Epitranscriptomics: RNA Modifications in Bacteria and Archaea

ABSTRACT

The increasingly complex functionality of RNA is contrasted by its simple chemical composition. RNA is generally built from only four different nucleotides (adenine, guanine, cytosine, and uracil). To date, >160 chemical modifications are known to decorate RNA molecules and thereby alter their function or stability. Many RNA modifications are conserved throughout bacteria, archaea, and eukaryotes, while some are unique to each branch of life. Most known modifications occur at internal positions, while there is limited diversity at the termini. The dynamic nature of RNA modifications and newly discovered regulatory functions of some of these RNA modifications gave birth to a new field, now often referred to as “epitranscriptomics.” This review highlights the major developments in this field and summarizes detection principles for internal as well as 5′-terminal mRNA modifications in prokaryotes and archaea to investigate their biological significance.

INTRODUCTION

Today, RNA is no longer seen as a mere intermediary, transferring information from genes to proteins. It has become more and more obvious that RNA plays additional biological roles in living organisms (1, 2). Such roles are highly diverse, including catalytic and gene-regulatory functions. While the catalytic roles in present-day biology appear limited, regulatory RNAs are abundant in all kingdoms of life and control central biological processes from cell cycle progression, differentiation, adaptation, and stress response to pathological processes, such as carcinogenesis or inflammation (3, 4).

This increasingly complex functionality of RNA is contrasted by its simple chemical composition. RNA is generally built from only four different nucleotides (adenine, guanine, cytosine, uracil), but a set of >160 chemical modifications decorate RNA molecules to support their diverse coding, structural, catalytic, and regulatory functions. The location, abundance, and distribution of various types of RNA modifications are dependent on organism as well as on environmental conditions. Many modified nucleosides are conserved throughout bacteria, archaea, and eukaryotes, while some are unique to each branch of life (2). Most known modifications occur at internal positions, while there is limited diversity at the termini (5, 6).

The main goal of this review is to provide an overview of bacterial and archaeal mRNA modifications. For the sake of completeness, we only briefly summarize rRNA and tRNA modifications that have been extensively reviewed elsewhere (7, 8). Thus, this review aims to bring together the latest developments in the identification and characterization of internal as well as 5′-terminal mRNA modifications in prokaryotes and archaea.

The difference in profile and type of RNA modifications is an important feature that distinguishes the three kingdoms of life. Our general understanding of the molecular functions of RNA modifications is largely based on investigations on tRNAs and rRNAs (9, 10). Compared to other RNA species, tRNAs possess the highest number of RNA modifications, on average, 14 modifications per tRNA. A comparison of modified tRNA species from Archaea, Bacteria, and Eukarya reveals a core set of 18 modifications that occur in tRNA in all three domains of life (11). Currently, ∼60 rRNA and tRNA modifications are known in bacteria (5, 12). These include base isomerization, base alteration, ribose 2′-hydroxyl group methylation, and complex modifications involving the sequential addition of modifications or large chemical groups (5). While most modifications seem constitutive, certain tRNA modifications are dynamic and depend on growth rate/phase or defined stress conditions (10).

Intensive research into tRNA and rRNA modifications has extended their scope and contributed heavily to the development of new methods to identify as well as to characterize RNA modifications.

The detection and identification of RNA modifications has a long history and reaches back to the early days of molecular biology. Traditionally, RNA modifications were identified by thin-layer chromatography (TLC), anion-exchange chromatography, or UV spectrophotometry (Fig. 1A) (13–15). Early discoveries of modified nucleotides relied on the complete digestion of a given RNA molecule into mononucleotides or dinucleoside monophosphates by use of different types of nucleases. The resulting hydrolysates were analyzed by TLC. A two-dimensional TLC on cellulose plates is the most frequently applied version, which has been used to identify at least 100 chemically distinct nucleotides and dinucleotides (16). The use of pre- or postradioactive labeling procedures greatly facilitated the identification of even tiny amounts of modified nucleotides. More recently, liquid chromatography coupled with mass spectrometry (LC-MS) has been used to discriminate modified nucleosides based on both retention time and mass-to-charge ratios (13, 17–19). However, one major drawback of both techniques is the relatively large amount (often micrograms) of pure and homogeneous sample that is required for reliable detection. Such amounts are feasible for highly abundant RNAs like tRNA, rRNA, or transfer-messenger RNA but hard to obtain for less abundant RNAs like mRNA or small regulatory RNA (sRNA) (16). Another important drawback of the digestion of RNA to single nucleotides is the unavoidable loss of the sequence information of those RNAs that actually carry an RNA modification. To overcome these bottlenecks, the rapid development of novel high-throughput or next-generation transcriptome-sequencing techniques (RNA-Seq) (20) helped researchers to identify the location of RNA modifications and to reveal distinct distribution patterns of these modifications throughout the transcriptome (Fig. 1B).

FIGURE 1.

Identification of RNA modifications. (A) Identification of RNA modifications by a combination of digest to single nucleotides and chromatography. Total RNA is digested by nucleases (nuclease P1) and single building blocks separated by TLC or MS coupled with LC. (B) Identification of RNA modifications and their associated transcripts. Modified RNA is specifically enriched by a protocol that is based on antibody treatment (immunoprecipitation), an enzymatic reaction, or a chemical treatment. Afterwards, enriched, modified RNA is converted into cDNA to produce a library for NGS. The reads are mapped to the genome to identify the sequence of the transcripts that bear a specific RNA modification. Known or possible 5′-terminal (C) or internal (D) mRNA modifications in bacteria and archaea.

RNA-Seq usually addresses the complexity of prokaryotic and eukaryotic transcriptomes and can be defined as the massively parallel analysis of cDNA molecules by high-throughput sequencing. In comparison to the previously established microarray technology, RNA-Seq offers great advantages including lower cost; transcript profiling at single-nucleotide resolution; as well as a high dynamic range, sensitivity, and discriminatory power (21). In classical RNA-Seq experiments, RNA is converted to a cDNA library and PCR amplified. Following next-generation sequencing (NGS), the obtained short sequence reads are mapped onto a reference genome (20, 22).

Depending on the scientific question, different RNA-Seq protocols were established that use either poly(A)-tailing of RNAs combined with oligo-d(T)-priming of cDNA, or cDNA synthesis from a ligated RNA adapter (23). In recent years, RNA-Seq generally helped to clarify the nature of transcriptomes, to determine gene expression changes, and to identify coregulated genes. With this technique, extensive information on transcriptional start sites (TSSs) (21, 24), untranslated regions (UTRs) of mRNA genes, sRNA genes, and unknown open reading frames was provided (22). Aside from expression profiling, RNA-Seq has been applied to study RNA-protein interaction and the sum of actively translated mRNAs by so-called ribosome profiling (25).

