How RNA modifications regulate the antiviral response

Summary

Induction of the antiviral innate immune response is highly regulated at the RNA level, particularly by RNA modifications. Recent discoveries have revealed how RNA modifications play key roles in cellular surveillance of nucleic acids and in controlling gene expression in response to viral infection. These modifications have emerged as being essential for a functional antiviral response and maintaining cellular homeostasis. In this review, we will highlight these and other discoveries that describe how the antiviral response is controlled by modifications to both viral and cellular RNA, focusing on how mRNA cap modifications, N6-methyladenosine, and RNA editing all contribute to coordinating an efficient response that properly controls viral infection.

RNA modifications are important at all levels of host antiviral response

The host antiviral innate immune response limits infection by sensing features specific to viruses and activating a cellular defense program. This program acts on viral particles in infected cells, and primes surrounding naïve cells to prevent infection, and as such, must be carefully regulated to maintain cellular homeostasis in the absence of viral infection. The sensors that initiate the innate immune response are pattern recognition receptors (PRRs), which are proteins that detect pathogen-associated molecular patterns (PAMPs). PAMPs are defined as any pathogen-derived biomolecule recognized by PRRs, however nucleic acid-derived PAMPs represent a large and important class of ligands, and are a main focus of this text. Typically, cellular nucleic acids do not contain PAMP characteristics, as host cells modify, process, and localize their nucleic acids in a number of ways to avoid PRR detection and aberrant immune responses¹. Once an RNA PAMP is sensed by a PRR, such as the cytosolic RNA binding proteins RIG-I or MDA5, a signaling cascade is triggered resulting in the expression of cytokines, most notably type I and III interferons (IFN). IFN is secreted and signals to surrounding cells inducing a gene expression program of IFN-stimulated genes (ISGs) that prepare the cells to defend against potential infection. These ISGs often act directly on viral RNA to limit replication². Importantly, many steps of this antiviral innate immune response, from PAMP sensing to gene expression, are highly regulated at the RNA level. As such, regulatory components of RNA are essential for a properly coordinated innate immune response. One mechanism of RNA regulation, RNA modification, has emerged as a critical component for controlling both how cells distinguish self from non-self-nucleic acids, as well as for controlling gene expression in the activation of the antiviral response.

RNA modifications are features on an RNA molecule altering the canonical AUGC bases. While the most commonly known modification is the 5′ N7-methylguanosine (m⁷G) mRNA cap, studies have revealed over 170 other modifications that can be added to RNA³. While the function of many of these modifications are not yet known, the better-studied examples are critical for many aspects of RNA function. Given that RNA modifications are deposited onto molecules by their own enzymes, recognized by specific proteins, and are often able to be removed, these modifications comprise an important layer of regulatory complexity in RNA biology. In the context of the innate immune response, RNA modifications have emerged as key regulatory factors of the antiviral response by playing roles in PAMP recognition, basal expression of antiviral genes, and induction of antiviral genes. The importance of RNA modifications in controlling aspects of the host immune response has proven to be critical for the recent emergence of mRNA vaccines. For example, in the mRNA-based COVID-19 vaccines, the presence of the modified nucleobase N1-methylpseudouridine on the antigen-encoding mRNA increases antigen production while limiting the initial innate immune response to this foreign RNA, allowing for the development of adaptive, protective immune responses to the antigens encoded within the vaccine mRNA^4–9. In this review, the functional roles and regulation of RNA modifications in the context of the antiviral innate immune response will be discussed, highlighting three of the most well-studied examples: modifications to the mRNA cap, N6-methyladenosine (m⁶A), and RNA editing.

5′ mRNA capping and the antiviral response

Capping of mRNA 5′ ends is a highly conserved process that controls mRNA function in both the cytoplasm and in the nucleus. The discovery that 5′ ends of mRNA undergo processing, referred to as mRNA capping, was made nearly 50 years ago¹⁰. In the nucleus, it is required for RNA to avoid degradation via 5′−3′ exonucleases, for efficient splicing, and for promoting nuclear export¹¹. Cytoplasmic 5′ mRNA caps recruit translation initiation machinery and contribute to translation efficiency¹². Importantly, they also shield cellular mRNAs from triggering cellular PRRs (discussed below). As such, this 5′ modification on both cellular and viral RNA plays important functions in host discrimination between self and nonself RNA and the host response to viral infection. This section will discuss how 5′ mRNA cap structures are formed, function in mRNA, their roles in PRR recognition of PAMPs, and how they promote host gene expression during viral infection.

