NUDT21 Links Mitochondrial IPS-1 to RLR-Containing Stress Granules and Activates Host Antiviral Defense

The innate immune system provides the first line of defense against viral infection. The initial step of this defense is detection of nonself cues known as pathogen-associated molecular patterns by specialized sensors, known as pattern-recognition receptors (PRRs), in host cells. Recognition of viral pathogen-associated molecular patterns by PRRs results in the activation of a series of mechanisms to combat viral propagation. In vertebrates, activation of PRRs induces the production of type I IFNs such as IFN-α and IFN-β and the subsequent expression of hundreds of IFN-stimulated genes that play a major role in restriction of viral replication within infected cells (1–3).

Retinoic acid–inducible protein–I (RIG-I)–like receptors are a family of PRRs consisting of DEAD box–containing RNA helicases that recognize viral RNA in the cytoplasm. Among RIG-I–like receptors (RLRs), RIG-I recognizes RNA molecules containing a 5′-triphosphate group as well as relatively short (<100 bp) dsRNAs, whereas melanoma differentiation–associated gene 5 (MDA5) recognizes relatively long (>2 kb) dsRNAs, with both types of dsRNA being derived from a wide range of RNA viruses (4–6). The binding of RLRs to such viral RNAs triggers their interaction with a key antiviral hub protein, IFN-β promoter stimulator–1 (IPS-1; also known as MAVS, CARDIF, and VISA) (7–10). IPS-1 is anchored to the mitochondrial outer membrane, and activated IPS-1 forms large prion-like aggregates on these organelles (11) that in turn activate transcription factors such as IFN regulatory factor 3 (IRF3), NF-κB, and activator protein–1, resulting in the expression of type I IFNs (8, 12–16). The pivotal role of IPS-1 in antiviral responses is exemplified by the finding that IPS-1–deficient mice are more vulnerable to viral infection than are wild-type (WT) mice (17, 18). The N-terminal caspase activation and recruitment domain (CARD) of RLRs is required for the interaction with IPS-1 and was indeed found to be essential for IFN induction (19).

Viral RNAs, RLRs, and RLR-associated proteins have been found to be localized to stress granules (SGs) in virus-infected cells. SGs are membraneless structures that are composed of translation-stalled mRNAs and proteins in which translation is generally inhibited (20). They are formed in response to various cellular stresses including viral infection. Viral RNA–mediated activation of dsRNA-activated protein kinase (PKR) results in the phosphorylation of eukaryotic initiation factor 2a, translational arrest, and SG nucleation (21). The RNA-binding proteins Ras-GAP SH3 domain–binding protein 1 (G3BP1), T cell–restricted intracellular Ag 1 (TIA1), and TIA1-related protein (TIAR) participate in the formation of SGs (22, 23), which is thought to involve liquid–liquid phase separation (24). SGs have been proposed to serve as a platform for RLR recognition of viral RNA and consequent activation of antiviral responses, with such SGs also having been termed antiviral SGs (25). Inhibition of SG formation by depletion of G3BP1 indeed suppressed type I IFN expression in response to infection with influenza A virus or Newcastle disease virus (NDV) (25, 26). The fact that many viral factors interfere with SG formation (27–31) also supports an antiviral function of SGs.

The mitochondrial localization of IPS-1 has been thought to be essential for its function. For instance, deletion of the C-terminal transmembrane (TM) domain of IPS-1 abrogated both its mitochondrial localization and the induction of type I IFN (8, 32). Caspase-mediated cleavage of IPS-1 that results in detachment of the TM domain also inactivates IPS-1 function (33, 34). Given the mitochondrial localization of IPS-1, it has remained unclear how RLRs within SGs encounter and activate IPS-1. Of interest in this regard, a fraction of IPS-1 appears to colocalize with TIAR, a marker of SGs, in cells infected with viruses or transfected with polyinosinic–polycytidylic acid [poly(I:C)], a synthetic analogue of viral dsRNA (26, 35), suggesting that viral RNA may induce the localization of IPS-1 to SGs and its association with RLRs. The mechanism that might underlie such an effect has remained unknown, however.

