If the 5’ cap fits (wear it) – Non-canonical RNA capping

ABSTRACT

RNA capping is a prominent RNA modification that influences RNA stability, metabolism, and function. While it was long limited to the study of the most abundant eukaryotic canonical m⁷G cap, the field recently went through a large paradigm shift with the discovery of non-canonical RNA capping in bacteria and ultimately all domains of life. The repertoire of non-canonical caps has expanded to encompass metabolite caps, including NAD, FAD, CoA, UDP-Glucose, and ADP-ribose, alongside alarmone dinucleoside polyphosphate caps, and methylated phosphate cap-like structures. This review offers an introduction into the field, presenting a summary of the current knowledge about non-canonical RNA caps. We highlight the often still enigmatic biological roles of the caps together with their processing enzymes, focusing on the most recent discoveries. Furthermore, we present the methods used for the detection and analysis of these non-canonical RNA caps and thus provide an introduction into this dynamic new field.

KEYWORDS:

• RNA
• RNA cap
• RNA modifications
• dinucleoside polyphosphate
• NAD
• epitranscriptomics

1. Introduction

RNA modifications are present in most types of RNA, and they have a wide variety of roles from affecting RNA stability and metabolism to facilitating their cellular roles, guiding RNA-protein interactions, and more [1,2]. The first RNA modifications were discovered more than half a century ago, and there are more than 170 known chemical modifications of RNA to date [3]. For the sake of clarity, they can be separated into three main categories based on their relative position within the RNA molecule. (i) Tail modifications are at the 3’ end of the RNA and they are usually formed through the addition of non-templated nucleotides and are a key component of quality control and RNA decay pathways [4]. Although tails formed by each of the four RNA nucleotides have been observed, the most common ones are polyadenylation and polyuridylation [5,6]. (ii) Internal RNA modifications are contained within the transcript and regulate properties of the RNA ranging from structure and stability to binding proteins and translational fidelity [7]. Internal modifications are most abundant in transfer RNAs (tRNA) and ribosomal RNAs (rRNA) [8]. (iii) The 5’ ends of RNA can be modified by the addition of nonencoded nucleotides known as 5’ RNA caps (Figure 1), with 7-methylguanosine (m⁷G) being the most abundant and most extensively studied RNA cap in eukaryotes [9]. This conventional m⁷G cap is considered the canonical form of the RNA cap due to its prevalence and well explored role, whereas the less common and more recently discovered cofactor, metabolite and other caps are considered non-canonical.

Figure 1. Structures of cap1 and all non-canonical RNA caps and cap-like structures known to date.

2. Canonical capping

2.1 m⁷G

Historically, methylated nucleotides were considered to be unique to rRNA and tRNA, and the 5’ends of all mRNAs were thought to be triphosphorylated both in eukaryotes and prokaryotes. Then about half a century ago, the m⁷G cap was detected in messenger RNAs (mRNA) transcribed from the viral genomes of the cytoplasmic polyhedrosis virus and the vaccinia virus (dsRNA and DNA viruses respectively) [10,11]. This finding was swiftly followed by the discovery of the m⁷G cap in eukaryotic mRNA from HeLa cells where it was linked to the protection of the 5’end of the RNA from 5’ exonucleases [12]. It was quickly ascertained that the main function of the m⁷G cap is twofold. First, it protects the 5’ end of the RNA from degradation by 5’ to 3’ exoribonucleases, contributing to the stability of the transcript. Second, it is required for the initiation of translation [13,14]. The effects of the m [7]G cap on the initiation of translation were then linked to a cap binding protein, later termed the eukaryotic translation initiation factor 4E (eIF4E). Since then, it has been shown that the m⁷G cap is an integral component of eukaryotic translation, serving as the recognition point for the initiation of translation and regulating 5’ end interactions [15]. The discovery of eIF4E started a new avenue of research into the cap through the studies of RNA cap-binding proteins [16]. 5’ RNA capping, however, remained a process that was thought to be exclusive to eukaryotes.

The next 45 years shed more light on the plethora of roles of the m⁷G cap in mRNA. To name a few – the addition of the m⁷G cap to mRNA occurs co-transcriptionally during RNA polymerase II (RNA pol II) transcription, and it almost immediately regulates the ongoing process. The m⁷G cap recruits transcription regulators such as Mot1p that help the formation of the pre-initiation complex. This facilitates transcription initiation creating a sort of synergistic loop to promote gene expression [17]. In addition, the mRNA cap facilitates the recruitment of the nuclear cap-binding complex (CBC). The protein complex specifically recognises the cap structure and, in turn, recruits other proteins to the mRNA [18]. The addition of the CBC is also responsible for the recruitment of the small nuclear RNAs (snRNA), spliceosome formation, and subsequent splicing of the pre-mRNA [19,20]. Interestingly, even though the m⁷G cap is at the 5’end of the mRNA, it also plays a role in the 3’end processing of the transcript. Specifically, it seems as though the CBC stabilises the cleavage complex at the polyA site of the RNA [21]. The CBC also has a role in the regulation of histone mRNAs, which are not polyadenylated and have a 3’-end stem loop instead. Upon CBC knockdown, however, the histone mRNAs start to acquire a polyA tail, suggesting that the m⁷G cap and the CBC are also responsible for the 3’ processing of this specific type of transcript [22].

