Viral RNA is recognized by innate immune receptors from two different families. In endolysosomal compartments, Toll-like receptors (TLRs) 3, 7 and 8 recognize either double-stranded RNA (dsRNA) or single-stranded RNA. In the cytoplasm, viral genomic RNA or transcriptional intermediates are recognized by DExD/H-box helicases RIG-I and MDA5. Recent structural studies of these RNA sensors have provided atomic-level insight into the recognition mechanism of viral RNA. TLR3 dimerizes around a straight 45-bp stretch of dsRNA, explaining the length requirement of at least 40 bp for dsRNA recognition. RIG-I recognizes blunt ends of dsRNA with 5′-triphosphate caps. Ligand binding releases RIG-I from a closed autoinhibited state, exposing the CARD signalling domains. MDA5 recognizes long dsRNA by cooperatively assembling into helical filaments. RNA recognition by RIG-I and MDA5 triggers assembly of their common downstream signalling adaptor MAVS from its inactive monomeric form into its active polymeric form. While RIG-I and MDA5 appear to activate MAVS via distinct oligomerization mechanisms, a common paradigm is emerging in innate immunity for signal transduction by oligomerization-dependent signalling platforms. Many open questions remain including the role of proteolytic activation in RNA recognition by TLR3 and how unanchored ubiquitin chains contribute to RNA recognition by RIG-I and MDA5.
Vertebrates rely on their innate immune system to detect viruses and other microbes. Innate immune receptors detect broadly conserved microbial structures throughout the cell including bacterial cell wall components, microbial nucleic acids and certain highly conserved proteins. The rapidly ensuing inflammatory response culminates in activation of the NF-κB and type I interferon signalling pathways. This response is the first line of defence against infection and controls the activation of the adaptive immune system.
At the cell surface and in endolysosomal compartments, Toll-like receptors (TLRs) are the principal family of molecular sentries for the innate immune recognition of microbial patterns (Janeway and Medzhitov, 2002). Microbial nucleic acids are one of the principal classes of ligand for innate immune receptors. A subfamily of TLRs recognizes microbial nucleic acids, in endolysosomal compartments exclusively (Ahmad-Nejad et al., 2002; Matsumoto et al., 2003; Latz et al., 2004; Nishiya et al., 2005). TLR3 recognizes double-stranded RNA (dsRNA) ligands longer than 40 bp (Alexopoulou et al., 2001; Leonard et al., 2008). TLR7 and TLR8 recognize single-stranded RNA (ssRNA) ligands (Diebold et al., 2004; Heil et al., 2004). TLR9 recognizes microbial DNA ligands (Bauer et al., 2001). TLRs bind ligands with a leucine-rich repeat (LRR) ectodomain. Ligand binding to the ectodomain induces a conformational change, which is transmitted across the membrane by a transmembrane helix to the cytoplasmic Toll/interleukin-1 receptor homology (TIR) domain, leading to the recruitment and activation of various signalling adaptors such as MyD88 (Kawai and Akira, 2010).
In the cytoplasm, RIG-I and MDA5, a pair of RNA helicases, recognize complementary sets of viral RNA ligands. RIG-I recognizes blunt or 5′-triphosphorylated ends of viral genomic RNA segments, while MDA5 recognizes long cytosolic dsRNA delivered or generated during viral infection (Hornung et al., 2006; Kato et al., 2006; Pichlmair et al., 2006). RIG-I and MDA5 each contain two N-terminal caspase recruitment domains (CARDs), a DExD/H-box helicase (consisting of two RecA-like helicase domains, Hel1 and Hel2, and an insert domain, Hel2i) and a C-terminal domain (CTD). RIG-I adopts a closed inactive conformation in the absence of RNA (Kowalinski et al., 2011). RNA binding through the helicase and CTD domains (Jiang et al., 2011; Luo et al., 2011) releases the CARDs, which then recruit unanchored lysine 63-linked polyubiquitin chains and activate the signalling adaptor MAVS (IPS-1) (Zeng et al., 2010). In contrast, MDA5 does not sequester its CARDs (Berke and Modis, 2012) and cooperatively assembles into ATP-sensitive filaments on dsRNA (Peisley et al., 2011; Berke and Modis, 2012; Berke et al., 2012). Moreover, the MDA5 CTD is required for cooperative filament assembly but not for RNA binding (Li et al., 2009; Takahasi et al., 2009; Berke and Modis, 2012). The MDA5 CARDs have been proposed to nucleate the assembly of MAVS into its active polymeric form (Hou et al., 2011; Berke and Modis, 2012) in a process that also involves free polyubiquitin chains (Jiang et al., 2012). The self-propagating ‘prion-like’ properties of MAVS polymers amplify signalling (Hou et al., 2011).
