The Chase for the RIG-I Ligand—Recent Advances

Abstract

Multicellular organisms evolved efficient host-defense mechanisms to sense viruses and to block their replication and spread. Invertebrates and plants mainly rely on RNA interference (RNAi) for antiviral defense. In mammals, the initiation of antiviral defense mechanisms is largely based on the detection of viral nucleic acids by innate receptors: retinoic acid–inducible gene I (RIG-I)–like helicases (RLHs) and Toll-like receptors (TLRs). RLHs play a pivotal role in sensing viral RNA and DNA in the cytoplasm of cells. RLHs, like Dicer of the RNAi pathway, belong to the phylogenetically conserved DExD/H-box family of helicases. Unlike TLRs, RLHs are functional in all somatic cells. Activation of RIG-I triggers antiviral responses including type I interferon (IFN), inflammasome activation and proapoptotic signaling. Here, we provide a comprehensive overview of the current literature on the ligand structures detected by RIG-I, and conclude with the molecular definition of the RIG-I ligand: short double-stranded blunt-end 5′-triphosphate RNA. The recent information on the RIG-I ligand now allows the design of short double-stranded RNA (dsRNA) oligonucleotides that are ideally suited alone or in combination with small-interfering RNA (siRNA) for the treatment of viral infection and cancer.

Introduction

Viruses represent the most abundant class of pathogens. Viruses manipulate the metabolism of the host cell to convert cells into virus-producing factories and cause dysfunction and damage triggering cell death. In response, organisms have evolved efficient host-defense mechanisms to sense viruses and block their replication and spread.¹ In both plants and invertebrates, RNA interference (RNAi) is an important antiviral defense mechanism in which double-stranded RNA (dsRNA) is detected by RNase III enzymes of the Dicer family and is cleaved into small-interfering RNA (siRNA) of 21–23 nucleotides. These are incorporated in the RNA-induced silencing complex, which directs the RNase H enzyme AGO to complementary sequences resulting in highly specific cleavage of viral RNA.

In contrast, mammals mainly rely on innate [i.e., type I interferon (IFN)] and adaptive immune responses to overcome viral infection. Although viral proteins can contribute to the induction of type I IFN in mammals the innate immune system senses mainly viral nucleic acids.¹ The innate receptors responsible for nucleic acid recognition belong to two families, the transmembrane Toll-like receptors (TLRs) that detect viral RNA or DNA in the endosomal compartment, and the cytoplasmic sensors retinoic acid–inducible gene I (RIG-I), MDA-5, and Lgp2, which form the RIG-I-like helicase (RLH) family.¹ It is interesting to note that DExD/H-box helicases comprise phylogenetically related members of the RIG- and Dicer-like helicase families, and that both sets of evolutionary conserved DExD/H-box helicases direct antiviral responses in both invertebrates and vertebrates. The concept that both RNAi and innate immunity are intricately related is further supported by the fact that TAR RNA-binding protein and PKR activator are not only involved in the regulation of type I IFN upon dsRNA recognition, but in fact are key components of the RNAi machinery.² Moreover, RIG-I is homologous to Dicer-related helicase-1 in Caenorhabditis elegans, and the RNAi machinery in most lower organisms (but not mammals) uses RNA-directed RNA polymerase (leading to the generation of 3pRNA) as an amplification mechanism. In addition, there is evidence for an involvement of micro RNA and RNAi components in the fine-tuning of innate immune responses. Thus, the RNAi and the RLH pathways are closely related with possible divergence from RNAi to innate immunity in recent evolution.

