Reversible RNA Modification N¹-methyladenosine (m¹A) in mRNA and tRNA

Abstract

More than 100 modifications have been found in RNA. Analogous to epigenetic DNA methylation, epitranscriptomic modifications can be written, read, and erased by a complex network of proteins. Apart from N⁶-methyladenosine (m⁶A), N¹-methyladenosine (m¹A) has been found as a reversible modification in tRNA and mRNA. m¹A occurs at positions 9, 14, and 58 of tRNA, with m¹A58 being critical for tRNA stability. Other than the hundreds of m¹A sites in mRNA and long non-coding RNA transcripts, transcriptome-wide mapping of m¹A also identifies >20 m¹A sites in mitochondrial genes. m¹A in the coding region of mitochondrial transcripts can inhibit the translation of the corresponding proteins. In this review, we summarize the current understanding of m¹A in mRNA and tRNA, covering high-throughput sequencing methods developed for m¹A methylome, m¹A-related enzymes (writers and erasers), as well as its functions in mRNA and tRNA.

The RNA modification N¹-methyladenosine

Shortly after the discovery of the base N¹-methyladenine (m¹A) in 1961 [1], Dunn et al. isolated 1-methyladenosine mononucleotide from RNA [2]. m¹A has been discovered in tRNA [3], rRNA [4], [5], [6], mRNA [7], [8], [9], [10], and mitochondrial (mt) transcripts [9], [10]. m¹A, N³-methylcytidine (m³C), and N⁷-methylguanosine (m⁷G) are the most commonly methylated nucleotides with a positive electrostatic charge under physiological conditions [11]. The electro-chemical interaction resulting from the positive charge of m¹A is critical for the function and structure of tRNA [11]. The methyl group on m¹A in mRNA blocks the Watson-Crick base pairing, affecting reverse transcription and protein translation [7], [8], [9], [10].

m¹A was first found in yeast tRNA^Phe [3]. Decades later, m¹A has also been found in 264 out of 564 tRNA sequences in bacteria, archaea, and eukaryota, which can occur at positions 9, 14, and 58 [12]. Among them, m¹A58 is conserved in bacteria, archaea, and eukaryota, and m¹A58 is critical for tRNA stabilization [12], [13]. A recent study has reported that m¹A58 can be reversed by the human AlkB homolog 1 (ALKBH1) demethylase [13].

In contrast to the high abundance of m¹A in tRNA, m¹A is present in mammalian mRNA with a low abundance (m¹A/A: 0.015%−0.054% in mammalian cells and up to 0.16% in mammalian tissues) [7], [8]. The development of m¹A-sequencing methods and new single-base resolution methods [7], [8], [9], [10] has facilitated unraveling the presence of m¹A in nuclear, cytosolic, and mt-encoded transcripts [9], [10]. For instance, Li et al. found 473 m¹A sites in mRNA and long non-coding (lncRNAs) transcripts in HEK293T cells. The majority of these m¹A sites are located within the 5′-untranslated region (5′ UTR) [9], which is consistent with previous findings [7], [8]. Interestingly, 22 m¹A sites have been identified in 10 of 13 mt transcripts, with 21 sites located in the coding sequence (CDS) and one in the 3′ UTR [9]. m¹A at CDS of mt transcripts affects mitochondrial translation, suggesting that m¹A might regulate the normal function of mitochondria.

m¹A sequencing

In order to comprehensively explore m¹A methylation in transcriptome, it is necessary to develop m¹A high-throughput sequencing. The low abundance of m¹A in mRNAs necessitates highly sensitive detection methods. In addition, m¹A can be rearranged to m⁶A under alkaline conditions (Dimroth rearrangement) [14]. Therefore, the signature of m¹A in mRNA may disappear during the preparation of mRNA fragmentation or mRNA digestion, which makes it difficult to reduce the noise in m¹A measurement and sequencing. m¹A blocks the Watson-Crick interface and effectively stalls reverse transcription (RT), thus inducing truncations or mutations in RT products [15], [16], [17], [18]. This feature can be employed to identify m¹A sites, and accordingly, the high-throughput m¹A sequencing methods were developed based on this idea.