Recent structural studies of innate immune receptors have provided atomic-level insight into the recognition mechanism of viral RNA both in the cytoplasm and in endocytic compartments. Here, we review how these studies have advanced our understanding of innate immune recognition of viral RNA.
Recognition of viral RNA in endolysosomal compartments
The signalling activity of the nucleic acid-sensing TLRs, in contrast to other TLRs, is regulated by proteolysis (Ewald et al., 2011). The ectodomain of TLR9 (TLR9-ECD) must be proteolytically activated by endosomal proteases in order for DNA ligand binding to produce an innate immune signal (Ewald et al., 2008; Park et al., 2008). TLR9 is cleaved between LRRs 14 and 15 (near residue 477) by asparagine endopeptidase and cathepsins to generate the functional receptor (Fig. 1A). TLR7 and TLR8 are processed in an analogous manner (Ewald et al., 2008). Likewise, TLR3 is primed for signalling in endolysosomes by proteolytic processing between residues 252 and 346 (Garcia-Cattaneo et al., 2012; Qi et al., 2012). In apparent contradiction with the requirement for TLR3 or TLR9 to be cleaved, the N-terminal cleavage fragment of these TLRs appears to be important for signalling. Several mutations within the N-terminal cleavage fragment of the TLR3 ectodomain (TLR3-ECD) have been shown to inactivate TLR3 (Liu et al., 2008). The crystal structure of TLR3-ECD has revealed that one of the dsRNA binding sites is located at the N-terminal end of the protein (Liu et al., 2008). Moreover, LRRs-2, -5 and -8 of TLR9, within the N-terminal cleavage fragment of TLR9-ECD, are required for signalling activation (Peter et al., 2009). To reconcile a contribution of the N-terminal cleavage fragment of TLR9-ECD in DNA recognition with the requirement for proteolytic activation of the ectodomain, it has been proposed that the cleavage products remain associated and that ectodomain cleavage allows TLR9 to undergo the ligand-induced conformational change necessary for receptor activation (Fig. 1B) (Li et al., 2012). As TLRs 3, 7 and 8 undergo the same proteolytic processing as TLR9, their signalling activation is likely to be regulated by a similar mechanism.
Recognition of dsRNA by TLR3
Toll-like receptor 3 recognizes viral genomic dsRNA or replication intermediates released in endolysosomal compartments (Akira et al., 2006), for example when the integrity of the viral capsid is compromised. TLR3 binds dsRNA in a sequence-independent manner, but an acidic environment (pH < 6.5) is critical for binding (de Bouteiller et al., 2005; Leonard et al., 2008). The minimum length of dsRNA required for TLR3-ECD binding is 40–45 bp, and receptor binding is positively cooperative (Leonard et al., 2008).
