RIG-I-Mediated Antiviral Responses to Single-Stranded RNA Bearing 5'-Phosphates

See allHide authors and affiliations

Science  10 Nov 2006:
Vol. 314, Issue 5801, pp. 997-1001
DOI: 10.1126/science.1132998


Double-stranded RNA (dsRNA) produced during viral replication is believed to be the critical trigger for activation of antiviral immunity mediated by the RNA helicase enzymes retinoic acid–inducible gene I (RIG-I) and melanoma differentiation–associated gene 5 (MDA5). We showed that influenza A virus infection does not generate dsRNA and that RIG-I is activated by viral genomic single-stranded RNA (ssRNA) bearing 5′-phosphates. This is blocked by the influenza protein nonstructured protein 1 (NS1), which is found in a complex with RIG-I in infected cells. These results identify RIG-I as a ssRNA sensor and potential target of viral immune evasion and suggest that its ability to sense 5'-phosphorylated RNA evolved in the innate immune system as a means of discriminating between self and nonself.

The innate immune response to viral infection is characterized by the rapid production of a range of cytokines, most prominently type I interferons (IFN-α/β) (1). Specialized plasmacytoid dendritic cells (pDC) produce IFN-α/β when RNA or DNA viral genomes in endosomes trigger toll-like receptors (TLRs) 7, 8, and 9 (2). Other cell types rely on cytoplasmic virus sensors such as the RNA helicases RIG-I and MDA5 (37), which are believed to be activated by dsRNA produced during viral replication or convergent transcription of viral genes (8). Viral recognition pathways can be targeted as a means of immune escape (9). For example, influenza A virus NS1 protein suppresses IFN-α/β production in animal models, cell lines, and primary cells (10), including conventional (nonplasmacytoid) dendritic cells (cDC), a critical cell type for the induction of adaptive immunity (11, 12). NS1 possesses an RNA binding domain at the N terminus (13) and some evidence suggests that it exerts its suppressive effect by sequestering dsRNA (10).

To address whether this is the case, we first confirmed that influenza A virus (A/PR/8/34 strain) inhibits IFN-α/β production through the action of NS1. cDC derived from murine bone marrow progenitors (BM-DC) (14) infected with a mutant virus lacking the protein (ΔNS1) produced 100 times more IFN-α than did cells infected with the parental wild-type strain (Fig. 1A) (11, 12). This was independent of the TLR7-TLR8-TLR9 adaptor MyD88 (Fig. 1A), suggesting that the effect occurred by means of the cytoplasmic pathway. We then assessed the extent to which dsRNA is generated during influenza replication. Consistent with recent data (15), no dsRNA was detected in BM-DC or in more permissive Vero cells (Fig. 1, B and C, and fig. S1) with either ΔNS1 or the wild-type virus, even though the cells were uniformly infected (fig. S2). In contrast, dsRNA was detectable upon transfection with poly inosinic:cytidylic acid (poly I:C), a synthetic dsRNA, or infection with encephalomyocarditis virus (EMCV), a picornavirus (Fig. 1, B and C, and fig. S1). We next examined the ability of NS1 to inhibit responses to Semliki Forest virus (SFV), which, similar to EMCV, generates high levels of dsRNA (16). Cells infected with a recombinant SFV encoding NS1 (SFV-NS1) expressed NS1 protein (fig. S3), but produced levels of IFN-α comparable to cells infected with a control recombinant virus or wild-type SFV (Fig. 1D). Similarly, transfection with NS1 had no effect on the induction of an IFN-β reporter in response to EMCVor SFV, although it potently inhibited the response to ΔNS1 influenza or Sendai virus (SeV) (Fig. 1E) (13). Notably, the latter two viruses generate minimal levels of dsRNA (Fig. 1, B and C) (15), but induce high levels of IFN-α (Fig. 1, A and E, and fig. S4), whereas EMCV and SFV induce high levels of dsRNA (Fig. 1, B and C, and fig. S3) (16), but lower levels of IFN-α (Fig. 1E and fig. S4). Collectively, these data indicate that neither IFN-α induction nor the inhibitory effect of NS1 correlates with the presence of dsRNA.