The m⁷G cap stabilises the RNA and protects the 5’ end from degradation by exonucleases and the removal of the cap or decapping is a key part of RNA turnover. The ability to control the decay of capped RNAs allows the cells to finetune gene expression and readily react to various stimuli [23]. A number of proteins have been implicated in mRNA decapping, namely members of the Nudix family of hydrolases. These enzymes hydrolyse nucleoside diphosphates linked to the moiety x (hence the name), where x stands for any moiety [24]. For instance, the human mRNA decapping enzyme 2 (hDcp2) is a Nudix protein cleaving the m⁷G cap from RNA [25]. There is a large number of Nudix enzymes, and they are extremely versatile in their choice of substrate [26]. Interestingly, Nudix enzymes have been shown to decap a number of non-canonical RNA caps as well, which will be further discussed throughout this review [27,28].

The m⁷G cap is also responsible for mRNA export from the nucleus, as the RNA export factor (REF) relies on the structure of the cap in order to bind to the transcript and transport it into the cytoplasm [29]. In the cytoplasm, the m⁷G cap and methylations on the first two nucleotides downstream, serve as a molecular signature helping the immune system distinguish between self and foreign RNA [30]. This link between the m⁷G cap, its methylated variants, and the innate immune response via RIG-I-like receptors is beyond the scope of this review, though it has been previously studied and described [31–33]. The same is true for the other roles of the m⁷G cap (translation, RNA decay, etc.) on mRNA, which have also been the subject of a number of reviews [34–36].

Long noncoding RNAs (lncRNA) are also RNA pol II transcripts and as such, they often have both the polyA tail and the m⁷G cap, which aid in nuclear export and RNA processing [37]. This is even more true for long intervening ncRNAs (lincRNA), though it is important to note that lincRNAs seemingly have their own unique processing mechanisms despite their similarity to mRNA [38]. MicroRNA precursors are also RNA pol II products, they are long unstructured transcripts also bearing the m⁷G cap [39]. The m⁷G is again responsible for the recruitment of export proteins; after cleavage and maturation, however, the cap also plays a role during the interaction with Argonaute proteins that preferentially bind to a 5’monophosphate [40]. While the effects of the m⁷G cap have common features in all of the aforementioned capped transcripts, the actual effects and further modifications of the cap vary greatly depending on the specific transcript, suggesting sophisticated means of regulation. Recently, it was also discovered that cells possess a cytoplasmic capping machinery, which adds another level of regulation of transcript fate [41]. The uncapped transcripts can be kept in a “frozen” state and activated by recapping in order to fulfil their roles [42].

2.2 TMG

The m⁷G cap is also a key component of the maturation of most small nuclear RNAs (snRNAs), specifically the ones that are transcribed by RNA polymerase II (Pol II) [43]. These methylated transcripts are exported into the cytoplasm, where they are stabilised by the SMN complex that facilitates several maturation steps [44]. These include additional methylations of the cap by the enzyme trimethylguanosine synthase 1 (TGS1) forming the 2,2,7-trimethylguanosine (TMG) cap. The TMG cap then initiates the import of the snRNA back into the nucleus by facilitating the assembly of the import complex [45]. Once back in the nucleus, the TMG cap is recognised by import receptors and facilitates spliceosomal assembly [46].

Several small nucleolar RNAs (snoRNA) are also capped. While the majority of snoRNAs are encoded in the introns of protein-coding genes and require splicing to be released, some have their own snRNA type genes [47]. Interestingly, these capped snoRNAs – specifically the U3 and U8 RNAs are retained within the nucleus of the cell despite being capped. In fact, these snoRNAs undergo hypermethylation in the nucleoplasm before their transport to the nucleolus, where they participate in rRNA biogenesis [48]. This nuclear retention is probably due to a discriminatory step during their maturation. The interaction with snoRNA binding proteins in the nucleus overrides the export signal of the cap, thus discriminating the hypermethylated transcripts from the ones being exported [49].

m⁷G capping usually occurs co-transcriptionally during RNA synthesis by by RNA pol II transcripts [50]. tRNAs are the products of RNA polymerase III (RNA pol III) and are transcribed as precursors (pre-tRNA). Pre-tRNAs are trimmed on both the 5’ and 3’ ends, nucleotides are then added to all the 3’ends as well as some 5’ends, and they are spliced and heavily modified [51]. Some pre-tRNAs, however, are also m⁷G-capped and subsequently hypermethylated. This protects them from being degraded by 5’exonucleases and provides an alternative means of tRNA export, although the exact steps and proteins implicated in this process are yet to be discovered [52].

3. Non-canonical cap-like structures

3.1 5'-γ-methyl-phosphate (γ-mNTP)

One of the earliest discovered non-canonical caps is the 5’ γ monomethyl phosphate. Unlike other caps discussed in this review, it is a simple cap-like structure without a flanking nucleotide-like moiety. Found in eukaryotic RNAs, it was initially identified on the RNA pol III transcript of U6 snRNA. It involves the addition of a methyl group to the 5’ triphosphate of uncapped RNAs [53]. U3 snRNA in plants, also transcribed by RNA pol III, has the same 5'-γ-methyl-phosphate cap, as do the small noncoding RNAs B2 and 7SK [54-56]. While the 7SK RNA has been shown to regulate transcription of RNA pol II through binding to the positive transcription elongation factor b (P-TEFb) while RNA B2 regulates Pol II by binding to the preinitiation complex and inhibiting it, it is unclear whether these functions are dependent on the 5'-γ-methyl-phosphate cap or not [57,58]. It has been shown that RNA pol III transcripts are often associated with the La protein that is necessary to stabilise them. This interaction is hindered by the 5'-γ-methyl-phosphate cap destabilising the transcript [59]. It remains to be seen if this cap-like structure has any further biological implications.