The crystal structure of TLR3-ECD in the absence of RNA revealed a horseshoe-shaped fold with N- and C-terminal capping domains (LRR-NT, LRR-CT) and 21 central LRRs (Bell et al., 2005; Choe et al., 2005). Eleven N-glycans were visible in the structure. Three years later, the structure of mouse TLR3-ECD in complex with a 46-bp dsRNA revealed the structural basis of RNA recognition by TLR3 (Liu et al., 2008). The dsRNA is sandwiched by two TLR3-ECD molecules forming an M-shaped assembly with the two LRR-NT domains 110 Å apart (Fig. 1C). The dsRNA interacts with two sites on each TLR3 subunit, including a contact with the glycan linked to Asn413. All TLR3-RNA contacts involve the ribose-phosphate backbone of the dsRNA, explaining why recognition is independent of RNA sequence. The TLR3-ECD/dsRNA interfaces are mostly hydrophilic and involve the side-chains of His39, His60, Asn515, His539, Asn541 and Arg544. This is consistent with the pH dependence of RNA binding, as at the reduced pH of the endosome histidine residues become protonated and positively charged, thereby enhancing the binding affinity for the RNA phosphate backbone.
The TLR3-dsRNA structure also revealed a ligand-induced dimer interface between TLR3-ECD monomers. The two LRR-CT domains form hydrogen bonds and salt bridges, bringing the two C-terminal residues within 25 Å of each other (Liu et al., 2008). The proximity of two C-termini brings the cytosolic TIR domains into place for signalling. The structure suggests a strict length requirement of at least 40 bp for dsRNA binding. It remains unclear how shorter dsRNA ligands (21–39 bp) activate TLR3 signalling (Kariko et al., 2004) and how proteolytic activation of TLR3 may affect ligand binding. It also remains to be determined how TLR3 senses dsRNA ligands longer than 90 bp, which may induce higher-order oligomerization of the receptor (Fig. 1D), as has been proposed for other TLRs (Ferrao et al., 2012). Indeed, evidence is mounting for a signalling mechanism that requires TLRs to form clusters of activated receptor dimers, which allow cytosolic adaptor molecules such as MyD88 for TLR7/8/9 or TRIF for TLR3 to recruit downstream effectors to form superhelical oligomeric signalling platforms (Lin et al., 2010; Ferrao et al., 2012; Luo et al., 2012) (Fig. 1D). In the so-called Myddosome, six MyD88 molecules, four IRAK-4 molecules and four IRAK-2 molecules assemble into a stable ternary complex upon TLR4 activation (Lin et al., 2010). As each activated TLR dimer binds two MyD88 molecules, a Myddosome could accommodate the TIR domains from three activated TLR dimers (Gay et al., 2011). TLR3 signalling depends on TRIF rather than MyD88. TRIF recruits TRAF6 and activates TAK1 for NF-κB activation (Kawai and Akira, 2008). Recent studies have shown that TRIF forms a multiprotein complex with TRADD, Pellino-1 and RIP1 (Kawai and Akira, 2010). Future studies are required to determine whether TRIF forms Myddosome-like signalling platforms in TLR3 signalling.
Recognition of ssRNA by TLR7/8
TLR7 and TLR8 recognize endosomal ssRNA, including ssRNA from vesicular stomatitis virus, influenza A virus and human immunodeficiency virus (Akira et al., 2006). TLR7/8 signalling can be modulated with synthetic small molecules, such as imidazoquinolines and nucleoside analogues (Hemmi et al., 2002; Heil et al., 2003). Biochemical studies have shown that residues in LRR8 and LRR17 of TLR8 are essential for signalling (Gibbard et al., 2006). However, the molecular basis of TLR7/8-ssRNA binding specificity is not well understood, because of the difficulty to obtain pure TLR7/8-ECD proteins and the absence of structural data.
Recognition of viral RNA in the cytoplasm
Recognition of short, capped dsRNA by RIG-I
Shortly after its initial discovery (Yoneyama et al., 2004), RIG-I was found to contain a CTD (Fig. 2A) whose ectopic expression caused a dominant negative phenotype in signalling assays leading to it being called the repressor or regulatory domain (Saito et al., 2007). The CTD has a similar structure to the small GTPase Exchange Factor Mss4 and imparts much of RIG-I's specificity by recognizing the end structure of RNA and a 5′-triphosphate cap (Cui et al., 2008; Takahasi et al., 2008). Blunt-ended dsRNA with a 5′-triphosphate cap is the most effective type of ligand for inducing RIG-I signalling. Ligands with 3′ overhangs have reduced effectiveness, and ssRNA or 5′ overhangs produce no signal (Schlee et al., 2009).
