Occurrence and Functions of m⁶A and Other Covalent Modifications in Plant mRNA^,

Covalent mRNA modifications play crucial roles in gene regulation and are required for plant embryonic and postembryonic development.

Abstract

Posttranscriptional control of gene expression is indispensable for the execution of developmental programs and environmental adaptation. Among the many cellular mechanisms that regulate mRNA fate, covalent nucleotide modification has emerged as a major way of controlling the processing, localization, stability, and translatability of mRNAs. This powerful mechanism is conserved across eukaryotes and controls the cellular events that lead to development and growth. As in other eukaryotes, N⁶-methylation of adenosine is the most abundant and best studied mRNA modification in flowering plants. It is essential for embryonic and postembryonic plant development and it affects growth rate and stress responses, including susceptibility to plant RNA viruses. Although the mRNA modification field is young, the intense interest triggered by its involvement in stem cell differentiation and cancer has led to rapid advances in understanding how mRNA modifications control gene expression in mammalian systems. An equivalent effort from plant molecular biologists has been lagging behind, but recent work in Arabidopsis (Arabidopsis thaliana) and other plant species is starting to give insights into how this essential layer of posttranscriptional regulation works in plants, and both similarities and differences with other eukaryotes are emerging. In this Update, we summarize, connect, and evaluate the experimental work that supports our current knowledge of the biochemistry, molecular mechanisms, and biological functions of mRNA modifications in plants. We devote particular attention to N⁶-methylation of adenosine and attempt to place the knowledge gained from plant studies within the context of a more general framework derived from studies in other eukaryotes.

Control of gene expression is of paramount importance in biology, and a variety of molecular mechanisms working at the DNA, pre-mRNA, mRNA, and protein levels have evolved to ensure that appropriate levels of activity of gene products operate at any given time. It has long been recognized that chemical modification of the transcription template, either of the DNA itself or of associated histone proteins, is key to the regulation of gene expression. Thanks to rapid progress in the last decade, it has now become clear that chemical modification of mRNA also plays crucial roles in the control of endogenous gene activity and in shaping the outcome of host-virus interactions. This Update provides a brief overview of mRNA modifications identified in plants and their viruses, and an account of what is known of their functional relevance and the molecular mechanisms underlying their functions. We also include a discussion of emerging controversies and outstanding questions in this field.

SEEING MRNA MODIFICATIONS: HIGH-THROUGHPUT SEQUENCING DELIVERS THE BREAKTHROUGH

Although chemically modified nucleotides have been known to exist in RNA, even in mRNA, since more than 40 years (Holley et al., 1965; Desrosiers et al., 1974; Perry and Kelley, 1974; Boccaletto et al., 2018), the mRNA modification field has only taken off with the recent introduction of high-throughput sequencing-based methods to map modified nucleotides in specific mRNAs transcriptome wide (see Box 1 for a historical context of the development of the field). Three main categories of such methods enjoy widespread use (see Linder and Jaffrey [2019] for a detailed review focused on methods). (1) Antibodies specific to a modification are used to immunoprecipitate fragmented mRNA, and RNA sequencing (RNA-seq) is applied to input and immunoprecipitated RNA fractions. Significant enrichment then identifies mRNA intervals likely to contain a modified nucleotide (Dominissini et al., 2012; Meyer et al., 2012; Edelheit et al., 2013; Schwartz et al., 2013; Luo et al., 2014; Delatte et al., 2016; Dominissini et al., 2016; Li et al., 2016; Cui et al., 2017). (2) Modified nucleotides are specifically derivatized using unique reactivity with an appropriate compound, and the presence of the derived nucleotide is read either as a mutation signature or as a stop upon reverse transcription (Squires et al., 2012; Carlile et al., 2014; Lovejoy et al., 2014; Schwartz et al., 2014a; Li et al., 2015; David et al., 2017; Enroth et al., 2019; Sun et al., 2019; Zhang et al., 2019b). The cross-linking immunoprecipitation (CLIP) variant miCLIP constitutes an important special case in this category: a modification-specific antibody is UV cross-linked to purified RNA, allowing mutation-dependent mapping of cross-link sites (Ke et al., 2015; Linder et al., 2015) as in other CLIP techniques (Ule et al., 2018). (3) Differential RNA cleavage at modified versus unmodified nucleotides is used to identify modified sites based on the percentage of sequence read-throughs and stops at each nucleotide (Birkedal et al., 2015; Garcia-Campos et al., 2019; Zhang et al., 2019d). In addition to those three types of methods, direct RNA sequencing by Oxford Nanopore Technology (Garalde et al., 2018) has recently been used successfully to map N⁶-methyladenosine (m⁶A) sites based on systematic base-calling errors at modified sites (Liu et al., 2019; Parker et al., 2019; Yang et al., 2019). In general, methods in categories 2 and 3 produce modification maps with nucleotide resolution, while methods in category 1 produce intervals containing a modification. It remains an active field of research to develop methods to detect modifications for two reasons. First, new methods for the detection of known RNA modifications are in high demand, because no single method developed thus far perfectly combines the desired sensitivity and specificity: high-confidence modification sites should ideally rely on results obtained by two orthogonal methods (Zhang et al., 2019b). Indeed, antibody-based methods in categories 1 and 2 have been used since 2012 to detect m⁶A sites (Dominissini et al., 2012; Meyer et al., 2012; Schwartz et al., 2013; Luo et al., 2014; Ke et al., 2015; Linder et al., 2015), but the recent development of RNase cleavage-dependent m⁶A mapping (Garcia-Campos et al., 2019; Zhang et al., 2019d) shows that the limited sensitivity of antibody-based methods led to substantial underestimation of the number of m⁶A sites in the transcriptome. Second, it is possible that some mRNA modifications await discovery, thus necessitating the development of methods for their detection.