3.2 5'-α –(di)methyl-phosphate (α-m-NMP)

A similar cap-like structure created by methylating or dimethylating 5' monophosphate RNA also exists on some eukaryotic RNAs. Deposited on the oxygens of the α-phosphate by an RNA methyltransferase containing the bicoid interacting 3 domain (BCDIN3D), this cap negatively regulates the maturation of micro RNAs (miRNA). The dimethylation of pre-miRNAs causes reduced processing by the enzyme Dicer leading to a lower level of mature miRNAs [60]. The upregulation of BCDIN3D and consequently the dimethylation of specific miRNAs has also been implicated in breast cancer [61]. The let-7 pre-miRNA is processed more effectively with the 5'-α -dimethyl-phosphate and a depletion of BCDIN3D reduced the levels of the mature let-7 miRNA. This reduction then leads to changes in the transcriptome and proteome of cancerous cells, preventing tumour formation [62]. In contrast, pre-miR-145 is not processed as effectively when methylated and increased levels of mature miR-145 have been shown to suppress the tumorigenic phenotypes in breast cancer cells, suggesting a completely opposite effect that warrants further research [60]. Apart from breast cancer, the upregulation of BCDIN3D and downregulation of the miRNA miR-195 − 3p was observed in cervical cancer. As such, BCDIN3D and its non-canonical capping reaction are of interest as therapeutic targets in a number of cancers [63].

Interestingly, BCDIN3 also specifically recognises the 5' monophosphate of tRNA^His to monomethylate it with a higher efficiency than the aforementioned miRNAs [64]. However, the levels of tRNA^His are not changed upon BCDIN knockout and the only effect seems to be a decreased affinity towards histidyl-tRNA synthetase [65]. Even though BCDIN specifically recognises tRNA^His discriminating it from other tRNAs, the role of this 5'-α-methyl phosphate cap remains unknown [66].

4. Non-canonical caps

The two non-canonical cap-like structures mentioned above are specific in that they only carry methyl groups on the 5' terminal phosphates. They do not have a flanking nucleotide-like moiety as is the case for the canonical m⁷G cap or the TMG cap. However, the m⁷G cap is structurally similar to nucleotide coenzymes such as nicotinamide adenine dinucleotide (NAD), flavin adenine dinucleotide (FAD), and coenzyme A (CoA). This has raised the question whether they may exert functions similar to those of 5' RNA caps. Indeed, Escherichia coli (E. Coli) RNA polymerase and T7 RNA polymerase were shown to incorporate these molecules in vitro, using them as non-canonical initiating nucleotides (NCIN) to start transcription [67–69]. The first confirmation that these noncanonical caps exist in vivo came 30 years after the first in vitro experiments. Using liquid chromatography combined with mass spectrometry (LC-MS), CoA, its derivatives, and NAD were identified in small RNA fractions from two bacterial species – E. Coli and Streptomyces venezuelae [70,71]. Not only does this increase the chemical diversity of 5' RNA caps, but it also showed that RNA capping was not limited to eukaryotes, which was still the consensus at the time [72]. However, the identity of the capped transcripts and their role within cells was still unknown.

4.1 NAD

While the NAD-RNA cap was first detected using LC-MS, the first sequencing method directly showing a metabolite cap and its identity came in 2015 with the development of NAD captureSeq [71,73] (Figure 2). Taking advantage of adenosine diphosphate-ribosyl cyclase (ADPRC) from the sea slug Aplysia californica, which specifically transglycosylates NAD with alkynyl alcohols, creating a “clickable” RNA. The enrichment of such RNAs identified them as a subset of small regulatory RNAs and mRNA-like 5'-terminal fragments in the bacterium E. coli. The decapping and degradation pathway of these RNAs was also discovered and it was shown that the NAD cap protects the transcripts from degradation. The NAD-RNA, however, can be decapped by the Nudix enzyme NudC, triggering its degradation [73]. Since then, NAD captureSeq has been applied to other bacteria, archaea, and eukaryotes ranging from Saccharomyces cerevisiae to Metazoa, identifying a variety of capped transcripts [74–77].

Figure 2. Simplified schematics of NAD captureSeq, ONE-seq, and NAD tagSeq.

In 2018, CapZyme-seq was developed as an alternative method, using the bacterial RNA decapping enzyme NudC and the fungal decapping enzyme Rai1 to remove the metabolite RNA caps leaving behind a 5' monophosphate that can then be ligated to a barcoded RNA adapter [76]. Concurrently, a sample is treated with RNA 5'-polyphosphatase (Rpp), which does the same to uncapped RNAs. The two samples are compared identifying the NAD-RNA. NudC and Rai1, however, also cleave other non-canonical caps, which may result in false positives [78].

NAD tagSeq, later improved to NAD tagSeq II, uses the same principle as NAD captureSeq. The ADPRC-treated sample is clicked to a 3’ azide RNA – the “tag”. The tagged samples can then be directly sequenced by nanopore technology [79] (Figure 2). NAD tagSeq II replaces the copper click chemistry with the gentler strain-promoted azide-alkyne cycloaddition (SPAAC) that avoids RNA fragmentation caused by copper and allowing for a lower input of starting material [80].

SPAAC-NAD-seq combines the methods mentioned above and adds a pre-treatment step with an m⁷G antibody which removes around 80% of the canonical cap. This is done because ADPRC exhibits some activity towards the m⁷G cap and thus results in false positive hits [73,81].

ONE-seq minimises the number of steps throughout the NAD-RNA sequencing procedure. ADPRC directly biotinylates NAD-RNAs using N-[2-(2-hydroxyethoxy) ethyl]-biotinamide (HEEB) as substrate. The biotinylated NAD-RNAs are captured by streptavidin beads, and NudC then cleaves the diphosphate bridge of the NAD cap, while potentially ADPRC ligated m⁷G RNAs linked through a triphosphate stay on the beads [82] (Figure 2).