A recent crystal structure of full-length duck RIG-I (dRIG-I) reveals the architecture of unliganded RIG-I (Kowalinski et al., 2011). In the structure, the CARDs are sequestered through an interaction of CARD2 with the Hel2i domain, thereby keeping the receptor inactive (Fig. 2B). Two phenylalanines in the Hel2i domain form key contacts with CARD2, and mutation of either phenylalanine results in constitutively active phenotypes (Gee et al., 2008). The CTD is disordered in the dRIG-I structure, indicating a degree of flexibility. RNA presumably binds first to the exposed CTD before displacing the CARD domains and forming contacts with the helicase domains, leading to release of the CARDs and signalling activation (Fig. 2B).
Crystal structures of RIG-I constructs lacking CARDs and bound to dsRNA reveal the basis of RNA recognition (Jiang et al., 2011; Kowalinski et al., 2011; Luo et al., 2011). Superfamily 2 helicases typically bind a single strand of nucleic acid through several conserved motifs. Threonines in motif V of Hel2 and motif Ib of Hel1 bind phosphates of the tracking strand in a 5′ → 3′ direction. In RIG-I, an additional motif, IIa, just upstream of motif II, binds a phosphate in the complementary strand, imparting double-stranded specificity to the helicase. Upon binding RNA, a conformational change closes the cleft between Hel1 and Hel2 causing the helicase to wrap tightly around the RNA like a C-clamp (Fig. 2C). The amount of movement may be dependent on the nucleotide state of the helicase domain, as a greater closure of Hel1-Hel2 is apparent with ADP·AlF4 (Kowalinski et al., 2011) or ADP·BeF3 (Jiang et al., 2011) bound. A long helix at the end of Hel2 with a sharp turn leading into a second helix integrates tightly with Hel1 and connects to the CTD. These ‘bridging’ or ‘pincer’ helices are thought to co-ordinate conformational changes between the helicase domains for appropriate ligand and nucleotide-dependent signalling by the CARDs (Civril et al., 2011; Jiang et al., 2011; Kageyama et al., 2011; Kowalinski et al., 2011; Luo et al., 2011).
Hel2i has also been proposed to play a role in dsRNA recognition. However, contacts between Hel2i and RNA in the RNA bound structures differ depending on nucleotide bound state (Jiang et al., 2011; Kowalinski et al., 2011; Luo et al., 2011). Interestingly, in a nucleotide (AMPPNP) bound structure of the mouse helicase domain without RNA (PDB entry 3TBK), the orientation of Hel2i and Hel2 is very similar to both an RNA-bound, nucleotide-free structure (PDB entry 2YKG) and the CARD-containing, RNA-free structure (PDB entry 4A2Q). Given that none of these structures have the maximal cleft closure seen for the ternary complex (PDB entries 3TMI and 4A36), both RNA and nucleotide binding may be required for Hel2i repositioning and CARD release for downstream signalling.
The role of ATPase activity in RIG-I signalling remains unclear. Mutation of the conserved lysine in motif I produces a dominant negative phenotype (Yoneyama et al., 2004; 2005; Bamming and Horvath, 2009). While all helicase motifs are required for ATPase activity, mutations of each motif generated different signalling phenotypes. In the case of a constitutively active motif III mutant, cross-talk was observed with endogenous wild-type RIG-I signalling (Bamming and Horvath, 2009). It remains unclear whether these mutations strictly affect hydrolysis or whether they also affect binding.
Although RIG-I binds RNA as a monomer, it was recently shown that unanchored Lys63-linked polyubiquitin chains induce the assembly of tetrameric RIG-I-RNA complexes (Jiang et al., 2012). Moreover, these free ubiquitin chains are required for RIG-I signalling (Zeng et al., 2010), and the downstream signalling adaptor MAVS must assemble from its monomeric form into a membrane-tethered polymer (or fibrils in the absence of the mitochondrial transmembrane anchor) to be activated, in a prion-like, self-propagating switch (Hou et al., 2011). Thus, in the current model of RIG-I signalling, oligomeric assemblies containing RIG-I, RNA and ubiquitin nucleate the assembly of MAVS into its active polymeric form (Fig. 2D).