We, in this study, identify nucleoside diphosphate–linked moiety X (Nudix)–type motif 21 as an IPS-1 interactor. Nudix-type motif 21 (NUDT21) is an RNA-binding protein that constitutes the CFIm complex in nucleus and regulates the choice of polyadenylation site by binding to the 5′-UGUA-3′ element in the 3′ untranslated region of transcripts (36–39). NUDT21-mediated alternative polyadenylation has been shown to influence various cell fate decision processes as well as tumorigenesis (40, 41). Virus infection was previously shown to induce genome-wide changes in polyadenylation site selection in host transcripts and was also reported to limit a host gene expression through NUDT21 (42, 43). Unexpectedly, we found that whereas NUDT21 is localized mostly to the nucleus, where alternative polyadenylation takes place, a fraction of NUDT21 is associated with mitochondria in resting cells but also localizes to SGs in response to poly(I:C) transfection. Moreover, NUDT21 was found to associate with IPS-1 and to play an important role in its localization to SGs as well as in the efficient induction of IFN expression in response to poly(I:C) transfection or to infection with encephalomyocarditis virus (EMCV). Our results thus suggest an unexpected role for NUDT21 in the recruitment of mitochondrial IPS-1 to SGs and in the consequent promotion of RLR-mediated activation of IPS-1 and IFN induction in virus-infected cells.

HeLa S3, HEK293T, and RAW264.7 cells were maintained in DMEM supplemented with 10% FBS and 1% penicillin–streptomycin. HEK293T cells were transfected with the use of the GeneJuice Transfection Reagent (Merck Millipore), whereas transfection of HeLa S3 cells with expression vectors or poly(I:C) was performed with Lipofectamine 2000 (Thermo Fisher Scientific). RAW264.7 cells were transfected with poly(I:C) with the use of Lipofectamine LTX Reagent with PLUS Reagent (Thermo Fisher Scientific).

The plasmids pEF–BOS–FLAG–IPS-1–encoding WT human IPS-1 and pEF–BOS–FLAG–RIG-I–encoding WT human RIG-I were kindly provided by M. Yoneyama (Division of Molecular Immunology, Medical Mycology Research Center, Chiba University, Chiba, Japan). A full-length cDNA for NUDT21 was amplified by PCR from a mouse cDNA library with the primers 5′-GGCAGATCTATGTCTGTGGTGCCGCCCAA-3′ and 5′-GGCGAATTCTCAGTTGTATATAAAATTGA-3′ (sense and antisense, respectively) and was subcloned into either the BglII and EcoRI sites of the pCS4 vector or the BamHI and EcoRI sites of pcDNA3.1. Human G3BP1 cDNAs with or without stop codon were amplified by PCR from pN1/G3BP1–near-infrared fluorescent protein (iRFP) (Okada laboratory plasmids, no. 129339; Addgene) with the sense primer 5′-GCCAGATCTATGGTGATGGAGAAGCCTAG-3′ and the antisense primers 5′-GCCGAATTCGGATCCTTACTGCCGTGGCGC-3′ or 5′-GCCGAATTCGGATCCCTGCCGTGGCGCAAG-3′, respectively, and were subcloned into the BamHI and EcoRI sites of pcDNA3. Myc epitope–tagged full-length human IPS-1 and Myc–IPS-1(ΔTM) cDNAs were amplified by PCR from an Myc–IPS-1 expression vector described previously (44) with the sense primer 5′-ACTGCGGCCGCATGAGCAAAAGCTCATTT-3′ and the antisense primers 5′-GCCTCTAGACTAGTGCAGACGCCGCCG-3′ or 5′-GCCTCTAGACTAGTCTACCTGGGATGCCA-3, respectively, and were subcloned into the NotI and XbaI sites of either pcDNA3 or the G3BP1 expression vector. Myc–IPS-1 (aa 1–98), Myc–IPS-1 (aa 99–200), and Myc–IPS-1 (aa 201–540) cDNAs were amplified by PCR from an Myc–IPS-1 expression vector with the sense primer 5′-GGCAGATCTATGCCGTTTGCTGAAGACAAGACC-3′ and the antisense primer 5′-GGCAGATCTCTACCGAGGCTGGTAGCTCTGGT-3′, the sense primer 5′-GGCAGATCTACCTCGGACCGTCCCCCAGAC-3′ and the antisense primer 5′-GGCAGATCTCTATGTGTCCTGCTCCTGATGCCCGCT-3′, or the sense primer 5′-GGCAGATCTATGGAACTGGGCAGTACCCACACAGCA-3′ and the antisense primer 5′-GGCAGATCTCTAGTGCAGACGCCGCCGGT-3′, respectively, and were subcloned into the BglII and EcoRI sites of the pCS4 vector. Poly(I:C) was obtained from GE Healthcare and was introduced into cells by transfection with Lipofectamine 2000 (Thermo Fisher Scientific).