The two newest methods are called NADcapPro and “intramolecular circularization of Noncanonical-Capped RNA” – circNC. These methods work orthogonally to eliminate the limitations of the other methods. The canonical cap is removed using an m⁷G decapping enzyme – the yeast scavenger mRNA decapping enzyme DcpS [83]. ADPRC and SPAAC are used to biotinylate the RNAs and streptavidin beads are used for the pulldown [84]. CircNC takes advantage of the NAD decapping enzyme retinoic acid induced 1 (Rai1) and circularises the leftover monophosphate RNA that is reverse transcribed and amplified [84]. The detailed workings and principles of the sequencing methods are beyond the scope of this article, but they have been thoroughly analysed and explained in a recent review [85].

NAD-RNA in prokaryotes

The number of identified NAD-RNAs and their roles vary greatly depending on the organisms and biological context in which they were analysed. In bacteria, the NAD cap shields the transcripts from degradation by 5' exonucleases [73]. Interestingly, NADH, the reduced form of NAD, serves as an initiating substrate for the E. coli primase DnaG, suggesting a possible role in the generation of Okazaki fragments and regulation of DNA replication [86]. NAD-RNA has also been shown to play a key role during the infection of E. coli by the T4 bacteriophage. During infection, the phage uses its adenosine diphosphate (ADP)-ribosyltransferase ModB to attach NAD-RNA molecules to proteins, essentially RNAylating them. The inactivation of ModB slows down the lytic cycle of the bacteriophage [87]. Another bacterium, Staphylococcus aureus (S. aureus), codes for a quorum sensing regulatory RNA termed RNAIII [88]. RNAIII has the 5' NAD cap in 10%-35% of its transcripts and the amount of the cap inversely correlates with the cytotoxicity of the bacteria. An increase in NAD-RNAIII causes a decrease in the expression of several toxins produced by the bacterium [89]. Studies in Bacillus subtilis showed that mRNA is increasingly capped with NAD in the late exponential growth phase. Depletion of the B. subtilis deNADing enzyme BsRppH positively or negatively affected the expression of 13% genes clearly connecting NAD-RNA to gene regulation [90]. NAD-RNA and its decapping machinery were also recently detected in archaea and mycobacteria, where the NAD cap probably serves as a degradation marker [74,91]. Both bacteria and archea also share toll-interleukin-1 receptor (TIR) domain-containing proteins that cleave the nicotinamide (NAM) moiety of NAD-RNA to start the NAD-RNA degradation process. Given the conservation of the TIR domain across species, this might present a novel conserved mechanism of deNAMing NAD-capped RNA and preparing it for further reactions [92].

NAD-RNA in eukaryotes

NAD capture Seq was also applied to S. cerevisiae, which identified NAD-capped mRNAs of both cellular and mitochondrial origin, with a preference for intron-containing transcripts [75]. The vast majority of NAD-RNA, however, are short 5' mRNA ends. RNA pol II uses NAD to start transcription from different promoters and this probably creates incomplete transcripts, preventing full-length RNA synthesis. It is important to note that NAD-mRNA is not translated and that there is a fast deNADing process suggesting that NAD-RNA is detrimental to yeast [93]. The mitochondrial RNA polymerase, however, is much more efficient at incorporating NAD into RNA, with a reported 60% of mitochondrial RNAs being NAD-capped. This could mean that NAD-RNA serves specific functions within different cellular compartments [94]. In yeast, apparently some RNAs can be NADylated post-transcriptionally, as the NAD modification appears on the 5' ends of 5'-processed snoRNAs and mRNAs. The same study also showed that NAD-RNA associates with ribosomes in mitochondria, yet again hinting at a different role for the NAD cap in different cellular compartments [84].

Apart from sequencing, studies of the protein interactome of NAD-RNA provide an alternative approach to uncover the function of NAD caps. Using immobilised 3’ biotinylated NAD-RNA to perform a protein pull-down in yeast, 3 main candidate proteins were identified using mass spectrometry. This method termed 5' NAD cap RNA Affinity Purification (NcRAP) detected the cytoplasmic 5'-3' exoribonucleases Xrn1, the nuclear 5'-3' exoribonucleases Rat1, and the Rat1 interacting partner Rai1. These proteins release NAD from the RNA, returning it to the cellular pool and establishing a sort of equilibrium between NAD-RNA and NAD [95]. In Arabidopsis thaliana, NAD-RNAs are also quite frequent and often transcribed from protein coding genes [79]. They appear to be present in nuclear and mitochondrial transcripts, but not in those from chloroplasts. The NAD-capped transcripts were reported to be also spliced, polyadenylated, enriched within the polysomal fraction, and associated with translating ribosomes. Whether NAD-mRNAs are translated in plants, however, remains to be seen [96]. NAD-RNA is also integral for the response of the plant to the hormone abscisic acid that noticeably changes the NAD-RNA transcriptome and the production of free NAD. The NAD cap destabilises the transcripts, and they are cleaved into small noncoding RNAs by the RNA-dependent RNA polymerase 6 (RDR6). This targeted degradation may be one of the factors regulating seed germination [97]. Interestingly, the main RNA deNADing enzyme “decapping exoribonuclease (DXO)” is not only responsible for the decapping and subsequent degradation of NAD-RNA, but it also activates the methyltransferase AtRNMT1 that methylates the mRNA cap to convert it into m⁷G. This creates a direct link between non-canonical and canonical capping, hinting at the importance of NAD-RNA and its interactome [98].