Recognition of long dsRNA by MDA5
In contrast to RIG-I, recognition of specific chemical structures by MDA5 has not been demonstrated. Rather, MDA5 discriminates for dsRNA longer than 1 kb from a distinct but partially overlapping set of viruses (Kato et al., 2008; Loo et al., 2008). Progressive digestion of 4 kb poly(I:C) dsRNA converts it from an MDA5 agonist to a RIG-I agonist (Kato et al., 2008). Structures of the MDA5 Hel1 domain (PDB entry 3B6E), Hel2i (PDB entry 3TS9) and MDA5 CTD (PDB entries 3GA3, 2RQB) reveal structural differences to their RIG-I counterparts (Li et al., 2009; Takahasi et al., 2009; Berke and Modis, 2012). The MDA5 CTD lacks a positively charge 5′-triphosphate binding pocket and the surface analogous to the RNA-binding surface of the RIG-I CTD is flatter in MDA5. Moreover, the MDA5 CTD has a lower affinity for dsRNA than the RIG-I or LGP2 CTDs. Instead, the MDA5 CTD is required for positive cooperativity of MDA5 binding to dsRNA. Deletion of the CTD (Berke and Modis, 2012) or replacement with the RIG-I CTD (Peisley et al., 2011) abolishes binding cooperativity.
RIG-I and MDA5 also diverge in the structure and function of their Hel2i domains. The Hel2i α2 helix, which in RIG-I interacts with CARD2 or RNA, is shorter in MDA5. A phenylalanine residue essential for binding CARD2 in RIG-I (F540 in the duck sequence) is not conserved in MDA5, suggesting that in the absence of RNA, MDA5 CARDs are regulated differently. Neither MDA5 Hel2i nor a construct with all three MDA5 helicase domains interacts with the isolated CARDs. Small angle X-ray scattering of full-length MDA5 shows that the protein has exists as a diverse ensemble of extended structures (Fig. 2E), consistent with a flexible but not completely random conformation of the 95-amino-acid CARD2-Hel1 linker (Berke and Modis, 2012). Together, these data demonstrate that in contrast to RIG-I, MDA5 has an open and flexible structure in the absence of RNA.
If the MDA5 CARDs are not kept inactive by intramolecular association, how then is MDA5 signalling regulated in the resting state? And how can dsRNA orders of magnitude longer than a single protein activate a signal whereas shorter dsRNA cannot? Evidence that MDA5 forms ATP-sensitive filaments on dsRNA provides insight on both of these questions (Peisley et al., 2011; Berke and Modis, 2012; Berke et al., 2012). Negative-stain electron microscopy shows that MDA5 forms filaments along dsRNA. Electron microscopy image reconstructions reveal helical filaments with ring-like asymmetric units 44 Å-wide separated by variable helical twists (Berke et al., 2012). An atomic model of MDA5 based on available crystal structures of RIG-I and MDA5 fragments fits well into the electron microscopy structure (Fig. 2F). Contacts between adjacent MDA5 molecules in the filament occur through the CTD, consistent with its role in dsRNA binding cooperativity. The regular array of MDA5 CARDs on the surface of the MDA5-dsRNA filaments could nucleate the oligomerization of MAVS, a necessary step for downstream signalling (Tang and Wang, 2009; Hou et al., 2011). Unanchored Lys63-linked polyubiquitin chains were recently shown to be required for activation of IRF3 by MDA5 (Jiang et al., 2012). It remains unclear whether the ubiquitin chains bind to MDA5 filaments or are associated with smaller oligomeric complexes as with RIG-I. Intriguingly, ATP concentrations within the narrow range observed in the cytoplasm (1–5 mM) progressively inhibit MDA5 filament formation (Berke and Modis, 2012) and some ATPase-deficient MDA5 mutants are constitutively active (Bamming and Horvath, 2009). Therefore, the ultimate length and lifetime of an MDA5 filament, and hence its signalling output, could be determined by the length of both the dsRNA ligand and the cellular ATP concentration, providing an elegant mechanism for tunable length discrimination (Fig. 2G). It is probably not a coincidence that the mitochondrion, which is responsible for cellular homeostasis and energy production, is emerging as a central hub of antiviral immune signalling. Indeed, upon sensing viral RNA, RIG-I is redistributed from the cytosol to mitochondrial subcompartments that become associated with stress granules, where RIG-I forms a ‘translocation complex’ with MAVS, the mitochondrial targeting chaperone 14-3-3ε and the TRIM25 ubiquitin ligase (Liu et al., 2012; Onomoto et al., 2012).