For cellular RNAs the conventional mRNA capping pathway modifies transcripts as they are being transcribed by RNA polymerase II. Nascent RNA initially contains a 5′-triphosphate (5′-ppp), which is converted to 5′-diphosphate (5′-pp) via RNA triphosphatase allowing for RNA guanylyltransferase and guanine-N7 methyltransferase to add and methylate a terminal guanosine base¹². The result is an m⁷G joined via a 5′−5′ triphosphate bridge, which is referred to as cap-0 (Fig. 1A)¹⁰. Cellular RNAs are then further modified by addition of a methyl group at the 2′-O-hydroxyl position of the first nucleotide by a cellular 2′-O-methyltransferase, CMTR1, to form the cap-1 RNA structure (Fig. 1A)¹³. Initial characterization of cap-1 in mRNA showed that all transcripts contain cap-1¹⁴. More recent findings using mass spectrometry confirm that 88% of mRNAs contain cap-1, while surprisingly the remaining 12% contain a unique methylation pattern where cap-0 mRNAs with a 5′ adenosine are methylated at the N6 position (m⁶A)¹⁵. Cap-1 mRNA can be further modified by an additional methyl group at the N6 position of the first nucleotide (only in adenosines), resulting in m⁶Am (Fig. 1A)¹⁶. This methylation is carried out by the host m⁶Am methyltransferase called PCIF1, also referred to as CAPAM^16–18. Both initial and current studies estimate that 17% to 30% of mRNA contains a m⁶Am in the first nucleotide^15,19. The general function of m⁶Am is not fully understood, however recent work suggests it is important for RNA stability²⁰. The methyltransferase for m⁶Am, PCIF1, has only recently been described, despite the m⁶Am modification being discovered several decades previous^14,16. Interestingly, cells are permissive to PCIF1 knockout, suggesting m⁶Am is not an obligate RNA modification. Indeed, mapping of m⁶Am in PCIF1 knockout cells showed that depletion of PCIF1 is correlated with decreased RNA half-life but does not necessarily confer downregulation of a given gene¹⁶. Along with addition of a 2′-O-methyl group at the first nucleotide (cap-1), a second 2′-O-methyl group can be added at the second nucleotide to form cap-2 (Fig. 1A). The addition of this second 2′-O-methyl group is catalyzed by the cellular methyltransferase CMTR2^21,22. Currently, the prevalence of the second 2′-O-methyl group in the cap in cellular mRNA is unknown, however, its function is understood to work synergistically with the first 2′-O-methyl group. Lastly, additional caps containing metabolites such as FAD and UDP have been detected in host RNA via mass spectrometry, although these structures comprise <1% of all cellular transcripts and their function is not yet known¹⁵. Overall, the 5′ ends of mRNA can be modified in a number of ways that influence several aspects of mRNA function.

Figure 1. 5′ mRNA capping regulates the antiviral response.

A) Schematic of the canonical 5′ mRNA cap structures and the enzymes involved in their processing (RNA triphosphatase (RNA TPase); RNA guanylyltransferase (GTase); guanine-N7 methyltransferase (N7 MTase); CTMR1, CMTR2, and PCIF1). B) Diagram showing roles of 5′ capping in antiviral response. Modification of incoming viral RNA can alter RIG-I binding and subsequent MAVS/TBK1 and IRF3 signaling to induce type I interferon (IFN). IFN signals to other cells via JAK/STAT and ISGF3 to induce IFN-stimulated genes (ISGs) including the cap-1 methyltransferase CMTR1 and the cap modification sensor proteins, IFIT1 and IFIT3, which inhibit viral translation.

mRNA caps play important roles in general mRNA function. In the nucleus, m⁷G (cap-0) modified RNA transcripts are bound by cap-binding complex that protects newly synthesized RNA from degradation by cellular 5′−3′ exonucleases²³, promotes splicing, facilitates 3′-end processing, and enhances mRNA export²⁴. Once exported to the cytoplasm, m⁷G (cap-0) provides further functionality to mRNA by facilitating cap-dependent translation by recruiting the translation initiation machinery²⁵. Other cap structures such as m⁶Am can prevent the decapping enzyme DCP2 from removing the caps and thus this modification prevents mRNA degradation by exonucleases²⁰. In their roles in regulating RNA metabolism and function, mRNA caps also play an integral role in the host cellular surveillance pathways that discriminate self from foreign RNA.

