Abstract
Virus infection in mammals elicits a variety of defense responses that are initiated by signals from virus-sensing receptors expressed by the host. These receptors include the ubiquitously expressed RIG-I–like receptor (RLR) family of RNA helicases. RLRs are cytoplasmic proteins that act in cell-intrinsic antiviral defense by recognizing RNAs indicative of virus presence. Here, we highlight recent progress in understanding how RLRs discriminate between the RNA content of healthy versus virus-infected cells, functioning as accurate sensors of virus invasion.
Viruses are obligate intracellular parasites that infect all organisms, from bacteria to humans. Their evolution represents a constant arms race with the host: Viruses need to reprogram host cells in order to produce progeny virus, but this is often successfully limited by the host antiviral defense, which in turn is frequently targeted by the virus, and so forth. Mammals possess the most multifaceted antiviral defense program. Their reaction to viral infection includes the rapid induction of antiviral proteins, natural killer cells, neutralizing antibodies and cytotoxic T cells. These immune responses are coordinated by signaling molecules, including the type I interferons (IFN-α and IFN-β) and the related type III IFN (IFN-λ). All nucleated cells can synthesize IFNs in response to virus infection, which implies the existence of cell-intrinsic mechanisms for sensing viral presence. Some of these mechanisms have been identified recently and involve signaling for IFN gene transcription by members of the RIG-I–like receptor (RLR) family of pattern recognition receptors in response to specific RNA “patterns” that are generated during virus infection (1).
The RLR family has three members: retinoic acid inducible gene I (RIG-I), melanoma differentiation-associated gene 5 (MDA5), and laboratory of genetics and physiology 2 (LGP-2). These cytoplasmic proteins all share a central DExD/H-box RNA helicase domain. RIG-I and MDA5 also have two N-terminal caspase activation and recruitment domains (CARDs). CARDs allow for the interaction of activated RIG-I or MDA5 with the adaptor protein mitochondrial antiviral signaling (MAVS, also known as IPS-1, VISA, and Cardif), which localizes to the outer mitochondrial membrane. MAVS relays the signal to kinases such as TANK-binding kinase 1 (TBK1) and IκB kinase ε (IKKε), which in turn activate transcription factors, including interferon response factor 3 (IRF-3), IRF-7, and nuclear factor κB (NF-κB), which coordinate IFN gene induction (1). As well as this pathway, a MAVS-independent function of RIG-I was recently described: RIG-I directly activates the inflammasome, a protein complex that cleaves pro-interleukin-1β (IL-1β) into mature IL-1β, a pro-inflammatory cytokine (2).
RIG-I is indispensable for IFN responses to many single-stranded RNA viruses. These include negative-stranded viruses of the orthomyxovirus (such as influenza A virus) and paramyxovirus (such as measles, mumps, and Sendai virus) families and positive-stranded viruses like hepatitis C or Japanese encephalitis viruses. That RIG-I–deficient mice are highly susceptible to infection with these viruses underscores the importance of that RLR in antiviral defense. Similarly, MDA5 is essential for protection from a different set of viruses, including picornaviruses (such as poliovirus and encephalomyocarditis virus). The largely non-overlapping pattern of virus susceptibility in mice deficient for either RLR implies that the two receptors possess distinct virus specificities, although some viruses can be dually recognized by either RIG-I or MDA5. Little is known about virus sensing by LGP2, which may instead primarily play a regulatory role (1). The virulence of some viruses, including some strains of influenza A virus (3), is due at least in part to a dysregulation of the innate immune response. Therefore, understanding how RLRs become activated may allow the development of new strategies for the containment of viral spread and prevention of disease, as well as help to understand the basic principles underlying self/virus innate immune discrimination. Here, we summarize the rapid progress made in the last few years toward defining ligands for RLRs (Fig. 1).
Putative RIG-I ligands. RIG-I has been reported to be triggered experimentally by a variety of RNA agonists. 5′-PPP–bearing RNAs are shown in green, RNAs without 5′-PPPs in blue, and RNAs that may have different 5′-end characteristics in orange. An antagonist is shown in black. Activated RIG-I promotes the induction of interferons and other pro-inflammatory cytokines via the mitochondrial adaptor MAVS (bold red arrows). MAVS-dependent induction of pro-interleukin-1β allows it to be processed into mature interleukin-1β by the inflammasome, which can be directly activated by RIG-I in a MAVS-independent manner.