In mammals, NAD cap also seems to be that of a degradation marker by promoting DXO-mediated RNA decay [76]. Apart from DXO, an alternative decapping pathway to DXO in human cell lines involves the cluster of differentiation 38 (CD38) [99]. These glycoproteins are present on the surface of a number of immune cells and have versatile roles in cellular signalling and adhesion, often using NAD as a substrate [100]. CD38 glycoproteins convert the NAD cap into an adenosine diphosphate ribose (ADPr) moiety that itself represents another type of non-canonical RNA cap [101]. NAD-RNA in human cell lines also responds to the infection by the human immunodeficiency virus (HIV-1). Specific transcripts such as the snRNA U1 lose the NAD cap and generally affect the NAD-RNA status of the cell [102]. Given the role of CD38 in antiviral signalling and HIV-1 infection, it is interesting to imagine NAD-RNA as a link connecting the two, however more research is necessary to test the possibility [101].

It has been shown that free NAD levels in some tissues decrease with age [103]. NAD-RNA levels and the identity of capped transcripts also change in mouse liver during ageing. Younger animals tended to have more NAD-RNA in genes connected to splicing and DNA repair, whereas older animals had more NAD-RNA associated with metabolic genes [82]. While the datasets of NAD-RNA are growing and improving with each new sequencing method and each newly discovered decapping enzyme, a lot remains unknown about the biological function or the functional interactome of NAD-RNA [27,28,70,80,82].

4.2 CoA

CoA was one of the first cofactor caps identified in RNA along with its derivate forms such as 3'-dephospho-CoA (dpCoA), succinyl- CoA, acetyl- CoA, and others in E. coli and S. venezuelae by use of LC-MS [70]. The low abundance of CoA-RNA in cells makes it a difficult target for identification. Later, dpCoA-RNA was discovered and quantified also in the small RNA fraction of S. aureus. Using a modified mercury affinity electrophoresis (APM-PAGE), which retards dpCoA-RNA migration, and the E. coli enzyme phosphopantetheine adenylyltransferase (PPAT) to radioactively label the RNA, the capped RNA was detected at a concentration of 14 pmol per μg of RNA [104]. The Nudix enzymes NudC and NudL are primarily responsible for the decapping of CoA-RNA in bacteria, with NudC being faster and promiscuous and NudL slower but specific for CoA-RNA [105,106].

In Arabidopsis thaliana, AtNUDT27 decaps CoA, the only plant Nudix enzyme as yet known to do so [27]. In mammals, however, CoA-RNA is the most readily accepted substrate for the Nudix enzymes, cleaved by six different enzymes, NUDT7, NUDT8, NUDT12, NUDT15, NUDT16, and NUDT19²⁸.

Given that dpCoA has a reactive thiol group, it is an ideal candidate for chemical conjugation using maleimides [107]. This led to the development of dpCoA tagSeq, which uses a maleimide-modified RNA to “tag” the dpCoA-RNA at the 5' end with a barcode sequence. Using the dpCoA tagSeq method, the authors identified more than 40 genes that code for RNAs that are subject to dpCoA modification in mouse livers [108]. It is important to note, however, that the vast majority of the identified sequences in this study were the results of false positives coming from chimeric reads between the sequences of highly abundant transcripts and the spike-in dpCoA-RNA added to the sample [109]. The authors of the original study added several data processing steps in order to improve the method, leaving 7 RNAs as potential dpCoA-capped transcripts that remain to be verified [110]. Unlike some of the other non-canonical caps that are only used as NCINs, dpCoA can be added as a posttranscriptional modification. The aforementioned E. Coli enzyme PPAT accepts 5' triphosphate RNA as a substrate for dpCoAylation in vitro, provided that the RNA has at least four unpaired nucleosides at the 5'end and the terminal nucleoside is an adenosine. Moreover, using model RNA sequences, PPAT was shown to add the dpCoA cap in vivo, although it is important to note that the enzyme PPAT prefers ATP as a substrate and high ATP concentrations inhibit the dpCoAylation of RNA [111]. A better sequencing method and validation of the dpCoA-RNA transcripts are still required to shed more light on their identities and functions.

4.3 FAD

FAD is another adenine-containing cofactor that is accepted by RNA polymerases as an NCIN and was detected in both prokaryotic and eukaryotic RNA [67,112]. Like NAD-RNA, FAD-RNA is decapped by the enzymes Rai1 and DXO, in a process termed deFADing that releases the free form of FAD and leaves behind a 5'monophosphate RNA [113]. FAD is also accepted by the E. Coli primase DnaG, just like NAD, further increasing the variability of primers initiating the synthesis of Okazaki fragments [86].

To better understand the role of FAD-RNA, a study of its interacting proteins was conducted. Using FAD cap RNA Affinity Purification (FcRAP) based on the previously used NcRAP, the 5'-3' exoribonucleases Rat1 and XRN1 were identified along with the previously mentioned Rai1 that associates with Rat1. The bacterial 5'-3' exoribonuclease has the same deFADing activity, suggesting that it may be a conserved ability of all 5'-3' exoribonucleases [114]. The spectrum of deFADing proteins was further expanded after a screen for mammalian Nudix enzymes showed that NUDT2 and NUDT16 also accept FAD-RNA as a substrate [28].