Conclusions and future directions
The structures of innate immune receptors reviewed here have been highly invaluable in providing an atomic-level understanding of the mechanisms of viral RNA recognition by innate immune receptors from the TLR and RIG-I-like receptor families. TLR3 dimerizes around 45 bp of dsRNA, explaining the 40-bp length requirement for dsRNA recognition. RIG-I recognizes blunt ends of dsRNA and 5′-triphosphate caps. Ligand binding releases RIG-I from a closed autoinhibited state, exposing the CARD signalling domains. MDA5 recognizes long dsRNA by cooperatively assembling into helical filaments. Both RIG-I and MDA5 trigger assembly of MAVS into its active polymeric form upon binding RNA, albeit via distinct mechanisms.
Many open questions remain. Although the TLR3-dsRNA complex is dimeric, TLR3 signals through TRIF, which forms a multiprotein complex with at least three other signalling molecules (TRADD, Pellino-1 and RIP1) (Kawai and Akira, 2010). Future studies are required to determine whether TRIF forms Myddosome-like signalling platforms in TLR3 signalling and, in this context, whether more than one TLR3 dimer is required to activate signalling. Moreover, structural data are still completely lacking for TLR7, TLR8 and TLR9, and the molecular basis of proteolytic activation of the nucleic acid-sensing TLRs remains unknown. For RIG-I and MDA5, an important open question is the mechanism by which unanchored ubiquitin chains contribute to signalling. A high-resolution structure of MDA5 in complex with dsRNA is also necessary to obtain a complete and detailed picture of the molecular determinants of RNA recognition.
Oligomeric assemblies as innate immune signalling platforms
A key conclusion from recent structural advances is that innate immune receptors assemble into large oligomeric signalling platforms, including TLRs, which have been traditionally viewed as dimerization-activated receptors. Signalling through assembly of large oligomeric arrays has now been observed in a number of immunological contexts, particularly in the mitochondrial pathway of apoptosis mediated by the Apaf-1 apoptosome and the PIDDosome (Park et al., 2007; Dickens et al., 2012; Reubold and Eschenburg, 2012). The oligomeric assemblies of RIG-I/polyubiquitin, MDA5, MAVS and TLR signalling adaptors reinforce the ongoing paradigm shift in our fundamental understanding of how innate and apoptotic signals are transduced, from prototypical linear signalling cascades to oligomeric platform-based signal amplification (Park et al., 2007; Lin et al., 2010; Mace and Riedl, 2010). All of these signalling platforms have in common that they assemble through domains from the Death Domain superfamily, which includes TIR domains and CARDs. However, some important questions remain regarding the molecular mechanism of these signalling platforms, in particular, how MAVS forms polymers (or fibrils) and how it binds the CARDs from MDA5 and RIG-I. Further studies are also necessary to determine how unanchored ubiquitin chains contribute to MDA5 and RIG-I signalling, and how TLRs interact with the Myddosome.
This work was supported by NIH grant P01 GM022778 to Y. M. and by a Burroughs Wellcome Investigator in the Pathogenesis of Infectious Disease Award to Y. M.