MRNA MODIFICATIONS IDENTIFIED IN PLANTS AND OTHER EUKARYOTES

In plants, there are now reports on mapping of three distinct covalent nucleotide modifications in the bodies of mRNA: m⁶A (Li et al., 2014c; Luo et al., 2014; Wan et al., 2015; Shen et al., 2016; Duan et al., 2017; Anderson et al., 2018; Miao et al., 2019; Parker et al., 2019), 5-methylcytidine (m⁵C; Cui et al., 2017; David et al., 2017; Yang et al., 2019), and pseudouridine (Ψ; Sun et al., 2019). Other types of mRNA modifications, including C-U editing of mitochondrial and plastidial mRNAs (Shikanai, 2006), alternative 5′-caps (Kiledjian, 2018; Wang et al., 2019), and untemplated addition of nucleotides to mRNA 3′-ends (De Almeida et al., 2018), will not be covered here. The discovery of mRNA modifications in plants is lagging behind that in animal cells, in which at least six additional modifications have been mapped and, in some cases, also functionally analyzed (Boccaletto et al., 2018). These modified nucleotides include inosine (I; Bass and Weintraub, 1988; Shevchenko and Morris, 2018), internal (as opposed to cap) 7-methylguanosine (m⁷G; Zhang et al., 2019b), cap-proximal 2′-O,N⁶-dimethyladenosine (m⁶A_m; Wei et al., 1975a; Linder et al., 2015; Mauer et al., 2017; Boulias et al., 2019), N¹-methyladenosine (m¹A; Dominissini et al., 2016; Li et al., 2016, 2017b; Safra et al., 2017), 4-acetylcytidine (ac4C; Arango et al., 2018), 5-hydroxymethylcytidine (hm⁵C; Delatte et al., 2016), and 2′-O-methylation (any nucleotide [Nm]; Furuichi et al., 1975; Wei et al., 1975b; Dai et al., 2017; Bartoli et al., 2018). For at least one of these modifications, m⁷G, plant homologs exist of the enzyme responsible for its introduction into tRNA (Alexandrov et al., 2002) and some mRNA sites (Zhang et al., 2019b). For others, such as ac4C and hm⁵C, the identified modifying enzymes do not have direct counterparts in plants.

Regardless of the existence of close homologs in plant genomes of RNA-modifying enzymes identified in mammalian cells, it is possible that chemical modifications in plant mRNA, not necessarily limited to the ones shown to occur in other eukaryotes, await discovery and functional characterization. Some studies do indeed provide evidence for the presence of additional modified nucleotides in plant mRNA. For example, liquid chromatography-mass spectrometry analysis of total hydrolysates indicates the presence of internal m⁷G in mRNA of several plant species (Chu et al., 2018). In addition, a thorough examination of sites of recurrent misincorporation by reverse transcriptase in Arabidopsis (Arabidopsis thaliana) RNA-seq data suggested the rare presence of several mRNA modifications with potential to alter Watson-Crick base pairing (Vandivier et al., 2015). These included 3-methylcytidine (m³C), whose presence in some mRNAs was validated by RNA immunoprecipitation with a m³C-specific antibody (Vandivier et al., 2015). It is worth noting, however, that the number of misincorporation sites was orders of magnitude higher in the uncapped mRNA pool undergoing degradation than in intact, capped mRNAs. Thus, rather than resulting from regulated modification, many modified nucleotides identified by this approach may be chemically damaged (e.g. by oxidation) such that they accumulate throughout the life of an mRNA until its degradation. It is also important to note that, despite their clear utility, detection methods of modified nucleotides involving total hydrolysis of poly(A⁺) fractions such as liquid chromatography-mass spectrometry and thin-layer chromatography may easily include contamination from tRNA and/or rRNA in which the relative, and certainly also the absolute, content of modified nucleotides other than m⁶A is manyfold higher than in mRNA (Boccaletto et al., 2018). The difficulty in obtaining pure mRNA is highlighted in a study by Legrand et al. (2017), in which total RNA subjected to two rounds of poly(A) selection followed by two consecutive rounds of rRNA depletion still contained 7.1% rRNA.

CARTOGRAPHY OF MRNA MODIFICATIONS IN PLANTS

Cartography of mRNA modifications (i.e. their mapping to specific sites in mRNAs) is very much in its infancy, particularly in plants, in which m⁶A stands out as the so-far best understood mRNA modification, with only limited knowledge available on Ψ and m⁵C. A recent report shows that Ψ exists in hundreds of mRNAs in Arabidopsis (Sun et al., 2019), but the functions of Ψ in mRNA or the identities of pseudouridine synthases responsible for mRNA pseudouridylation remain unknown. Three different studies have mapped hundreds of m⁵C sites in Arabidopsis mRNA, but there is discrepancy in assessments of whether m⁵C is enriched in coding sequences or in 3′-untranslated regions (UTRs), perhaps as a consequence of the different tissues used or because of actual inconsistencies between results obtained by the two different m⁵C mapping methods employed (Cui et al., 2017; David et al., 2017). These studies also showed that the tRNA methyltransferase NSUN2/TRM4B may catalyze C5-cytidine methylation in mRNA, because some mRNA-m⁵C sites are lost in trm4b mutants (Cui et al., 2017; David et al., 2017). In addition, quantitative differences in the extent of m⁵C modification of specific mRNAs between different tissues were noted, perhaps pointing to regulatory properties of this modification (David et al., 2017). In one case, orthogonal methods were used for m⁵C mapping, yielding a high-confidence set of m⁵C sites (Yang et al., 2019). Interestingly, this report also showed that m⁵C facilitates mRNA long-distance transport through the phloem (Yang et al., 2019), but the underlying mechanisms of this important phenomenon, including the identities of possible m⁵C-binding proteins, remain unknown. We focus the remainder of this Update on m⁶A, simply because, at present, its importance is more clearly established and the molecular mechanisms involved are better studied than for any other mRNA modification.

