Long Double-Stranded RNA Induces an Antiviral Response Independent of IFN Regulatory Factor 3, IFN-β Promoter Stimulator 1, and IFN¹

Viral infection of susceptible cells leads to the induction of a cellular antiviral response aimed at limiting virus replication and spread. dsRNA, a byproduct in the replication cycle of virtually all viruses (1), is a potent inducer of innate and adaptive immune responses. Three different families of pattern recognition receptors, the TLRs, the retinoic acid-inducible gene I (RIG-I)³ -like receptors (RLRs) and the nucleotide oligomerization domain-like receptors (NLRs) bind dsRNA and initiate cellular signaling pathways (2). The TLRs and RLRs elicit type 1 IFN and cytokine production while NLRs promote IL-1β maturation through caspase 1 activation. In nonprofessional innate immune cells such as fibroblasts, TLR3 and the RLRs RIG-I and melanoma differentiation-associated gene 5 (MDA-5) bind dsRNA and mediate antiviral signaling through the adaptors Toll/IL-1R domain-containing adaptor inducing IFN-β (TRIF) and IFN-β promoter stimulator 1(IPS-1), respectively (2). Both adaptors subsequently activate the cellular transcription factors NF-κB and IFN regulatory factor (IRF) 3, leading to IFN-β and cytokine production. Similar to the cytoplasmic RLRs, protein kinase regulated by RNA (PKR) also binds dsRNA and activates IRF3 through IPS-1 (3). IFN-β signaling through the JAK-STAT pathway, which includes IRF9 as an essential component, leads to the induction of IRF7 and amplification of the cellular antiviral response through the induction of IFN-α species and IFN-stimulated genes (ISGs) (4).

Recent studies have begun to characterize the dsRNA-binding properties of TLR3, PKR, MDA-5, and RIG-I. Binding of dsRNA to TLR3 depends strongly on pH and dsRNA length, suggesting that TLR3 signaling initiates within an endosome through a pH-dependent binding mechanism and that sufficient dsRNA length (∼45 bp) is required to bind to both the N-terminal and C-terminal regions of the TLR3 ectodomain (5, 6, 7, 8). A similar minimal length dependence was observed for dsRNA to bind two PKR monomers and elicit PKR autophosphorylation and activation (9, 10). With the cytosolic RLRs RIG-I and MDA-5, both dsRNA origin and length influence recognition and binding. RIG-I preferentially recognizes RNA from a variety of RNA viruses while MDA-5 binds the dsRNA mimic polyinosinic:polycytidylic acid (poly (I:C)) and picornavirus RNA (11). Furthermore, RIG-I and MDA-5 were shown to preferentially bind to short and long dsRNA molecules, respectively (12). Regardless of which cellular protein senses the dsRNA, current models suggest that activation of all sensors leads to IRF3 activation and subsequent ISG and type 1 IFN induction.

The importance of IRF3 in the early and late phases of type 1 IFN expression and antiviral immunity was demonstrated through the generation of IRF3^−/− and IRF3^−/−9^−/− mice (4). Mice deficient for IRF3 were more vulnerable to viral infection and IFN production was reduced 20- to 50-fold. In the absence of both IRF3 and IRF9, type 1 IFN production was completely ablated (4). Since IRF7 gene expression is dependent on type 1 IFN signaling, levels of this transcription factor were significantly diminished in cells lacking IRF3 and absent in cells lacking IRF3 and IRF9 following viral infection. These studies led to a model in which constitutive expression of low levels of type 1 IFN in uninfected cells supports basal expression of equally low levels of IRF7. Upon viral infection, activation of IRF3 initiates a positive feedback loop resulting in robust IFN-β, IRF7, IFN-α, and ISG induction (13). Subsequent studies in IRF7^−/− and IRF3^−/−7^−/− mice confirmed the requirement for IRF3 and IRF7 in virus-mediated IFN-β and IFN-α production, respectively (14).

