N6-Methyladenosine modification affects nearly all aspects of RNA biology, including stability, translation, splicing, nuclear export, and localization. Methylation modification, adding adenosine to N6 to form m⁶A, is catalyzed by a large heterogeneous complex of proteins that are named as “writer,” including METTL3, METTL14, or Wilms tumor 1-associated protein and KIAA1429 (Meyer and Jaffrey, 2017; Shi et al., 2019). In contrast, demethylases enzymes like fat mass and obesity-associated protein (FTO) or α-ketoglutarate-dependent dioxygenase AlkB homology 5 (ALKBH5) designated as “eraser” remove the methyl group (Jia et al., 2011; Zheng et al., 2013). The YTH domain family of proteins (YTHDC1, YTHDC2, YTHDF1, YTHDF3, and YTHDF3) and others named as “reader” recognize and bind to the m⁶A modification site to directly regulate the posttranscriptional functions of modified RNAs (Shi et al., 2019). m⁶A modifications are typically identified within the DRA^mCH motif (D = G/A/U, R = A/G, and H = A/C/U); however, given the fact that only some of the DRACH motifs in eukaryote transcriptome are modified, there might exist some mechanisms for site-selective modification (Dominissini et al., 2012).

Another kind of RNA base methylation is the C5-methylation of RNA cytosine-m⁵C. m⁵C widely exists in cytoplasmic and ribosomal RNA (rRNA), tRNA, mRNA, and some non-coding RNAs (Lewis et al., 2017; Bohnsack et al., 2019). In eukaryotes, m⁵C is catalyzed by enzymes of the NOL1/NOP2/SUN domain (NSUN) family and DNA methyltransferase family protein (DNMT2), a homolog of DNA methyltransferase (Reid et al., 1999). Recent studies showed m⁵C was present in numerous virus genomes and might have non-negligible effects on antiviral innate immunity (Winans and Beemon, 2019; Wnuk et al., 2020). For instance, a high level of m⁵C modifications in HIV-1 genomic RNA (gRNA) promoted the expression of viral genes by regulating splicing and the translation efficiency of viral mRNAs (Courtney et al., 2019). Silencing or inactivation of the major writer NSUN2 of m⁵C reduced the m⁵C abundance in HIV-1 transcripts and inhibited virus replication by disrupting the alternative splicing and the followed translation of HIV-1 mRNA (Kong et al., 2020).

Cellular mRNA conventionally has a triphosphate at 5ʹ end (5ʹ-ppp), which is converted to 5ʹ-diphosphate (5ʹ-pp) by RNA triphosphatase (Ramanathan et al., 2016). This conversion resulted in mRNA capping by guanylyltransferase and guanine-N7 methyltransferase (Shatkin, 1976). After adding a terminal guanosine base, the mRNA transcripts possess an m⁷G joined via a 5ʹ, 5ʹ-triphosphate bridge, designated as cap-0 (Shatkin, 1976). When a cellular 2ʹ-O-methyltransferase, CMTR1, further modifies the mRNA, a methyl group is added at the 2ʹ-O-hydroxyl position of the first nucleotide to form cap-1 RNA structure (Belanger et al., 2010). Meanwhile, a second 2′-O-methyl group can be added at the second nucleotide to form cap-2, catalyzed by another cellular methyltransferase CMTR2 (Werner et al., 2011; Smietanski et al., 2014). mRNA capping is considered one of the key factors in regulating RNA metabolism and function (Topisirovic et al., 2011), including stabilizing the mRNA and serving as a chemical marker to discriminate self from foreign RNA, the latter of which may interfere with the innate immune sensing of viral derived RNA (Hocine et al., 2010; Ramanathan et al., 2016). Some viruses, such as West Neil virus (WNV) or Dengue virus (DENV), encode 2ʹ-O MTases that catalyze 2ʹ-O-methyl adenosines inside the virus genome (Dong et al., 2012; Chang et al., 2016). Interestingly, this internal adonosine 2ʹ-O-methyl activity requires the same K-D-K-E motif as that for 2ʹ-O methylation of the 5ʹcap (Yap et al., 2010; Dong et al., 2012). Given that many viruses possess cap structures in their RNA components, this type of modification is supposed to play a pivotal role in antiviral innate immunity.