The first-generation m¹A sequencing methods have been reported by two independent groups [7], [8] (Figure 1). The procedures of these two methods are largely similar. They both use the m¹A antibody to enrich m¹A-modified mRNA fragments and take advantage of the characteristics of m¹A to induce mismatch or truncation during the RT process. However, there are also differences between these two methods. Dominissini et al. use Dimroth rearrangement to change m¹A to m⁶A and then analyze m¹A sites based on mismatch rates. Li et al. use E. coli AlkB to demethylate m¹A to normal A and then calculate m¹A sites based on truncation position. The resolution of the first generation sequencing method is around 130 nucleotides [7], [8]. To identify m¹A sites more precisely for a more accurate picture of m¹A distribution, researchers have recently developed the second-generation m¹A sequencing method.

Figure 1.

Schematic outline of the two generations of m¹A sequencing methods

RT, reverse transcription; TGIRT, thermostable group II intron reverse transcriptase.

A highly processive reverse transcriptase thermostable group II intron reverse transcriptase (TGIRT) is used in the second-generation m¹A sequencing method to optimize the RT step [9], [10] (Figure 1). TGIRT exhibits excellent read-through efficiency and relatively high mutation frequency at the site of m¹A, resulting in a higher signal-to-noise ratio for detecting the precise position of m¹A, and therefore achieving a single-base resolution.

Using the second-generation m¹A sequencing method, Li et al. [9] identified 740 m¹A sites in the transcriptome of 293 T cells, including 418 and 55 sites found in mRNA and lncRNA transcripts, respectively. The majority of these m¹A sites in mRNA are within the 5′ UTR, which agrees with the previous findings [7], [8]. Interestingly, Li et al. find 24 m¹A sites at the first nucleotide (cap+1) and three sites at the second nucleotide (cap+2) of 5′ end of the transcripts. Ribosome profiling demonstrates that sites in the 5′ UTR and cap+1 position, but not those in the CDS or 3′ UTR, are correlated with higher translation efficiency. This observation that there is a correlation between m¹A sites in the 5′ UTR and translation efficiency is in agreement with a previous report [7].

According to the RNA location, writer enzyme involved, and sequence-structural features, Li et al. divide the m¹A sites in mRNA transcripts into three categories. These include tRNA methyltransferase 6/61A (TRMT6/61A)-dependent m¹A sites in nuclear-encoded mRNAs (53 sites), TRMT6/61A-independent m¹A sites in nuclear-encoded mRNAs, and m¹A sites in mt-encoded mRNAs (22 sites). TRMT6/61A-dependent m¹A sites are barely enriched in the 5′ UTR and conform to a GUUCRA tRNA-like motif and T-loop-like structures. m¹A sites in mt-mRNAs are mainly located in CDS, and play a role in inhibiting mitochondrial translation. Among the ten m¹A-modified mt-RNAs, m¹A modifications on 5 mt-RNAs are installed by mt-tRNA methyltransferase TRMT61B. Would the random mismatches occurring during the RT process of TGIRT yield false positive results? Li et al. used a method called reverse calculation to test this possibility. They take the sample of demethylase (−) (i.e., m¹A) as background and the sample of demethylase (+) (i.e., A) as signal to call m¹A peaks. As a result, they have called 17 peaks, confirming the high accuracy of TGIRT enzyme. Using primer extension assay, they also verify several m¹A sites in mt mRNA, confirming again the authenticity and accuracy of their method [9].

Surprisingly, about the same time, another group adopted a similar approach but reached a very different conclusion [10]. Safra et al. identified only 15 m¹A sites in mRNAs and lncRNAs, with 9 in cytosolic mRNAs, 1 in cytosolic lncRNAs, and 5 in mt-mRNAs. The 10 m¹A sites in cytosolic transcripts are installed by TRMT6/61A, whereas one m¹A site in mt-mRNA ND5 is methylated by TRMT10C. They also find that m¹A within 5′ UTR or CDS of cytosolic mRNAs leads to translational repression, indicating that m¹A might be related to ribosomal scanning or translation.