Specific features of the 5′ ends of mRNA can be sensed by the antiviral innate immune system, which include cytoplasmic PRRs that detect non-self RNA (Fig. 1B). For example, RIG-I was first described as an innate immune sensor of dsRNA, such that its binding to dsRNA results in the induction of IFN for the antiviral response²⁶. We now know that RIG-I also can discriminate between specific RNA modifications in the mRNA caps^26–29. For example, RIG-I is activated by mRNA that is not fully capped and contains 5′-ppp, 5′-pp, 5′-p, 5′-OH^30–32. However, while RIG-I does bind to m⁷G-capped mRNA (cap-0) the presence of of 2′-O-methylation at the first nucleotide dramatically reduces RIG-I binding (cap-1)^32–34. Thus, cap modifications can alter how viral or self RNA are sensed by RIG-I for the antiviral response.

Other cytosolic PRRs, such as those in the IFN-induced proteins with tetratricopeptide repeats (IFIT) family, sense changes to modifications at the 5′ ends of mRNA. The IFITs are RNA binding proteins that restrict infection of specific viruses and thus are important for the IFN-induced antiviral response³⁵. The human IFITs include four proteins (IFIT1, IFIT2, IFIT3, and IFIT5), with IFIT1 and IFIT3 having roles in sensing 5′ cap structures. IFIT1 binds RNA that lacks 2′-O-methylation (cap-0, Fig. 1A). IFIT3 helps to stabilize the interaction of IFIT1 with RNA and this inhibits ribosome binding resulting in decreased translation^36–39. While IFIT1 has high affinity for RNA that lacks 2′-O-methylation, at high levels it can bind to some cap-1 RNA⁴⁰. Thus, 2′-O-methyl groups (cap-1 or cap-2) prevent RNA recognition by the IFIT1/IFIT3 complex, and thus this modification is important for protecting viral or self RNA from translational inhibition by the IFIT1/IFIT3 complex. Interestingly, recent studies suggest that proteins induced by IFN sense the absence of the cap modification m⁶Am in vesicular stomatitis virus to inhibit replication⁴¹. These data suggest that m⁶Am on viral RNA protects it from being sensed by ISGs that ultimately inhibit viral replication.

The importance of the RNA modification status of the mRNA cap in the antiviral innate immune response is illustrated by the fact that many RNA viruses possess the ability to evade sensing by cytosolic RNA sensor proteins. For example, both flaviviruses and coronaviruses, among others, encode enzymes with m⁷G and 2′-O-methyltransferase activity to cap their RNA, and viruses deficient in these activities are highly-sensitive to IFN^42,43. Influenza viruses encode an enzyme that cleaves the 5′ ends of cellular mRNAs, which include the cellular caps, and utilizes them to begin transcription of viral RNAs in a process called “cap-snatching”, which ensures their 5′ ends have modifications that prevent their RNA from being sensed⁴⁴. Additionally, several viruses prevent recognition by RLRs in their uncapped RNA by utilizing extensive secondary structure or adding proteins to their 5′ ends^45,46.

While cap structures were historically thought of as obligate structures^15,19, recent evidence suggests that cells may regulate capping of their own mRNA to generate a functional antiviral response. For example, in humans, the 2′-O-methyltransferase CMTR1, also referred to as ISG95, is one of the many genes upregulated in response to IFN^13,47,48. Interestingly, when CMTR1 is depleted, the IFN-induced protein expression of specific ISGs is reduced resulting in a suboptimal antiviral response⁴⁸. This suggests that in the presence of IFN, either certain RNAs are capped with higher priority than others or that a subset of RNAs do not require cap 2′-O-methylation (cap-1) for efficient translation, perhaps because they encode mechanisms to evade detection and translational inhibition by the IFIT1/IFIT3 complex. Thus, cellular capping enzymes, and their resulting mRNA caps, are important for proper gene expression during immune stimulation, as highlighted by recent work in T cell activation⁴⁹.