Abs to NUDT21 were obtained from Proteintech; those to Myc (9E10), to p38, to TIAR, and to TOMM20 were from Santa Cruz Biotechnology; those to IPS-1 for immunostaining as well as those to phospho-p38, to phospho-JNK, to cleaved caspase-3, to cleaved poly(ADP-ribose) polymerase (PARP), and to phospho-IRF3 were from Cell Signaling Technology; those to FLAG (M2) were from Sigma-Aldrich; those to hemagglutinin were from Roche Diagnostics; those to IPS-1 for immunoblot analysis were from Abcam; those to cytochrome C were from BD Pharmingen; and those to CFIm68 were from Bethyl Laboratories. Abs to RIG-I were kindly provided by M. Yoneyama (Chiba University).

Knockdown of NUDT21 was achieved with the use of Stealth RNA interference (Thermo Fisher Scientific) in HeLa S3 cells or the use of Silencer Select small interfering RNAs (siRNAs) (Thermo Fisher Scientific) in RAW264.7 cells. Cells were transfected with siRNA oligonucleotides with the use of the Lipofectamine RNAiMAX reagent (Thermo Fisher Scientific) and were used for experiments after incubation for 72 h. The siRNA sequences used in this study are follows: human NUDT21 1, 5′-UGAACCUCCU-CAGUAUCCAU-AUAUU-3′ and 5′-AAUAUAUGGA-UACUGAGGAG-GUUCA-3′; human NUDT21 2, 5′-GCACCAGGAU-AUGGACCCAU-CAUUU-3′ and 5′-AAAUGAUGGG-UCCAUAUCCU-GGUGC-3′; mouse NUDT21, 5′-CCUCCUCAGU-AUCCGUAUAT-T-3′ and 5′-UAUACGGAUA-CUGAGGAGGT-T-3′; and negative control siRNAs for HeLa S3 cells (catalog no. 12935300; Thermo Fisher Scientific) and RAW264.7 cells (catalog no. AM4637; Thermo Fisher Scientific) were also used.

Liquid chromatography and tandem mass spectrometry were performed as previously described (45).

Immunoblot analysis was performed as described previously (44). In brief, cells were lysed with a solution containing 20 mM Tris-HCl (pH 7.5), 150 mM NaCl, 10 mM β-glycerophosphate, 5 mM EGTA, 1 mM Na₄P₂O₇, 5 mM NaF, 0.5% Triton X-100, 1 mM Na₃VO₄, 2 mM DTT, and aprotinin (0.5 mg/ml); cell lysates were fractionated by SDS-PAGE on a 10% gel; and the separated proteins were transferred to a polyvinylidene difluoride membrane. The membrane was incubated first with primary Abs for 24 h at 4°C and then with HRP–conjugated secondary Abs (GE Healthcare) for 1 h at room temperature. After a wash with a solution containing 50 mM Tris-HCl (pH 8), 150 mM NaCl, and 0.05% Tween 20, the membrane was processed for detection of peroxidase activity with chemiluminescence reagents and an Image Quant LAS4000 instrument (GE Healthcare).

HEK293T cells or HeLa S3 cells were harvested 20 h after transfection with plasmids encoding FLAG–IPS-1 and Myc–NUDT21 or that encoding FLAG–RIG-I, respectively. Cells were then lysed with a solution containing 20 mM Tris-HCl (pH 7.5), 150 mM NaCl, 10 mM β-glycerophosphate, 5 mM EGTA, 1 mM Na₄P₂O₇, 5 mM NaF, 0.2% Triton X-100, 1 mM Na₃VO₄, 2 mM DTT, and aprotinin (0.5 mg/ml). The cell lysates were incubated with rotation at 4°C first for 45 min with Abs to FLAG and then for 45 min with protein A–Sepharose (GE Healthcare). The resulting immunoprecipitates were subjected to immunoblot analysis.