In addition to transcriptome-wide analysis, RNA-Seq protocols were developed that allow the identification of biologically relevant internal RNA modifications. RNA-Seq has been combined with immunoprecipitation techniques (26–29), chemical derivatization (30, 31), or enzymatic treatments (32, 33) (Fig. 1B), facilitating the identification of several RNA modifications in prokaryotes, archaea, and eukaryotic organisms (34). The multitude of established RNA modifications were first designated as “RNA epigenetics” in 2010 (35). The dynamic nature of internal RNA modifications and newly discovered regulatory functions of some of these RNA modifications gave birth to a new field, now often referred to as “epitranscriptomics” (36–38).

In this review, several mRNA modifications that were identified at the 5′ terminus of RNA or at internal positions of mRNAs will be discussed in detail (Fig. 1C and D). Much attention has been drawn by eukaryotic mRNA modifications. However, transcriptome-wide studies to characterize internal or 5′-terminal mRNA modifications in bacteria and archaea have surfaced only in the past 5 years. This review describes on the one hand the well-known 5′-triphosphorylated (5′-PPP), 5′-monophosphorylated (5′-P), and 5′-hydroxylated (5′-OH) termini. On the other hand, the newly identified 5′-diphosphorylated RNA (5′-PP) and NAD-capped RNA (5′-NAD-RNA) are highlighted (Fig. 1C). In contrast to 5′-terminal RNA modifications, a significantly larger number of internal RNA modifications are known. Therefore, this review focuses on five mRNA modifications (N⁶-methyladenosine [m6A], 5-methylcytosine [m5C], inosine [I], pseudouridine [Ψ], and 2′-O-methylated nucleotides [Nm]) that have been identified in bacterial or archaeal mRNAs or are very likely to be present in these organisms.

HIGH-THROUGHPUT SEQUENCING APPROACHES TO IDENTIFY INTERNAL RNA MODIFICATIONS

The combination of RNA-Seq with immunoprecipitation or chemical treatments facilitated the identification of several internal RNA modifications: m6A, m5C, I, Ψ, and Nm (Table 1; Fig. 1) (5). The existence of these universal modifications has been proven already in large numbers of archaeal, bacterial, and eukaryal tRNAs (reviewed in references 8, 10, 11, 39, and 40). However, the presence of these modifications in mRNA is poorly understood. In this section, the current scientific knowledge about m6A, m5C, inosine, Ψ, and Nm, which are present in all kingdoms of life, is summarized.

TABLE 1.

Internal RNA modifications

Modification	Structure	Detection	Enzymes	Molecular rolesa

^a Abbreviations: RNAi, RNA interference; miRNA, microRNA.

I

Adenosine-to-inosine (A-to-I) editing is one of the most prevalent types of RNA editing, especially in higher eukaryotes. Inosine is well studied in tRNAs in all domains of life and is mainly present at position 34, which is the first nucleotide of the anticodon (wobble position); position 37 (following the anticodon); and position 57 (at the TΨC-loop) of the tRNA. Interestingly, at position 34, inosine is the final modified base, while at positions 37 and 57, inosine is further modified by methylation. A-to-I editing is generally performed by two groups of RNA adenosine deaminases that are active on either tRNA (adenosine deaminases acting on tRNAs [ADATs]) or mRNA (adenosine deaminases acting on mRNAs [ADARs]) (34, 41, 42). Inosine at position 57 has only been identified in archaea. In contrast, A-to-I editing at position 34 has been confirmed in bacteria and eukarya.

The conversion of adenosine to inosine in mRNA plays numerous roles in modulating gene expression, including recoding codons, altering alternative splicing, and regulating microRNA biogenesis and function (43). Inosine base-pairs with cytidine, rather than thymidine, in reverse transcription and is thus read as guanosine in cDNA. Based on this phenomenon, high-throughput RNA-Seq was performed and enabled transcriptome-wide identification of A-to-I edited sites. However, the simple conversion of RNA into cDNA and genome mapping of A to C results in background noise caused by single-nucleotide polymorphisms, somatic mutations, pseudogenes, and sequencing errors. Therefore, a new biochemical technique called inosine chemical erasing (ICE)-Seq was established to selectively identify A-to-I edited sites (Table 1; Fig. 2D). In this approach, acrylonitrile is used to selectively react with inosines in RNA, forming N¹-cyanoethylinosine (ce¹I) (44). Because ce¹I inhibits first-strand cDNA extension, inosine-containing RNA is eliminated, while unmodified RNA gives rise to full-length cDNA that is analyzed by high-throughput sequencing. Using ICE-Seq, Sakurai et al. have successfully identified >19,000 novel editing sites in the human adult brain transcriptome (45). ICE-Seq is suitable for identifying A-to-I editing in any organism. Its application to RNA isolated from bacteria or archaea may contribute to the identification of editing sites in mRNAs isolated from these organisms.

FIGURE 2.

Methods to detect internal RNA modifications. (A) Identification of m5C-modified mRNA by bisulfite sequencing. Selective conversion of cytosine to uracil by bisulfite ions, whereas m5C remains as cytosine. After reverse transcription, a cDNA library is prepared, analyzed by high-throughput sequencing, and mapped to the genome. Cytosine is mapped as thymine, whereas m5C residues read as cytosine. (B) Ψ-Seq: identification of pseudouridine-modified RNA. mRNA is treated with CMC, which selectively reacts with Ψ and causes a stop during reverse transcription. cDNA libraries are amplified and sequenced. The reads from a CMC-treated and nontreated control are compared to map pseudouridine-modified RNA. (C) Identification of methylated adenosine by m6A-Seq. Total RNA is fragmented to 100-nucleotide-long RNAs. m6A-specific antibodies are used to immunoprecipitate RNA. RNA is reverse-transcribed to cDNA and analyzed by high-throughput sequencing. These reads produce a peak whose summit reflects an underlying m6A residue. (D) Identification of adenosine-to-inosine editing: ICE-Seq. ICE is based on cyanoethylation of inosine by acrylonitrile combined with reverse transcription and high-throughput sequencing. Inosine pairs with cytosine (control), whereas the chemical modification of inosine results in a stop during reverse transcription. Erased G signals originate from inosines in the sequence map of cDNAs and are finally used to identify A-to-I editing sites. (E) Mapping of 2′-O-methylated nucleotides (Nm) by Nm-Seq. Fragmented RNA is subjected to iterative oxidation-elimination-dephosphorylation cycles that remove 2′-hydroxylated nucleotides in the 3′-to-5′ direction. Internal 2′-O-methylation sites stay intact, and fragments ending with 2′-hydroxyl are finally blocked by an incomplete oxidation-elimination cycle. 2′-O-methylated RNA fragments are ligated to an adapter. After library construction and high-throughput sequencing, reads are mapped to the genome. At 2′-OMe sites, an asymmetric coverage profile is observed whose uniform 3′ end corresponds to the methylation position.