How m⁶A affects the antiviral response

The RNA modification m⁶A also plays important roles in regulating numerous aspects of the antiviral response. Similar to mRNA caps, this modification was first described nearly a half century ago^19,50. However, the function of m⁶A remained unclear until recently, when the proteins that add, remove, and bind this modification were discovered; and sequencing techniques to profile this modification across the transcriptome were developed^51–54. These and other studies have revealed that m⁶A has roles in nearly all aspects of RNA biology, including stability, translation, splicing, nuclear export, and localization. The m⁶A RNA modification is added to the methyl group at the N⁶ position of adenosine by large heterogenous network of proteins that make up the m⁶A methyltransferase complex (Fig. 2A). The core proteins in this complex are the methyltransferase protein called METTL3, as well as its major interaction partners METTL14 and WTAP, which are required for m⁶A addition in cells⁵². METTL3 methylates mRNAs at a specific RNA motif, DRA^mCH (D=G/A/U, R=A/G and H=A/C/U). Interestingly, while these motifs appear at high frequency in mRNA, only a fraction of DRACH motifs are methylated, suggesting the existence of another layer of specificity for m⁶A deposition^54–56. While it has been suggested that all the features that dictate whether a specific RNA motif is methylated or not depend upon RNA sequence alone⁵⁷, the features that underlie selectivity of the m⁶A methyltransferase complex for specific DRA^mCH motifs remains an open question. The m⁶A methyltransferase complex contains many proteins aside from the core components, several of which have unique roles for targeting of RNA⁵². For example, VIRMA is important for 3′UTR-m⁶A, while RBM15 appears to be important for poly(U)-adjacent m⁶A^58,59. The m⁶A methyltransferase complex can also be recruited to specific mRNAs based on the transcription and chromatin states of their genes. For example, METTL14 can target the m⁶A methyltransferase complex to specific mRNAs via H3K36 methylation of histones, and METTL3 can target the m⁶A methyltransferase complex to specific genes via their promoters and chromatin accessibility^60–62. m⁶A modification can be altered at the level of deposition, or “writing”, in specific contexts, including during viral infection, suggesting that the accessory proteins in the m⁶A methyltransferase complex, as well as cellular chromatin and transcription states, are likely subject to regulation^63,64. METTL16 also has m⁶A methyltransferase activity, but it only modifies specific RNAs, such as U6 spliceosomal RNA and MAT2A RNA^65,66. The various functions of m⁶A on RNA are enacted by the binding of m⁶A “reader” proteins. Canonically, these include the YTH domain proteins, YTHDF1, YTHDF2, and YTHDF3, as well as YTHDC1 and YTHDC2^67,68. Other proteins have been identified that are lacking YTH domains and bind at or in close proximity to m⁶A sites, as well as proteins that are repelled by m⁶A, with the list of these proteins and how they alter RNA metabolism continually growing^69,70. As will be discussed below, the downstream consequences of m⁶A on cellular or viral RNA, which are a result of the RNA binding proteins that interface with m⁶A on these RNAs, can affect the host innate immune response.

Figure 2. m⁶A regulates the antiviral response.

A) Schematic of adenosine (A) to N6-methyladenosine (m⁶A) modification with m⁶A machinery “writer,” “eraser,” and “reader” proteins indicated. B) Diagram showing roles of m⁶A in antiviral response. Modification of incoming viral RNA can inhibit RIG-I binding and subsequent MAVS/TBK1 and IRF3 signaling to induce type I IFN. The transcript encoding IFN-β is m⁶A-modified and subject to YTHDF2-dependent degradation. IFN signals to other cells via JAK/STAT and ISGF3 to induce transcription of ISG mRNA, which is m⁶A-modified and subject to YTHDF-dependent regulation. ISGs, such as ISG20, and possibly others, regulate m⁶A RNA.