While this seems to show an intricate layer of FAD-RNA decapping and regulation, the first actual biological role of a FAD-RNA came from studies on Hepatitis C virus (HCV). HCV is a positive-strand RNA virus that is under selective pressure to start its negative strand synthesis with an A residue, and it is well known that HCV requires FAD for replication in cell cultures [115,116]. Using the Arabidopsis thaliana Nudix pyrophosphohydrolase 23 (AtNUDX23) that specifically cleaves FAD, the authors modified CapZyme seq and showed that HCV RNA is capped with FAD in up to 75% of its transcripts. The sequencing results were also validated using LC-MS. The authors then proceeded to show that HCV replication is started using FAD as an NCIN and different viral strain transcripts were capped with different intensities. Apart from initiating viral replication, the FAD cap also protected the virus from innate immune recognition. Transfecting cells with FAD-RNA showed that the cap evades the nucleic acid sensor Retinoic acid-inducible gene I (RIG-I) and prevents the stimulation of other innate immune genes [117]. It remains to be seen if other pathogens also employ this strategy of masking their transcripts with metabolite RNA caps. To date, however, this is the only shown biological role of the FAD RNA cap.

4.4 Dinucleoside polyphosphates

Dinucleoside polyphosphates (Np_nNs) are another type of molecule similar to the m⁷G RNA cap in structure and just like NAD and the other metabolites serving as RNA caps, they are present in all cells across all domains of life [118]. They are often called alarmones because their production increases under various stress conditions and their roles range from second messengers to regulators of cellular biosynthetic pathways [119,120].

Np_nN RNA caps were discovered in the small RNA fraction of E. coli using LC-MS [121]. It was also shown that various RNA polymerases accept Np_nNs as NCINs and that the E. Coli enzymes RNA 5'-pyrophosphohydrolase (RppH) and bis(5'-nucleosyl)-tetraphosphatase (ApaH) function as decapping enzymes [122,123]. The LC-MS analysis enabled the distinction between various lengths of the polyphosphate chains of Np_nNs, revealing that the caps are further regulated by methylations that protect them from cleavage by RppH. These methylations occur predominantly under stress conditions such as starvation, probably stabilising the capped transcripts, and return to normal when the stress conditions stop. Concurrently, dinucleoside tetraphosphate caps were also discovered using in vivo radioactive ³²P labelling of E. coli RNAs, cap cleavage, and subsequent boronate polyacrylamide gel electrophoresis (PAGE) analysis and northern blotting. The presence of the cap in the cells also increased during disulphide stress, though the study did not explore their methylation status [124]. Apart from incorporation of Np₄N caps by RNA polymerase and their removal by decapping enzymes, lysyl-tRNA synthetase LysU was also identified as a capping enzyme, capable of installing the cap on 5'triphosphorylated transcripts [124].

Interestingly, an Np_nN with a shorter polyphosphate chain was previously discovered in E. coli. The RNA 3'-Phosphate Cyclase (RtcA) uses 5' monophosphorylated RNA as a substrate for adenylation in vitro, generating an AppN (Ap₂N)-capped transcript or even an Ap₂N-capped DNA. The addition of this cap is performed in a ligase-like manner leaving behind an activated molecule [125]. Preadenylated RNA or DNA can then be further ligated by other ligases, even in an ATP-independent manner [126]. Any other biological role of this AppN cap has not yet been discovered. A screen of Arabidopsis thaliana Nudix enzymes revealed that AtNUDT7 and AtNUDT27 both decap Ap₂A-RNA, suggesting that the presence of this cap is subject to regulation [27].

The most abundant Np_nN, diadenosine tetraphosphate (Ap₄A), was also discovered as an RNA cap in mammalian cells. Ap₄A-capped RNAs were found in the fraction of RNAs longer than 200 nucleotides in two cell lines; human embryonic kidney cells (HEK293T) and rat basophilic leukaemia cells (RBL-2H3) [127]. NuP1 was again used to digest the RNA and LC-MS to confirm the presence of the Ap₄A RNA cap. The human nudix enzyme Nudt2 and DXO were identified as Ap₄A-RNA decapping enzymes. In an attempt to elucidate the role of Ap₄A-RNA, it was shown that RNAs transfected into cells are stabilised when carrying this cap. The Ap₄A-RNA, however, was not translated, nor did it interact with eukaryotic translation initiation factors and, more importantly, did not elicit an immune response. Cells transfected with Ap₄A-RNA failed to induce any of the innate immune genes associated with nucleic acid sensing in the cell. This suggests that the Ap₄A cap is a natural component of the cellular transcriptome but it does not show if Ap₄A-RNA has a specific role in mammalian cells [127]. Ap₄A-RNA can also be decapped by the plant enzymes AtNUDT6 and AtNUDT7, though the former preferentially cleaves the free molecule and the latter prefers the capped RNA as a substrate. The presence of Ap₄A-RNA in plants, however, has not been confirmed [27]. In general, Np_nNs seem to provide an additional layer of transcriptional regulation, the exact mechanisms, however, are yet to be discovered.

4.5 UDP-Glucose

Nucleotide sugars were also recently discovered to function as non-canonical RNA caps. Using a transcription template under control of an acnA promoter at which UTP is utilised as an initiating nucleotide, E. coli polymerase readily accepted both uridine diphosphate glucose (UDP-Glc) and uridine diphosphate N-acetylglucosamine (UDP-GlcNAc) [128]. Both uridine diphosphate sugars are common bacterial cell wall precursors present in prokaryotic cells in high quantities [129].

UDP-Glc is also the activated form of glucose used as a precursor for carbohydrate activation or for glucosyl transfer reactions in all organisms, meaning it’s also present in eukaryotic cells [130]. Indeed, the UDP-Glc and UDP-GlcNAc RNA caps were also discovered in eukaryotes using the mass spectromentry-based method CapQuant [112]. Quite large amounts of these caps were also detected in viral RNAs, specifically the dengue viral particles, in which UDP-GlcNAc-RNA was more abundant than in eukaryotic or prokaryotic cells. The concentration of UDP-GlcNAc-RNA in the sample was 10x higher than that of UDP-Glc-RNA [115]. This could be explained by dengue virus usurping the host machinery to glycosylate its own proteins or by a general increase in UDP-GlcNAc that occurs during certain viral infections [131,132].