Recently, Bar-Yaacov et al. identified 15 novel A-to-G RNA-editing events in prokaryotic mRNAs (46). Instead of using ICE-Seq, they sequenced in parallel RNA and DNA isolated from Escherichia coli and identified editing events as base difference between DNA and RNA sequence. The editing itself, performed by the tRNA-specific adenosine deaminase TadA, recodes the protein sequence, which potentially affects protein function and cell physiology (46).

m5C

Another recently characterized RNA modification is m5C. The methylation of cytosine in DNA (m5dC) is a widespread epigenetic marker and has been extensively studied in genomic eukaryotic DNA (47). This modification is usually detected in DNA by bisulfite treatment (48). Bisulfite causes the deamination of unmethylated cytosine to uracil, while leaving methylated cytosine intact. However, the conversion of m5dC is performed under harsh reaction conditions (high pH) that are detrimental to the stability of phosphodiester bonds in RNA. Thus, a modified bisulfite treatment protocol was developed that allowed m5C-site detection in tRNA and rRNA (49) (Table 1; Fig. 2A). In 2012, the first m5C methylome was obtained in a transcriptome-wide manner by combining the modified bisulfite treatment and high-throughput sequencing (50). m5C has been identified in noncoding RNAs including tRNA and rRNA. Especially in tRNAs, m5C stabilizes the secondary structure and changes the anticodon stem-loop conformation (39). Moreover, >8,000 potential m5C sites in human mRNA were identified. However, due to possible incomplete conversion of regular cytosine in double-stranded RNA regions and to the presence of other cytosine modifications resistant to bisulfite treatment, it has been suggested that these sites include potential false positives from stochastic nonconversion events (51, 52).

The first studies to identify m5C in bacteria and archaea were reported in 2013. RNA bisulfite sequencing was used on total RNA isolated from both Gram-positive (Bacillus subtilis) and Gram-negative (E. coli) bacteria, an archaeon (Sulfolobus solfataricus), and a eukaryote (Saccharomyces cerevisiae) (28), followed by massively parallel sequencing to map m5C. In this study, the known m5C residues, identified in rRNAs in previous studies, were confirmed for all organisms. However, m5C seemed to be absent from any of the E. coli and B. subtilis mRNAs. Intriguingly, m5C has been mapped in S. solfataricus mRNAs. In addition, a first consensus motif (AUCGANGU) was identified that directs methylation in this archaeon. These results were the first evidence for mRNA modifications in archaea, suggesting that this mode of posttranscriptional regulation extends beyond the eukaryotic domain (28).

Methylation of the cytosine is performed by specific m5C methyltransferases (m5C-MTases), which transfer methyl groups from S-adenosylmethionine (SAM) to form m5C (53). The first characterized m5C-MTase family member is RsmB, which mainly modifies bacterial 16S rRNA (54). Thus far, >30 homologous proteins have been identified by sequence analysis. Two RNA methyltransferases, NSun2 and DNMT2, have been found to catalyze m5C methylation in higher eukaryotes (55, 56). Recently a cysteine-to-alanine mutation (C271A) (57, 58) in human NSUN2 has been applied to identify m5C sites (33). The mutation inhibits the release of the enzyme from the protein-RNA complex, resulting in a stable covalent bond between NSun2 and its RNA targets (33). Combining the methyltransferase activity with cross-linking by UV light successfully identifies the targets of NSun2 throughout the transcriptome. In addition to the described protocols, m5C sites could be analyzed and verified by m5C RNA immunoprecipitation (m5C-RIP) approaches (28). Using a monoclonal antibody that specifically binds 5-methylcytosine, m5C sites of S. solfataricus mRNA, identified by bisulfite sequencing, could be validated. Such immunoprecipitation procedures allow the detection of low-abundant methylated RNAs without the requirement of extremely deep sequencing and can confirm the results obtained by high-throughput sequencing.

m6A

The most abundant internal modification in eukaryotic mRNA is methylation at the N⁶ position of adenosine. In fact, 0.2 to 0.6% of all adenosines in mammalian mRNA are methylated at position 6 (59). This modification was identified using a combination of radioactive labeling and chromatography >40 years ago (60). However, a transcriptome-wide profile of m6A in mammalian cells was not generated until the development of new methods in 2012, which are based on m6A-specific immunoprecipitation and high-throughput sequencing called m6A-Seq (26) and MeRIP-Seq (27) (Table 1; Fig. 2C). In both methods, purified mRNA is sheared into 100-to-150-nucleotide RNA fragments. m6A-containing RNAs are immunoprecipitated by an m6A-specific antibody. The enriched m6A-containing RNA fragments are then subjected to library construction and high-throughput sequencing. Both methods have identified ∼10,000 m6A peaks in the mammalian transcriptome and showed for the first time that m6A modifications are enriched in 3′ UTRs, peaking sharply near the stop codon. However, identifying the precise location of m6A within a transcript remained a challenge (34). Recently, UV-induced RNA-antibody cross-linking strategies have been adapted into the m6A-Seq and MeRIP-Seq protocols, allowing identification of base-resolution m6A methylomes. This technique, called photo-cross-linking-assisted m6A sequencing (PA-m6A-Seq) (61), is based on the incorporation of 4-thiouridine (^4SU) into RNA during cell growth. After m6A immunoprecipitation, the recovered m6A-containing RNA can be cross-linked to the anti-m6A antibody under UV light (366 nm) specific for ^4SU. Moreover, m6A-CLIP (cross-linking and immunoprecipitation) and miCLIP methods were developed that are based on photo-cross-linking as well (62, 63). Briefly, RNA fragments are immunoprecipitated and cross-linked to the antibody by UV light (254 nm). The protein-RNA cross-linking sites lead to patterned mutational or truncation profiles during reverse transcription, thereby revealing the precise position of m6A. These approaches have enabled important discoveries in the field of epitranscriptomics. Over the last years the biosynthesis and the removal of m6A by writers (METTL3 or METTL14) or erasers (FTO [64] and ALKBH5 [65]) have been studied in eukaryotic organisms in detail.