In the context of innate immune response, m⁶A is important for the initial sensing of RNA PAMPs by PRRs (Fig. 2B). Similar to inclusion of the m⁷G and 2′-O-methylation on the caps on viral RNA, internal m⁶A modification on viral RNA prevents its recognition by PRRs. The earliest direct investigation of m⁶A on PRR activation showed that RNA modifications, including m⁶A, on in vitro transcribed RNA decreased the sensing of this RNA by Toll-like receptors⁴. This direct influence of m⁶A upon PRR binding was revisited when internal m⁶A bases in dsRNA probes inhibited RIG-I binding⁷¹. Similarly, the addition of m⁶A to in vitro transcribed RNA derived from HCV, the polyU/C-rich region of the 3’UTR (HCV PAMP), also prevented RIG-I activation⁷². These studies have now been followed by several reports supporting a role of m⁶A in preventing RIG-I sensing of viral or self RNA, including circular RNA^73–79. It is generally not clear how m⁶A-modification of the few adenosines in the HCV PAMP RNA or even in other RNA prevents RIG-I binding or activation. One hypothesis is that m⁶A alters the dsRNA structure of these RNAs and that the loss of RIG-I binding is due to loss of the double-stranded nature of these RNAs. Indeed, m⁶A can alter RNA structure^80–83, and could be altering cellular dsRNA or the small panhandle structures known to be present in in vitro transcribed RNA, such as in the HCV PAMP^72,84. A role for m⁶A in altering dsRNA and preventing RIG-I activation is also supported by RNA virus studies. For example, during vesicular stomatitis virus infection, siRNA depletion of METTL3 (reduced m⁶A) increases viral dsRNA, and this results in more RIG-I bound to viral RNA⁸⁵. Another hypothesis is that m⁶A prevents RIG-I activation by recruiting cellular RNA binding proteins that inhibit dsRNA formation or directly impedes RIG-I binding without fully altering RNA structure, as has been suggested for in vitro transcribed viral RNA or with circular RNA^28,76,78. Indeed, internal modifications on in vitro transcribed RNA can prevent RIG-I translocation along the RNA, which is required for its activation^71,86. Thus, further studies are required to understand exactly how the presence of m⁶A on viral or host RNA prevents RIG-I activation. Interestingly, ISGs have been shown to sense m⁶A on viral RNA to inhibit infection. For example, the interferon-induced exonuclease, ISG20, selectively degrades m⁶A-modifiied HBV RNA⁸⁷. While few examples are known of ISGs targeting m⁶A, it seems likely that other ISGs will be discovered that recognized modified viral transcripts to inhibit infection (Fig. 2B). Together, these examples show the importance of m⁶A in altering how cells detect PAMPs.

m⁶A also plays diverse roles in controlling gene expression during the antiviral innate immune response, sometimes helping to promote its activation, while other times being important for turning off this response (Fig. 2B). Some of the initial studies that described the function of m⁶A on cellular RNA found that depletion of the m⁶A demethylase ALKBH5 resulted in induction of signaling that induces ISGs⁸⁸. This finding suggested to us and others that m⁶A would likely be a key player in some aspect of IFN induction or signaling. Indeed, signaling proteins important for IFN induction, specifically MAVS, TRAF3, and TRAF6, all have m⁶A. During vesicular stomatitis virus infection, these transcripts are demethylated by ALKBH5 which interacts with the RNA helicase DDX46 for this demethylation. This leads to nuclear retention of these transcripts, which reduces their expression and thus prevents IFN induction⁸⁹. Another study has found that m⁶A on TRAF6 allows it to bind to YTHDF1, which works with DDX60 to promote its translation in the immune response to bacteria⁹⁰. These two studies support a model in which m⁶A promotes TRAF6 expression. However, it is not clear if ALKBH5/DDX46-demethylation of TRAF6 is induced by viral infection, nor is it known if YTHDF1/DDX60-promoted translation of TRAF6 is signal dependent. Another dead box helicase protein, DDX5 can also cooperate with METTL3 to promote m⁶A-modification of transcripts that encode proteins important for IFN induction, DHX58, p65, and IKKγ, which leads to their export from the nucleus⁹¹. This increased nuclear export decreases their transcript levels due YTHDF2-mediated degradation. This is in direct contrast to the study described above that showed that increased nuclear export of MAVS, TRAF3, and TRAF6 promoted their expression⁸⁹. While in both studies, m⁶A promoted nuclear export, the key difference was that DHX58, p65, and IKKγ were bound by YTHDF2, while MAVS, TRAF3, and TRAF6 were not. The mechanism for why certain transcripts are targeted by specific m⁶A-binding proteins, including those in the YTHDF family, remains unclear, but likely depends on a combination of contextual factors such as other RNA binding proteins, RNA structure, and RNA localization^54,92. Lastly, m⁶A controls the expression of cytokines important for the antiviral response. The mRNA encoding IFN-β, called IFNB1, is modified by m⁶A when it is transcribed, after which it can bind to YTHDF2 to be targeted for degradation (Fig. 2B)^93,94. This mechanism of regulation by m⁶A ensures that the induction of IFNB1 is only ever transient. Thus, having m⁶A on IFNB1 prevents its prolonged induction and limits the potential for autoinflammation. In summary, m⁶A modification of innate immune signaling mRNA can directly influence the expression of that gene, and therefore, the immune response.