Fig. 1.

Neither NS1 inhibition nor the induction of IFN-α/β correlates with viral dsRNA. (A) BM-DC from C57BL/6 (wt) or myd88–/– mice were cultured overnight in medium alone (mock) or with wild-type or ΔNS1 influenza virus. Data are average IFN-α levels of triplicate samples. (B and C) dsRNA in Vero cells 6 hours after infection with influenza virus or EMCV or after transfection with poly I:C measured by enzyme-linked immunosorbent assay (B) or flow cytometry (C). Numbers in (C) indicate the percentage of cells in gate. SSC, side scatter. (D) Same as (A), but cells were infected with wild-type SFV or recombinant SFV encoding NS1 or an irrelevant protein [ovalbumin (OVA)]. (E) Induction of luciferase activity in human embryonic kidney (HEK) 293 cells cotransfected with IFN-β reporter plasmids together with NS1-encoding plasmid or control empty vector and subsequently infected with the indicated viruses. Error bars in (A), (B), and (D) show means + SD. n.d., none detectable.

SeV and ΔNS1 influenza are recognized by RIG-I, whereas EMCV is recognized by MDA5 (6, 7). Therefore, we investigated whether the virus-specific effects of NS1 reflected its ability to interact with RIG-I. Consistent with this possibility, RIG-I tagged with green fluorescent protein (GFP) or a hemagglutinin peptide (HA) coprecipitated with NS1 from postnuclear lysates of influenza-infected cells (Fig. 2, A and B). In addition, the cytoplasmic fraction of NS1 colocalized with GFP–RIG-I in infected cells (Fig. 2C). In contrast, MDA5 did not associate with NS1 in infected cells (Fig. 2B). These results suggest that NS1 selectively targets RIG-I rather than dsRNA during influenza virus infection.

Fig. 2.

The NS1 protein of influenza A virus interacts with RIG-I but not MDA5. (A and B) 293T cells were transfected with pGFP–RIG-I (A) or with pHA–RIG-I or pHA-MDA5 (B), and 12 hours later they were infected or not infected with influenza virus, as indicated. At 24 hours, cells were lysed and analyzed by Western blot (WB) for the presence of NS1 and GFP (A) or NS1 and HA (B) in total cell lysates (lower panels) or after immunoprecipitation (IP) with an antibody to NS1 (upper panels). (C) 293T cells were transfected with GFP–RIG-I, infected with influenza virus at 16 hours, and stained for NS1 at 24 hours. Shown are GFP–RIG-I, NS1, and the merged image.

Given the lack of dsRNA in infected cells, we addressed whether RIG-I might be activated directly by the influenza ssRNA genome. Transfection with genomic RNA extracted from influenza virions (flu vRNA) induced potent activation of the IFN-β reporter (Fig. 3A) and production of IFN-α and interleukin (IL)–6 by BM-DC at levels comparable or superior to those obtained by transfection with poly I:C (Fig. 3C). This was not due to the generation of progeny virus because flu vRNA transfection did not result in viral replication (Fig. 3B), consistent with the evidence that vRNA from negative strand viruses is not infectious (17). The response to flu vRNAwas RIG-I dependent, given that it was inhibited by dominant negative RIG-I or by small interfering RNA (siRNA)–mediated knockdown of RIG-I in either mouse or human cells (Fig. 3D and fig. S5). RIG-I blocked responses to SeV but not EMCV (Fig. 3D), as expected (6), confirming the specificity of RIG-I reduction. NS1 suppressed the response to flu vRNA, and this was partly relieved by two point mutations previously reported to attenuate NS1 binding to RNA (18) (mutNS1, Fig. 3A). Finally, the addition of purified RIG-I to flu vRNA led to the formation of complexes with high molecular weight (Fig. 3E), demonstrating that RIG-I directly binds the influenza genome. Thus, RIG-I recognizes influenza ssRNA genomes and signals for cytokine production unless suppressed by NS1.