Although the prototypic response to viral infection and generation of dsRNA in nonprofessional innate immune cells is the activation of IRF3 and subsequent production of type 1 IFN, induction of ISGs or an antiviral response has been observed in the absence of either IRF3 or IFN production. Infection of IRF3^−/− mouse embryonic fibroblasts (MEFs) with Newcastle disease virus induced a panel of IRF3-independent direct response genes, including IFN-inducible p200 family proteins p203 and p204 and Mx2 (15). Although small interfering RNA-mediated knockdown of IRF3 in a human hepatoma cell line dampened ISG induction in response to poly(I:C) and Sendai virus infection, it had no effect on ISG56 and MxA induction in response to Sin Nombre virus particle treatment (16). Similarly, expression of a dominant negative form of IRF3 in 293 cells failed to impact on the ability of poly(I:C) to block hepatitis C virus replication (17). Finally, consistent with previous observations (4), IRF3^−/− mice were more vulnerable to infection with West Nile virus (WNv) than their wild-type (WT) counterparts (18), despite the ability of this virus to elicit type 1 IFN production in knockout animals (18, 19). Although these studies suggest that IRF3 is important, but not essential, for virus-mediated type 1 IFN production, IRF3 has been shown to play a critical role in the induction of ISGs and an antiviral response in the absence of IFN production (20, 21, 22). This activity stems from the ability of IRF3 to bind directly to the promoter region of a subset of ISGs (23).

Despite recent advancements in characterizing viral dsRNA-mediated antiviral immunity in nonprofessional innate immune cells, two issues remain outstanding. First, the biological consequence of differential dsRNA recognition remains unclear, particularly since downstream signaling pathways converge on IRF3. Second, although both IRF3 and type 1 IFN play central roles in mediating the antiviral response to dsRNA and can each function in the absence of the other, it is unclear whether dsRNA can induce antiviral activity in the absence of both IRF3 and type 1 IFN. In this study, we provide evidence that the innate antiviral response to viral dsRNA is length dependent in WT mouse fibroblasts and that the length dependence is exacerbated in the absence of IRF3. Furthermore, we provide evidence of an IRF3- and type 1 IFN- independent antiviral response that is more robustly activated by longer dsRNA molecules.

MEFs were derived from WT C57BL/6, IRF3^−/− (4), IRF3^−/− and IRF9^−/− (4), IPS-1^−/− and IPS-1^+/+ littermate control mice and were maintained in α-MEM supplemented with 10% FBS, 100 U · ml⁻¹ penicillin, 100 μg · ml⁻¹ streptomycin, and 2 mM l-glutamine. Experiments were performed with cells at passages three to eight. All cells were incubated at 37°C in a humidified 5% CO₂ incubator. Vesicular stomatitis virus (VSV) expressing GFP (VSVgfp; provided by B. Lichty, McMaster University, Hamilton, Ontario), HSV-1 (strain KOS) expressing GFP (HSV-1gfp) (24), and WNv (Kunjin subtype; provided by M. Diamond, Washington University, St. Louis, MO) were propagated on Vero cells (American Type Culture Collection).

dsRNA was synthesized by in vitro transcription using a Megascript RNAi kit (Ambion). One microgram of PCR fragments amplified from portions of the cloned WNv genome was used as a template (Table I). The dsRNA lengths of 200, 500, and 1000 bp (v200, v500, v1000) were derived from the E protein sequence and a 3000-bp dsRNA fragment was derived from the NS3-NS4B sequence (v3000). The average length for the poly(I:C) (GE Healthcare) used in this study was ∼4000 bp as determined by marker size comparison using agarose gel electrophoresis and a 1-kb Plus DNA ladder (Fermentas).

Cells were treated with dsRNA in equal molar amounts to ensure an equal number of molecules per dsRNA length, unless otherwise noted. dsRNA treatments were performed in serum-free OptiMEM medium (Life Technologies) in the presence of 50 μg/ml DEAE-dextran (Pharmacia) for 1 h followed by additional indicated amounts of time in full growth medium, unless otherwise noted. DEAE-dextran is a cationic polymer that binds negatively charged nucleic acids and enables a closer association between the negatively charged cell membrane and the nucleic acid of interest (25). In all experiments, DEAE-dextran was used in dsRNA-untreated controls to ensure that the polymer alone was not influencing subsequent cellular responses.