Cells were fixed with 4% formaldehyde for 10 min at 37°C, permeabilized with 0.2% Triton X-100 in PBS for 5 min, and incubated for 24 h in PBS containing 2% FBS and 2% BSA (blocking buffer). In Fig. 1C, PBS containing 10% house serum was used as a blocking buffer. They were then exposed at room temperature first for 1 h to primary Abs in blocking buffer and then for 1 h to Alexa Fluor–conjugated secondary Abs (Thermo Fisher Scientific) and Hoechst 33342 in blocking buffer. Mowiol was used as mounting medium. Images were acquired with a TCS SP5 confocal microscope (Leica) and were processed with ImageJ (National Institutes of Health).

Manders M1 and M2 coefficients for colocalization were calculated with Coloc 2 of Fiji. A statistical significance test was derived by Costes (46). For the experiment shown in Fig. 2C, samples were prepared in the same way as for immunofluorescence analysis described above, with the exceptions that ProLong Diamond (Thermo Fisher Scientific) was used as mounting medium and that images were acquired with a TCS SP8 confocal microscope (Leica). Three-dimensional images were acquired to meet the Nyquist condition (pixel size of 40.6 nm for x and y and of 100 nm for z) and were deconvoluted with Huygens software (Scientific Volume Imaging). Manders M1 and M2 for the colocalization of IPS-1 and TOMM20 were then calculated. For the experiment shown in Fig. 2E, samples were again prepared in the same way as for immunofluorescence analysis with the exception that ProLong Diamond (Thermo Fisher Scientific) was used as mounting medium. Three-dimensional images were acquired to match the Nyquist condition (pixel size of 50 nm for x and y and of 100 nm for z) and were deconvoluted with Huygens software (Scientific Volume Imaging). Manders M1 coefficient for the colocalization of IPS-1 and TIAR was then calculated. The SG volume per cell was measured with the three-dimensional object counter (threshold, 25; size filter, 1) of Huygens software in images for which the background intensity of the cytosol had been subtracted.

Total RNA was obtained from cells with the use of RNAiso (TaKaRa). Reverse transcription was performed with 1 μg of total RNA and ReverTra Ace qPCR RT Master Mix with gDNA Remover (Toyobo). The resulting cDNA was subjected to real-time PCR analysis in a Roche LightCycler with the use of a KAPA SYBR Fast qPCR Kit (Nippon Genetics). The abundance of each target mRNA was normalized by that GAPDH mRNA. The PCR primers (sense and antisense, respectively) were 5′-ACTCCTCCACCTTTGACGCT-3′ and 5′-TCCTCTTGTGCTCTTGCTGG-3′ for human GAPDH, 5′-CTGGCTGGAATGAGACTATTGTT-3′ and 5′-CTTCAGTTTCGGAGGTAACCTG-3′ for human IFN-β, 5′-AACCTGAACCTTCCAAAGATGG-3′ and 5′-TCTGGCTTGTTCCTCACTACT-3′ for human IL-6, and 5′-ATGAGCACTGAAAGCATGATCC-3′ and 5′-GAGGGCTGATTAGAGAGAGGTC-3′ for human TNF-α, 5′-TTGCATTGGTAACTGGTGGAGA-3′ and 5′-TTATGTTCCTTAGGCTTTGTAATATGTG-3′ for human NUDT21, 5′-TGGGTGTGAACCACGAG-3′ and 5′-AAGTTGTCATGGATGACCTT-3′ for mouse GAPDH, and 5′-TCCACCAGCAGACAGTGTT-3′ and 5′-CTTTGCACCCTCCAGTAATAGC-3′ for mouse IFN-β.

HEK293T cells seeded in 24-well plates (7.5 × 10⁴ cells per well) were transiently transfected for 20 h with 8 ng of a reporter plasmid encoding Renilla luciferase under the control of the human IFN-β gene promoter (p125-RLuc) together with 5 ng of an expression plasmid encoding FLAG-tagged full-length IPS-1 and either 50 or 500 ng of a plasmid for Myc-tagged IPS-1(ΔTM). The activity of Renilla luciferase in total cell lysates was measured with the use of a Dual-Luciferase Reporter Assay System (Promega) and was normalized by that of firefly luciferase derived from 80 ng of a control plasmid.

EMCV and NDV were kindly provided by M. Yoneyama (Chiba University). The cells were incubated in culture medium with 0.1 (+) or 0.2 (++) PFU of EMCV, or with 0.3 (+) of NDV for 2 h, and then replaced in EMCV- or NDV-free culture medium for 10 h.