Virus sensing is highly discriminative given that RLRs are localized in the cytoplasm, where host RNAs abound; yet, signaling occurs only in infected cells. Thus, for RLRs to be activated, they must detect RNA bearing a molecular pattern not found under normal conditions. Such patterns may be chemical modifications of RNA (or the absence of such modifications), specific secondary or tertiary RNA conformations, particular sequences, or the annealing of two complementary RNA strands so as to form double-stranded RNA (dsRNA). Furthermore, the agonistic RNA may be of viral or cellular origin. Early studies demonstrated that RIG-I–dependent IFN production can be triggered by transfection of certain synthetic and natural RNAs into cells and provided the first insight into how RIG-I might discriminate virus from host RNA. These studies showed that RNA transcribed in vitro by phage polymerases (IVT-RNA) is a potent RIG-I agonist (4, 5). IVT-RNAs have an uncapped 5′-triphosphate (5′-PPP) that is required for their IFN-inducing activity. 5′-PPPs promote the binding of RIG-I (4, 5), activation of its adenosine triphosphatase (ATPase) activity, and conformational changes that allow RIG-I dimerization and exposure of CARDs for interaction with MAVS (6, 7). Furthermore, structural analysis indicates that the C terminus of RIG-I folds into a domain that recognizes uncapped 5′-PPPs on RNA (6, 7). This can explain why RIG-I does not respond to host cytoplasmic RNA because the latter lacks 5′-PPPs; for example, mRNAs are capped, and nuclear processing of ribosomal and tRNAs removes or modifies 5′-PPP groups before they reach the cytoplasm.
Surprisingly, two recent studies report that chemically synthesized RNAs bearing 5′-PPP do not trigger RIG-I, whereas the same RNAs can do so when made by means of in vitro transcription (8, 9). This apparent contradiction can be explained by the fact that IVT-RNA preparations often contain small amounts of unexpected RNA species. These include RNAs made in error by phage polymerases that switch from transcribing the DNA template to copying their own RNA product. When this occurs, the polymerase effectively extends the 3′ end of the RNA into a self-complementary molecule that folds into a hairpin. Only these hairpins, and other base-paired RNAs present in IVT-RNA preparations, act as agonists for RIG-I (8, 9). Furthermore, annealing of inert 5′-PPP RNA made through chemical synthesis to a complementary RNA oligonucleotide lacking 5′-PPP restores RIG-I activation, particularly if a blunt end is formed (8, 9). These data indicate that base-pairing at the 5′-end of RNA, together with a 5′-PPP, is required for RIG-I activation. RIG-I can translocate on base-paired RNA, and this probably contributes to its signaling activity (10).
These findings are consistent with earlier observations that viral RNA genomes extracted from influenza A or rabies virus particles can trigger RIG-I (4, 5). Those viruses have a 5′-PPP–bearing RNA genome, and enzymatic removal of the 5′-PPP abolishes stimulatory activity (4, 5). Some viruses, such as Hantaan virus, Crimean-Congo hemorrhagic fever virus, and Borna disease virus, have genomes with 5′-monophosphate ends, and RNA extracted from these virus particles does not activate RIG-I (11). The RNA genomes of influenza A, rabies, and other viruses that are recognized by RIG-I have complementary 5′ and 3′ ends and adopt a “panhandle” conformation (12). Therefore, these RNAs provide the two features read by RIG-I: a 5′-PPP and base-pairing at the 5′-end. It is worth remembering that base-paired does not mean double-stranded; by definition, dsRNA requires two (complementary) RNA molecules. However, the IVT-RNA hairpins and viral genomic RNAs that trigger RIG-I are single-stranded, even if they form intra-molecular base pairs. This distinction is important: The RNA genome of RIG-I–dependent viruses provides all features required for RIG-I activation in a linear RNA molecule without the need for viral replication. We therefore suggest that the use of the term double-stranded be restricted to the situation in which two complementary RNA molecules anneal, such as after the replication of some viruses.
As well as a secondary structure at the 5′-PPP end, other properties of RNAs may also influence recognition by RIG-I. For example, incorporation of modified bases such as pseudouridine—which are often found in cellular RNAs—into IVT-RNA decreases its stimulatory potential (5). A sequence motif in the 3′ nontranslated region of the hepatitis C virus genome was suggested to be required for RIG-I activation, together with the 5′-PPP end of the genome (13). Surprisingly, some RNAs without 5′-PPPs have also been reported to trigger RIG-I. These include chemically synthesized dsRNA oligonucleotides as well as shortened forms of poly inosinic: polycytidylic acid (poly I:C), which is an analog of dsRNA (7, 14). Furthermore, products of host RNA cleavage by ribonuclease (RNase) L, which bear 5′-hydroxyl- and 3′-monophosphate ends, were suggested to contribute to RIG-I activation (15). Finally, phosphorothioated single-stranded DNA oligonucleotides (containing a sulfur-substituted internucleotide bond) were recently identified as RIG-I antagonists (16).