BIOCHEMICAL FRAMEWORK FOR THE FUNCTION OF M⁶A

Occurrence of m⁶A in Plant Transcriptomes

In agreement with early biochemical studies of animal, viral, and plant mRNA (Wei et al., 1976; Dimock and Stoltzfus, 1977; Schibler et al., 1977; Wei and Moss, 1977; Canaani et al., 1979; Nichols and Welder, 1981), transcriptome-wide mapping of m⁶A has identified RR[m⁶A]CH (R = A/G, H = A/C/U) as the most significantly enriched motif in m⁶A peaks of all eukaryotes analyzed to date (Dominissini et al., 2012; Meyer et al., 2012; Schwartz et al., 2013; Ke et al., 2015; Linder et al., 2015; Lence et al., 2016; Zhao et al., 2017; Li et al., 2019), including plants (Luo et al., 2014; Wan et al., 2015; Shen et al., 2016; Duan et al., 2017; Wang et al., 2017; Anderson et al., 2018; Miao et al., 2019; Parker et al., 2019). This finding strongly supports the existence of a conserved mechanism of m⁶A deposition in eukaryotic mRNA and is in perfect agreement with the fact that the adenosine methyltransferase complex is conserved across eukaryotes but not in prokaryotes, in which m⁶A occurs in a completely different sequence context in mRNA (Box 2; Deng et al., 2015). In addition to its presence in a defined consensus motif, m⁶A was predominantly found in the 3′-UTR of mRNAs by most of the above-mentioned studies in eukaryotes, again supporting a highly conserved mechanism of adenosine methylation, while it is distributed more or less evenly along prokaryotic transcripts (Deng et al., 2015). Nonetheless, recent reports have raised the question of whether different m⁶A motifs exist in plants (Li et al., 2014c; Anderson et al., 2018; Wei et al., 2018b; Luo et al., 2019; Miao et al., 2019; Zhang et al., 2019a), a debate that is detailed in Box 2 and illustrated in Figure 1. It is our judgment that the current evidence most strongly supports predominant use of the pan-eukaryotic RR[m⁶A]CH motif. In spite of this, there does appear to be a difference in the functional categories of m⁶A-containing mRNAs between plants and cultured animal cells. In plants, many m⁶A-containing mRNAs encode ribosomal and photosynthesis-related proteins, mitochondrial electron transport factors, and other basic metabolic enzymes (Luo et al., 2014; Wan et al., 2015; Shen et al., 2016; Wang et al., 2017; Anderson et al., 2018), while such housekeeping factors appear to be depleted from sets of m⁶A-containing mRNAs in yeast and mammalian cell cultures (Schwartz et al., 2014b; Ke et al., 2017). It is a question of outstanding importance to define if and how the labeling of these many housekeeping transcripts by m⁶A contributes to plant growth and development. Finally, a recent report deposited in bioRxiv shows that primary microRNA transcripts in Arabidopsis also contain m⁶A and proposes that the modification regulates microRNA biogenesis (Bhat et al., 2019), as previously proposed in animals (Alarcón et al., 2015; Berulava et al., 2015).

Figure 1.

Schematic representation of the m⁶A pathway and the functions of its characterized components. m⁶A writers are depicted on a blue field, readers on green, and erasers on orange. The canonical m⁶A consensus motif (RRACH) is chosen for the general representation, but alternative motifs (UGUAY and GGAU) recently proposed as plant specific are indicated in a separate box (see Box 2 for details). An endogenous m⁶A target is depicted as a pre-mRNA to highlight the connection of m⁶A writing to transcription. YTHDF m⁶A readers (ECT2, ECT3, ECT4, and probably additional ECTs) are represented as binding to the same transcript for convenience, but there are still no data clarifying whether different ECTs can bind in cis or not. If so, they could compete or have synergistic effects, perhaps interacting with each other as proposed for animal YTHDFs (Shi et al., 2017). Asterisks represent putative additional readers/erasers (listed in the gray boxes of Fig. 2), as many homologs in the plant YTH and ALKBH families remain uncharacterized. AMV, Alfalfa mosaic virus; CPSF, cleavage and polyadenylation specificity factor. PAS, polyadenylation signal.

Writing m⁶A

mRNA-modifying enzymes are often referred to as writers, by analogy with signal transduction systems in which writers catalyze the formation of a regulatory posttranslational modification, readers act as effectors by binding to the modified amino acid, and erasers remove the modification to reset signaling (Lim and Pawson, 2010). We adopt this nomenclature here but note that its appropriateness is intensely debated, because it is difficult to obtain direct proof that the same mRNA molecules undergo reversible cycles of methylation and demethylation (Ke et al., 2017; Mauer et al., 2017, 2019; Meyer and Jaffrey, 2017; Rosa-Mercado et al., 2017; Darnell et al., 2018; Wei et al., 2018a; Shi et al., 2019). Despite the simple biochemistry of the m⁶A writer reaction, nucleophilic substitution resulting in transfer of the methyl group from S-adenosylmethionine (SAM) to the N⁶-amine of adenosine, the major eukaryotic mRNA adenosine methyltransferase turns out to be extraordinarily complicated. Initial attempts at its purification from mammalian cells resulted in the definition of two subcomplexes, ∼200-kD MT-A and ∼875-kD MT-B, both required for full methyltransferase activity, even in vitro (Bokar et al., 1994). While we now know that MT-A contained the catalytic core consisting of a dimer of the two methyltransferase-like proteins METTL3 (Bokar et al., 1997) and METTL14 (Liu et al., 2014), the exact composition of the originally defined MT-B complex is still unclear. Both MT-A subunits have direct homologs in plants: METTL3 corresponds to MTA (Zhong et al., 2008) and METTL14 corresponds to MTB (Růžička et al., 2017). Several additional factors either fully or partly required for m⁶A deposition in vivo have been identified. Many of these factors are also conserved and include the following proteins, denoted with plant name first and mammalian homolog following a slash (see Figs. 1 and 2 for additional details): the splicing factor FKBP12 Interacting Protein37 (FIP37/WTAP; Vespa et al., 2004; Zhong et al., 2008; Liu et al., 2014; Ping et al., 2014; Schwartz et al., 2014b; Shen et al., 2016; Růžička et al., 2017); the protein VIRILIZER (VIR/KIAA1429), originally identified in Drosophila melanogaster (Niessen et al., 2001; Ortega et al., 2003; Schwartz et al., 2014b; Růžička et al., 2017); the putative ubiquitin E3 ligase HAKAI/CBLL1 (Horiuchi et al., 2013; Růžička et al., 2017); the Zn finger protein ZC3H13 (Guo et al., 2018; Knuckles et al., 2018; Wen et al., 2018) without orthologs in plants (Balacco and Soller, 2019); and the RNA-binding proteins RBM15A/B (Horiuchi et al., 2013; Yan and Perrimon, 2015; Lence et al., 2016; Patil et al., 2016). Interestingly, the closest plant homolog of RBM15A/B is FLOWERING LOCUS PA (FPA), an RNA-binding protein required for the control of flowering time (Schomburg et al., 2001). However, the homology between FPA and RBM15A/B is confined to the RNA recognition motifs and the overall degree of identity is low. Although the possible involvement of FPA in m⁶A deposition remains uninvestigated, mutants in FPA and m⁶A pathway components share at least one molecular phenotype: partial loss of function causes transcriptional read-through and chimeric RNA formation (Hornyik et al., 2010; Duc et al., 2013; Pontier et al., 2019). Nonetheless, it is clear that the requirement of FPA for m⁶A writer function, if any, cannot be absolute, because in contrast to most knockout mutants in m⁶A writer components (Vespa et al., 2004; Zhong et al., 2008; Růžička et al., 2017), null alleles of fpa complete embryogenesis (Schomburg et al., 2001).