Recent work by our lab has shown that m⁶A can also promote the expression of genes that have direct antiviral function^63,95. We showed that a subset of ISGs are m⁶A-modified and that their translation is enhanced by YTHDF1. As such, when we depleted METTL3, IFN was less effective at inducing an antiviral response, as the expression of key antiviral ISGs was abrogated ⁹⁵. Similarly, others have also now shown that m⁶A and YTHDF1 promote the expression of genes required for the intestinal immune response to bacteria⁹⁰. Further, in investigating how the m⁶A epitranscriptome is altered during Flaviviridae infection, we found that specific host transcripts had altered levels of m⁶A, and some of these changes occurred due to activation of signaling induced by Flaviviridae, such as ER stress and RIG-I signaling⁶³. In addition, many of the genes that change m⁶A-modification during viral infection encode proteins that regulate the outcome of viral infection. Importantly, while the genes encoding these proteins were induced by infection, many were not canonical antiviral ISGs, revealing that m⁶A can play diverse roles in the host response to infection by acting on a variety of host response genes.

A-to-I editing and the antiviral response

Perhaps the most well-studied modification to RNA in the context of the antiviral host response is adenosine to inosine (A-to-I) editing (Fig. 3A). Unlike cap modifications and m⁶A modifications, A-to-I editing is much simpler to detect at the sequence level, as modified inosine bases introduce point mutations (A to G) when reverse transcribed. Therefore, A-to-I editing can be evaluated using conventional RNA sequencing pipelines, allowing for robust evaluation of editing frequencies and even identity of sites by mining previous data sets with sufficient read-depth. Indeed, A-to-I editing has been studied in the context of many models, notably cancer and immunology, as reviewed recently^96–98. RNA editing enzymes and the RNA modifications they catalyze are extremely important for maintaining an effective immune response and preventing host RNA from improperly activating host RNA sensors.

Figure 3. A-to-I editing regulates the antiviral response.

A) Schematic of adenosine and the known enzymes involved in its modification to inosine. B) Diagram showing roles of A-to-I editing in the antiviral response. A-to-I editing of incoming viral RNA can inhibit binding by PRRs that sense dsRNA and subsequent MAVS/TBK1/IRF3 signaling that induces type I IFN or inhibits translation. IFN signals to other cells via JAK/STAT and ISGF3 signaling induce transcription of ISGs, including ADAR1 p150. ADAR1 p150 A-to-I edits newly transcribed endogenous host dsRNA (Alu elements) to prevent detection by PRRs and aberrant IFN induction.

A-to-I editing is carried out primarily by ADAR1, which belongs to a family of adenosine deaminases acting on RNA (ADARs), of which there are 3 in humans: ADAR1, ADAR2, and ADAR3⁹⁹. ADAR1 and ADAR2 have enzymatic activity, while ADAR3 is catalytically inactive¹⁰⁰. ADAR1 accounts for the vast majority of RNA editing in cells^101,102, and it exists in two isoforms, ADAR1 p110 (short) and p150 (long), which arise from the alternative splicing of mutually exclusive transcription start sites^103,104. Importantly, while ADAR1 p110 is basally expressed, in response to IFN, it is rapidly ubiquitinated and degraded while p150 is induced^104,105. The importance of this regulation of the p110 and p150 isoforms is not fully understood, but speculative models suggest temporal regulation of ADAR1 is important for the antiviral innate immune response^98,106. ADARs recognize their RNA substrates through their dsRNA binding domains, and then they carry out editing by flipping out bases and modifying them via a separate deaminase domain^99,107. Interestingly, the most common RNA substrate for A-to-I editing are host Alu elements, endogenous retroelements that can form dsRNA^{102,108–110}. ADARs have a preference of editing dsRNA but generally do not target specific sequences¹¹¹. In line with this, all adenosines within dsRNA Alu elements can be edited, while adenosines in other transcripts are generally not edited¹⁰⁸. A-to-I editing in dsRNA regions, such as Alu elements, introduces I-T wobble-pairs that result in less stable RNA structure¹¹². This destabilization can alter how RNA binding proteins recognize RNA and has important implications for how dsRNA binding proteins, such as RLRs, recognize RNA.