An exciting new therapeutic perspective is to tackle these two most fundamental antiviral pathways that evolved jointly for the treatment of viral infection and cancer. Based on the recent insight into the RNA structures involved in both pathways, such so-called “bifunctional siRNA” molecules are now in place for clinical development.³

Immunorecognition of Viral Nucleic Acids

In vertebrates, antiviral responses are initiated through the detection of viral nucleic acids by innate pattern recognition receptors. Among the TLRs that detect viral nucleic acids, TLR7, TLR8, and TLR9 are mainly expressed in immune cell subsets, whereas TLR3 is expressed on immune cells and many nonimmune cells.^4,5 TLRs detect viral RNA or DNA based on certain molecular characteristics and the cellular localization of nucleic acids. Upon activation TLRs induce cytokines including type I IFN and interleukin-12 and Th1 chemokines. Initially, TLR3 was thought to be the only pattern recognition receptor capable of detecting long dsRNA. In 2004, Yoneyama et al.⁶ identified a truncated complementary DNA coding for a caspase recruitment domain (CARD) belonging to the RIG-I by screening a complementary DNA library for IFN-β-inducing genes. RIG-I is a cytosolic DExD/H box-containing RNA helicase, a group of helicases that unwind dsRNA in an ATP-dependent manner. It is composed of an N-terminal tandem CARD domain followed by a helicase domain and the C-terminal “regulatory domain” with a zinc coordination site. Similar to other known dsRNA-sensing antiviral proteins in the cytosol including PKR, OAS, and RNAse L, RIG-I could be precipitated by the artificial dsRNA poly(I:C) (ref. 6, reviewed in ref. 5). Overexpression of RIG-I was shown to induce interferon regulatory factor-3 dimerization and activation of the IFN-β gene upon stimulation with dsRNA [poly(I:C)] or infection with a (−) single-stranded RNA virus such as New Castle disease virus. Mutation within the ATPase region or deletion of the CARD domain led to inactivation of the RIG-I response to poly(I:C) in HEK 293T cells.⁶ RIG-I without the CARD domain exhibited a dominant negative effect on dsRNA-stimulated signaling. The CARD domain of activated RIG-I binds to IFN promoter stimulator 1 (IPS-1) (also known as Cardif, MAVS, or VISA), which in turn activates the interferon regulatory factor-3 kinase TBK1. In parallel to RIG-I, Yoneyama et al. reported the presence of the homologous proteins MDA-5 and Lgp2 (ref. 6). Although MDA-5 possess the same domain structure as RIG-I, Lgp2 lacks a CARD domain. Consequently coexpression of MDA-5 led to enhanced IFN-β promoter induction upon poly(I:C) transfection, whereas Lgp2 exhibited similar suppressive effects as CARD-depleted RIG-I. Further studies dissected the roles of RIG-I and MDA-5 (refs. 7,8). It was shown that MDA-5 plays an essential role in the detection of picornaviruses such as encephalomyocarditis virus or Theiler's virus. In addition, MDA-5 emerged to be indispensable for the induction of type I IFN by poly(I:C) when administered intravenously into mice or when transfected into mouse embryonic fibroblasts (MEFs), peritoneal macrophages, or dendritic cells in vitro. Conversely, RIG-I was shown to detect long dsRNA generated by in vitro transcription using phage polymerase⁷ and all classes of RNA viruses tested (Table 1) except picornaviruses, and was not essential for poly(I:C)-dependent type I IFN induction.^{7,9,10,11,12,13,14,15}

Table 1. Viruses recognized by RIG-I and evading RIG-I recognition.