In contrast, studies describing the potential presence of m6A modifications in bacterial or archaeal RNA were lacking. m6A has been well described in the rRNA in bacteria (66). In E. coli, A1618 and A2030 of 23S rRNA are methylated by methyltransferases RlmF and RlmJ, respectively (67, 68). The deletion and overexpression of RlmF results in a loss of cell fitness and growth (67), while an RlmJ mutant shows mild phenotypes under various growth conditions. In order to identify m6A modifications in bacterial mRNA, Deng et al. performed a highly sensitive LC-MS-based measurement of m6A/A ratios on purified mRNAs isolated from seven model bacteria (E. coli, Pseudomonas aeruginosa, Pseudomonas syringae, Staphylococcus aureus, B. subtilis, Anabaena sp. PCC 7120, and Synechocystis sp. PCC 6803) (63). The highest m6A/A ratios were detected in the Gram-negative species E. coli, P. aeruginosa, and P. syringae (>0.2%). In E. coli, 265 m6A peaks were identified, corresponding to 213 mRNAs and 15 sRNAs. In P. aeruginosa, 105 m6A peaks were determined that were mapped to 68 mRNAs and 2 sRNAs. Interestingly, the majority (70%) of m6A sites occur within open reading frames and have a GCCAG consensus sequence. This contrasts with eukaryotes, where m6A sites are mapped around stop codons and in 3′ UTRs with an RRACU motif (26, 27, 69). Moreover, the majority of identified m6A-modified genes in E. coli and P. aeruginosa are involved in energy metabolism or belong to the class of sRNAs. It has been suggested that m6A plays a potential functional role in these processes (63). In contrast to eukaryotic organisms, the bacterial or archaeal RNA methyltransferases are still unknown. In E. coli, no METTL3 or METTL14 homologs have been identified yet (63). Moreover, unlike in eukaryotes, where m6A is a dynamic modification (64), m6A/A ratios seem to be stable under a variety of growth conditions. This suggest that prokaryotes and eukaryotes regulate m6A modification by different mechanisms that have to be explored (70).

Ψ, an Abundant Internal mRNA Modification

The most abundant RNA modification is Ψ, which is often referred to as “the fifth ribonucleoside” (71, 72). Ψ is generated via isomerization of uridine, catalyzed by two distinct mechanisms: the RNA-dependent mechanism with the box H/ACA RNPs and the RNA-independent mechanism with Ψ synthases (34). Historically, Ψ was first identified in rRNA and tRNA in all kingdoms of life (reviewed in references 8 and 72). Further occurrences of Ψ are known in snRNAs of various eukaryotes (5). Ψ’s functional relevance is well documented in rRNAs, where pseudouridylation is required for ribosome biogenesis and translational fidelity. Moreover, Ψ-modified snRNAs have been shown to contribute to pre-mRNA splicing (73). Many Ψ residues present in rRNAs and snRNAs are highly conserved across species. However, it has been shown that pseudouridylation can be induced in response to cellular stress and differentiation in yeast (74). These findings suggest that inducible pseudouridylation could be a lot more widespread and may provide a dynamic regulatory mechanism for RNA function in all kingdoms of life (2, 75).

To identify and to map Ψ-modified positions across the entire transcriptome with single-nucleotide resolution, different high-throughput sequencing methods (Pseudo-Seq, Ψ-Seq, PSI-Seq, and CeU-Seq) have been published (34, 75) (Table 1; Fig. 2B). These methods are based on selective chemical labeling of Ψ by the carbodiimide N-cyclohexyl-N′-β-(4-methylmorpholino)ethylcarbodiimide p-toluenesulfonate, known as CMC. The modified CMC-Ψ adduct results in a termination of cDNA synthesis. Next, cDNA strands are processed into cDNA libraries and analyzed by high-throughput sequencing. Finally, reads are mapped to the genome to detect Ψ sites across the entire transcriptome at single-base resolution.

Using these methods, Ψ was also found to be present in eukaryotic mRNA (76), although the biological function of such mRNA pseudouridylations remains enigmatic. Nevertheless, Ψ is abundant in mammalian mRNA, with a Ψ/U ratio of about 0.2 to 0.6% in human cells and mouse tissues (77). However, proof for the existence of Ψ modifications in bacterial or archaeal mRNA is lacking (70).

Nm

Next to the already described Ψ, the methylation of the 2′ hydroxyl group of the ribose (Nm, for unspecified 2′-O-methyl nucleoside) moiety is the most abundant RNA modification in rRNA (78). 2′-O-methylation changes the biophysical and biochemical properties of the nucleotides that carry this modification. For instance, 2′-O-methylation alters the hydration sphere around the oxygen and impacts sugar edge interactions and sugar pucker conformation. Moreover, 2′-O-methylation increases the stability of RNA against alkaline hydrolysis. 2′-O-methylation has been extensively studied for a number of years in order to identify functional and mechanistic links between this modification and specific biological pathways. For instance, 2′-O-methylation in RNA modulates the biogenesis and activity of the ribosome (79). However, the detailed roles of 2′-O-methylation in rRNA are not yet well understood (80).

The identification of 2′-O-methylation sites traditionally relied on the property of Nm to pause reverse transcription in the presence of limiting amounts of dNTPs (81). This mapping method was recently adapted to a high-throughput sequencing format to identify eukaryotic 2′-O-methylated rRNAs (2OMe-seq) (82).

In addition, a new method, called RiboMeth-Seq, was developed that allows the detection of relatively small changes (>10%) in Nm profiles (83). RiboMeth-Seq is based on the resistance of 2′-O-methylated ribose to alkaline hydrolysis, leading to underrepresentation of these positions among read starts or ends to provide a negative readout of the methylation landscape. However, other methods to detect Nm positions, present in relatively rare RNA molecules (e.g., mRNA), Nm-Seq and RibOxi-Seq (84, 85), were established based on the different chemical properties of nucleosides with 2′-OH and 2′-OMe (Table 1; Fig. 2E). The Nm-Seq protocol applies classical periodate oxidation of ribose 2′,3′ vicinal diols to yield a dialdehyde intermediate that undergoes spontaneous β-elimination under mildly basic conditions (85). After dephosphorylation of the resulting 3′ monophosphate, another oxidation-elimination-dephosphorylation (OED) is carried out. The progressive shortening process comes to a halt when a 2′-O-methylation site is reached, as it lacks vicinal diols. After several iterative cycles, the library is enriched in fragments ending with Nm, which are preferentially ligated to the 3′ adapter and analyzed by high-throughput sequencing.

The development of a variety of different sequencing techniques facilitated the identification of Nm modifications in eukaryotes. Ribose-methylated bases have been identified in rRNA, tRNA, snRNA, microRNA, and mRNA (59). In addition, Nm has been found to be associated with the eukaryotic m7G caps (86) and is involved in host pathogen responses (87, 88).

However, as of yet, 2′-O-ribose methylation has not been observed in archaeal and bacterial mRNA. Only modifications in tRNA (88) or rRNA have been described. In archaea, 2′-O-methylation of nascent rRNA molecules is triggered by RNP complexes containing small C/D box s(no)RNAs, which identify targets for methylation of the ribose. It has been hypothesized that 2′-O-methylation contributes to the folding, structural stabilization, assembly, and function of the 23S rRNA within the large subunit of the ribosome in archaea (89).

In bacteria, 2′-O-methylation modifications are relatively rare and introduced by dedicated site-specific or region-specific methyltransferases. A 2′-O-methylation of guanosine at position 18 (Gm18) of E. coli tRNA^Tyr has attracted great interest. This methylation was found to suppress innate immune activation in human innate immune cells (88). However, little is known about Nm sites in archaeal and bacterial mRNA. The application of Nm-Seq or RibOxi-Seq to new species will contribute to a better understanding of the biological relevance of this RNA modification.