WT MEFs were seeded on glass coverslips to ∼60% confluency and allowed to attach overnight. Cells were then infected with Kunjin virus at a multiplicity of infection (MOI) of 10 for 1.5 h, after which the virus was removed and fresh medium was added to the wells. As a positive control, cells were treated with 100 μg · ml⁻¹ poly(I:C) for 8 h. At specific time points, cells were fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton X-100, and blocked in blocking buffer (3% BSA, 3% goat serum, and 0.02% Tween 20 in PBS). Coverslips were incubated with the mouse anti-dsRNA Ab J2 (English and Scientific Consulting) at a 1/200 dilution, followed by an Alexa Fluor 488-labeled anti-mouse secondary Ab, also at 1/200 (Molecular Probes). All Ab dilutions were performed in blocking buffer. Nuclei were stained with Hoechst stain. Images were captured using a Leica DM-IRE2 inverted microscope.

RNA was harvested from cells infected with Kunjin virus or treated with dsRNA as indicated using TRIzol reagent (Invitrogen). Twenty micrograms of RNA was electrophoresed in a 10% nondenaturing polyacrylamide gel and transferred onto a nylon membrane (Hybond N+; GE Healthcare) using a semidry transfer apparatus. The membranes were blocked with 5% milk in TBS and blotted with the J2 Ab followed by a goat anti-mouse IgG secondary Ab. The dsRNA was visualized by the ECL system. In vitro-transcribed RNA was electrophoresed on the same gel as a size comparison for samples.

RNA was isolated from dsRNA-treated cells using TRIzol reagent and 2.5 μg of RNA was DNase treated using a DNA-free kit (Ambion). RNA integrity and quantity were measured using the Agilent 2100 Bio-Analyzer (Agilent). Three hundred nanograms of high-quality total RNA was reverse transcribed using the RT² First Strand Kit (SABiosciences). The cDNA was applied to the Mouse IFNs and Receptors RT² Profiler PCR Array (SABiosciences) as per the manufacturer’s instructions. The arrays were run on the Applied Biosystems PRISM 7900HT Sequence Detection System using Sequence Detector Software version 2.2. Data were analyzed using the online SABiosiences RT² Profiler PCR array data analysis tool based on the ΔΔC_t method. Specifically, gene expression was normalized to five housekeeping genes (Gusb, Hprt, Hspcb, Gapdh, and β-actin) and expressed as fold change over the control group (cells treated with DEAE-dextran alone).

Three hundred nanograms of total RNA was reverse transcribed with 0.2 ng of random 6-mer primer and 50 U of Superscript II (Invitrogen) in a total reaction volume of 20 μl. Real-time quantitative PCR was performed in triplicate, in a total volume of 25 μl, using Universal PCR Master Mix and gene-specific TaqMan primers (Applied Biosystems). Data were analyzed using the ΔΔC_t method. Specifically, gene expression was normalized to the housekeeping gene (Gapdh) and expressed as fold change over the control group (cells treated with DEAE-dextran alone).

Cells were seeded in 12-well dishes and treated with serial dilutions of nM quantities of dsRNA for specified lengths of time. Cells were subsequently infected with 0.1 PFU/cell VSVgfp for 1 h in serum-free medium. This amount of virus was determined to be the maximal dose for which signal saturation in untreated cells did not occur. The viral inoculum was then removed and replaced with DMEM containing 1% methylcellulose. GFP fluorescence intensity was measured 24 h later on a Typhoon Trio (GE Healthcare) and quantified using ImageQuant TL software. A dose-response curve was generated for each dsRNA molecule and an EC₅₀ value was calculated using GraphPad PRISM software. No differences in VSVgfp infection rates in untreated cells were observed (supplemental Fig. 1).⁴

Data are expressed as means ± SEM (unless otherwise indicated). Statistical analysis was performed using a one-way ANOVA with Tukey’s post hoc test for pairwise comparisons, a Dunnett’s post test for comparisons with control treatments, and an unpaired t test in the case of a comparison between two values. All statistical analyses were performed using GraphPad InStat. A value of p < 0.05 was considered statistically significant.