Figure 2.

Biological functions of the m⁶A pathway in plants as inferred by the mutant phenotype of its characterized components in Arabidopsis. The mammalian homologs of each factor is indicated. Arabidopsis bona fide writers are shaded in blue, erasers in orange, and readers in green. Arabidopsis genes marked with asterisks (as also labeled in Fig. 1) inside gray-shaded boxes are orthologs of m⁶A pathway components in other organisms or paralogs of such genes in Arabidopsis, but their possible or expected roles as m⁶A players in plants have not been verified experimentally. For this reason, columns 4 and 5 are omitted in the gray boxes corresponding to putative erasers and readers. Column 3 is also omitted for simplicity as it can be summarized as follows: AtALKBH proteins are homologs of the mammalian ALKBH1 to ALKBH8 and FTO family (Mielecki et al., 2012); all ECTs belong to the YTHDF clade (YTHDF1–YTHDF3 in mammals), while At4g11970 presents homology with the YTHDC clade (YTHDC1 and YTHDC2 in mammals; Scutenaire et al., 2018). In addition to knockout, knockdown, and overexpression lines, transgenic plants expressing point mutants of ECT2 (12–14), ECT3 (12), ECT4 (12), and CPSF30 (15) with impaired ability to bind m⁶A, or a catalytically inactive ALKBH10b (9), are also described in the indicated references and behave like null mutants for the phenotypes described in all cases. References are as follows: 1, Zhong et al., 2008; 2, Bodi et al., 2012; 3, Růžička et al., 2017; 4, Shen et al., 2016; 5, Vespa et al., 2004; 6, Parker et al., 2019; 7, Schomburg et al., 2001; 8, Kim et al., 2008; 9, Duan et al., 2017; 10, Martínez-Pérez et al., 2017; 11, Mielecki et al., 2012; 12, Arribas-Hernández et al., 2018; 13, Scutenaire et al., 2018; 14, Wei et al., 2018b; 15, Pontier et al., 2019; 16, Li et al., 2014a.

Although the MTA/MTB methyltransferase accounts for the vast majority of m⁶A sites in Arabidopsis (Zhong et al., 2008; Shen et al., 2016; Růžička et al., 2017; Anderson et al., 2018), additional m⁶A methyltransferases exist. The essential mammalian enzyme METTL16 methylates structured RNAs containing the nonamer UAC(A)GAGAA (Pendleton et al., 2017). Its few identified targets include the MAT2A mRNA encoding SAM synthetase, and the embryo-lethal phenotype of mettl16 mutants may indeed result from disrupted SAM homeostasis (Pendleton et al., 2017; Warda et al., 2017; Mendel et al., 2018). Interestingly, a clear METTL16 homolog, FIONA1 (FIO1), exists in plants. fio1 loss-of-function mutants are viable and fertile but exhibit increased period length in the circadian clock (Kim et al., 2008), a phenotype shared with mutants homozygous for the hypomorphic vir-1 allele (Růžička et al., 2017; Parker et al., 2019). The biochemical functions of plant FIO1/METTL16, including as an m⁶A methyltransferase, remain unexplored, however.

It is a relevant question how cells avoid that the specificity of the m⁶A methyltransferase be undercut by direct incorporation of m⁶ATP during transcription. After all, degradation of m⁶A-containing RNA would result in m⁶AMP that could be converted into an RNA Polymerase II substrate (m⁶ATP) through salvage pathways. The answer appears to lie in a combination of two factors (Chen et al., 2018): first, the existence of a conserved enzyme with specific m⁶AMP deaminase activity; second, the fact that adenylate kinase has poor activity toward m⁶AMP, resulting in poor conversion of m⁶AMP into m⁶ADP. Interestingly, knockout of the Arabidopsis m⁶AMP deaminase MAPDA1 leads to reduced root growth, suggesting that avoidance of m⁶A recycling may be of importance in vivo (Chen et al., 2018).

Erasing m⁶A

One of the spectacular early findings in the m⁶A field was the discovery that enzymes in the ALKBH family have m⁶A demethylase activity (Jia et al., 2011; Zheng et al., 2013), suggesting the regulatory capacity of the modification for the first time. Several of these enzymes catalyze oxidative dealkylation of N-methylated nucleotides (Ougland et al., 2015). The Arabidopsis genome encodes 13 members of the ALKBH family of oxidases (Mielecki et al., 2012), two of which (ALKBH9B and ALKBH10B) have been shown to have m⁶A demethylase activity in vitro and have m⁶A-related functions in vivo (Figs. 1 and 2; Duan et al., 2017; Martínez-Pérez et al., 2017). While such demethylases are important for the control of flowering time (ALKBH10B) and susceptibility to viruses (ALKBH9B) and have an impact on m⁶A/A ratios in viral RNA (ALKBH9B) and, more subtly, in endogenous mRNA (ALBH10B; Duan et al., 2017; Martínez-Pérez et al., 2017), their exact functions in vivo remain ill defined.

Reading m⁶A

At least two different properties of RNA may change upon chemical modification of nucleotides. (1) The structure may change as a consequence of altered base pairing properties or rigidity of structure. In turn, this may create or influence the accessibility of binding sites for proteins, or even small molecules, in nearby sequences. (2) The modification itself may create a binding site for an RNA-binding protein with specific affinity for the modified nucleotide.