5′-TERMINAL MODIFICATIONS OF RNA

In eukaryotes as well as in prokaryotes, the 5′ end of newly synthesized RNAs bears a triphosphate derived from the first transcribed nucleotide. Depending on the cellular processing, the 5′ termini can be converted into a diphosphate, monophosphate, or hydroxyl function. Recent evidence indicates that the 5′ status is an important determinant for molecular recognition by RNA-processing enzymes (90–92), for instance, the 5′-monophosphate-assisted RNA cleavage by RNases (93).

Efforts toward detecting modifications at the 5′ terminus of RNA started with the identification of the eukaryotic cap. In the mid-1970s, biochemical analyses revealed that certain viral and eukaryotic mRNAs have a cap—a methylated guanosine residue linked to the 5′ end of mRNA through an inverted 5′-to-5′ triphosphate bridge (94, 95). Functions associated with the 5′ cap are pre-mRNA splicing, 3′-poly(A) addition, overall stability, nuclear exit to the cytoplasm, protein synthesis, and mRNA turnover induced by decapping (96). The methylated guanosine and the neighboring nucleotides can carry further modifications (e.g., O- and N-methylations). Decapping and the conversion to 5′-monophosphorylated RNA inhibit translation initiation and trigger RNA degradation by 5′-to-3′ exonucleases (e.g., Xrn1) (97–100). In contrast to the eukaryotic kingdom, it was believed that most primary transcripts in bacteria and archaea possess a 5′-end triphosphate moiety that is derived from the first transcribed nucleotide (101, 102). In addition, certain primary bacterial transcripts were identified that possess a 5′ hydroxyl group and are generated by primer-dependent transcription initiation in E. coli and Vibrio cholerae (103, 104).

Primary Transcripts: 5′-PPP-RNA, 5′-P-RNA, and 5′-OH-RNA

The annotation of TSSs is essential for analyzing promoters, 5′ UTRs, and operon architecture and for discovering novel transcripts. Differential RNA-Seq (dRNA-Seq) approaches, selectively analyzing primary transcripts with 5′-PPP termini, have been developed in recent years. These TSS mapping techniques generally combine enzymatic treatments and adapter ligation with RNA-Seq techniques (21, 24) to identify primary transcripts (5′-PPP-RNA) (Fig. 3A). Briefly, total RNA is treated with 5′-P-dependent terminator exonuclease (TEX), which specifically degrades 5′-P-RNAs. 5′-PPP-RNAs are not degraded and are therefore enriched in the sample. Afterwards, the RNA is treated with tobacco acid pyrophosphatase (TAP), converting 5′-PPP-RNA into 5′-P-RNA. The 5′-P-RNA is finally ligated to an adapter and analyzed by high-throughput sequencing. The sequencing of dRNA-Seq libraries yields characteristic enrichment patterns of the cDNA coverages at the 5′ end of primary transcripts, which can be used to annotate TSS. dRNA-Seq has been applied to >30 organisms, including diverse bacteria and some archaea (21). In addition, Ettwiller et al. developed another technique, called Cappable-seq, to identify TSS at single-base resolution (102). They perform an enzymatic labeling of 5′-PPP termini with biotin, which enables the subsequent enrichment and high-throughput sequencing of primary transcripts.

FIGURE 3.

Identification of 5′-terminal RNA modifications. (A) Differential RNA-Seq to identify primary transcripts. Total RNA is treated with a 5′-P-dependent exonuclease that degrades specifically 5′-P-RNA. 5′-PPP-RNA is converted into 5′-P-RNA enzymatically by TAP or similar enzymes. The 5′ end is ligated to an adapter sequence and the RNA reverse-transcribed into cDNA, which is analyzed by high-throughput sequencing. The reads are mapped to the genome to identify TSSs. (B) Schematic representation of the NAD captureSeq protocol that allows the identification of NAD-capped RNAs. Total RNA is treated with ADPRC from A. californica, which specifically catalyzes the transglycosylation reaction of NAD with 4-pentyn-1-ol. The product of the reaction is biotinylated by CuAAC (click reaction). The NAD-capped RNA is captured as well as enriched on streptavidin beads and ligated to an adapter. After on-bead reverse transcription and a second adapter ligation, the obtained cDNA is amplified by PCR and submitted to high-throughput sequencing.

In addition to 5′-PPP termini, primary transcripts can carry 5′-OH or 5′-P termini that are formed by primer-dependent transcription initiation. RNA polymerases can use 2-to-∼4-nucleotide RNAs, so-called “nanoRNAs” (105), to prime transcription initiation in vivo (106). For the identification of these transcripts carrying 5′ monophosphate or a 5′ hydroxyl group, several NGS strategies were developed. 5′-Hydroxyl RNA sequencing was established, which enables the specific capture of 5′-OH termini (107). The E. coli RtcB RNA ligase attaches an oligonucleotide linker to 5′-hydroxylated RNAs that are finally analyzed by NGS. Moreover, 5′-P-RNAs are sequenced by 5′-RNA-Seq methods, which are based on the ligation of an adapter to 5′-monophosphorylated RNAs (104, 108).

5′-NAD-Capped RNA

The absence of 5′-capped RNA was considered as one of the hallmarks of prokaryotic gene expression. Until recently, the only known structure in prokaryotic RNA that resembles a cap was the covalent RNA adenylate intermediate involved in RNA ligation pathways (5′-AppRNA), in which an adenosine is linked to the RNA fragment via a 5′,5′ pyrophosphate bridge (109). Biological roles of these AppRNAs, besides ligation, have not been reported.

In 2009, the Liu group provided evidence that RNA can contain nucleotide-linked cofactors like the ubiquitous redox cofactor NAD (110) or coenzyme A (CoA) (Table 2) (111). Using an MS-based technique that requires complete hydrolysis of RNAs to the single-nucleotide level, they identified NAD in total RNA isolates from E. coli and Streptomyces venezuelae (110). Another publication in 2009 reported that CoA is linked to RNA species, and experimental evidence suggested attachment at the 5′ end of the RNA (111). However, both studies combined hydrolysis of the RNA with MS to detect the nucleotide-linked cofactors. Therefore, the sequences that bear the NAD modification initially remained unknown. Approaches that specifically enrich for NAD-modified RNA from total RNA were lacking.

TABLE 2.

Cofactor-capped RNAs

Modification	Structure	Detection	Enzymes	Molecular roles

The techniques that had been successfully applied to identify and analyze internal RNA modifications, such as cross-linking or immunoprecipitation, could not be directly transferred to identify NAD-modified RNA. Antibodies offer too weak binding affinities for NAD, and cross-linking approaches generally result in a nonspecific enrichment of RNA. Thus, to isolate NAD-RNA from biological samples, a new capture technique was required, which had to be specific for the NAD moiety without damaging the RNA molecules. Inspired by the identification of cofactor-modified RNA in 2009 (110), Cahovà et al. developed a NAD-specific chemoenzymatic capture approach, called NAD captureSeq (Fig. 3B), which enabled specific enrichment of NAD-modified RNA from total RNA and the analysis of the enriched RNA by high-throughput sequencing (32, 112).