Although dsRNA was thought to be produced in virtually all virally infected cells, a study using a dsRNA-specific Ab suggested that dsRNA can be detected following infection with positive-strand RNA, dsRNA, or DNA viruses but not with negative-strand RNA viruses (26). However, a recent study showed that VSV, a negative-strand RNA virus, is capable of detectable dsRNA production (12), suggesting that viruses of all classification likely produce detectable amounts of dsRNA of various lengths. WNv (strain Kunjin) is a positive-strand RNA virus that has been shown by immunofluorescence microscopy to produce dsRNA following infection of Vero monkey kidney epithelial cells; however, the length of the dsRNA produced was not determined (27). In the present study, abundant dsRNA was also detected by immunofluorescence microscopy in Kunjin virus-infected WT MEFs by 8 h after infection using the J2 anti-dsRNA Ab (Fig. 1,A). As a control, poly(I:C) was used. Furthermore, in both MEF and Vero cells, dsRNA of high molecular weight (>3000 bp) could be detected in extracts as late as 48 h after infection by immunoblot analysis using the same Ab (Fig. 1 B). These data confirm the production of long, stable dsRNA molecules in WNv-infected cells.

To assess the biological consequence of dsRNA length, dsRNA molecules of specific lengths (200, 500, 1000, and 3000 bp) were in vitro-transcribed from cloned fragments of the WNv genome. These particular lengths were chosen because similar-sized dsRNA molecules have been observed in virally infected cells, such as those infected with reovirus, VSV, encephalomyocarditis virus (12), and WNv, as determined from the present study. Discrete bands of the appropriate size were detected on a 1% agarose gel (Fig. 2,A). The 3000 bp molecule consistently separated into two fragments. In contrast, poly(I:C) ranged in length from 500 to 5000 bp, with the average length determined to be ∼4000 bp. To ensure that the in vitro transcribed dsRNA molecules were stable within treated cells, similar to that observed with viral infection, immunoblot analysis using the J2 Ab was performed on extracts harvested 6 h after dsRNA treatment. Intracellular dsRNA was relatively stable, particularly the 200, 500, and 1000 bp species (Fig. 2 B). Although degradation of the 3000 bp species was evident at this time point, the predominant species observed were ∼2000–3000 bp. Thus, in vitro-transcribed dsRNA is relatively stable within treated cells, over a range of dsRNA lengths. Furthermore, dsRNA of all lengths entered into cells with equal efficiency (supplemental Fig. 2 and data not shown) and by 1 h after treatment, all cell-associated dsRNA had internalized (data not shown).