These two studies share some similar findings, while disagreement arises in terms of the number of m¹A sites in human mRNAs. Safra et al. suggest that m¹A is a rare modification in human mRNAs based on their observation that only a dozen m¹A sites have been identified using their method [10]. In contrast, Li et al. have identified hundreds of m¹A sites, and thus they consider m¹A a prevalent modification in human mRNAs [9]. Why do the numbers of m¹A sites identified in these two studies differ so much? As speculated in the technology preview [19], differences in experimental procedures may confer differences in the quality of sequencing datasets. For instance, 1) compared to AlkB demethylation, the alkaline conditions used for Dimroth rearrangement might affect RNA integrity and rearrangement efficiency of m¹A to m⁶A; 2) a competitive elution step at m¹A antibody immunoprecipitation is omitted in [10], together with the lack of optimization of the RT step of TGIRT; 3) a random 10-nt barcode (unique molecular identifier, UMI) in the DNA 3′-adaptor is used in [9] to remove repetition caused by PCR and improve the accuracy of detection, while lack of UMIs could lead to trouble in data analysis and mutation rate calculation [10]. In addition, the lower number of raw sequencing reads in the study [10] could also explain the fewer m¹A sites identified.

m¹A writers: m¹A methyltransferases

tRNA m¹A methyltransferase (MTase) has been extensively studied, since m¹A was first found in tRNA. Given m¹A58 is dominant and conserved across the three domains of life, studies of tRNA m¹A MTases started from looking for m¹A58 MTase.

tRNA m¹A58 MTase was initially characterized using partially purified protein fractions from bovine liver, which exhibits m¹A MTase activity in vitro toward E. coli tRNA₂^Glu [20]. The genes encoding m¹A58 MTase were firstly identified in Saccharomyces cerevisiae [21]. The S. cerevisiae m¹A58 MTase comprises two subunits, tRNA adenine-N¹-methyltransferase non-catalytic subunit and tRNA adenine-N¹-methyltransferase catalytic subunit, encoded by two essential genes Trm6 and Trm61, respectively [21]. Trm61 is responsible for AdoMet-binding and catalytic function as an enzyme, while Trm6 is critical for tRNA binding [22]. The purification and characterization of the m¹A58 MTase from S. cerevisiae [21], [22] has facilitated the identification of other m¹A58 MTases from humans and other organisms. In eukaryota, m¹A58 MTase consists of Trm6 and Trm61 orthologs, whereas prokaryotic m¹A58 is installed by a single protein encoded by orthologs of Trm61 [23]. TRMT61A and TRMT6 are the human othologs of yeast Trm61 and Trm6, respectively, which are responsible for m¹A58 modification of cytoplasmic tRNAs [24] (Figure 2).

Figure 2.

The methylation and demethylation of m¹A in tRNA and mRNA

In nucleus, m¹A modifications in pre-tRNA and specific pre-mRNA are installed by tRNA m¹A methyltransferase complex TRMT6/61A and erased by AlkB homolog proteins ALKBH1 and ALKBH3, respectively. In mitochondrion, mt-tRNA and a subset of mt-mRNA are methylated by mt-tRNA m¹A methyltransferase TRMT61B and TRMT10C. mt, mitochondrial.