Functionally, A-to-I editing alters RNA function in a number of ways. Initial characterization of A-to-I editing observed that the modification results in the recoding of glutamate receptors due to CAG glutamate codons being read by the ribosome as CGG codons¹¹³. Since this initial discovery, several examples of RNA editing recoding transcripts have been identified, mainly in neural models, along with numerous (>1000) potential recoding sites identified bioinformatically^114,115. In general, the alteration of protein sequence by recoding is not a frequent occurrence, given that the majority editing sites fall in noncoding regions. Outside of recoding, A-to-I editing has been attributed to splicing regulation, miRNA binding, circular RNA biogenesis, and most relevant to this review, recognition of RNA by PRRs^{102,116–120}.

The importance of A-to-I editing in the antiviral innate immune response initially became apparent when deficiencies and mutations in ADAR1 were shown to cause the inflammatory disorder Aicardi-Goutiéres syndrome (AGS)^121,122. AGS is characterized by constitutive induction of IFN and caused by mutations in genes that encode proteins involved in nucleic acid metabolism¹²³. As the RNA editing activity of ADAR1 is essential to prevent aberrant IFN induction, this suggests that cellular RNAs must be edited to shield them from PRRs¹²⁴. Indeed, deletion or mutation of MDA5 or its downstream signaling partner MAVS prevents an aberrant IFN response induced by reduced ADAR1 expression^{109,124–126}. ADAR1 deficiency can also lead to activation of other dsRNA PRRs, such as the dsRNA kinase PKR for translational shutdown, and RIG-I for IFN induction^102,127. Depletion of RIG-I and PKR only partially rescue the aberrant immune activation caused by deletion of ADAR1, suggesting that MDA5 is a primary sensor of non-edited ADAR1 substrates^102,126. Together, these studies reveal that ADAR1 editing of cellular RNA is essential to prevent PRR activation and the associated IFN response (Fig. 3B). Importantly, as ADAR1 limits IFN expression in the absence of viral infection, the substrate of ADAR1 in these contexts is cellular RNA. As mentioned above, we now know that source of this self-RNA edited by ADAR1 are usually endogenous Alu retroelements, which base pair to make dsRNA structures^109,110. Thus, ADAR1 is essential to prevent aberrant IFN induction by cellular RNA.

Given the importance of A-to-I editing to prevent sensing by the antiviral innate immune system, one would hypothesize that ADAR1 could be co-opted to edit viral RNA that often contains double-stranded regions. Indeed, several RNA viruses are edited by ADAR1 including: measles virus, hepatitis D virus (HDV), HIV, and even SARS-CoV-2^128–132. HDV provides a rare example of ADAR1-induced protein recoding, where editing of its RNA by ADAR1 results in an UAG stop codon being recoded to UIG, read as UGG, leading to translational read through and expression of a protein important for viral replication^130,133. ADAR1 can promote HIV replication by editing the viral RNA, but the mechanism for this increased replication is not understood¹³². On the other hand, ADAR1 restricts measles virus, and measles virus has A-to-I editing¹³⁴. However, it is not known if the antiviral function of ADAR1 is through editing of the viral RNA. Beyond these examples, the importance of A-to-I editing in viral replication remains unclear. While much like 5′ mRNA capping and m⁶A modification, A-to-I editing can have effects on viral replication, decoupling whether altered editing of viral RNA or altered host gene expression result in the phenotype generated by ADAR1 inactivation is often not achievable due to technical limitations of the experimental systems used. The implications and current understanding of A-to-I editing in the context of viral infection have been extensively reviewed elsewhere^135,136, and conclude that while A-to-I editing and ADARs are important factors for viral replication, the underlying mechanisms are diverse and necessitate further research.

Other RNA modifications and the antiviral response

This review highlighted three of most well-known RNA modifications, but these represent only a fraction of the over 170 known distinct chemical modifications, including glycans, found on RNA molecules^3,137. Other RNA modifications not yet discussed in this text that likely regulate the antiviral innate immune response include pseudouridine (ψ), 5-methylcytosine (m⁵C), and N4-acetylcytidine (ac4C)^138–141. The RNA modification ψ can ablate RIG-I binding to PAMPs and it has been detected on viral RNA^71,72,142. Further, it can regulate basic RNA processes such as splicing and translation, making it a probable regulator of both host and viral RNA during the antiviral response^139,143. Similarly, ac4C can promote cellular RNA translation, has been detected on the transcripts of several RNA viruses, and was recently shown to promote stability of specific HIV transcripts^142,144,145, suggesting that it also could regulated aspects of the antiviral response. m⁵C has also emerged as having roles in viral infection. This modification has long been known to be present in tRNA and rRNA, with its roles on mRNA only more recently appreciated^146–148. On murine leukemia virus RNA, the m⁵C modification is essential for proper nuclear export and translation¹⁴⁹. This mechanism was found to be mediated in part by the m⁵C binding protein called ALYREF, which facilitates nuclear export of transcripts. This functionality, combined with studies detailing a pro-translational effect of m⁵C on mRNA transcripts indicates a potent ability to regulate host gene expression through the presence of this modification. Indeed, studies have begun to describe how m⁵C and the associated machinery can be found to be involved with expression of cell cycle regulators including p21, and even immunity related proteins such as IL-17A^150,151, suggesting that it too could regulate the expression of genes in the antiviral response.