Fig. 3.

Influenza vRNA triggers innate responses in aRIG-I–dependent manner. (A) HEK293 cells were cotransfected with IFN-β reporter plasmids and NS1 expression plasmid (NS1 or mutNS1) or empty vector (control). After 24 hours, cells were transfected with flu vRNA (0.2, 0.04, or 0.008 μg) and luciferase activity was measured at 38 hours. Western blots show the presence of NS1 or tubulin. (B) Lysates from cells transfected with flu vRNA or control mouse mRNA were probed with polyclonal antibody against influenza proteins. Influenza-infected cells served as positive (pos) control. (C) IFN-α and IL-6 accumulation in overnight culture supernatants from BM-DC that were mock treated or transfected with flu vRNA (1 and 0.2 μg) or poly I:C (0.5 μg). n.d., none detectable. Error bars show means + SD. (D) Inhibition of responses to flu vRNA by RIG-I reduction. Human HEK293 cells and mouse NIH 3T3 cells were cotransfected with IFN-β reporter plasmids and siRNAs specific for mouse or human RIG-I. After 72 hours, cells were transfected with flu vRNA (0.2 or 0.04 μg) or were infected with SeV or EMCV. Luciferase activity was measured at 86 hours. (E) Cold electrophoretic mobility shift assay analysis of RIG-I and flu vRNA interaction. Flu vRNA (1.08 nM) was incubated with the indicated concentrations of purified FLAG–RIG-I or bovine serum albumin (BSA) and resolved by electrophoresis. RNA was visualized by SYBR Green staining.

Mouse mRNA, total mammalian RNA [consisting of ∼70% ribosomal RNA (rRNA)], and mammalian or bacterial tRNA did not elicit IFN responses (fig. S6), suggesting that RIG-I recognition is specific for viral RNA. Influenza vRNA is uncapped (17), and phosphorylated 5′ termini present in siRNA and ssRNAs generated by in vitro transcription have been reported to induce IFN-α/β when transfected into cells (19). We confirmed the latter observation (fig. S7) and tested whether flu vRNA recognition similarly depends on the presence of 5′-phosphates. Treatment with calf intestinal phosphatase (CIP) completely abrogated the stimulatory properties of flu vRNA (Fig. 4, A and B). This was due to phosphatase activity rather than nonspecific effects of CIP, given that it could be blocked by inorganic phosphate or EDTA (fig. S8). Furthermore, CIP did not affect the ability of vRNA to stimulate TLR7-dependent IFN-α production from pDC-containing cell populations (fig. S8). Vesicular stomatitis virus (VSV) also has an uncapped genome (17) and is recognized by RIG-I (6). Similar to influenza, transfection with VSV vRNA induced an IFN response, which was completely abrogated by previous CIP treatment (Fig. 4, C and D). Consistent with the evidence that EMCV is not recognized by RIG-I (6, 7), EMCV vRNA failed to induce a response when transfected into cells at amounts comparable to flu or VSV vRNA (Fig. 4C). These data suggest that cells use RIG-I to respond to the phosphorylated 5′ termini of uncapped ssRNA viral genomes.