Quantitative data are presented as means ± SEM or ± SD and were compared with Student t test or paired t test. A p value <0.05 was considered statistically significant.

In a search for regulators of IPS-1, we performed coimmunoprecipitation analysis to identify IPS-1–associated molecules. FLAG epitope–tagged IPS-1 expressed in HEK293T cells was thus immunoprecipitated with Abs to FLAG, and the resultant precipitates were analyzed by a highly sensitive direct nanoflow liquid chromatography–tandem mass spectrometry system. We identified NUDT21 among the proteins that coprecipitated with FLAG–IPS-1. To confirm this result, we performed coimmunoprecipitation analysis with HEK293T cells expressing FLAG–IPS-1 and Myc epitope–tagged NUDT21 and found that Myc–NUDT21 coprecipitated with FLAG–IPS-1 (Fig. 1A). Furthermore, FLAG epitope–tagged RIG-I expressed in HeLa S3 cells was immunoprecipitated with Abs to FLAG, and we found that both endogenous IPS-1 and NUDT21 coprecipitated with FLAG–RIG-I (Fig. 1B). Given that RIG-I is one of the representative IFN-stimulated genes whose expression is strongly enhanced after viral infection, these results support the notion that NUDT21 forms a complex with IPS-1 as well as RIG-I during antiviral responses.

To examine which domains of IPS-1 interact with NUDT21, we generated Myc-tagged fragments of human IPS-1 that contain either CARD (aa 1–98), the proline-rich domain (aa 99–200), or the C-terminal domain, including the TM domain (aa 201–540). We found that hemagglutinin epitope–tagged NUDT21 coprecipitated with Myc–IPS-1 fragments containing CARD or the proline-rich domain, but not with that containing the C-terminal domain (Supplemental Fig. 1). These results suggested that IPS-1 associates with NUDT21 through its N-terminal domains, including CARD and the proline-rich domain.

These results suggestive of a physical interaction between IPS-1 and NUDT21 were unexpected given the previously described localization of NUDT21 in the nucleus and that of IPS-1 to the mitochondrial outer membrane (8). Immunofluorescence analysis of HeLa S3 cells indeed revealed that most NUDT21 was localized to the nucleus (Fig. 1C). However, we detected a fraction of the NUDT21 signals in the cytoplasm (Fig. 1C). Knockdown of NUDT21 by transfection of the cells with a siRNA resulted in a reduction in these cytoplasmic and nuclear signals, supporting their specificity for NUDT21 (Fig. 1C). The NUDT21 signals in the cytoplasm overlapped with those for the mitochondrial marker cytochrome C (Fig. 1C). By contrast, CFIm68, a cofactor of NUDT21 in its nuclear function, did not appear to localize to mitochondria (Supplemental Fig. 2). Together, these results suggested that a fraction of NUDT21 localizes to mitochondria, where it may associate with IPS-1.

We next transfected HeLa S3 cells with the viral dsRNA mimic poly(I:C) to examine its possible effect on the intracellular distribution of NUDT21. NUDT21 immunoreactivity was detected as granule-like aggregates in the cytoplasm at 6 h after poly(I:C) transfection, and these granule-like signals were again attenuated by NUDT21 knockdown (Fig. 1D). Moreover, we found that these cytoplasmic signals of NUDT21 colocalized with the SG marker TIAR (Fig. 1D). Indeed, 26.89 ± 3.65% of the area of TIAR-positive foci overlapped with NUDT21 immunoreactivity (Fig. 1E), suggesting that NUDT21 becomes localized to SGs in the presence of cytoplasmic dsRNA and consequent activation of the RLR pathway.