RIG-I Ligand: 5′-Triphosphate and Sequence

The first hint toward a type I IFN–inducing modification of RNA came from the laboratory of John Rossi developing siRNA approaches against human immunodeficiency virus infection. While developing siRNA against human immunodeficiency virus, Rossi's group observed that siRNA that was generated by in vitro transcription using phage polymerase, but not synthetic siRNA, induced type I IFN in a variety of human cell lines (HEK293, HeLa, K562, CEM, Jurkat; see Table 2).¹⁶ As template-dependent RNA transcription generally occurs primer independent from the 5′- to the 3′-end of RNA, a triphosphorylated end is left at the 5′-end of de novo–generated RNA transcripts of all known RNA polymerases including phage polymerase. Kim et al. reported that removal of the triphosphate at the 5′-end either by RNase T1 (removed the 5′-end 3p-GGG, which was single stranded in their case) or by phosphatase treatment was sufficient to completely abrogate IFN-inducing activity of in vitro–transcribed RNA.¹⁶ In another study, Hornung et al. searched for RLH ligands by screening RNA molecules for their ability to induce IFN-α in human primary monocytes lacking the ability to produce IFN-α upon TLR-ligand stimulation.¹¹ In these studies, the surprising observation was made that 5′-triphosphate RNA (Table 2) induced similar quantities of IFN-α in human primary monocytes as the TLR9 ligand CpG DNA or the TLR7 ligand RNA did in the major type I IFN–producing cell, the plasmacytoid dendritic cell. Removal of the triphosphate at the 5′-end abolished IFN-α-inducing activity of RNA in monocytes but not in plasmacytoid dendritic cells, which are capable of recognizing the same 5′-triphosphate RNA molecule via TLR7. Variation of the 5′-modification of RNA revealed that neither a diphosphate nor a monophosphate modification was sufficient to reconstitute the maximum amount of IFN-α induction seen with 5′-triphosphate RNA (3pRNA) in monocytes. Conversely modifications of bases (pseudouridine, 2-thio-uridine) and backbone (2′-O-methyl-uridine) abolished IFN-α-inducing activity of 3pRNA. Furthermore, Hornung et al. identified RIG-I as the receptor responsible for immunorecognition of 3pRNA. At the same time, Pichlmair et al. found that detection of influenza virus RNA depends on the presence of at least one phosphate at the 5′-end of RNA and that the presence of long dsRNA was dispensable¹² (Table 2).

Table 2. Putative RIG-I ligands generated by enzymatic polymerization or cleavage. Due to the use of enzymatic polymerization, the RNA molecules in the table are not molecularly defined and additional RNA molecules may be present.

Later, Saito et al. concluded from their data that RIG-I recognizes (+)RNA viruses [e.g., hepatitis C virus (HCV)] in a sequence-dependent manner.¹⁷ In their work, small domains of the genome of HCV were made by in vitro transcription and assessed for RIG-I activity. The authors claimed that a 100-nt U- or A-rich region 8,000 nt downstream of the 5′-triphosphate end is responsible for RIG-I activation (Table 2). Interestingly, in their work a poly U elicited a similar IFN response as a poly A sequence. In order to demonstrate the in vivo relevance of their finding, they compared the IFN response of a full-length HCV genomic transcript to a transcript lacking the whole 3′-nontranslated region (230 nt) including the putatively immune stimulatory sequence (PU/UC, 105 nt). Although the genome lacking the 3′-nontranslated region was nonimmunostimulatory, the impact of this deletion on virus translation and replication was not tested in their study. As positive (+)ssRNA virus genomes are translated and represent highly infectious agents, a correct interpretation of these results is challenging. Gondai et al.¹⁸ discovered that extension of the 5′-end by more than one G-abolished type I IFN (Table 2) by in vitro–transcribed short-hairpin RNAs. Both studies suggested that the sequence composition of an RNA contributes to their ability to activate RIG-I.

RNase Cleavage Products

The above-mentioned findings were in conflict with the initial finding that RIG-I can be activated by poly(I:C).⁶ Poly(I:C) is a dsRNA polymer that is generated by polynucleotide phosphorylase and that possesses monophosphates at the 5′-end.¹⁹ In order to dissect recognition patterns leading to the recognition by MDA-5 or RIG-I, Kato et al.²⁰ observed that the ability of poly(I:C) to activate either RIG-I or MDA-5 is size dependent (Table 2). In their study, they digested poly(I:C) with RNaseIII (generating RNA fragments with 5′-monophosphates and 2-nt 3′-overhangs) and size-fractionated RNA fragments were transfected into wild type MEF or MEF deficient for RIG-I or MDA-5. They observed that high-molecular-weight poly(I:C) (7 kb) was preferentially recognized by MDA-5, whereas fractions containing shorter poly(I:C) (300 bp) were exclusively recognized by RIG-I. In another study, Malathi et al.²¹ reported that activation of the antiviral endoribonuclease RNase L by 2′,5′-linked oligoadenylate cleaves self-RNA into small RNA cleavage products that trigger type I IFN production (Table 2). Interestingly, both, MDA-5 and RIG-I were found to be involved in the recognition of such short (<200 nt) RNaseL cleavage products of cellular self-RNA. RNase L which is inducible by type I IFN cleaves ssRNA to RNA fragments with 5′-OH and 3′-monophosphate ends. A structural motif responsible for the immunorecognition of RNA was not identified in their work.