The NAD captureSeq protocol combines enzymatic transglycosylation with “click” chemistry (113, 114) and can be split into three parts: (i) isolation of NAD-modified RNA, (ii) cDNA preparation, and (iii) validation of the NGS hits. To enrich for 5′-NAD-modified RNA, total RNA is treated with ADP-ribosyl cyclase (ADPRC) from Aplysia californica. This enzyme, previously described to replace nicotinamide with other N-heterocyclic nucleophiles (115), was surprisingly found to be capable of catalyzing the transglycosylation reaction of NAD with alkynyl alcohols. This enzyme can be applied to introduce selectively a “clickable” pentynyl handle in place of the nicotinamide. The product of the reaction is biotinylated by copper-catalyzed azide-alkyne cycloaddition (CuAAC). To ensure specific isolation of NAD-modified RNA from total RNA, the biotinylated, formerly NAD-modified RNA is captured and enriched on streptavidin beads. In the next step, enriched RNAs that were 5′-NAD-modified before are converted into cDNA. Finally, a combination of adapter ligation and on-bead reverse transcription can be applied to obtain a cDNA library that is submitted to high-throughput sequencing and mapped to the genome (112).

Based on the identification of NAD in total RNA isolated from the well-studied bacterium E. coli by Chen et al. in 2009 (110), the NAD captureSeq protocol was applied to E. coli total RNA for the first time. To identify enriched NAD-RNA in the NAD captureSeq sample, two negative controls that lack either ADPRC (minus ADPRC) or the pentynol were applied. In both cases a “clickable” pentynol cannot be attached to the NAD moiety, which makes biotinylation by CuAAC reaction impossible. The comparison of the NGS reads from the fully treated sample with those from appropriate controls revealed strong enrichment of a specific set of sRNAs, e.g., GadY, GcvB, ChiX, McaS, RNAI, and CopA. Those enriched candidates were reported to act by different pathways (116–126). Interestingly two sRNAs were enriched (RNAI and CopA) that are encoded on plasmids and control the plasmid replication, a mechanism that has been studied extensively since the 1980s (118, 120). Intriguingly, other sRNAs were not found to be enriched. Another set of enriched sequences represented 5′ fragments of different mRNAs, e.g., gatY, pgk, hdeD, and leader peptides ilvL and hisL, encoding either enzymes involved in cellular metabolism or leader peptides with known regulatory functions (32). These 5′ fragments were homogeneous in size and likely represent unknown sRNAs that exist in the cell in a NAD-modified state (32). Interestingly, tRNAs and rRNAs were not enriched in the NAD captureSeq approach.

The results of the NAD capture experiments were confirmed by different biochemical analyses, including enzymatic digests and MS. Especially for RNAI, the NAD modification was validated by LC-MS and quantified by a biochemical, ligation-based assay to 15% (32).

To identify characteristic features of the enriched RNAs, the NAD-modified RNAs were analyzed for common structural and sequence properties (32). The only feature that most enriched genes share is the TSS (+1 position) (32). The promoters predicted to generate the enriched NAD-modified RNAs possess an adenosine at the start site, which is in agreement with published TSS mapping data (24, 127). However, using bioinformatic predictions and database scans, no common sequence or structural features in promoters, transcripts, or transcription factor binding sites were identified for the NAD-modified RNAs (32).

Inspired by the identification of NAD-modified RNA in E. coli, the NAD captureSeq protocol was applied to total RNA isolated from yeast (128) and human cells (129). Intriguingly, a subset of eukaryotic RNAs (mRNA, snoRNAs, and small Cajal body RNAs [scaRNAs]) has been reported to possess a NAD-cap, which was confirmed by a biochemical shift assay. However, the presence of eukaryotic NAD-capped RNA has not yet been validated by MS.

The identification of NAD-RNA in archaea or bacterial species other than E. coli is still lacking and needs to be investigated.

Biosynthesis of NAD-Capped RNA

RNA is a central molecule in all kingdoms of life, as it connects the storage of genetic information in the form of DNA to its translation into proteins. DNA is thereby transcribed into RNA via DNA-dependent RNA polymerases. In eukaryotes, RNA modifications including the synthesis of the mRNA cap are generally generated post- or cotranscriptionally (Fig. 4A) (130, 131). In contrast to the eukaryotic m7G-cap synthesis, the incorporation of NAD seems to occur during transcription initiation (128, 132, 133). One observation in support of this assumption is the nucleotide sequence of those RNAs that carry a NAD modification. All NAD-RNAs that were identified by the NAD captureSeq approach possess as first nucleotide an adenosine. NAD contains an adenosine substructure that is recognized by the polymerase. During transcription initiation, NAD competes with ATP for the incorporation into the RNA chain at the +1 position. Systematic investigation revealed that the promoter nucleoside immediately upstream of +1 is a key determinant of coenzyme incorporation (132). Crystal structures of RNA polymerase in the presence of NAD and biochemical analysis confirmed that the promoter influences the NAD-capping efficiency (132, 134).

FIGURE 4.

Biosynthesis and removal of 5′-terminal RNA modification. (A) Synthesis of m7G-capped RNA in eukaryotes. 5′-PPP-RNA is synthesized by the RNA polymerase. The 5′-γ-phosphate of the nascent pre-mRNA is hydrolyzed by an RNA triphosphatase to 5′-PP-RNA. In the next step, a guanine monophosphate nucleoside is transferred to the 5′-diphosphate mRNA end by RNA guanylyltransferase to generate G-PPP-RNA. Finally, the guanine N7 position (blue) is methylated by SAM to form m7G-PPP-RNA by SAM. (B) Schematic representation of the synthesis of primary transcripts and cofactor-capped RNA in E. coli. Bacterial RNA polymerase is able to initiate transcription with a nucleotide triphosphate or an adenosine-containing cofactor, such as NAD, to generate 5′-PPP-RNA or NAD/cofactor-capped RNA. (C) Decapping of m7G-capped RNA in eukaryotic organisms. A decapping enzyme complex including Dcp2 removes the cap and converts the RNA in 5′-P-RNA that is targeted for degradation by 5′-dependent exonucleases like Xrn1. (D) 5′-End processing in E. coli. NAD-RNA is decapped by NudC into 5′-P-RNA. RppH processes primary transcripts and 5′-PP-RNA into 5′-P-RNA, which triggers RNase E-mediated degradation.

Interestingly, a recent report from Julius et al. provides further insight into the mechanism of cofactor capping (135). The authors show in vitro that adenine-containing cofactors (NAD, NADH, FAD, and CoA) can all be incorporated at the +1 position by the E. coli RNA polymerase. Moreover, they expanded the repertoire of potential capping molecules to include uridine-containing precursors of oligosaccharide and cell wall biosynthesis (UDP-glucose and UDP-N-acetylglucosamine) into their capping studies. However, the existence of NADH-RNA, FAD-RNA, UDP-glucose-RNA, and UDP-N-acetylglucosamine-RNA in vivo remains to be shown.