The interaction of NUDT21 with IPS-1 as well as its mitochondrial localization and its appearance at SGs in response to dsRNA exposure prompted us to examine whether NUDT21 regulates the localization of IPS-1. Consistent with previous studies (26, 35), we found that a fraction of IPS-1 colocalized with TIAR in cells transfected with poly(I:C), whereas most IPS-1 appeared to colocalize with the mitochondrial import receptor subunit TOM20 homolog (TOMM20) in both control and poly(I:C)-transfected HeLa S3 cells (Fig. 2A). We calculated Manders coefficients for the colocalization of IPS-1 and TIAR (47). Manders M1 (sum of TIAR signal intensity overlapping with IPS-1 versus total TIAR signal intensity) was significantly increased by poly(I:C) transfection (Fig. 2B), whereas Manders M2 (sum of IPS-1 signal intensity overlapping with TIAR versus total IPS-1 signal intensity) did not differ significantly between cells with or without poly(I:C) exposure (Fig. 2B). These results suggested that a fraction of IPS-1 becomes localized to SGs in response to the presence of cytoplasmic dsRNA. We also found the 34.21 ± 5.88% of the area of TIAR-positive foci overlapped with IPS-1 at 6 h after poly(I:C) transfection, indicating that IPS-1 is not distributed to all SGs. To examine IPS-1 localization at a higher resolution, we stained endogenous IPS-1 and TOMM20 at 6 h after poly(I:C) transfection in HeLa S3 cells expressing a fusion protein of G3BP1 and iRFP and obtained z-stack images deconvoluted with Huygens software. A fraction of IPS-1 signals was found to localize to G3BP1-iRFP foci in poly(I:C)-transfected cells, whereas TOMM20 was detected exclusively outside of G3BP1-iRFP foci (Fig. 2C). We then examined whether poly(I:C) transfection affects the colocalization of IPS-1 and TOMM20. Although Manders M1 (sum of IPS-1 signal intensity overlapping with TOMM20 versus total IPS-1 signal intensity) was not significantly altered by poly(I:C) transfection, Manders M2 (sum of TOMM20 signal intensity overlapping with IPS-1 versus total TOMM20 signal intensity) was slightly but significantly reduced in cells exposed to poly(I:C) (Fig. 2D), suggesting that a fraction of IPS-1 separates from TOMM20 in response to poly(I:C) stimulation. Together, these results indicated that a fraction of IPS-1, but not of another mitochondrial protein (TOMM20), becomes localized to SGs in response to the presence of cytoplasmic dsRNA.

We next examined the effect of NUDT21 on the distribution of IPS-1 in cells transfected with poly(I:C). We thus calculated Manders coefficients for IPS-1 and TIAR in z-stack images deconvoluted with Huygens software and found that Manders M1 (sum of TIAR signal intensity overlapping with IPS-1 versus total TIAR signal intensity) was markedly reduced by NUDT21 knockdown (Fig. 2E, 2F). Importantly, NUDT21 knockdown did not affect the overall (mitochondrial) distribution of IPS-1 in control (nontransfected) cells (Supplemental Fig. 3A), indicating that the effect of NUDT21 on the localization of IPS-1 was stimulus dependent. By contrast, NUDT21 knockdown did not significantly alter the volume of TIAR-positive foci (SGs) per cell (Fig. 2G) or the abundance of IPS-1 (Supplemental Fig. 3B). We also found that NUDT21 knockdown did not significantly affect the ratio of SG-containing cells (Fig. 2H) or RIG-I accumulation at SGs in response to poly(I:C) transfection (Fig. 2I). Collectively, these results suggested that NUDT21 plays a role in IPS-1 localization to SGs in response to the presence of cytoplasmic dsRNA, whereas it does not overtly affect SG formation or RIG-I accumulation at SGs.

The interaction between RLRs and IPS-1 triggers activation of downstream signaling pathways in response to virus infection (7–10). Given that our data implicated NUDT21 in localization of IPS-1 to SGs, which contain RLRs, we next asked whether NUDT21 contributes to antiviral responses mediated by RLRs and IPS-1 such as the induction of IFN-β and the proinflammatory cytokines IL-6 and TNF-α in HeLa S3 cells transfected with poly(I:C). We found that NUDT21 knockdown markedly suppressed the increase in the amount of IFN-β mRNA induced by poly(I:C) stimulation (Fig. 3A, 3D). Knockdown of NUDT21 in the mouse macrophage cell line RAW264.7 also suppressed upregulation of IFN-β mRNA in response to poly(I:C) transfection (Fig. 3F, 3G). The upregulation of proinflammatory cytokines IL-6 and TNF-α mRNAs in response to poly(I:C) transfection in HeLa S3 cells was also significantly attenuated by depletion of NUDT21 (Fig. 3B, 3C). These results thus suggested that NUDT21 is necessary for the efficient induction of IFN-β and proinflammatory cytokines in response to the presence of cytoplasmic dsRNA.