After the pseudouridine (ψ) was firstly identified in plant turnip yellow mosaic virus (TYMV) in 1998 (Becker et al., 1998), the follow-up research continuously indicated the abundant ψ in RNA viruses, especially in positive-sense RNA viruses (McIntyre et al., 2018). Psedouridine occurs through isomerization of uridine-to-5-ribosyl uracil by pseudouridine synthases (PUS; Markham and Smith, 1951). Similar to ψ, the deamination of adenosine to inosine (A-to-I) that depends on the catalyzing of adenosine deaminase acting on RNA (ADAR) family is also considered RNA editing or uncanonical nucleotides (Chen et al., 2000; Bass, 2002; George et al., 2014; Chung et al., 2018; Eisenberg and Levanon, 2018). The difference is that inosine generally acts similarly to guanosine (G), whereas ψ remains the original capacity of uridine to some extent. Although U to ψ conversion does not change the Watson–Crick base-pairing with adenosine, in certain cases, ψ enables base pairing with any other nucleotides (Samuel, 2011; Pfaller et al., 2021). Both ψ and A-to-I editing may significantly convert RNA biology, including changing the coding preference of viral RNA dependent RNA polymerases, mediating alternative splice and even affecting RNA structures (Netzband and Pager, 2020; Pfaller et al., 2021).

Modified nucleotides may stabilize the functional RNA structures by reinforcing the hydrogen bond between Watson–Crick pairs, resulting in augmented thermal stability and reduced dynamics (Serra et al., 2004; Zhou et al., 2016; Frye et al., 2018). Under other circumstances, Watson–Crick base pairs consisting of modified nucleotides may induce an alternative folding representing significant alterations in the RNA secondary or tertiary structures, negatively affecting RNA stability (Durbin et al., 2016; Zhou et al., 2016; Roundtree et al., 2017; Wei and He, 2021). Moreover, the posttranscriptional introduction of modified nucleotides can affect RNA intermolecular interactions with other encountered molecules such as DNA partners, RNA binding proteins, or other RNAs (Zhao et al., 2017; Nachtergaele and He, 2018; Shi et al., 2019). Since all functions are regulated by structure to a certain degree, viral RNAs carrying modified nucleotides (or uncanonical nucleotides) commonly represent functional differences in virus life-cycle, thus mediating virus infection in host cells. Besides, some RNA viruses those complete their life-cycle in cytoplasm influence host cell genomic transcription inside the cell nucleus, for example, ZIKV infection affects some endogenous genes’ trancription which occured inside the cell nucleus (Gokhale et al., 2020). In this case, epigenetic regulations may facilitate the virus to overcome the spatial barrier. To sum up, some of the demonstrated RNA modifications in virus genome as well as the correlative functions are listed in Figure 1. Notably, many viruses use epigenetic modifications as crucial tools to evade antiviral innate immune response.

Retinoic acid-inducible gene I, melanoma differentiation-associated protein 5 (MDA5), and laboratory of genetics and physiology 2 (LGP2) are the three major homologous helicases of RLRs (Wu et al., 2013; Yu et al., 2018; Thoresen et al., 2021). RIG-I and MDA5 displayed similar component structures, including caspase activation domain in N-terminal, recruitment domains (CARDs) for communicating with downstream signals, a DExD/H-box helicase domain with RNA binding and ATP hydrolysis activity, and a C-terminal domain (CTD; Yoneyama et al., 2004; Jiang et al., 2011; Kowalinski et al., 2011; Yu et al., 2018). The pathogen-associated molecular patterns (PAMPs) motifs of RIG-I include exposed 5ʹtriphosphate (5ʹppp) or diphosphate of double-strand RNAs (dsRNA), panhandle structures of viral genomic RNA, and uridine-rich sequences, while MDA5 recognizes long dsRNA such as poly (I:C; Yoneyama et al., 2004; Hornung et al., 2006; Kato et al., 2006; Cui et al., 2008; Liu et al., 2018). RIG-I is expressed at a low level in non-infected cells, usually referred to as a resting state with RNA-binding and helicase domains covered by RDs (Kowalinski et al., 2011; Luo et al., 2011). Following recognition of PAMPs, RIG-I undergoes a conformational change that provides room for closer interaction with more PAMPs RNAs and begins to release the CARDs for MAVS interaction and signaling (Pichlmair et al., 2006; Peisley et al., 2013; Thoresen et al., 2021). When the complex consisting of RIG-I, MAVS, and other cytosolic proteins translocate from the cytoplasm to the associated mitochondrial membrane, RIG-I CARDs interact with the MAVS CARD to catalyze the filament formation of MAVS and then activates TBK1 and IKKε to initiate downstream signaling (Pichlmair et al., 2006; Peisley et al., 2013; Goubau et al., 2014).