For m⁶A, there is evidence for both types of function in mammalian cells (Wang et al., 2014; Liu et al., 2015), although category 2 appears to be more prevalent (Patil et al., 2018). The best studied proteins with specific m⁶A-binding capacity contain a so-called YT521-B Homology (YTH) domain of ∼140 amino acids (Imai et al., 1998; Hartmann et al., 1999; Stoilov et al., 2002; Zhang et al., 2010) that recognizes the N⁶-methyl group on adenosine via a highly conserved hydrophobic binding pocket containing three aromatic side chains surrounding the methyl group (the aromatic cage; Li et al., 2014b; Luo and Tong, 2014; Theler et al., 2014; Xu et al., 2014; Zhu et al., 2014). Two phylogenetic classes of YTH domains can be defined, YTHDF and YTHDC. In mammals, YTHDF-containing proteins are predominantly cytoplasmic, bind to all m⁶A sites in mRNA, and the three paralogs (YTHDF1–YTHDF3) share high amino acid similarity throughout their entire length, including in their long N-terminal intrinsically disordered regions (IDRs). On the contrary, the YTHDC domain is found in two very different proteins in mammals: YTHDC1 is nuclear and binds to some sites in mRNAs and nuclear non-coding RNAs, whereas YTHDC2 is enriched in perinuclear regions of the cytoplasm and its mRNA-binding profile shows little overlap with m⁶A sites (Patil et al., 2016; Hsu et al., 2017; Kretschmer et al., 2018; Zaccara et al., 2019). YTHDC2 is specific to mammals and contains several other folded domains in addition to the YTHDC domain (Bailey et al., 2017; Wojtas et al., 2017; Kretschmer et al., 2018), while YTHDC1 has long N- and C-terminal IDRs (Patil et al., 2016). YTHDF- and YTHDC1-type proteins are found in many eukaryotes (Balacco and Soller, 2019), including plants, whose genomes encode more YTH domain proteins than other organisms (Li et al., 2014a; Scutenaire et al., 2018). For example, 13 YTH domain-containing proteins are encoded in Arabidopsis (Fig. 2) compared with five in humans. The 13 Arabidopsis YTH-domain proteins can be divided into 11 YTHDF proteins called EVOLUTIONARILY CONSERVED C-TERMINAL REGION1 (ECT1) to ECT11, one classical YTHDC1-type protein (At4g11970), and one YTHDC protein, unusual in that it also contains additional folded domains: it is CPSF30, the 30-kD subunit of the cleavage and polyadenylation specificity factor involved in pre-mRNA cleavage and 3′-end formation (Zhang et al., 2008; Thomas et al., 2012; Bruggeman et al., 2014; Pontier et al., 2019).

Several lines of evidence suggest that while m⁶A-binding specificity resides in the YTH domain, the effector function, at least of YTHDF proteins, resides in the IDR. For example, tethering the IDR of YTHDF2 to reporter mRNAs is sufficient to cause their localization to P-bodies in mammalian cells (Wang et al., 2014). The IDR of YTHDF2 also interacts with the deadenylase complex CCR4-NOT (Du et al., 2016) and, via the adaptor protein HRSP12, with the endoribonuclease RNase P/MRP (Park et al., 2019). However, since the affinity of isolated YTH domains for m⁶A-modified RNA is modest (0.1–5 μm; Li et al., 2014b; Luo and Tong, 2014; Theler et al., 2014; Xu et al., 2014, 2015; Zhu et al., 2014), it is possible that the IDR participates in RNA binding in vivo, as suggested by the loss of mRNA binding in vivo of a mutant in human YTHDF3 containing a deletion in the IDR (Zhang et al., 2019c). In this regard, it is noteworthy that three Arabidopsis YTHDF proteins (ECT5, ECT9, and ECT10) contain an amino acid substitution expected to result in 10-fold higher affinity toward m⁶A RNA than other YTH domains, based on structural and biochemical analyses of analogous mutants in human YTHDF1 (Xu et al., 2015; Scutenaire et al., 2018). These proteins may, therefore, differ in requirements for their IDRs for RNA interactions compared with other YTH proteins, or they may simply bind with higher affinity. It is an interesting property of both mammalian YTHDFs and plant ECTs that they are able to undergo phase transition in vitro to a condensed liquid or gel-like phase (Arribas-Hernández et al., 2018; Fu and Zhuang, 2019; Gao et al., 2019; Ries et al., 2019). It is of considerable interest to determine whether such phase separation properties underlie biological functions and the subcellular localization of YTHDF proteins. To date, YTH domain proteins are the only characterized class of m⁶A readers in plants (Figs. 1 and 2). In mammalian cells, several other RNA-binding proteins have been proposed to function as m⁶A readers (Balacco and Soller, 2019), including the translation initiation factor eIF3 that is recruited to m⁶A-containing 5′-UTRs to stimulate cap-independent translation initiation (Meyer et al., 2015).

MOLECULAR ROLES OF M⁶A IN PLANTS

Cytoplasmic Roles

What effect does the presence of m⁶A in an mRNA have on its stability and translatability? In yeast and mammalian cell culture, the evidence is strong that it accelerates mRNA decay (see e.g. Herzog et al. [2017] or Ke et al. [2017] for animals, Bushkin et al. [2019] for yeast, or Zaccara et al. [2019] for the most recent review), as proposed in early studies (Sommer et al., 1978). Accelerated mRNA decay in mammals involves YTHDF proteins, perhaps in particular YTHDF2 (Wang et al., 2014; Ivanova et al., 2017; Zhang et al., 2017; Zhao et al., 2017; Paris et al., 2019), but whether the mechanism relies mostly on CCR4-NOT deadenylase recruitment (Du et al., 2016), perhaps concurrent with sorting to P-bodies (Wang et al., 2014), enhanced endonucleolysis (Park et al., 2019), or a combination remains unclear. m⁶A may also stimulate translation, in this case via other YTHDF proteins (Wang et al., 2015; Li et al., 2017a; Shi et al., 2017). In particular, YTHDF1 promotes the translation of m⁶A targets via interaction with the eukaryotic initiation factor eIF3 (Wang et al., 2015) in a process that may be of particular biological relevance in neurons (Shi et al., 2018; Weng et al., 2018). A cytoplasmic role of METTL3 itself has also been proposed in the activation of translation, also via eIF3 recruitment (Lin et al., 2016; Choe et al., 2018).