Human mt-tRNAs are known to contain m¹A at positions 9 and 58 [24], which are catalyzed by TRMT10C and TRMT61B, respectively [25], [26] (Figure 2). In addition, m¹A is also found at position 1322 of 28S rRNA, which is catalyzed by the human nucleolar protein nucleomethylin (NML; also known as RRP8) [27].

m¹A is present in human nuclear-encoded mRNA and mt-mRNA [9]. tRNA m¹A MTases are found to be able to write m¹A in some mRNA transcripts as well. For instance, TRMT6/61A can install m¹A sites within a GUUCRA tRNA-like motif with T-loop-like structure in some nuclear mRNAs, whereas TRMT61B methylates half of the identified m¹A sites in mt-mRNA transcripts [9] (Figure 2). In addition, TRMT10C can add m¹A at the 1374 position of mt-mRNA ND5 [10] (Figure 2). Apparently, more m¹A mRNA MTases may be identified in future.

m¹A erasers: m¹A demethylases

m¹A is discovered to be the second reversible RNA modification. The first reversible RNA modification, m⁶A, is demethylated by two AlkB family proteins, i.e., fat mass and obesity-associated (FTO) and ALKBH5 [28], [29]. Similarly, two AlkB family proteins ALKBH3 and ALKBH1 have been found to demethylate m¹A [8], [13] (Figure 2).

E. coli AlkB is an Fe^II/α-ketoglutarate (α-KG)-dependent dioxygenase that repairs N-alkylated DNA lesions [30], [31]. In mammals, nine homologs of AlkB, i.e., ALKBH1-ALKBH8 and FTO, have been identified [32], [33]. ALKBH3 and ALKBH2 are known DNA-repair proteins that protect the genomic integrity of mammalian cells [34], [35]. Unlike ALKBH2 that repairs DNA alkylation lesions in double-stranded DNA (dsDNA), ALKBH3 repairs N-methylated bases, for instance, m¹A, m³C, N³-methylthymine (m³T), and N¹-methylguanine (m¹G), in single-stranded DNA (ssDNA) or ssRNA [35]. ALKBH3 is coupled with the helicase activating signal cointegrator 1 complex subunit 3 (ASCC3) to unwind dsDNA for dealkylation [36]. ALKBH3, also known as prostate cancer antigen-1 (PCA-1), is highly expressed in a few human cancers and promotes apoptotic resistance and angiogenesis in prostate cancer and pancreatic cancer [37], [38]. Due to its demethylation activity of m¹A in vitro, expectedly, ALKBH3 is confirmed to function as an m¹A mRNA demethylase in vivo [8]. Notably, a recent study reports that ALKBH3 can function as a tRNA demethylase to promote protein synthesis in cancer cells [39]. ALKBH3 displays demethylation activities toward m¹A, m³C, and m⁶A in tRNA to enhance protein translation efficiency in vitro, whereas knockdown of ALKBH3 increases m¹A levels in tRNA and decreases protein synthesis in cancer cells [39].

ALKBH1 shows the highest sequence homology with E. coli AlkB and exhibits enzymatic activities toward a wide range of substrates. For instance, ALKBH1 has weak demethylation activity toward m³C in ssDNA [40]. ALKBH1 also acts as a histone dioxygenase during neural development, which specifically removes the dimethylation of K118 or K119 on histone H2A [41]. However, Wu et al. did not observe the demethylation activity on histone H2A in ALKBH1^−/− embryonic stem cell lines. Rather, they found ALKBH1 is a DNA N⁶-methyladenine (m⁶dA) demethylase that regulates epigenetic gene silencing [42]. ALKBH1 also exhibits apurinic/apyrimidinic (AP) lyase activity, cleaving DNA at abasic sites via a β-elimination mechanism [43]. Interestingly, ALKBH1 has been recently identified as an m¹A demethylase in cellular tRNAs [13]. Knockdown of ALKBH1 increases the m¹A methylation level and the cellular level of the targeted tRNA_i^Met. These data indicate that the function of m¹A58 is to stabilize tRNA_i^Met, leading to the attenuated initiation of protein translation. Demethylation of other targeted tRNAs by ALKBH1 affects translation elongation by decreasing the usage of tRNAs in protein synthesis [13]. Furthermore, ALKBH1 can hydroxylate 5-methylcytosine (m⁵C) to 5-formylcytosine (f⁵C) in cytoplasmic tRNA^Leu and mt-tRNA_i^Met, resulting in an increase in mitochondrial protein translation [44], [45]. The diverse roles and enzymatic activities of ALKBH1 are in agreement with its multiple cellular localizations in the nucleus, cytoplasm, and mitochondria.