Recent evidence suggests that recognition of dsRNA by PKR, TLR3, RIG-I, and MDA-5 is length dependent (9, 12, 28). However, in fibroblasts, signaling pathways activated by these proteins converge on IRF3 to mediate IFN and ISG induction. Little is understood regarding the biological significance of differential recognition of dsRNA molecules, particularly given that recognition of dsRNA by these four proteins leads to IRF3 activation. To specifically address this issue, we used equal molar amounts of dsRNA of various lengths to ensure that the only variable was the dsRNA length and not the number of dsRNA molecules. However, when equal concentrations (by weight) of dsRNA were tested, length- dependent gene induction as described below was still observed (data not shown). WT MEFs were treated with 1.5 nM poly(I:C) or dsRNA of 200, 1000, or 3000 bp lengths. This amount of dsRNA was determined to elicit a complete antiviral response in WT MEFs (see Fig. 4). RNA harvested 6 h after treatment was reverse transcribed and the cDNAs were assayed using mouse IFNs and receptors RT² Profiler PCR Arrays (supplemental Fig. 3). In addition to housekeeping controls, this array contains 84 genes whose expression is controlled by or involved in signal transduction mediated by IFN ligands and receptors. Included on the array are 21 IFN and cytokine species, 35 IFN and cytokine receptors, 8 IRFs, 2 IRF-binding proteins, and 18 ISGs. Shown in Fig. 3 are ISGs whose expression profile changed with treatment (with Adar serving as a representative of an ISG whose expression profile did not change), all IRFs, all type 1 IFN species, and cytokines whose expression profile changed with treatment. No receptors are reported since none showed changes in expression profile with treatment. A length- dependent induction of a subset of ISGs, IRFs, and IFNs was observed, with dsRNA of the 200 bp length eliciting a modest induction and poly(I:C) demonstrating the most robust induction (Fig. 3,A). When IRF3^−/− MEFs were treated with 3 nM poly(I:C) or dsRNA of various lengths (the concentration of dsRNA required to elicit a complete antiviral response in IRF3^−/− MEFs; Fig. 4), the same subset of ISG, IRF, and IFN transcripts were induced (Fig. 3 B), also in a length-dependent fashion. However, dsRNA of the 200 bp length was uniformly diminished in its capacity to stimulate transcription of most genes represented on the array, while treatment with poly(I:C) and dsRNA of the 3000 bp length dsRNA stimulated gene transcription to a similar capacity. Of note, IRF7 and IFN-β levels remained elevated in IRF3^−/− MEFs, consistent with previous observations (4, 13).

When IRF3^−/−9^−/− MEFs were treated with 8.5 nM poly(I:C) or dsRNA of various lengths (the highest concentration of dsRNA possible under these experimental conditions), ISG, IRF, or IFN transcript accumulation was not observed, even after 24 h of treatment (Fig. 3,C). The levels of housekeeping genes in mock-treated cells were similar in all cell types (Fig. 3,D), confirming the integrity of the RNA. Furthermore, the levels of the remaining IRF family members were similar in all cell types (Fig. 3 E). These data suggest that although IRF3 is not essential for dsRNA-mediated ISG, IRF, and IFN induction per se, there appears to be a length-dependent requirement for IRF3, particularly with smaller lengths of dsRNA. Furthermore, IFN signaling (via IRF9-associated pathways) is required for dsRNA-mediated ISG, IRF, and IFN induction in the absence of IRF3.

To confirm that the expression of ISGs, IRFs, and IFNs was more dependent on IRF3 following stimulation with short dsRNA molecules as opposed to long dsRNA molecules, we performed quantitative RT-PCR analysis of Cxcl10, Ifit1, IRF7, and IFN-β1 in WT and IRF3^−/− MEFs following treatment with short (200 bp) and long (3000 bp) dsRNA molecules. As shown in Fig. 4,A, induction of all four transcripts was significantly reduced in IRF3^−/− MEFs relative to WT MEFs following stimulation with short dsRNA molecules. In contrast, a similar level of induction was observed following stimulation with long dsRNA molecules in both cell types (Fig. 4 B).

To confirm the dsRNA length-dependent nature of ISG and IFN induction and to determine the biological outcome in the presence and absence of IRF3, we performed standard antiviral assays to ascertain the effective concentration of dsRNA that prevents viral replication by 50% (EC₅₀). WT and IRF3^−/− MEFs were treated with serially diluted nM concentrations of in vitro-transcribed dsRNA of specific lengths or poly(I:C) for 6 h, followed by challenge with VSVgfp under the viral promoter. GFP fluorescence was monitored 24 h after infection and the EC₅₀ values were determined. VSVgfp replication was similar in untreated WT and IRF3^−/− cells, indicating that these cells are equally susceptible to VSVgfp infection (supplemental Fig. 1). A statistically significant difference in the EC₅₀ values for dsRNA molecules of different lengths was observed in both WT and IRF3^−/− MEFs (Fig. 5, A and B). A length-dependent antiviral response was also observed in both WT and IRF3^−/− MEFs when induced by equal weights of dsRNA (data not shown). When the EC₅₀ values for a given length of dsRNA were compared between WT and IRF3^−/− MEFs, there was a statistically significant difference in the ability of dsRNA of 200 and 500 bp to elicit an antiviral response, whereas dsRNA of 1000 and 3000 bp and poly(I:C) elicited a similar antiviral response in the two cell types (Fig. 5 C). Furthermore, poly(I:C) was able to limit the replication of HSV-1, a large DNA virus, indicating that this antiviral response is not unique to VSV (supplemental Fig. 4). These data support what was observed at the transcript level; the antiviral response to dsRNA is length dependent and the antiviral response to shorter dsRNA molecules is more reliant on IRF3, whereas the antiviral response to larger dsRNA molecules is predominantly IRF3 independent.