RNA Polymerase III Transcripts of Exogenous DNA

Recently, two groups independently observed that dsDNA transfected into the cytosol of cells promotes a TLR9-independent type I IFN response.^22,23 A siRNA-mediated knockdown of IPS-1 in 293T cells led to the reduction of dsDNA-induced type I IFN response suggesting an IPS-1 dependent pathway of dsDNA recognition.²² However, murine IPS-1/MAVS-deficient cells still responded to dsDNA.²⁴ Of note, in the above-mentioned experiments the heteropolymer dAdT was used as synthetic dsDNA analog. Later Cheng et al.²⁵ reported that human cell lines (Huh7, HEK293) raise an IPS-1- and RIG-I-dependent type I IFN response upon transfection of dAdT but not of plasmid DNA. The mechanism responsible for this seemingly confusing result was found later by two independent groups.^26,27 Ablasser et al. and Chiu et al. discovered that dAdT serves as a template for the endogenous RNA polymerase III leading to 5′-triphosphorylated AU-polymers that are detected by RIG-I (Table 2). In cells that are unable to detect dsDNA sequences other than dAdT (most tested nonimmune cells such as HEK293 cells)²⁶ polymerase III, RIG-I represents the only pathway to detect cytosolic dsDNA. As some DNA viruses and intracellular bacteria raise IPS-1- or RIG-I-dependent type I IFN response in nonimmune cells,^26,27 it was proposed that in these cells the innate immune response to intracellular pathogen-derived dsDNA generally occurs via RIG-I-dependent sensing of polymerase III transcripts of such DNA.

Blunt-End dsRNA

Before the discovery of the 5′-triphosphate group as a pattern recognition motif for RIG-I, Marques et al.²⁸ proposed that synthetic blunt-end dsRNA oligonucleotides are detected by RIG-I (Table 3). The glioblastoma cell line T98G was transfected with blunt-end siRNA or siRNA with 3′-overhangs. They found that short synthetic dsRNA with blunt ends but not short synthetic dsRNA with 3′-overhangs induced a type I IFN response as detected by measuring p56, a type I IFN–induced protein. The same effect was seen for MRC-5 cells. The use of siRNA-mediated gene silencing of RIG-I in T98G suggested the involvement of RIG-I. Confusingly, HeLa cells and HT1080 cells, which show high p56 induction in response to in vitro–transcribed RNA, were not responsive to short blunt-end synthetic dsRNA. In HT1080 cells, the response to short blunt-end synthetic dsRNA could be recovered by priming with type I IFN. Based on the results with T98G cells, the RIG-I motif was defined as blunt-end dsRNA longer than 23 bp with higher activity when the RNA contained two blunt ends and with higher tolerance toward 5′-overhangs compared to 3′-overhangs (Table 3).

Table 3. RIG-I ligands generated by chemical synthesis. With the use of chemical synthesis, the RNA molecules depicted are molecularly well-defined.