Despite our limited knowledge about the physiological roles and pathways involved in the regulation of most of the many diverse modifications to mRNA, many will likely play important roles in the interplay between virus and host. Indeed, mass spectrometry has found that viral infection changes the levels of a wide variety of RNA modifications in cellular RNA¹⁴². Further, RNA extracted from virions of multiple diverse viruses is decorated with a wide variety of modifications¹⁴². For many of these modifications, very little is known about their biological function or how they are added to RNAs. While there are inherent challenges to studying many of these modifications due to technical limitations detecting their abundance and location on specific mRNAs, once these limitations are overcome, we are likely to see new roles for a variety of RNA modifications in controlling a functional and protective antiviral innate immune response.

Discussion

RNA modifications play an essential role in the antiviral innate immune response at nearly every level. Herein we provided examples of the most well-studied RNA modifications known to regulate antiviral innate immunity: 5′-capping, m⁶A, and A-to-I editing. For each of these modifications, we provide a primer for how they are added and regulated, as well as examples of how they influence antiviral immunity, ranging from altering PRR recognition of PAMP RNA, cytokine production, and expression of genes involved in the antiviral response. We highlight the role of RNA modifications in preserving innate immune homeostasis, specifically relevant for m⁶A and A-to-I editing, which prevent aberrant PRR activation in response to host-derived PAMPs, and for the IFN-inducible the 2′-O-methyltransferase, CMTR1, which catalyzes the addition of cap-1 to ISGs to prevent their translation inhibition by IFIT proteins and ensure a functional antiviral response^{48,77,102,109}. While many advancements have been made in exploring the roles of RNA modifications in antiviral innate immunity, much is yet to be learned, including contributions of less-studied modifications to the antiviral response, defining new antiviral proteins that sense modifications, and if different modifications regulate each other for immune responses.

Novel modes of immune regulation by RNA modifications have been uncovered by new technologies that can detect and quantify RNA modifications. However inherent technical limitations to current experimental approaches have made it difficult to fully understand the breadth of RNA modification-mediated regulation of the host response. Specifically, in the case of m⁶A and A-to-I editing, a reliance on short-read sequencing for mapping modification sites means that it is often not feasible to determine if a transcript is modified at multiple sites simultaneously. This hurdle would be relevant in cases where specific modification sites have different function, such as for m⁶A in HCV RNA, where in the IRES the modification promotes replication, but in the E1 gene it inhibits virion production^152,153. Thus, RNA modifications located throughout a transcript could differentially regulate function, and the ways in which this could occur are unknown. Understanding these mechanistic nuances will shed light on how specific modifications regulate transcripts and their downstream consequences in the antiviral response. Relatedly, accurately quantifying changes in some RNA modifications, such as m⁶A, is cumbersome, as these detection techniques often rely on immunoprecipitation based methods for detection^154,155. However, new approaches, such as the use of long-read direct sequencing that directly detects RNA modifications, antibody-independent modification mapping, and more sophisticated detection and analyses pipelines, are beginning to overcome these barriers^155–161.

RNA modifications are emerging as key factors that control many aspects of human biology as mutations in RNA modifying enzymes are linked to a diverse array of diseases including cancer, developmental, neurological, metabolic, cardiovascular, and, as discussed extensively here, immunological diseases¹⁶². Indeed, therapeutic strategies to target the machinery of each of the three modifications discussed in this review are currently being developed, and hold great promise for preventing disease, especially for targeting m⁶A in acute myelogenous leukemia^163–166. RNA modifications beyond those described are an essential component of safe and effective mRNA vaccines^4–8. As the field of RNA modifications in regulating innate immunity, the host response, and disease in general, progresses, we expect RNA modification biology to become a common component in the development of therapeutic platforms.