We failed to detect significant IRF, ISG, or IFN induction in response to dsRNA or poly(I:C) in IRF3^−/−9^−/− MEFs (Fig. 3,C), consistent with previous reports (4, 13). To determine whether these cells were capable of eliciting an antiviral response in the absence of IFN or ISG induction, we performed a standard antiviral assay using VSVgfp following treatment with dsRNA and poly(I:C). Although we were unable to determine EC₅₀ values for each length of dsRNA, given that a complete antiviral response was not observed at the highest concentration of dsRNA permissible, we did observe a significant reduction in the ability of VSVgfp to replicate in cells treated for 24 h (Fig. 6). No significant antiviral activity was observed following a 6-h treatment (data not shown). At dsRNA lengths of 200, 500, and 1000 bp, VSVgfp replication was reduced by ∼60% (56.7 ± 2.2%, 58.0 ± 1.7%, and 60.8 ± 1.1%, respectively), whereas dsRNA of the 3000 bp length and poly(I:C) reduced VSVgfp replication by ∼70 and 90% (67.4 ± 3.4% and 85.7 ± 1.3%, respectively), respectively. A statistically significant difference was observed with the latter two treatments. These data suggest that dsRNA can elicit an antiviral response in the absence of the type I IFN system, including IRF3. As observed in IRF3^−/− MEFs, treatment with poly(I:C) significantly inhibited the replication of HSV-1 (supplemental Fig. 4). Since dsRNA also activates NF-κB and elicits the induction of cytokines and chemokines, we assessed induction of cytokines and chemokines using pathway-specific PCR arrays. However, we did not detect significant induction of any genes under conditions that elicit an antiviral response (data not shown). The nature of this cellular response is unknown, but is currently under investigation. Although dsRNA length dependence was observed, this dependence was not as robust as observed for the cellular antiviral response mediated by IRF3 and type 1 IFNs.

Previous studies in MEFs harvested from IPS-1 null mice demonstrated that IPS-1 is essential for IFN-β production following treatment with virus or dsRNA (29). The lack of IFN-β production was attributed to impaired IRF3 and NF-κB activation in these cells. Given that neither IRF3 nor type 1 IFN are necessary for cellular antiviral activity, particularly in response to long dsRNA molecules, we tested whether IPS-1 is necessary under similar conditions. Similar to results observed in IRF3^−/−9^−/− MEFs, dsRNA of any length failed to elicit a significant antiviral response following a 6-h treatment (data not shown). However, following a 24-h treatment, dsRNA of 3000 bp and poly(I:C) induced an antiviral response with EC₅₀ values of 0.51 nM ± 0.16 and 0.23 nM ± 0.04, respectively; Fig. 7). Short dsRNA molecules of 200 bp lengths failed to induce an antiviral response in the absence of IPS-1. Conversely, dsRNA of 200 and 3000 bp lengths and poly(I:C) all induced an antiviral response in IPS-1^+/+ littermate controls with EC₅₀ values of 0.64 ± 0.23 nM, 0.22 ± 0.05 nM, and 0.17 ± 0.04 nM, respectively (Fig. 7). dsRNA entry was similar between IPS-1^+/+ and IPS-1^−/− MEFs (data not shown) as was VSVgfp infectivity in untreated cells (supplemental Fig. 1).