The E. Coli enzymes MurA, MurB, and MurC have been shown to modify the sugar cap by attaching a clickable moiety in vitro, presenting a potential way of pulling dow the sugar-capped RNAs. Human nudix protein NUDT5 was then identified as a UDP-GlcNAc-RNA decapping enzyme, leaving the decapped RNA ready for ligation [133]. NUDT22, is another member of the Nudix hydrolase family, that has been repeatedly implicated in RNA decapping, is also known to cleave UDP-Glc into UMP and glucose 1-phosphate, making it a potential decapping enzyme for these nucleotide sugar caps, though its activity on the RNA remains to be tested [134]. A specific sequencing method is needed to identify the capped transcripts and shed more light on their roles within the cell.

4.6 Adenosine diphosphate ribose

The addition of adenosine diphosphate ribose (ADPr) was long thought to be a uniquely posttranslational modification of proteins, affecting their functions and interactions. Installed by a family of enzymes called poly(ADP-ribose) polymerases (PARPs), the modification was shown to play a role in DNA repair, antiviral responses, cell cycle, and more [135]. The 5' phosphorylated ends of RNA can also be modified in this manner yielding a 5'-phospho-ADPribose, introduced by 2'-phosphotransferase (Tpt1) enzymes [136,137]. A number of human PARPs were shown to ribosylate the phosphorylated 5' ends of RNA as well, namely PARP10, PARP11, and PARP15. The ADPr-RNA is resistant to 5' phosphatase degradation and the addition of the cap by PARP10 was shown to be reversible as several human ADP-ribosyl hydrolases decap the RNA. Interestingly, two viruses - Venezuelan equine encephalitis virus (VEEV) and severe acute respiratory syndrome coronavirus (SARS-CoV) - code for their own hydrolases capable of decapping ADPr-RNA [138]. Most of the experiments, however, were done in vitro using recombinant and not full-length proteins. Thus, corresponding functional in vivo studies are required [138]. Subsequent studies revealed that there is a difference in substrate specificity between TRPT1, the human homologue of Tpt1 that only modifies RNAs with purines at the 5' end and PARPs that modify purines and pyrimidines. ADPr-RNA was also detected in vivo, something that was difficult before, given the usual acidic phenol RNA extraction and the susceptibility of ADPr-RNA to acidic pH. In view of the localisation of some PARPs to stress granules, it was proposed that they can ribosylate mRNAs, which are then decapped and canonically recapped by the cytoplasmic capping machinery. However, further research is needed to verify this hypothesis [139].

5. Conclusion

RNA capping is an intrinsic process that plays an integral role in post-transcriptional gene regulation and RNA metabolism. Originally thought to be limited to the canonical m⁷G cap and restricted only to eukaryotic RNA Pol II transcripts, the discovery of non-canonical RNA caps in various organisms has greatly expanded our understanding of RNA modifications. While the m⁷G cap has been meticulously studied and implicated in a wide variety of cellular processes, the roles of non-canonical RNA caps are still somewhat unknown and warrant further research in order to ascertain their functions within the cell.

The wide array of non-canonical RNA caps is chemically diverse, encompassing 5' RNA ends ranging from 5'-γ-methyl-phosphate, cofactors such as NAD, CoA, FAD, metabolites and alarmones of the dinucleoside polyphosphate family, to nucleotide sugars like UDP-glucose or ADP-ribose [70,121,128,138]. While these caps have been identified across all domains of life suggesting they are evolutionarily conserved, their significance in the regulation of biological processes has yet to be determined Table 1.

Table 1. Summary of known cap-like structures and non-canonical RNA caps.