RIG-I Protein Structure and Function

Cui et al.²⁹ performed interaction studies of synthetic blunt-end dsRNA and of in vitro–transcribed triphosphorylated ssRNA with purified recombinant full-length RIG-I and domain deletion mutants. The ATPase activity of several mutants was compared for synthetic blunt-end dsRNA and in vitro–transcribed ssRNA (ivt3P-ssRNA). The results from these experiments demonstrated that ATPase of full-length RIG-I is activated upon interaction with ivt3P-ssRNA, whereas the response to nonphosphorylated dsRNA was marginal. In contrast, the ATPase of RIG-I lacking the CARD domain was activated by both ivt3P-RNA and dsRNA. Interaction studies with recombinant RIG-I protein or the regulatory domain confirmed the requirement of the 5′-triphosphate group for binding.²⁹ The crystal structure predicted a triphosphate-binding site in the regulatory domain (amino acids 802–925) of RIG-I.²⁹ Structure-guided mutational analysis of RIG-I function in HEK293 cells suggested a positively charged grove as the likely 5′-triphosphate-binding site with K858 being essential for binding to and activation by 5′-triphosphate RNA. On the other hand, a residual ATPase activity of the isolated RIG-I-DECH domain with synthetic dsRNA pointed to a dsRNA-interacting site in the DECH domain.²⁹ Takahasi et al.³⁰ applied an NMR approach to analyze the structure of the RIG-I regulatory domain and its interaction with dsRNA and ivt3P-ssRNA in solution. NMR titration experiments suggested a positively charged “cleft” to be involved in binding of both RNA ligand types. Structure-guided mutational analysis of RIG-I function in RIG-I-transfected RIG-I^−/− MEFs confirmed that K858 together with K861 (KK858/861AA double mutant) was crucial for both dsRNA- and ivt3P-ssRNA-induced type I IFN induction. A KK878/880AA mutant impaired type I IFN induction. They described a RIG-I dependent response to synthetic 5′- and 3′-monophosphorylated dsRNA oligonucleotides in a murine cell line and in MEFs as analyzed by interferon regulatory factor-3 dimerization and IFN-β promoter reporter assays (Table 3). In contrast to Marques et al.,²⁸ Takahasi et al. reported that dsRNA with no phosphorylated ends did not elicit type I IFN.³⁰ Similar to Marques et al.,²⁸ 3′-overhangs (at the 5′-monophosphorylated end) were not tolerated; 5′-overhangs were not tested (Table 3). For 3′-monophosphorylated dsRNA, dsRNA with 2-nt 3′-overhangs induced a type I IFN response; however, no comparison to other end structures (blunt, 5′-overhang) was performed in their study. As monophosphorylation did not appear to enhance the binding of dsRNA to RIG-I protein, it was proposed that the mechanism of monophosphorylation-dependent RIG-I-inducing activity may be due to increased RNA stability in vivo.³⁰ In contrast to earlier findings,²⁸ the immunologic activity of dsRNA with 3′-overhangs at the 5′-monophosphorylated end were negatively correlated with increased capacity to serve as a helicase substrate for RIG-I. In light of recent work from Taniguchi's group,³¹ HMGB proteins may facilitate RIG-I binding to its substrate, and thus the molecular RIG-I ligand interaction seems not yet fully elucidated.

Synthetic Short Blunt-End Double-Stranded 5′-Triphosphate RNA

As phage polymerase in vitro–transcribed RNA is restricted to a conserved starting nucleotide (G; or A followed by G), this method cannot be used for the identification of optimal 5′-end sequences of 5′-triphosphate RNA for RIG-I recognition. We established a protocol that allows to make highly pure synthetic 5′-triphosphate RNA.³² Surprisingly, in contrast to all published studies using in vitro–transcribed RNA, synthetic single-stranded triphosphate RNA (3P-ssRNA) did not induce type I IFN in human monocytes. The “same” RNA sequence generated by in vitro transcription (ivt3P-ssRNA) was highly stimulatory. Reversed cloning and sequencing of ivt3P-ssRNA revealed the presence of double-stranded hairpin species and complementary sequences generated by template-dependent phage RNA transcription. The result suggested that RIG-I was not activated by the intended single strand but by unexpected aberrant side products. Indeed, transcription reaction conditions, which do not permit the generation of unintended complementary RNA, abolished RIG-I-inducing activity of in vitro–transcribed ssRNA.³² Consequently, addition of a complementary ssRNA to synthetic 3P-ssRNA recovered RIG-I stimulation. Further, we demonstrated that base pairing of the nucleoside carrying the 5′-triphosphate was required (no 5′-overhang tolerated), whereas a 2-nt 3′-overhang at the 5′-triphosphate end impaired RIG-I activation by >70% (Table 3). The structure at the nonphosphorylated end had no substantial impact on RIG-I activation, if the length of the dsRNA was at least 19 bp. Small (3 nt) bulge loops were well tolerated. All four nucleotides were accepted at the 5′-end of the RIG-I ligand; however, although pppA, pppG, and pppU slightly differed in RIG-I activation (A > G > U), pppC induced ~50% less type I IFN. Of note, most genomic viral RNA (vRNA) start with pppA (exceptions HCV: pppG; Ebola: pppU). No vRNA starting with pppC is known so far.