Fig. 4.

ssRNAs containing phosphorylated 5′ ends bind RIG-I and activate antiviral responses. HEK293 cells transfected with IFN-β reporter and Renilla luciferase control plasmids (A and C) or BM-DC (B and D) were transfected with different amounts (0.6, 0.2, 0.06, or 0.02 μg) of mouse spleen mRNA or flu vRNAs, or with vRNA from 8 × 107 plaque-forming units VSV (C) or fivefold serial dilutions thereof (D). RNAs were pretreated with CIP or not (NoCIP). Luciferase activity [(A) and (C)] or IFN-α and IL-6 [(B) and (D)] were measured after overnight culture. (E) In vitro transcribed biotinylated RNA was treated with or without CIP, bound to streptavidin beads, and incubated with lysates from 293T cells transfected with GFP–RIG-I. Data show protein eluted from the beads by washing with buffer containing the specified NaCl concentrations. (F) Lysates of 293T cells cotransfected with pFLAG–RIG-I and pCAAGS-NS1 or pCAAGS-mutNS1 were incubated with or without 7SK-as RNA that had been either mock or CIP treated. After immunoprecipitation (IP) with the antibody to NS1, the presence of RIG-I and NS1 was analyzed by Western blot (WB).

Finally, we assessed whether 5′ phosphorylation contributes to RIG-I binding. In vitro transcribed RNA formed a complex with RIG-I in cell lysates, which was less resistant to salt extraction when the RNA was pretreated with CIP (Fig. 4E). In vitro binding assays that used purified RIG-I confirmed that complexes with CIP-treated RNA were less stable (fig. S9). Notably, the addition of control but not CIP-treated RNA to cell lysates promoted the formation of a complex containing NS1 and RIG-I (Fig. 4F), mimicking the association seen in infected cells (Fig. 2). RIG-I associated only weakly with the NS1 mutant (mutNS1, Fig. 4F), which binds ssRNA poorly when compared with wild-type protein (fig. S10). Thus, RIG-I preferentially forms stable complexes with RNA that contains phosphorylated 5′ ends and NS1 is recruited to such complexes at least in part through its RNA binding domain.

The ability to sense viral presence is critical for initiating innate and adaptive immunity to viral infection. We found that virus recognition can be accomplished by RIG-I–mediated sensing of ssRNA viral genomes bearing 5′-phosphates. This can be blocked by viral antagonists such as the NS1 protein of influenza A virus, which is found in a complex with RIG-I (supporting online material text). Our results demonstrate that the repertoire of antiviral defense strategies includes the detection of cytoplasmic ssRNA, explaining how some viruses that produce little or no dsRNA (15) can be efficiently recognized, even before viral replication (20). Consistent with the evidence that RIG-I can bind to dsRNA in vitro (4, 5), our data do not exclude the possibility that the complex of RIG-I and ssRNA is stabilized by the presence of intramolecular double-stranded regions, such as the panhandle structures that are found at the ends of the influenza genome (17). However, this is not sufficient to induce RIG-I activation unless the ssRNAs also contain phosphorylated 5′ termini. Notably, many RNA viruses have uncapped RNAs bearing 5′-triphosphates. In picornaviruses, a notable exception, the vRNA is covalently linked to a small protein, VPg (17), perhaps explaining why EMCV cannot be recognized by RIG-I (6, 7). The 5′-phosphates are also absent from self-mRNA as a result of the addition of a 7-methyl-guanosine cap and may be largely inaccessible in rRNA and tRNA, through association with ribosomal proteins or formation of cloverleaf structures containing 3′ overhangs. Thus, the cytoplasmic presence of RNA containing accessible 5′-phosphates allows discrimination between self and viral RNA, indicating that, similar to dsRNA, 5′ phosphate–bearing ssRNA constitutes a viral “pathogen-associated molecular pattern” (21). This finding, added to the recent discovery of innate sensing of cytoplasmic DNA (2224), suggests a parallel between cytosolic and endosomal viral recognition, with MDA5, RIG-I, and the cytosolic DNA receptor constituting functional homologs of TLR3, TLR7, TLR8, and TLR9. Similar to virologists, the innate immune system may therefore have learned to classify viruses by their genomes.

Supporting Online Material

Materials and Methods

SOM Text

Figs S1 to S10


References and Notes

View Abstract

Navigate This Article