Roles of m¹A in mRNA and tRNA

m¹dA in DNA is considered as a form of DNA damage modification, which leads to false base pairing and genomic mutations, and thus it has to be repaired [46]. On the other hand, endogenous enzymes are present to install the modification of m¹A in RNA and play important roles in tRNAs and mRNAs.

The idea that m¹A58 might be vital to tRNA structure was first proposed in studies on the 3-dimensional structure of S. cerevisiae initiator tRNA_i^Met [47]. Hydrogen bonds between adenosines A20, A54, and A60 play important roles in stabilizing yeast tRNA_i^Met. In this substructure, m¹A58 is linked to A54 and A60 via hydrogen bonds, which is also structurally critical for stabilization [47]. The presence of m¹A58 in all eukaryotic initiator tRNAs implies that m¹A58 is indispensable for tRNA structure and stability [12], which has been experimentally validated in human cells [13]. Knockdown of ALKBH1, the m¹A58 eraser, increases the cellular level of tRNA_i^Met [13], supporting the function of m¹A58 to maintain a stable structure of tRNA_i^Met. Furthermore, m¹A58 in human tRNA₃^Lys is important for HIV replication. An 18-nucleotide sequence in the HIV-1 genome (GenBank accession No.: NC 001802) is complementary to the last 18 nucleotides of human tRNA₃^Lys. This sequence can be used as the primer binding site (PBS) for a virus to form a hybrid with tRNA₃^Lys and synthesize its minus strand cDNA in the presence of reverse transcriptase. To precisely reproduce the 18 nucleotides of tRNA₃^Lys in mature cDNA, m¹A58 tRNA₃^Lys is required to terminate the reverse transcription process when the last 18 nucleotides of tRNA₃^Lys have been copied [48], [49], [50].

As m¹A is a newly-discovered modification in mRNA and lncRNA, its function has not been extensively explored yet. Up till now, only the function of m¹A at CDS in mt-mRNA has been confirmed in preventing the effective translation of modified codons due to the Watson-Crick disruptive nature of m¹A [9], [10]. Ribosome profiling data analysis suggest that m¹A at 5′ cap and 5′ UTR in nuclear mRNA might play a role in promoting translation [9].

Conclusion and perspective

The reversible m¹A modification in tRNA and mRNA uncovers new mechanisms underlying the epitranscriptomic regulation of gene expression. The presence and functions of m¹A in nuclear transcripts had been a subject of debate lately. Hopefully, the discovery of new m¹A writers, erasers, and readers in the future will provide convincing evidence to help resolving such argument. Although tRNA MTases have been found to modify a subset of m¹A sites in mRNA, the writer responsible for the majority of m¹A sites in mRNA is still unknown. Since ALKBH3 is a demethylase of mRNA m¹A, it is interesting to investigate whether its function in cancers is related to its demethylation activity toward mRNA m¹A, and whether other erasers for mRNA m¹A exist. Analogous to the positive charged cap m⁷G modification bound by a specific RNA-binding protein, mRNA m¹A could be recognized by reader proteins as well [51]. Compared to the extensively-studied m⁶A modification, our knowledge about m¹A in mRNA is limited. Functions of m¹A in mRNA and the underlying mechanisms need to be further investigated. It is anticipated that m¹A in mRNA and lncRNA might regulate RNA processing or protein translation in various manners: 1) m¹A is specifically bound by a reader; 2) m¹A controls RNA structure to affect protein-RNA interaction; 3) m¹A at CDS affects the base-pairing between codon and anticodon. It might be worth exploring whether m¹A occurs and plays important regulatory roles in other types of RNAs as well.

Competing interests

The authors declare that they have no competing interests.

Handled by Xiu-Jie Wang