Removal of 5′-Terminal RNA Modifications

RNA capping is considered a hallmark of eukaryotic gene expression, in which a 5′,5′-triphosphate-linked 7-methylguanosine protects mRNA from degradation and modulates maturation, localization, and translation (96). The eukaryotic m7G-cap is removed by various decapping enzymes, such as Dcp2/Dcp1, thereby initiating different RNA decay pathways by exonucleases (136) (Fig. 4C). Dcp2 belongs to the class of Nudix enzymes. The acronym “Nudix” was coined in 1996 and is derived from the chemical nature of the substrates of the catalyzed reaction, namely, the hydrolysis of nucleoside diphosphates (NDP) linked to another compound X, generating nucleoside monophosphates (NMPs) and the phosphorylated compound X (137). Nudix hydrolases occur in all kingdoms of life and are usually small proteins (16 to 21 kDa). They are known to hydrolyze nucleoside triphosphates, nucleotide sugars, dinucleoside polyphosphates, dinucleotide coenzymes, and capped RNAs (138–140).

In E. coli, 13 different Nudix hydrolases are known to date. The first Nudix enzyme that has been described to be active on 5′-terminal RNA modifications is the RNA pyrophosphohydrolase RppH (Fig. 4D) (140, 141). E. coli RppH was discovered to convert triphosphorylated RNA to monophosphorylated RNA in vitro (141). It was assumed that the conversion of a 5′-terminal triphosphate to a monophosphate occurs in a single step by pyrophosphate removal (142). However, a recent study from Luciano et al. shows that 5′-PPP-RNA is first transformed (likely by another, not yet identified hydrolase) into diphosphorylated RNA, which accumulates to a high cellular concentration. The 5′-PP-RNA is the preferred substrate of RppH and converted into 5′-monophosphorylated RNA (142). These results indicate that a previously unrecognized event modulates 5′-end-dependent mRNA degradation in E. coli and suggest an important role for 5′-PP-RNAs in other cellular processes and organisms other than E. coli.

The conversion to 5′-P-RNA triggers their degradation by RNase E, which is an endoribonuclease (101). RNase E was recently identified to sense specifically the 5′-modification status of RNAs (triphosphate versus monophosphate), which determines their stability and processing (92, 143, 144). It has been shown that, similarly to triphosphorylated RNA, 5′-NAD-modified RNA is much less susceptible to cleavage by RNase E than 5′-P-RNA. Thus, NAD represents the first prokaryotic RNA cap (70, 145) that protects the RNA against processing by Nudix RNA pyrophosphatase RppH. Another Nudix enzyme, the NADH-pyrophosphohydrolase NudC (146), was identified to specifically convert NAD-RNA into 5′-P-RNA, thereby triggering RNase E-mediated decay (32) (Fig. 4D). Crystallographic and biochemical mutation analysis identified the conserved Nudix motif as the catalytic center of NudC, which needs to be homodimeric, as the catalytic pocket is composed of amino acids from both protomers. NudC is single-strand specific and has a purine preference for the 5′-terminal nucleotide (147, 148). The enzyme strongly prefers NAD-RNA over NAD and binds to a diverse set of cellular RNAs in an unspecific manner (147). The decapping activity of NudC could provide the bacterium E. coli with an additional mechanism to selectively initiate degradation for a subset of cellular RNAs orthogonal to RppH processing of 5′-PPP and PP-RNAs.

Characterization and Quantification of 5′-Terminal RNA Modifications

The discovery of differently modified 5′ termini opens up new questions regarding their biological role(s) (37, 70, 145, 149). To examine the modified RNA itself and the enzymes associated with these 5′-terminal RNA modifications, several in vitro and in vivo assays were established in the past (32, 141, 142, 150, 151).

The relative quantification of 5′-PP-RNAs in vivo is performed by a phosphorylation assay by capping outcome (PACO) (142) (Fig. 5A). This method is based on the substrate specificity of the RNA guanylyltransferase that is usually part of the eukaryotic capping machinery. The guanylyltransferase transfers a GMP selectively to RNAs with a 5′-PP end (152). After successful capping of 5′-PP-RNA to 5′-G-PPP-RNA, residual 5′-P-RNA and 5′-PPP-RNA is dephosphorylated to 5′-OH-RNA by alkaline phosphatase treatment. Afterwards, the 5′-G-PPP-RNA is enzymatically decapped by pyrophosphohydrolases, such as “cap-clip,” thereby generating a 5′-P end. Thus, only those RNAs that previously carried a 5′-PP end now have a 5′-P terminus and can therefore be ligated to an adapter. After adapter ligation, a gel shift of the original 5′-PP-RNAs can be detected by Northern blot (142). By allowing simultaneous detection of both the ligation product and the unligated transcript, Northern hybridization makes it possible to quantify their relative abundance and thus to calculate the percentage of the transcript that is diphosphorylated in vivo and in vitro. In a similar fashion, 5′-P-RNAs are detected and relatively quantified in vivo by a PABLO assay (phosphorylation assay by ligation of oligonucleotides) (150).

FIGURE 5.

Methods to relatively quantify and characterize 5′-terminally modified RNAs and their decapping enzymes in vitro and in vivo. (A) Schematic representation of the identification and relative quantification of 5′-PP-RNA (PACO assay) in vivo. The 5′ end of 5′-PP-RNA is capped with a guanosine. Subsequent treatment with alkaline phosphatase removes the exposed 5′-terminal phosphates of 5′-PPP-RNA and 5′-P-RNA but not the protected phosphates of the guanylylated G-PPP-RNA, which is then converted to 5′-P-RNA by treatment with a pyrophosphohydrolase. After adapter ligation and Northern blot analysis, the amount of 5′-PP-RNA can be quantified by a shift. (B) APB PAGE to quantify 5′-NAD-capped RNA in vivo. APB interacts with cis-diols of the ribose, which results in retardation of NAD-capped RNA during gel electrophoresis. (C) Identification of decapping enzymes by specific radioactive labeling of RNA in vitro. NAD-capped RNA is transcribed by the bacteriophage T7 RNA polymerase in the presence of radioactively labeled NAD. This technology enables specific radioactive labeling of each RNA with a single radioactive mark specifically located at the 5′ termini. After a decapping reaction, e.g., by NudC, 5′-P-RNA is generated. The accessible radioactive phosphate is removed by alkaline phosphatase treatment. The RNA is analyzed by PAGE.