In plants, transcriptomic analyses have also been applied to analyze the effect of m⁶A on mRNA abundance (Fig. 3). Since knockout of m⁶A writer subunits is embryonically lethal (see "Functions in Development" below; Zhong et al., 2008), plants expressing either MTA or FIP37 from embryo-specific promoters in the respective knockout backgrounds have been used to obtain postembryonic tissues with 80% to 90% loss of m⁶A (Bodi et al., 2012; Shen et al., 2016). Studies using either these plants or the partial loss-of-function vir-1 allele (Růžička et al., 2017) showed a slight tendency for m⁶A targets to be less abundant in tissues of plants with reduced levels of m⁶A (Shen et al., 2016; Anderson et al., 2018; Parker et al., 2019). The same tendency was observed in knockouts of the m⁶A reader ect2 (Wei et al., 2018b). Altogether, those observations suggest that m⁶A may stabilize mRNAs, potentially by protection against endonucleolytic cleavage that occurs in m⁶A-depleted plants around sites of m⁶A deposition (Anderson et al., 2018). On the other hand, overaccumulation of m⁶A-containing mRNAs that encode important developmental regulators has also been found in m⁶A-depleted mutants (Shen et al., 2016). Moreover, ALKBH10B-mediated m⁶A demethylation of transcripts encoding key flowering-related genes is associated with their increased abundance and stability (Duan et al., 2017). Thus, the existence of m⁶A-induced mRNA destabilization in plants should not be ruled out.

Figure 3.

Plant studies reporting on effects of m⁶A on target RNA accumulation and, in a few cases, decay rates upon global inhibition of transcription.

We note that the interpretation of the transcriptome profiles reported of wild-type and m⁶A-deficient plants (Fig. 3) suffers from limitations that may preclude clear conclusions to be drawn on whether m⁶A actually stabilizes or destabilizes mRNA. First, they use mRNA isolated from all cells of seedlings, while analysis of activity of the MTA promoter suggests that m⁶A deposition may not be active in all cells (Zhong et al., 2008). Thus, in total extracts, the fraction of a particular mRNA target whose abundance can be directly influenced by the loss of methyltransferase activity is diluted by contributions from cells that do not even express the methyltransferase. That is especially relevant for assessing the overall abundance of m⁶A targets in plants, as they predominantly include ubiquitously expressed housekeeping genes (Anderson et al., 2018). Furthermore, as knockdown of MTA, FIP37, or VIR causes global changes in plant morphogenesis that may affect the expression domains of m⁶A-containing mRNAs, it compromises the ability of such experiments to yield conclusive results. Second, steady-state measurements comparing a stable mutant with a wild type do not provide kinetic information, as degradation and synthesis rates will inevitably be confounded. In conclusion, it is at present not clear if mRNAs are stabilized or destabilized by m⁶A (Fig. 3). There is limited direct evidence for either outcome, and the possibility of target- and/or cell-specific outcomes of mRNA methylation simply has not yet been investigated. Ideally, transcriptome-wide pulse-chase analyses that do not disrupt global transcription [e.g. using thiol(SH)-linked alkylation for the metabolic sequencing of RNA; Herzog et al., 2017] should be employed in the relevant cells rather than bulk tissue of the wild type versus mta/mtb/fip37/vir knockdowns or ect knockouts to reveal the effect of m⁶A deposition and ECT binding on transcript stability. Finally, the possible effects of m⁶A on translational control in plants remain entirely unexplored.

Nuclear Roles

In addition to the intensely studied effects on mature mRNA, roles of m⁶A in pre-mRNA processing have also been reported. For example, m⁶A is required for sex determination in flies, because the sex determination pathway relies on alternative splicing of the sex-lethal (sxl) pre-mRNA, a process that in turn depends critically on m⁶A deposition around the alternatively spliced sxl exon (Haussmann et al., 2016; Lence et al., 2016). In plants, a recent study has uncovered a special role of m⁶A in mRNA 3′-end formation and transcription termination at recently duplicated genes: at such loci, read-through transcription into the downstream gene is observed in m⁶A-deficient plants and, interestingly, in plants containing point mutations to abrogate the aromatic cage in the YTH domain of CPSF30 (Pontier et al., 2019). It is possible that m⁶A-directed CPSF30 function stimulates pre-mRNA cleavage and RNA Polymerase II termination more generally but that the effect of losing this reinforcement only becomes visible at loci with recent gene duplications (Pontier et al., 2019). In agreement with the idea of m⁶A-dependent 3′-processing in plants, Parker et al. (2019) recently reported that lack of m⁶A is associated with a shift to usage of more proximal poly(A) sites and transcriptional read-through, and Luo et al. (2019) observed correlation between the presence of m⁶A and alternative polyadenylation site usage in maize (Zea mays).