It is well established that type 1 IFNs produced early following virus infection play a critical role in the immune defense against most viruses by limiting virus replication and spread (30). Although various structural components of a virus can elicit type 1 IFN production, the most potent viral inducer is dsRNA, particularly in nonprofessional innate immune cells such as fibroblasts. Although it is recognized that the majority of viral dsRNA is bound by viral proteins, suggesting that little “free” viral dsRNA exists within a cell, the characterization of multiple cellular dsRNA-binding proteins predicts that either entire dsRNA molecules, or portions of these molecules, are available for recognition and binding by cellular proteins. At least four cellular proteins recognize viral dsRNA and mediate type 1 IFN production: TLR3, PKR, RIG-I, and MDA-5. Although these proteins preferentially bind dsRNA based on its cellular localization, viral origin, and/or length, the outcome of dsRNA binding is signaling through the adaptors TRIF (TLR3) or IPS-1 (PKR, RIG-I, and MDA-5) to activate IRF3. Activated IRF3 can induce ISGs directly, in the absence of IFN production, or cooperate with additional transcription factors such as IRF1, IRF7, and NF-κB to elicit type 1 IFN production.

Although length plays a role in determining which cellular proteins preferentially bind a given dsRNA molecule, the biological significance of dsRNA length is unclear, particularly given that IRF3 activation ensues regardless of the dsRNA-binding protein. To investigate this issue, we produced dsRNA of various lengths from the WNv genome, as the antiviral response to WNv involves TLR3, PKR, RIG-I, and MDA-5 (31, 32, 33) and WNv produces stable dsRNA in fibroblasts following infection, thus confirming the biological relevance of dsRNA species from this viral origin. In this study, we provide evidence that dsRNA induces a length-dependent antiviral response, with long dsRNA molecules inducing significantly higher levels of ISG, IRF, and IFN transcripts than short dsRNA molecules when equal nM amounts were compared. The length dependence was confirmed upon determination of EC₅₀ values in a standard antiviral assay.

Previous studies have shown that activation of TLR3 and PKR requires a minimal length of dsRNA, either to span multiple dsRNA binding sites within a single TLR3 molecule (5, 6, 7, 8) or to elicit dimerization of two PKR molecules (9, 10). Although activation of RIG-I and MDA-5 following dsRNA binding does not appear to have similar requirements, it has been established that MDA-5 preferentially binds long dsRNA molecules and poly(I:C) (11, 12). Accordingly, a simple explanation for the data presented in this manuscript is that MEFs express elevated levels of MDA-5 relative to RIG-I and thus long dsRNA molecules induce a greater biological response. However, untreated MEFs express low to undetectable levels of both MDA-5 and RIG-I, but rapidly up-regulate the expression of both ISGs following virus or dsRNA treatment (34), similar to what is seen with PKR. Another possible explanation is that MDA-5 binds dsRNA with a higher affinity than RIG-I. To date, comprehensive studies to specifically address this possibility have not been performed. An alternative explanation is that longer dsRNA molecules facilitate better multimerization of MDA-5 compared with shorter dsRNA lengths with RIG-I. There is a general precedence with cellular signal transduction that multimerization of pathway components leads to enhanced signal transduction and downstream biological activity. Regardless, the focus of the current study was not to determine which dsRNA-binding protein(s) mediates an antiviral response, but rather to delineate the biological outcome of differential dsRNA binding, particularly given that all characterized pathways converge on IRF3. Indeed, it is likely that per dsRNA molecule, long dsRNA molecules have an increased capacity to activate multiple dsRNA-binding proteins (of the same or different species), thus potentiating the downstream biological response.