Using commercial synthetic 3P-ssRNA, results obtained by Schmidt et al. confirm the requirement of dsRNA at the 5′-triphosphate end.³³ In conflict with our data, they reported the tolerance of a 1-nt 5′-overhang at the 5′-triphosphate end; furthermore in their work the minimum length of the double strand tolerated is 10 bp. It remains unclear from these data whether RIG-I indeed tolerates a 10mer dsRNA for several reasons: only three different sizes of triphosphate-dsRNA (15, 10, and 5 bp, blunt at the triphosphate end) were tested; the 10mer triphosphate-dsRNA induced a stronger response than the 15mer (error bars are missing in the figure), and the triphosphate-dsRNA positive control is missing. Of note, the chemical identity of the 5′-triphosphate RNA used in their study has not been confirmed. Furthermore, data by Schmidt et al. suggesting that longer (>1 nt) 5′-triphosphate overhangs in hairpin RNAs are tolerated are misleading as those experiments were performed using in vitro–transcribed 3P-RNA and the presence of complementary RNA cannot be excluded by size fractionation.

In both studies 5′-monophosphorylated and nonphosphorylated dsRNAs did not induce substantial amounts of type I IFN in human monocytes.^32,33 In agreement with this, Habjan et al.¹⁴ observed that some Bunyaviruses and Borna disease virus circumvent RIG-I-mediated detection by 5′-processing of their genomic RNA to obtain monophosphorylated 5′-ends. Thus, recognition of monophosphorylated RNA by RIG-I appears rather weak.

The Molecular Definition of the RIG-I Ligand Compared to Earlier Literature

Based on synthetic well-defined RNA structures, short double-stranded blunt-end 5′-triphosphate RNA was defined as RIG-I ligand. Due to the fact that the generation of RNA molecules by in vitro transcription leads to aberrant and unexpected RNA molecules,³² the interpretation and the conclusions of a number of studies in the literature including our own study¹¹ need to be reconsidered. Clearly an ssRNA molecule with a triphosphate group at the 5′-end as suggested by a number of groups^{7,9,10,11,12,13,14,15,16} is not sufficient for RIG-I activation. Furthermore, the finding of Gondai et al.¹⁸ is easily explained by the 5′-overhang at the triphosphate end of RNA, which is not allowed by RIG-I. In their work, the introduction of extra Gs at the 5′-triphosphate end generated single-stranded or mismatched 5′-triphosphate ends, which are not detected by RIG-I. Thus, the inhibition of RIG-I activity was not due to the specific sequence but to the structure of the 5′-triphosphate end of the hairpin RNA.

The observation of Saito et al.¹⁷ that both poly U and poly A rich 5′-triphosphate RNA molecules elicit similarly strong RIG-I responses can be explained by the fact that both types of RNA occur in the phage polymerase transcription reaction, which the authors intended to use for the generation of the one or the other RNA type. Indeed, the poly A rich RNA used by Saito et al.¹⁷ turned out to be nonstimulatory when separated from side products in a gel chromatography.³³ Furthermore, poly A and poly U are highly repetitive sequences that form weak or no secondary structures. As a consequence, hybridization of poly A and poly U single strands to poly(A:U) containing dsRNA is facilitated compared to mixed G/C rich sequences.