BIOLOGICAL ROLES OF M⁶A IN PLANTS

Functions in Development

The study of plants provided one of the key discoveries in the development of the mRNA modification field: using Arabidopsis, Fray and coworkers (Zhong et al., 2008) were able to prove in 2008 that m⁶A is crucial for ontogenesis in a multicellular eukaryote, as embryos defective in the homolog of the m⁶A methyltransferase identified in yeast and mammals (MTA) arrested at the globular stage. Subsequent phenotypic analyses of hypomorphic alleles or postembryonic knockdowns of mta, mtb, fip37, and vir showed that reduction of m⁶A results in slower growth, abnormal organ definition, loss of apical dominance, increased number of trichome branches, defective gravitropic responses, and aberrant development of lateral roots and vasculature (Bodi et al., 2012; Růžička et al., 2017; see Fig. 2 for a comprehensive description of knockout and knockdown lines). Crucial additional insights were gained from the work of Shen et al. (2016), who showed that stem cell differentiation in the shoot apical meristem depends on m⁶A: in plants depleted of FIP37 postembryonically, the shoot apical meristem dramatically increases in size and forms aberrant leaf primordia with a significant delay compared with the wild type. Interestingly, these strong differentiation phenotypes are associated with expansion of the organizing center of the meristem expressing the key transcription factor WUSCHEL (At2g17950), whose mRNA is modified by m⁶A (Shen et al., 2016). Another m⁶A target, the mRNA of the meristematic transcription factor SHOOT MERISTEMLESS (At1g62360), also overaccumulates in fip37 knockdown plants, apparently as a consequence of increased mRNA stability (Shen et al., 2016). Thus, m⁶A in plants is clearly required for meristem function, perhaps in particular for the step of stem cell differentiation, reminiscent of the requirement for m⁶A for embryonic stem cell differentiation in mammals (Batista et al., 2014; Geula et al., 2015). m⁶A is also involved in the maintenance of circadian and seasonal plant rhythms, as mutants defective in m⁶A deposition or removal exhibit lengthening of the circadian period (Parker et al., 2019) and late flowering (Duan et al., 2017), respectively. Finally, recent phenotypic and molecular studies of plants lacking one or several m⁶A readers provide additional mechanistic insights into the role of m⁶A in development (Arribas-Hernández et al., 2018; Scutenaire et al., 2018; Wei et al., 2018b). Such advances are discussed in detail below.

Functions in Stress Adaptation

Many mechanisms of posttranscriptional gene regulation, perhaps most prominently small RNA-based gene regulation, are employed to mediate rapid changes in gene expression required for stress adaptation (Sunkar et al., 2007), and it would, therefore, not be surprising to find such roles of the m⁶A-YTH system. Although early studies based on publicly available microarray data on Arabidopsis and rice (Oryza sativa) YTH domain proteins pointed to their induction by various abiotic and biotic stresses (Li et al., 2014a), relatively little has been done to test their functional implication in such responses. Scutenaire et al. (2018) found that ECT2 relocalizes to cytoplasmic stress granules upon heat stress. Similarly, Arribas-Hernández et al. (2018) observed that ECT2 and ECT4, and to a lesser degree ECT3, relocalize to cytoplasmic foci distinct from P-bodies upon osmotic stress. These studies show that reader proteins respond to stress, but they do not establish whether this is inconsequential or important for the adaptive response. Similar correlative evidence between m⁶A levels and a biotic stress response in tobacco (Nicotiana tabacum) plants has been reported. In response to tobacco mosaic virus infection, m⁶A/G ratios and expression of m⁶A writers decreased whereas mRNA levels of one potential demethylase increased (Li et al., 2018). The available evidence does not, however, allow a distinction between whether such changes are brought about by the virus to promote infection or by the host to limit infection, or even whether they are relevant for the host-virus interaction at all. In animals, examples of both positive and negative regulation of innate and adaptive immunity by m⁶A have been reported (see Williams et al. [2019] for review), and similar complexity might be expected for different pathogens or cell types in plants.

A role of m⁶A in selective stabilization of mRNAs responsive to salt stress has recently been suggested (Anderson et al., 2018). This study showed that transcripts involved in salt and osmotic stress were hypermethylated under salt treatment and more highly expressed in treated versus nontreated samples. On the contrary, mRNAs with a tendency to lose m⁶A marks in response to the treatment were related to more general processes like photosynthesis, and their abundance decreased upon salt stress. Nevertheless, normalization of relative m⁶A contents in treated versus untreated samples is not trivial when considering on/off genes such as salt stress-response genes. In addition, since m⁶A peaks are more easily mapped in abundant transcripts with antibody-based mapping approaches (Shen et al., 2016; Garcia-Campos et al., 2019), a bias toward the detection of de novo methylation of salt-responsive transcripts highly expressed after treatment cannot be excluded. Thus, orthogonal mapping approaches with the ability to accurately quantify m⁶A stoichiometry (e.g. MAZTER-seq; Garcia-Campos et al., 2019) will be valuable to address the dynamics of m⁶A marks in response to stress. In addition, genetic studies, ideally using sophisticated conditional loss-of-function systems of m⁶A writers and readers, will be required to evaluate the physiological relevance of the m⁶A regulatory pathway in stress adaptation.

Methylation of Viral RNA as an Immune Response in Plants

Modification of viral RNA by m⁶A can modulate the outcome of host-virus interactions in mammals. In some examples, such as the cytoplasmic (+)-strand RNA virus Zika, m⁶A attenuates viral infectivity (Gokhale et al., 2016; Lichinchi et al., 2016). In other cases, including some m⁶A sites in HIV, the presence of m⁶A on viral RNA enhances viral gene expression (Kennedy et al., 2017), making it clear that there is no general outcome of N⁶-adenosine methylation in host-virus interactions (Williams et al., 2019). In plants, RNA of Alfalfa mosaic virus (AMV) has been shown to contain m⁶A (Martínez-Pérez et al., 2017). In this case, the multifunctional AMV coat protein recruits the m⁶A eraser ALKBH9B, resulting in decreased m⁶A levels in viral RNA (Fig. 1). This m⁶A erasure dramatically increases systemic infection, indicating that m⁶A has strong antiviral effects in the case of AMV (Martínez-Pérez et al., 2017). It is not yet clear at which precise points in the infection cycle m⁶A acts (e.g. initial translation of viral RNA, transcription, cell-to-cell movement, or stability of viral RNA), and the molecular mechanisms underlying its clear antiviral effect remain unexplored. This includes important questions concerning possible links between nonself RNA recognition and N⁶-adenosine methylation, the identity of the m⁶A writer, whether reader proteins are involved, and if so how.

MOLECULAR MECHANISMS UNDERLYING BIOLOGICAL ROLES OF M⁶A IN PLANTS

The clear biological importance of m⁶A in plants immediately begs mechanistic questions. Given the biochemical framework established for at least some m⁶A-dependent functions, we can now ask these questions in more precise terms. These include the following. (1) How much of the observed m⁶A effects is mediated by readers? (2) How many readers are involved? (3) How much is explained by YTH domain proteins? (4) Which mRNAs are targeted, and what are the effects of reader binding to the relevant targets?