Despite the wealth of information on the types of RNA that can activate RIG-I, the natural RIG-I agonist (or agonists) responsible for inducing IFN in virus-infected cells remains unclear. This is because all of the data identifying RIG-I stimulatory RNAs were obtained in nonphysiological experimental settings, such as transfection of naked RNA into cells or in biochemical assays that measured RIG-I binding, ATPase activity, or conformational changes. Total RNA extracted from infected cells can trigger RIG-I (14, 17); however, the specific agonist within such pools has not been identified. Candidates include viral genomes, viral transcripts, replication intermediates, or host RNA cleaved by RNase L. IFNs were originally discovered through the treatment of chicken cells with high doses of heat-inactivated influenza A virus (18). This treatment delivers viral RNA genomes to the cytosol in the absence of virus replication, suggesting that those genomes are sufficient to trigger RIG-I. Thus, RIG-I may act in infected cells primarily by sensing incoming viral genomes or ones generated during viral replication.
Much less is known about the nature of the RNAs that act as agonists for MDA5. Poly I:C is prepared by annealing inosine and cytosine homopolymers that have 5′-diphosphate and 5′-monophosphate ends. Transfection of poly I:C into cells triggers MDA5-dependent IFN induction, and a recent report shows that a minimum length of poly I:C is required for efficient MDA5 activation: Shortening poly I:C to around 1000 nucleotides or less is reported to convert it into a RIG-I agonist (14). These observations have been interpreted to indicate that MDA5 recognizes long dsRNA generated during infection. Indeed, dsRNA accumulates in cells infected with viruses that are recognized by MDA5 (4, 14, 17). When we size-fractionated total RNA from cells infected with encephalomyocarditis or vaccinia viruses, however, we found that only a high-molecular-weight RNA stimulated MDA5 upon transfection into reporter cells (17). This RNA was larger than most of the dsRNA generated during infection and was composed of both single- and double-stranded portions, suggesting a weblike conformation (17). Poly I:C may similarly adopt a branched structure given that the inosine and cytosine polymers have varying lengths. The definition of synthetic RNAs recognized by MDA5 and the characterization of MDA5 agonists purified from infected cells will help elucidate how that RLR can discriminate virus from self RNA.
An exciting recent development is the realization that the function of RLRs goes beyond the sensing of RNA viruses; they can also drive IFN responses to cytoplasmic DNA (19, 20). Infection with DNA viruses or some bacteria can deliver DNA to the cytoplasm, and this can be mimicked experimentally by transfection of the DNA polymer poly dA:dT. Although poly dA:dT cannot be sensed directly by RLRs, it is transcribed by cytosolic RNA polymerase III into an uncapped RNA that triggers RIG-I (19, 20). Consistent with observations using IVT-RNA or viral RNA genomes, 5′-PPPs and base-pairing are both required for RIG-I activation via the RNA polymerase III pathway (19, 20). This pathway appears to contribute to IFN induction during infection with the bacterium Legionella pneumophilia (20) and with DNA viruses such as adenovirus, herpes simplex virus, and Epstein-Barr virus (19, 20). Vaccinia virus is another DNA virus that triggers RLRs, in this case by generating an RNA agonist for MDA5 (17).
How do these findings illuminate the arms race between virus and host? In the case of negative-strand RNA virus-sensing (such as influenza A virus), RIG-I appears to recognize those features of the viral RNA genome that are indispensable for virus replication: The 5′-PPP end and the panhandle act as a promoter for the viral polymerase, and the virus cannot alter this pattern without sacrificing its own replication (12). Instead, viruses fight back by encoding proteins that inhibit RLRs or downstream signaling pathways. For example, influenza A virus can inhibit RIG-I by means of its NS1 protein, whereas hepatitis C virus cleaves MAVS off mitochondria (1). Much remains to be clarified as to how virus is sensed by the infected cell and how this is translated into an innate immune response. What beside IFN induction is regulated by RLR activation? What are the molecular patterns recognized by MDA5? What sensors detect cytoplasmic DNA (and how are they regulated at mitosis, when the nuclear envelope breaks down)? Do polymorphisms or mutations in RLRs and downstream adaptors affect human susceptibility to virus infection? Can aberrant activation of RLRs lead to detrimental autoreactive responses? Fifty years after the discovery of IFNs (18), the struggle between virus and host is only just beginning to reveal its molecular secrets.