To date, the evidence for sequence motif–dependent recognition by RIG-I is scarce. Our results suggest a slight preference for A over G and U with substantially lower activity of C at the 5′-end in blunt-end 5′-triphosphate-dsRNA. Interestingly, triphosphate-A (pppA) is the most abundant first nucleotide in genomes of RNA viruses (exceptions are Ebola: pppU; and Hepatitis C: pppG), and no viral genome starts with a pppC. It is anticipated that sequence motifs play a more prominent role in nonperfectly matched dsRNA. The availability of synthetic 5′-triphosphate RNA now allows to address this issue in greater detail.

Detection of Viral RNA by RIG-I

How does the molecularly defined RIG-I ligand that requires at least short dsRNA relate to the recognition of viral RNA of ssRNA viruses? Not only dsRNA viruses but also (+)ssRNA viruses produce substantial amounts of cytosolic dsRNA during the replicative life cycle.³⁴ Thus, in accordance with the requirement of dsRNA for RIG-I recognition, dsRNA and (+)ssRNA viruses are detected by RIG-I (see Table 1). In contrast, negative-strand RNA [(−)ssRNA] viruses (Table 1) do not form dsRNA during infection;^12,34 nevertheless they are detected by RIG-I. However, the antibody used to demonstrate the absence of dsRNA^12,34 only detects dsRNA longer than 40 bases.³⁵ Like other negative-strand RNA viruses, influenza contains highly complementary 5′- and 3′-sequences that form a short (about 15 bp) double-strand structure with a perfect blunt end (so-called panhandle).³⁶ For influenza virus, these panhandle structures serve as an RNA transcription initiation site for the viral RNA polymerase complex in the nucleus of the host cell. The influenza panhandle structure (Figure 1) would not be detected by a dsRNA-specific antibody that requires a length of dsRNA >40 bases; however, panhandles of negative-strand RNA viruses meet many requirements for RIG-I recognition as defined in our recent work:³² dsRNA with a blunt end and a triphosphate group at the 5′-end. Recently, Dauber et al.³⁷ could show that the influenza panhandle is accessible for PKR in the cytosol of the host cell after export from the nucleus. As PKR and RIG-I appear to bind similar RNA structures,^32,38 it is tempting to speculate that RIG-I may also recognize the influenza panhandle structure. Indeed, there is recent evidence that RIG-I detects genomic RNA of influenza virus.³⁹ However, it seems likely that negative-strand viruses evolved sequences at the 5′- and 3′-ends of their genomes forming panhandles that largely evade immunorecognition by RIG-I. However, this hypothesis awaits formal evidence by using well-defined synthetic RNA oligonucleotides.

Figure 1.

Panhandle configuration of negative-strand RNA viruses. In silico hybridizations of the 5′- and the 3′-end of different negative-strand RNA viruses. MeV, measles virus; SeV, Sendai virus; VSV, vesicular stomatitis virus.

So far only one (−)ssRNA virus that replicates in the cytosol (La Crosse virus) has been examined for its production of dsRNA using the 40-bp dsRNA-specific antibody.³⁴ In general, the genomes of all (−)ssRNA viruses have the potential to form panhandle structures due to complementary 5′- and 3′-sequences (Figure 1). For Bunyaviridae including La Crosse virus, it was shown by electron microscopy and psoralen-crosslinking that 5′- and 3′-end of the viral genome form a panhandle structure in vivo.⁴⁰ The panhandles of La Crosse virus consist of a 24–27 bp dsRNA with few mismatches (Figure 2)⁴⁰ meeting nearly all requirements for RIG-I recognition.³² In contrast, viral particles of Sendai virus and measles virus contain predominantly linear nucleocapsids, suggesting that formation of a panhandle is prevented by encapsidation with structural proteins in vivo.⁴¹ However, still a minority of circular genomes is visible in viral particles of measles virus as examined by electron microscopy.⁴¹ On the other hand, Sendai virus and vesicular stomatitis virus are known to produce defective interfering (DI) viral genomes during life cycle.^42,43,44 Three types of DI genomes were classified:⁴⁴ DI genomes with internal deletions (see Figure 2, Deletion), 5′-promoter duplications leading to completely complementary 5′–3′-ends (see Figure 2, Panhandle) and “snap back” or hairpin DI genomes, consisting of a completely dsRNA hairpin of 100–1,000 bp (Figure 2; snap back). At least the “snap back” and the “panhandle” DI genome type most likely represent RIG-I ligands. Indeed, data by Strähle et al.⁴⁵ strongly suggest that activation of RIG-I by Sendai virus is correlated with the presence of snap back DI genomes (DI-H4), which occur during infection and do not show encapsidation and therefore can form panhandle structures in vivo.

