in mammals, contains high levels of m6A (Patil et al., 2016). YTHDC1 recognizes m6A residues on XIST and is required for XIST function, while knockdown of RBM15 or METTL13 impairs X-mediated inactivation, indicating that m6A is required for XIST function (Patil et al., 2016). XIST was also found to interact with METLL16 (Warda et al., 2017), but the function of this interaction is still unknown. METLL16 also binds to the 3′-region of the cancer-associated ncRNA metastasis-associated lung adenocarcinoma transcript 1 (MALAT-1). The presence of m6A in MALAT-1 has been shown to alter RNA structure to facilitate binding of HNRNPC, a pre-mRNA processing protein (Liu et al., 2015).

In Arabidopsis, m6A marks have also been found on ~10% of tRNAs, all rRNAs, and many snRNAs and snoRNAs (Wan et al., 2015; Shen et al., 2016), but the function remains to be determined.

Catabolism of m6A degradation products

Although the methyl group of m6A can be removed by erasers, m6A levels also decrease due to RNA decay. Until recently, the fate of the resulting N6 methylated AMP (N6-mAMP) has remained enigmatic. Witte and co-workers now demonstrated the presence of an N6-mAMP deaminase in Arabidopsis (Chen et al., 2018). Removal of the N6-mAMP methyl group yields IMP. Because RNA polymerases are able to act on N6-mAMP, this demethylation step may prevent mis-incorporation of modified adenosine into nascent transcripts. Subsequently, N6-mAMP deaminase activity was also found in HeLa cells. Furthermore, the authors demonstrated that adenylate kinase strongly prefers AMP over N6-mAMP, providing another safeguard mechanism against untargeted incorporation of m6A (Chen et al., 2018).

Conclusions

Almost half a century after their first mention, m6A modifications in mRNAs have moved center stage in plant RNA biology. Significant progress has been made to describe m6A sites transcriptome-wide and identify the molecular players involved in installing, erasing, and interpreting the marks. Nevertheless, we only begin to appreciate the distribution of m6A marks in the transcriptome, the proteins involved in m6A metabolism, and the spectrum of processes affected by m6A in plants. The similar m6A consensus motifs in mammals and plants suggest that m6A methylation may be conserved but also distinct differences were found.

So far, the m6A methylome has been determined by m6A-seq in Arabidopsis (Luo et al., 2014; Wan et al., 2015; Shen et al., 2016). In mammals, the inherent limitation to resolve m6A peaks with a resolution of about 200 nucleotides has been overcome by incorporating a UV fixation step (Chen et al., 2015; Ke et al., 2015; Linder et al., 2015). In plants, UV irradiation has been successfully employed in Arabidopsis to crosslink RNA and protein in vivo despite previous reservations about the efficiency of UV crosslinking in the presence of UV absorbing pigments in plant tissues, (Zhang et al., 2015; Reichel et al., 2016; Meyer et al., 2017). Thus, it should be feasible to increase the resolution of m6A profiling also in plants. Moreover, the comparison between different reports is hampered by the fact that different facets of a protocol are applied, that the determination of m6A peaks is accomplished by different bioinformatics pipelines, and that the quality of the antibody can be quite variable (Schwartz et al., 2013; Linder et al., 2015). Overall, how the methylosome achieves selectivity for specific transcripts and defined sites within the transcripts remains to be clarified.

Initial experiments point to extensive remodeling of the m6A methylome during plant development and in response to stress (Wan et al., 2015; Anderson et al., 2018). To fully appreciate the regulatory potential, temporal and spatial changes have to be determined systematically across tissues and in response to abiotic and biotic factors which have a prominent effect on plants as sessile organisms.

Furthermore, there is a gap in understanding the molecular consequences of m6A modification on mRNA processing and function, e.g. how processing factors are recruited. The impact of m6A depends on the timing of deposition in the cell (Martinez and Gilbert, 2018). METTL3 associates with RNA polymerase II, linking the addition of the m6A mark to transcription (Slobodin et al., 2017). Deposition in nascent RNA can affect all downstream processing steps including RNA stability in the cytoplasm. A first investigation of the m6A landscape in separate chromatin-associated, nucleoplasmic and cytoplasmic fractions in HeLa cells showed an overlap of ~90% (Ke et al., 2017). However, methylation levels of distinct sites are likely to vary.

Whereas mammals have 5 YTF domain proteins, Arabidopsis harbors 13, raising the question about their function. For example, mutations of the close ECT2 homolog ECT4 enhanced the phenotype of ect2 ect3 (Arribas-Hernández et al., 2018). A detailed characterization of their divergent expression patterns and a description of phenotypes of higher order mutants will unveil possible specific and redundant functions of all ECTs. Although the m6A machinery is increasingly well characterized, novel proteins are still being discovered in mammals, such as the novel reader protein Prrc2a (Wu et al., 2019). This indicates that additional proteins may emerge as writers, readers, and erasers of the m6A mark in Arabidopsis. Of note, a plethora of proteins with the capability of binding RNA have been identified in three recent mRNA interactome capture experiments, many of which have not yet been assigned a function in RNA metabolism (Marondedze et al., 2016; Reichel et al., 2016; Zhang et al., 2016; Köster et al., 2017).