We next examined whether NUDT21 mediates activation of the transcription factor IRF3 or the MAPKs p38 and JNK, all of which are essential factors for activation of the promoters of IFN-β and inflammatory cytokine genes in response to cytoplasmic dsRNA (15, 48). Knockdown of NUDT21 suppressed the increase in the abundance of phosphorylated (activated) forms of IRF3 as well as of p38 and JNK induced by poly(I:C) transfection (Fig. 3E), suggesting that NUDT21 promotes the activation of these signaling molecules that is essential for the transcription of IFN-β and inflammatory cytokine genes. In addition, NUDT21 knockdown suppressed the cleavage of caspase-3 and PARP induced by poly(I:C) transfection (Fig. 3E), suggesting that NUDT21 also plays a key role in the induction of caspase activation, and perhaps cell death, in response to the presence of cytoplasmic dsRNA. Together, these results indicated that NUDT21 mediates antiviral cellular responses induced by the presence of cytoplasmic dsRNA.

To investigate further the role of NUDT21 in antiviral responses triggered by RLRs and IPS-1, we asked whether NUDT21 knockdown affects type I IFN induction in cells infected with viruses. To this end, we studied the effects of EMCV and NDV, RNAs of which released into the cytoplasm are recognized predominantly by the RLRs MDA5 and RIG-I, respectively (6). We found that NUDT21 knockdown significantly suppressed the EMCV-induced phosphorylation of IRF3 and increase in IFN-β mRNA abundance (Fig. 4A, 4B). Depletion of NUDT21 also significantly attenuated the NDV-induced increase in the amount of IFN-β mRNA (Fig. 4C). Together, these results implicated NUDT21 in the optimal induction of IFN-β in response to viral infection.

Disruption of SGs has been shown to attenuate IFN induction by dsRNA (25, 30), although SGs appear to be dispensable for antiviral responses in some instances (49). We therefore investigated whether forced targeting of IPS-1 to SGs might enhance IFN induction in poly(I:C)-transfected cells. We first constructed an expression vector for a fusion protein containing the SG protein G3BP1 and Myc-tagged IPS-1. However, expression of G3BP1–Myc–IPS-1 resulted in formation of abnormal aggregates within HeLa S3 cells that appeared to include mitochondria as shown by TOMM20 immunostaining (data not shown). To avoid such mitochondrial aggregation, we generated a fusion protein containing G3BP1 and an Myc-tagged form of IPS-1 that lacks the C-terminal TM domain (Fig. 5A). We found that this G3BP1–Myc–IPS-1(ΔTM) fusion protein was present mostly in the cytosol of HeLa S3 cells in the absence of poly(I:C) transfection but preferentially localized to SGs in the presence of poly(I:C) (Fig. 5B). We then measured the level of IFN-β mRNA in cells expressing G3BP1, Myc-tagged IPS-1(ΔTM), or G3BP1–Myc–IPS-1(ΔTM) at 9 h after poly(I:C) transfection (Fig. 5C, 5D). Forced expression of G3BP1 resulted in slight enhancement of the increase in the amount of IFN-β mRNA induced by poly(I:C) transfection, consistent with a previous observation (50). We also found that expression of Myc-tagged IPS-1(ΔTM) slightly enhanced the increase in IFN-β mRNA abundance induced by poly(I:C). Expression of G3BP1–Myc–IPS-1(ΔTM) resulted in a markedly greater increase in the level of IFN-β mRNA in poly(I:C)-transfected cells compared with that induced by G3BP1 or Myc-tagged IPS-1(ΔTM) (Fig. 5D). Importantly, the amount of G3BP1–Myc–IPS-1(ΔTM) in the cells was markedly lower than that of Myc–IPS-1(ΔTM). These results together suggested that the localization of IPS-1 to SGs enhances IFN induction in the presence of cytoplasmic dsRNA.

In the innate immune system, the site of receptor–ligand interaction is often dissociated from that of signal transduction, likely to prevent aberrant activation of antipathogen programs in the absence of infection (51). However, these distinct compartments must encounter each other soon after ligand detection. We have now identified NUDT21 as a link between RLR-containing SGs and mitochondrial IPS-1 as well as an essential mediator of antiviral responses. Our study has therefore unveiled the existence of a cellular machinery that links the site of pathogen recognition and that of antiviral signal transduction.