Figure 2.

Types of viral genomes within defective interfering particles of (−)ssRNA viruses. Transcriptional errors lead to generation of shortened (ΔC) and 100% complementary sequences, depicted as B′ and C′.

Replication of negative-strand RNA viruses requires a highly conserved promoter at both ends of the genome. For vesicular stomatitis virus, it was shown that additional nucleotides introduced at the 5′-end of the vesicular stomatitis virus (−)ssRNA genome are eliminated during replication and that extra nucleotides at the 3′-end are not tolerated at all.⁴⁶ Thus, blunt-end 5′-triphosphate short dsRNA formation, the consequence of the requirement of two conserved promoters, may represent a negative-strand virus–associated molecular pattern detected by RIG-I. Even if evolution of viruses has forced an adaptation of the panhandle structure of some viruses in order to circumvent RIG-I recognition, RIG-I may still detect the errors of viral RNA polymerases similar to phage RNA polymerase in vitro transcripts.^32,33 The answer to this question requires a systematic analysis of viral panhandle structures by using synthetic 5′-triphosphate RNA oligonucleotides.

Even if virus-infected cells contain RIG-I ligands, viruses can escape RIG-I recognition by encoding information that abrogates RIG-I signaling. Examples are the HCV NS3/4A protease targeting the key signaling molecule of RIG-I, IPS-1,⁴⁷ the human respiratory syncytial virus nonstructural protein NS2 interacting with RIG-I,⁴⁸ and the influenza A virus NS1 protein forming a complex with RIG-I and IPS-1.¹²

Conclusion

Although a detailed understanding of RIG-I ligands presented by RNA viruses will reveal new viral escape mechanisms, the therapeutic development of well-defined 5′-triphosphate RNA RIG-I ligands for antiviral therapy is intuitive. The identification of the exact molecular structure required for RIG-I activation now allows to generate defined synthetic 5′-triphosphate RNA ligands for RIG-I with reproducible and predictable induction of potent antiviral responses. Synthetic 5′-triphosphate RNA has a number of advantages over 5′-triphosphate RNA generated by in vitro transcription: first, purity and identity of synthetic 5′-triphosphate RNA is independent of the respective sequence. Second, unlike with in vitro transcription the synthetic approach allows to modify the base composition at the 5′-triphosphate end of the RNA as exemplified by defining adenosine or guanosine as the preferred 5′-nucleosides. Third, large-scale synthesis based on standard RNA chemistry allowing the addition of modifications can be used. Fourth, the biological activity of synthetic 5′-triphosphate RNA is superior to in vitro–transcribed RNA. These advantages of synthetic 5′-triphosphate RNA will greatly improve the development of such RNA oligonucleotides for the treatment of viral infection and cancer. Synthetic 5′-triphosphate RNA oligonucleotides could be used to stimulate appropriate antiviral responses in order to protect cells from viruses that manage to escape viral recognition and in which the RIG-I pathway is still intact, such as hepatitis B virus. The introduction of target-specific silencing into such RNA oligonucleotides (bifunctional siRNA) may further improve the antiviral efficacy. Such bifunctional siRNA have proven efficient antitumor agents in tumor models including melanoma.³ Besides silencing of specific oncogenes, activation of RIG-I induces tumor cell apoptosis⁴⁹ and activates both the type I IFN and the inflammasome pathway⁵⁰ thereby recruiting innate effector cells. These mechanisms may act in concert to tackle the tumor from different biological angles.