Another well-characterized PRRs, TLRs, are widely distributed invertebrates. TLRs are anchored in the cell membrane as type I transmembrane proteins (Akira et al., 2006). The ectodomain (N-terminal) of TLRs consists of several leucine-rich repeat (LRR), which connect to the C-terminal Toll/interleukin-1 receptor (TIR) domain by transmembrane (TM) domain (Akira et al., 2001; Kawasaki and Kawai, 2014). Studies have shown that most TLRs function as homology dimers (Kawai and Akira, 2011). Two TIR domains became close to forming a competent signaling state that recruits the adapter proteins (O’Neill et al., 2013). Nearly all the activated TLRs can trigger proinflammatory gene expression despite functioning in specific aspects of antiviral immunity (Kawai and Akira, 2011). To date, 10 TLRs have been identified in human cells. Four of them functioned as immune sensors by detecting pathogens-derived nucleotides (Kawasaki and Kawai, 2014). TLR3 recognizes long dsRNA and recruits TRIF as its dedicated adapter protein. Phosphorylated TRIF provides a signaling hub for IRF3 phosphorylation by TBK1, which then activates downstream signaling pathways of TRIF (Matsumoto et al., 2011; Oshiumi et al., 2011; Liu et al., 2015). TLR7 and TLR8 detect RNA debris as short RNA segments, while TLR9 enables sensing short DNA fragments that contain CG dinucleotide motifs (Chan et al., 2015; de Oliveira Mann and Hornung, 2021). TLR7, 8, and 9 can recruit the adapter protein MyD88 to form a complex known as the Myddosome. Myddsosome interact with IκB kinase and TGF-beta-activated kinase 1 (TAK1) complex to initiate NF-κB and MAPK signaling, respectively (Motshwene et al., 2009). Interestingly, the complex can also trigger IRF activation that depends on TASL that is only expressed in specific cells (Heinz et al., 2020). TASL is also capable of IRF phosphorylation, while in this case, IRF5 and IRF7, as well as IRF3, may be activated to drive antiviral gene expression (Wust et al., 2021).

Although several studies indicate the important roles of TLRs that usually sense long dsRNA inside endolysosome or outside the cells in antiviral innate immune response, they have rarely been found to be regulated by RNA modifications, partly because many RNA viruses expose their genomic dsRNA in the cytoplasm (Akira et al., 2001; Alexopoulou et al., 2001; Heil et al., 2004). Some studies have implied that the epigenetic marks of viral RNA interfere with the innate immune signaling pathway by preventing TLRs activation. For instance, the 2ʹ-O-methyl marks on coronavirus RNAs avoid the recognition of TLR7 to evade the activation of the IFN signaling pathway, while this effect may also be achieved through MDA5 sensing signals (Zust et al., 2011). Coronaviruses replicating in MDA5 or TLR7 deficient mice are detected to the same extent as in IFNR-deficient mice. By employing in vitro modified RNA oligos, an early study showed that m⁶A limited the capacity of RNAs to activate TLR3, TLR7, and TLR8, while m⁵C and Ψ blocked the activation of TLR7 and TLR8 (Kariko et al., 2005). Recent studies applied CRISPER tools to map the function of m⁶A and demonstrated that m⁶A could suppress macrophage activation through TLR mediated signaling (Tong et al., 2021a). However, more evidence of virus RNA modifications regulating TLR mediated pathways in innate immune response remains to be discovered.

The biological function of m⁶A is mainly regulated by a methyltransferase (writer), demethylase (eraser), and m⁶A binding protein (reader; Tong et al., 2018). Many studies have shown that RNA m⁶A modification plays an important role in innate immune response, while the exact roles of m⁶A in regulating antiviral IFN signaling displays in opposite aspect (Gokhale et al., 2016; Guo et al., 2020). In some cases, m⁶A modifications in the virus genome promote the IFN and ISGs induction, whereas, under other circumstances, m⁶A modification occurs to turn off the antiviral innate immune response. For example, the m⁶A modifications at specific sites in the HBV transcript restricts the virus replication through IFN α-mediated response. Although HBV is a DNA virus, it replicates through transitional pre-genomic RNA (pgRNA). m⁶A modification of A1907 in HBV pgRNA is the key regulator of IFN α-mediated pgRNA decay. Further investigation showed that ISG20 selectively degraded the m⁶A HBV transcripts that are strictly regulated by m⁶A reader YTHDF2 (Liu et al., 2017; Imam et al., 2020).