Of particular interest, the dsRNA length dependence was exacerbated in the absence of IRF3. When comparing either individual transcript induction or EC₅₀ values for a given length of dsRNA, a significant difference was observed between the ability of short, but not long, dsRNA molecules to induce ISGs, IRFs, and IFNs or to protect WT vs IRF3^−/− MEFs from viral infection. These results confirm that IRF3 is not essential per se for ISG and IFN production (4, 13), but suggest that IRF3 plays a more critical role in response to recognition of short dsRNA molecules. Long dsRNA molecules (3000 bp and poly(I:C)) were equally capable of preventing virus replication in the presence and absence of IRF3. These data suggest that either a novel transcription factor with a similar activity to IRF3 exists or that in the absence of IRF3, another member of the IRF family can compensate for the loss of IRF3. Recent biochemical and structural studies of the IFN-β promoter show that IRF binding is critical for cooperative association of the IFN-β enhanceosome components and IFN-β transcription (35, 36). Although IRF1, IRF5, and IRF7 have been implicated in type 1 IFN production, it is unlikely that these IRFs are responsible for the effects seen in IRF3^−/− MEFs. We failed to detect differences in the basal expression of the IRF mRNA species in the various MEFs, and previous studies showed that ectopic expression of IRF1 fails to restore type 1 IFN induction in the absence of IRF3 (4) and that IRF5 activation does not occur in response to treatment with poly(I:C) (37). Furthermore, IRF7 expression in fibroblasts is dependent on type 1 IFN signaling (Ref. 4 and Fig. 3) and IRF7 predominantly activates IFN-α promoters, while robust IFN-β expression was observed in the absence of IRF3 following treatment with dsRNA or poly(I:C). Little is known regarding alternative pathways of dsRNA-mediated type 1 IFN induction involving proteins other than IRFs. Although NLRs have been shown to bind to dsRNA (38) and are important in mediating antiviral immune responses (39), activation of type 1 IFN has not been observed. Furthermore, it is possible that pattern recognition receptors capable of binding dsRNA that signal through alternative transcription factors exist. Thus, it remains to be elucidated how dsRNA, particularly large species, mediate type 1 IFN production in the absence of IRF3.

A further surprising observation from these data is the ability of dsRNA to control virus replication in the absence of IPS-1, IRF3, and type 1 IFN production. Although IRF3-dependent, IFN-independent as well as IRF3-independent, IFN-dependent antiviral responses have been observed, to our knowledge this is the first observation of antiviral activity mediated by dsRNA in fibroblasts under conditions where we fail to detect induction of IRFs, IFNs, or ISGs. Furthermore, we failed to detect induction of cytokines and chemokines under similar conditions. Although we were unable to elicit a complete antiviral response in IRF3^−/−9^−/− MEFs at the highest concentration of dsRNA permissible within our experimental system (8.5 nM), we observed a reduction in virus replication ranging from ∼60 to 90%, depending on dsRNA length. Consistent with data generated in IRF3^−/− MEFs, the antiviral response was more robust following treatment with long dsRNA molecules. This antiviral response reduced virus replication of both a RNA virus (VSV) and a DNA virus (HSV-1), albeit with different efficiency. Although a 6-h treatment with dsRNA was sufficient to significantly inhibit HSV-1 replication, a 24-h treatment was required to significantly inhibit VSV replication. These data highlight the complex nature of virus-host interactions.

A similar reduction in virus replication was made in MEFs deficient for the mitochondria-associated adaptor IPS-1. Although previous studies have shown that IPS-1 is important for activation of IRF3 and NF-κB and the subsequent induction of IFN-β, we observed that treatment with long dsRNA molecules was able to induce an antiviral response capable of controlling VSV infection in the absence of IPS-1. In both IRF3^−/−9^−/− and IPS-1^−/− MEFs, however, an extended treatment time (24 h) was required to elicit a biological response. These data suggest that the cellular pathway(s) that functions independently of IPS-1, IRF3, and type 1 IFN preferentially respond to long dsRNA molecules and require sufficient stimulation to accumulate and/or activate constituent signaling components. It is not clear at this time whether a novel dsRNA-binding protein participates in this antiviral response or whether the known dsRNA proteins signal through alternative pathways to block virus replication. Studies are underway to address this issue and determine the cellular genes that are induced by dsRNA under these conditions. Overall, these findings further characterize the requirement (or lack thereof) of IRF3 and the type 1 IFN system in the innate antiviral immune response following recognition of dsRNA.

We thank A. Thompson and D. Cummings for technical assistance and B. Lichty and M. Diamond for reagents.

The authors have no financial conflict of interest.