To study cofactor-capped RNA in vitro and in vivo, acryloylaminophenyl boronic acid (APB) gel electrophoresis was recently applied to separate NAD- or FAD-modified RNAs from unmodified RNAs (PPP-/PP-/P-RNA) (151) (Fig. 5B). APB is copolymerized with polyacrylamide and forms a relatively stable complex with 1,2-cis-diols (153), which occur naturally at the 3′ end of RNA and in the nicotinamide riboside of NAD-modified RNA at the 5′ end. The transient formation of diesters between the immobilized boronic acid and the diols causes lower mobility of the modified RNAs, which is visualized by a shift on the gel (151). APB affinity gels can be used to study cofactor-modified RNAs with small amounts of material, and to rapidly screen for their relative abundance in total RNA while avoiding complex sample treatments (151).

Another method to study and to identify enzymes that are processing 5′-modified RNA is the specific radioactive labeling of 5′ termini (32, 154). For this purpose, RNA can be prepared by in vitro transcription using the bacteriophage T7 RNA polymerase radioactive nucleotides. To study the processing of 5′-modifications, compounds that contain radioactively labeled phosphorus (³²P or ³³P) are best suited. Today, various nucleotides like α-AMP (³²P-A), γ-ATP (³²PPP-A), or NAD (5′-N-P³²P-A) are commercially available that are specifically incorporated at the 5′ end of the RNA by the T7 RNA polymerase (155). These radioactive initiator nucleotides enable specific tagging of each RNA with a single radioactive label specifically located at the 5′ terminus (32, 154, 156, 157). Thus, a cleavage or processing of the 5′ end of the RNA by an enzyme can be directly followed by a decrease of the radioactive signal. In the context of the identification of decapping enzymes, a specific radioactive labeling of NAD-capped RNA was used (Fig. 5C). To study enzymes that might convert NAD-RNA into 5′-³²P-RNA, alkaline phosphatase was added to the reaction mixture to remove the [³²P]phosphate, which is only accessible after enzymatic processing of the NAD-RNA. The decapping activity of NudC and different NudC mutants was examined by such classical biochemical assays in vitro (32, 147, 154).

INFLUENCE OF RNA MODIFICATIONS ON THE TRANSLATION OF PROTEINS

RNA is a central molecule in all kingdoms of life, as it connects the storage of genetic information in the form of DNA to its translation into proteins. In eukaryotes, the 5′ terminus of mRNAs is generally modified by a methylated guanosine (m7G) cap, which is crucial for the initiation of protein synthesis (158–160). The protein synthesis itself can be divided into four parts—initiation, elongation, termination, and ribosome recycling—and is regulated by the interplay of the ribosome, tRNAs, mRNA, ribosome, RNA binding proteins, as well as noncoding RNAs. Studies that examine the influence of co- and posttranscriptional (m)RNA modification on protein synthesis have evolved in recent years. Since the identification of m6A, the modifications m5C, Ψ, and 2′-O-methyl have been found within the open reading frames of mRNAs, where their presence suggests an influence on the fate of the mRNA, ranging from maturation to its translation and degradation (161, 162).

The influence of the bacterial NAD-cap on translation is not known yet. In eukaryotic cells, NAD-capped mRNA was not translated (129).

Recently, tRNA modifications were identified to be linked to translation efficiency and decoding fidelity (163, 164) and to be involved in fine-tuning of stress-related genes by driving codon-biased translation (163). To investigate the influence of RNA modifications on translation, in vitro translation assays are generally performed (161, 165). In studies published by Hoernes et al., m5C, m6A, or 2-O-methylated nucleotides were introduced at each of the three positions within a codon of a bacterial mRNA and analyzed concerning their influence on translation. The data indicate astonishingly versatile effects on protein synthesis depending not only on the type of the RNA modification but also on the codon position (165). Incorporation of single modifications in specific codon contexts reduced protein production in E. coli by 20 to 70% or resulted in truncated proteins through site-specific ribosome stalling at modified codons. To affect gene function through regulated rewiring of the genetic code is one exciting potential for mRNA modifications. So far, few studies describing an influence of RNA modification on translation are published. However, the development of several approaches to identify the precise location of the RNA modification sets the foundation to study the effect of epitranscriptomics on translation in vitro and in vivo (52).

OUTLOOK

The recent advances in the field of RNA modifications clearly show that the epitranscriptome and its modifying enzymes form a complex constellation that holds widely diverse functions. Posttranscriptional RNA modifications allow additional control of gene expression, serving as powerful mechanisms that eventually affect protein translation. The boost in the development of techniques to detect RNA modifications has revealed complex networks of modified RNAs and proteins that hold widely diverse and partly unknown functions.

Despite these major achievements, the epitranscriptome has mostly been studied in eukaryotic organisms. In contrast, detailed studies about RNA modifications and their writers, readers, and erasers in archaea or bacteria have received little attention. Because there are substantial differences between prokaryotic and eukaryotic mRNA metabolism, the functional consequences of mRNA modifications are likely to differ as well. m6A has been reported in mRNA from E. coli and P. aeruginosa, while m5C has been mapped in S. solfataricus mRNAs, and I was detected in 15 mRNAs from E. coli. 2-O-methylated nucleotides and Ψ have so far been identified in bacteria only in rRNAs and tRNAs. To elucidate the function of internal as well as 5′-terminal RNA modifications, knowledge about the exact location (detection) as well as the amount present (absolute and relative quantification) is required. Right now, there are hurdles to overcome in the fast-developing field of epitranscriptomics. First, there is a need for orthogonal methods to detect the RNA modifications. Moreover, we need robust and sensitive methods that need less input and are able to detect even less abundant RNA modifications. However, it is obvious that the development of such tools remains challenging especially in the context of >160 known RNA modifications. But is there a technique to identify different epitranscriptomic marks within the same transcript? Recently, two single-molecule methods—SMRT sequencing and the Oxford Nanopore Technology (ONT)—have demonstrated specific and base-resolution detection of m6A in synthetic RNA molecules (34, 166, 167). These approaches could potentially be used to detect multiple RNA modifications simultaneously in a single transcript (167). However, the rapid development of techniques to determine RNA modifications has resulted in a boost of data that are mapped to the genome and have to be independently validated. Still, the error rate of Illumina, SMRT, and ONT sequencing might lead to misinterpretation of the data. Moreover, chemical treatments of RNA as well as the application of antibodies are prone to creation and inclusion of experimental artifacts. To overcome these bottlenecks, careful validation of the RNA-Seq data and direct proof of the identified RNA modification are needed. The development of MS technology (13, 168, 169), different quantification tools like SCARLET (170), ligation-based assays like PABLO or PACO (142, 150), and gel-retardation assays (151) set the foundation to determine modification stoichiometry (9).

Finally, the main question that remains to be answered is the analysis of the biological function(s) of an RNA modification. The majority of functional studies have been performed in eukaryotic organisms. Studies of readers, writers, and erasers in bacteria and archaea are generally the exception. Thus, research on classical model organisms like E. coli, Salmonella enterica serovar Typhimurium, or B. subtilis should be expedited to identify the relevance of RNA modifications in these kingdoms of life as well.