Contrary effects were found in encephalon myocarditis virus (EMCV), herpes simplex virus type 1 (HSV-1), and VSV infection. In these cases, YTHDF3 can inhibit the expression of ISGs by promoting the translation of transcriptional inhibitor FOXO3 (Zhang et al., 2019). RAW264.7 cells with YTHDF3 gene deletion have extensive antiviral activity against RNA and DNA virus, and this activity is mediated by the IFNAR1 signal (Zhang et al., 2019). Notably, m⁶A modification, in this case, regulated the host cell transcripts to inhibit antiviral innate immune response instead of affecting viral RNAs. Indeed, this viral infection-induced host cell m⁶A epitranscriptome diversity has commonly been found to regulate the antiviral innate immune response. During VSV infection, m⁶A modifications in MAVS, TRAF3, and TRAF6 are demethylated by ALKBH5 through interacting with the RNA helicase DDX46, which leads these three transcripts to retention in nuclei. Abolished expression of these three transcripts prevents efficient IFN induction (Zheng et al., 2017). Similarly, human cytomegalovirus (hCMV) infection affects host m⁶A modification machinery, including METTL14 and ALKBH5, reducing the IFNβ production. When knocking down the expression of METTL14, the production of IFNβ and subsequent signaling depending on the JAK/STAT pathway are enhanced, which decreases the production of infectious hCMV virion in infected cells (Rubio et al., 2018).

Interestingly, besides the direct influences of m⁶A modification on HBV pgRNA, it also has been indicated that m⁶A modification of tumor suppressor phosphatase and tensin homolog (PTEN) transcript is affected by HBV infection through invaliding PI3K/AKT pathway and inhibiting IRF-3 nuclear export (Kim et al., 2021). Other studies also indicated that DHX58, p65, and IKKγ, which bind to YTHDF2, are mediated by m⁶A modification, potentially interfering with IFN induction during virus infection (Lichinchi et al., 2016). Besides, YTHDF1, METTL3, and METTL14 have also been found to increase the expression of ISGs like IFITM1 in an m⁶A binding-dependent manner, which further indicated the m⁶A methyltransferase complex might promote the antiviral activity of type I IFN (McFadden et al., 2021).

Excluding the direct regulations of host transcripts by m⁶A modification, interactions between RNA and RBPs may also be affected by m⁶A modifications that subsequently affect antiviral IFN response (Bidet et al., 2014). For example, it has been found that during DENV-2 infection, three conserved RBPs, G3BP1, G3BP2, and CAPRIN1, are regulatory factors necessary for antiviral IFN response by promoting the efficient translation of PKR and IFITM2 mRNAs (Arguello et al., 2017; Edupuganti et al., 2017).

As one of the most important signaling pathways in innate immune responses, IFN producing and effecting are likely regulated by diversity factors, probably due to numerous protein enzymes evolving in the IFN signaling pathway. For example, NSUN2, the methyltransferase of m⁵C, has multiple effects on RNA biogenesis, including converting vault ncRNA to vtRNA (Bohnsack et al., 2019; Kong et al., 2020). The vtRNA has been shown to promote Influenza A virus (IAV) replication in A549 cells and mouse lungs through repressing PKR activation and the subsequent effects of interferon (Li et al., 2015; Wnuk et al., 2020). Similarly, DNMT2 has been reported to be required for efficient IFN responses in Drosophila C virus or Sindbis virus infected Drosophila (Durdevic et al., 2013; Bhattacharya et al., 2017).

Other studies also indicate an important role of ADARs, the enzymes mediating A-to-I editing, in modulating innate immune response during virus infection (Pfaller et al., 2021). ADAR1 has a proviral effect on Measles virus (MeV) and VSV infection that depends on PKR activation (Nie et al., 2007; Pfaller et al., 2015), while the suppression of innate immune response by ADAR2 is supposed to rely on STAT1 in the case of Chikungunya virus (CHIKV) and Venezuelan equine encephalitis virus (VEEV; Schoggins et al., 2011; Clavarino et al., 2012). In the case of other viruses, such as BoDV, IAV, and Yellow fever virus (YFV), the mechanisms of ADAR-regulated IFN response remain indistinct (Pfaller et al., 2021). Interestingly, during HIV-1 infection, ADAR1 and ADAR2 may have opposite effects on virus replication, through the forming of DNA:RNA heteroduplex or antiviral innate immune response, respectively (Clerzius et al., 2009; Doria et al., 2009; Pujantell et al., 2017).