A few recent studies have started to wrestle with the considerable problem of doing genetics with 11 different, potentially redundant, ECT/YTHDF proteins (Li et al., 2014a). Two studies showed that single knockouts of ECT2 led to weak and stochastic increases in trichome branching (Scutenaire et al., 2018; Wei et al., 2018b), resembling one of the defects previously described in postembryonic mta knockdown mutants (Bodi et al., 2012). No other developmental phenotypes could be observed, suggesting that YTHDF reader function is partly involved in specifying branch numbers on trichomes but failing to make a case for reader function in the processes of clear biological importance unveiled by the studies of m⁶A-deficient plants (Fig 2; Vespa et al., 2004; Zhong et al., 2008; Bodi et al., 2012; Shen et al., 2016; Růžička et al., 2017). In contrast, the simultaneous knockout of three YTHDF proteins, ECT2, ECT3, and ECT4, offered substantial additional insights (Arribas-Hernández et al., 2018). This work demonstrated that the slow development and aberrant leaf morphology in postembryonic mta knockdown plants can be largely recapitulated upon knockout or inactivation of m⁶A binding of ECT2/3/4, strongly suggesting that the control of these important developmental processes by m⁶A requires this specific set of YTH domain proteins (Figs. 2 and 4). This study also revealed that both ECT2 and ECT3 play similar roles in the definition of trichome branch numbers and that the combined ect2/ect3 knockout has a much stronger defect than either single mutant, even exceeding the severity observed upon knockdown of m⁶A (Bodi et al., 2012; Arribas-Hernández et al., 2018). Thus, at least for the phenotypes of slow postembryonic growth, defective leaf morphogenesis, and definition of trichome branch numbers, the YTHDF proteins ECT2 and ECT3, in some cases also ECT4, are responsible for a considerable part of the effects dictated by m⁶A deposition. For the effects of m⁶A in other developmental processes (see Fig. 2 for a compilation), we do not have the answers yet.

Figure 4.

Phenotypic comparison between mta knockdown plants with low levels of m⁶A (AmiR-MTA; Shen et al., 2016) and the triple mutant ect2/ect3/ect4 defective in m⁶A reader function (Arribas-Hernández et al., 2018). Notice the similarity in the developmental delay (number of leaves at each stage), reduced stature (although aggravated in AmiR-MTA plants), and identical leaf shape (white arrows on photographs and silhouettes of the first true four leaves): triangular blade with more serrations than in wild-type (Col-0 WT) leaves. Plants were grown side by side in Percival growth chambers at 21°C/18°C (day/night) and with a long-day (16 h) light regime. DAG, Days after germination. Bars = 1 cm.

Only a single study has attempted to answer our question 4. Wei et al. (2018b) used formaldehyde cross-linking followed by ECT2 immunoprecipitation and sequencing to identify ECT2 mRNA targets. Nearly one-third of all expressed genes were identified as mRNA targets using this approach. These targets also included the known trichome regulators TRANSPARENT TESTA GLABRA1 (At5g24520), IRREGULAR TRICHOME BRANCH1 (At2g38440), and DISTORTED TRICHOME2 (At1g30825), whose mRNAs were less stable in ect2 mutants than in the wild type, perhaps accounting for their ∼0.6 to 0.8 times lower levels, as measured by quantitative reverse transcription PCR from RNA of whole seedlings. Hence, a model was proposed in which ECT2 stabilizes these mRNAs and their decreased expression in ect2 mutants underlies the increased trichome branch number phenotype (Wei et al., 2018b). As appealing as this model may seem at first sight, some problems are apparent with the evidence supporting it. First, the study did not recover the consensus RR[m⁶A]CH motif identified in multiple m⁶A-seq studies (Box 2) as enriched around ECT2 cross-linking sites. This question is more pertinent because formaldehyde, not UV light, was used as a cross-linker. Upon reaction with Lys side chains, formaldehyde forms a Schiff base intermediate that readily reacts with nucleophiles in either proteins or RNA, thereby facilitating indirect cross-links to RNA via other proteins (Hoffman et al., 2015). Thus, many targets identified by formaldehyde-CLIP may not be direct. Second, the transcriptomic analysis measuring decay rates and steady-state abundance were done with entire seedlings, not necessarily informative for gene expression changes in the few cells that matter for the described phenotype: trichomes at an early stage of development. Third, the function of the genes whose decreased abundance in ect2 mutants is proposed to underlie the stochastic increase in trichome branches does not match with the morphological and molecular phenotypes observed (Box 3; Koornneeff et al., 1982; Hülskamp, 2004; Saedler et al., 2004; Zhang et al., 2005; Pattanaik et al., 2014; Arribas-Hernández et al., 2018; Scutenaire et al., 2018; Wei et al., 2018b). For these reasons, we consider it fair to conclude that the evidence is strong that the YTHDF proteins ECT2, ECT3, and ECT4 mediate some m⁶A-dependent effects on growth and development in a manner that requires m⁶A binding. On the other hand, how precisely they do so remains much less well defined at present.

CONCLUDING REMARKS

We are just starting to elucidate how mRNA modifications and m⁶A in particular are at the core of plant development (see Advances Box), response to abiotic stress, and antiviral defense. Although it has not been addressed yet, modulation of growth in response to other phytopathogens and herbivory is likely to be another function. With m⁶A-YTH axes as the fundamental regulatory units, we now have the guidelines and the tools to find precise answers to how m⁶A exerts its regulatory functions. Plants might be particularly adept at exploiting the regulatory capacity of m⁶A, since they grow and develop throughout their life span in the face of a changing environment. They adapt their growth pace in different organs to shape their bodies in unique ways according to environmental cues: root- and stem-branching patterns, leaf shape and size, number of reproductive organs, and overall architecture are dynamically remodeled according to light, water, and nutrient availability, herbivory, and pathogen attack. The expanded families of YTHDF and ALKBH members in plants (Fig. 2) may reflect this necessity (see Outstanding Questions Box). By tissue- or stimulus-dependent expression of these growth regulators, plants may be able to balance the growth rate and final size of different organs in response to different conditions. Potentially, additional mRNA modifications could be combined with m⁶A to establish a sophisticated network for growth control (see Outstanding Questions Box). A challenge of the next few years will be to clearly define the molecular workings of these potent gene regulatory systems to enable precise tests of this hypothesis of dynamic plant growth control via mRNA modification systems.