Non-canonical RNA cap	Organisms	Type of RNA	Methods of detection	Incorporation into RNA/ capping mechanism (the most probable mechanism)	Decapping	Biological role
(γ-m-NTP)	Plants	ncRNA: U3, U6 snRNAs	isotopic labeling^Citation54	unknown methyltransferase	unknown	RNA stability^Citation57
	Mammalian cells	ncRNA: 7SK, B2	isotopic labeling, northern blot^Citation53	methyl phosphate capping enzyme MePCE^Citation56	unknown	binding to La protein^Citation59
(α-m-NMP)	Mammalian cells	miRNA^Citation60, tRNA^Citation64	isotopic labeling^Citation60	methyl phosphate capping enzyme BCDIN3D^Citation60	unknown	miRNA maturation, carcinogenesis^Citation60
NAD	Bacteria	sRNA^Citation73, mRNA^Citation90	LC-MS^Citation71, sequencing (e.g.^Citation73)	RNA polymerase^Citation69	NudC (E. coli)^Citation73, BsRppH (B. subtilis)^Citation90	RNA stability, quorum sensing RNAIII – cytotoxity^Citation89, RNAylation by bacteriophage protein ModB^Citation87
	Archaea	mRNA, tRNA	LC-MS^Citation74,Citation91, sequencing^Citation74	RNA polymerase (?)	thermal degradation – Saci Nudt5^Citation74	Degradation mark^Citation74
	Yeast	mRNA, mtRNA^Citation94	LC-MS^Citation93, sequencing^Citation93,Citation94	RNA polymerase^Citation69	Rai1/Rat1, Xrn1^Citation95	Degradation mark
	Plants	mRNA	sequencing^Citation79	RNA polymerase (?)	AtNudt19, AtNud27^Citation27	mark for generation of regulatory RNA^Citation97
	Mammalian cells	mRNA^Citation76, snRNA, snoRNA^Citation101, mitochondrial mRNA^Citation84, viral transcripts (DENV-2^Citation112)	LC-MS^Citation112, sequencing (e.g.^Citation76,Citation84)	RNA polymerase (?), posttranscriptional capping (?)	Nudt12^Citation28, Nudt13, Nudt16, DXO^Citation76	U1 snRNA binding to target^Citation101, correlation with aging in mice^Citation82
FAD	Bacteria	Unknown	LC-MS^Citation112	RNA polymerase^Citation67	RNase AM^Citation117	Unknown
	Yeast	Unknown	LC-MS^Citation112	RNA polymerase (?)	Rai1^Citation113, RNase AM^Citation114	Unknown
	Mammalian cells	Viral transcripts (HCV^Citation117, DENV-2^Citation112)	LC-MS^Citation112, sequencing^Citation117	Viral RNA polymerase^Citation117	DXO, Nudt2^28, Nudt16^28, XrnI^Citation114 (Plant AtNudt23 used for sequencing^Citation117)	Evasion from innate immune recognition^Citation117
CoA	Bacteria	Unknown	LC-MS^Citation70, mercury affinity electrophoresis^Citation104	RNA polymerase^Citation69 and phosphopantetheine adenylyltransferase^Citation111	NudC^Citation105, NudL^Citation106	Unknown
	Mice livers	mRNAs^Citation108–110	sequencing^Citation108–110	RNA polymerase (?)	Nudt2,7, 8, 12, 15, 16 and Nudt19^Citation28	Unknown
NpnNs	Bacteria	mRNA^Citation124	LC-MS^121, Northern blot^Citation124	RNA polymerase	RppH, ApaH	RNA stability
	Mammalian cells	unknown	LC-MS^Citation124	RNA polymerase (?)	Nudt2, DXO^Citation127	Unknown, recognized as self^Citation127
UDP-Glc and UDP-GlcNAc	Bacteria	Unknown	LC-MS^Citation112	RNA polymerase^Citation128	?	Unknown
	Yeast	Unknown	LC-MS^Citation112	RNA polymerase (?)	?	Unknown
	Mammalian cells	Unknown, viral transcripts (DENV-2)^Citation112	LC-MS^Citation112	RNA polymerase (?)	NUDT5^Citation133, Nudt22?^Citation134	Unknown
ADP-ribose	Bacteria	Unknown	Isotopic labeling^Citation136	TRPT1 and MACROD homologues in bacteria^Citation137,Citation138	ADP-ribosyl hydrolases?	Unknown
	Archaea	Unknown	Isotopic labeling^Citation136	2'-phosphotransferase Tpt1^Citation136	ADP-ribosyl hydrolases?	Unknown
	Mammalian cells	Unknown	Immunoblotting^Citation139	PARPs, 2'- phosphotransferase TRPT1^Citation138,Citation139	human and viral (VEEV and SARS-CoV) ADP-ribosyl hydrolases^Citation138	Unknown

To date, investigation of non-canonical RNA caps has implicated them in a number of cellular processes. These include RNA stability, post-transcriptional regulation, and immune evasion of pathogens. For instance, the NAD cap protects the RNA from degradation by 5' to 3’ exonucleases in bacteria but marks it for degradation in eukaryotes and has also been shown to affect the life cycles of certain pathogens [87]. FAD is another example, where its presence on HCV viral transcripts facilitates their circumvention of the innate immune response [120]. In fact, all of the cofactor caps were shown to interact with the cellular nucleic acid sensor RIG-I. In vitro prepared dsRNAs capped with NAD, FAD and dpCoA were purified and transfected into cells and a marked increase in antiviral signalling was observed. It is possible that the increase in signalling might have been caused by the dsRNA itself regardless of the RNA cap. This cannot be excluded because the HCV study showed the opposite results [117,139]. All this seems to underline the importance of non-canonical RNA capping.

Moreover, the development and improvement of innovative sequencing methods, such as NAD captureSeq or the modified CapZyme-seq have made it possible to identify and characterise non-canonically capped transcripts in a number of organisms. This has provided further insight into the mechanisms and the potential biological role of capping. Thanks to the NAD sequencing protocol, for example, we now know that the NAD cap in bacteria is mainly present on short regulatory RNAs or that there are preferred transcription start sites and consensus promoter sequences used for NAD capping [73,78].

Despite all these advancements in the field, many of the questions about the actual roles and functional mechanisms of non-canonical RNA caps remain unanswered. First and foremost, are they just stoichiometric incorporations of molecules readily accepted by RNA polymerases, or do they exert specific functions within the cellular transcriptome? To clarify this central question, future research should focus on trying to determine their functional interactome and the identity of the capped transcripts. To put it in a nutshell, it is imperative to develop specific sequencing methods for the different caps and identify the proteins that cap, decap, and interact with the capped RNA. More research into these caps promises to unravel the mechanisms through which they act in vivo and may potentially lead to innovative therapeutic strategies pertaining to RNA regulation and function.

Acknowledgments

We are grateful to members of the Cahova Group (Pavel Vopalensky, Ondrej Nesuta, and Maria Bianca Mititelu) for their help and advice. Figures were Created with BioRender.com. We acknowledge funding from the European Research Council Executive Agency (ERCEA) under the European Union’s Horizon Europe Framework Programme for Research and Innovation (grant agreement No. 101041374 – StressRNaction) and the Operational Programme Johannes Amos Comenius (OP JAC) project RNA for Therapy, reg. No. CZ.02.01.01/00/22_008/0004575 co-financed by the EU.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the Ministry of Education, Youth and Sports [OP JAC, RNA4Therapy, CZ.02.01.01/00/22_008/0004575]; HORIZON EUROPE European Research Council [No. 101041374 – StressRNaction].