What is the mechanism by which NUDT21 mediates type I IFN production? Although we cannot exclude the possibility that NUDT21 regulates type I IFN expression through alternative polyadenylation in the nucleus, we propose a model whereby NUDT21 promotes antiviral responses by recruiting IPS-1 to antiviral SGs on the basis of the following observations: 1) both mass spectrometry and coimmunoprecipitation analyses indicated that NUDT21 forms a complex with mitochondrial IPS-1. 2) A fraction of NUDT21 was found to be localized to mitochondria in resting cells and became localized to SGs in response to poly(I:C) transfection. 3) Another component of the CFIm complex, CFIm68, appeared to be localized only in the nucleus, not being detected in the cytoplasm, suggesting that NUDT21 has a cytoplasmic function independent of the CFIm complex. And 4) knockdown of NUDT21 suppressed the change in the localization of IPS-1 from mitochondria to SGs, but not the formation of SGs, in response to poly(I:C) stimulation. Further studies are required to understand in more detail how NUDT21 regulates IPS-1 localization in response to poly(I:C) transfection.

Caspase-mediated IPS-1 cleavage at a juxtamembrane site has been proposed to inactivate the function of IPS-1 (33, 34). However, an IPS-1 mutant lacking the TM domain was found to form prion-like fibers (“seeds”) that could convert native IPS-1 into functional aggregates (11). The role of IPS-1 cleavage therefore remains controversial. We found that expression of an IPS-1(ΔTM) mutant significantly enhanced the activity of the IFN-β gene promoter only in the presence of full-length IPS-1 (Supplemental Fig. 4). Moreover, our results showed that forced localization of IPS-1(ΔTM) to SGs enhanced type I IFN induction by poly(I:C) transfection. On the basis of these observations, we propose a two-step model: in the early stage of virus infection, the caspase-cleaved form of IPS-1 cooperates with full-length IPS-1 to form large aggregates on the mitochondrial surface that associate with RLRs in SGs and thereby elicit antiviral responses. In the late stage, caspases cleave the remaining intact IPS-1 molecules and thereby terminate IPS-1–mediated immune responses. This model may reconcile the apparent discrepancy in the effects of caspase-mediated cleavage on IPS-1 function mentioned above.

The precise nature of the physical interaction between SGs and IPS-1 aggregates remains unclear. Antiviral SGs are membraneless organelles that are formed by liquid–liquid phase separation within the cytoplasm. We observed that a fraction of IPS-1 appeared to form fiber-like structures rather than being evenly distributed within SGs (Fig. 2A). This observation and the previous findings on the prion-like aggregation of IPS-1 (11) suggest that IPS-1 aggregates may form a subcompartment within SGs, perhaps by undergoing a liquid-to-solid phase transition (which typically underlies the formation of solid prion-like aggregates or crystals), or that they may lie adjacent to SGs and associate with RIG-I in SGs at the interface. In these two cases, the presence of NUDT21 may facilitate functional solidification of IPS-1 in SGs or functional association between RIG-I in SGs and IPS-1 aggregates, respectively.

Lysine-63–linked polyubiquitination by the ubiquitin ligase TRIM25 has been implicated in the activation of RIG-I and subsequent formation of large IPS-1 aggregates (11, 52, 53). It would thus be of interest to examine the relation between such TRIM25-mediated polyubiquitination and NUDT21 in the formation of IPS-1 aggregates and their association with SGs. Of note, RIG-I has been shown to associate exclusively with either TRIM25 or IPS-1, with the two complexes being localized to distinct compartments (54). Together with our observation that IPS-1 is not distributed among all SGs in a cell, this finding suggests that RIG-I might shift from the TRIM25 compartment to the IPS-1 compartment in association with IPS-1 activation (54) and that NUDT21 may facilitate this transition.

In closing, we have, in this study, demonstrated the cytoplasmic localization of NUDT21 and its unexpected role in regulation of antiviral proteins in the cytoplasm in addition to its well-established localization to nucleus and function in alternative polyadenylation. We therefore propose that NUDT21 may function in broader biological contexts, at least at mitochondria and SGs, than anticipated, and our results provide a basis for the development of a new target for clinical intervention in viral propagation.

We thank M. Okajima and K. Takechi (Graduate School of Pharmaceutical Sciences, The University of Tokyo) for technical assistance; M. Yoneyama (Chiba University) for providing the pEF–BOS–FLAG–IPS-1 plasmid, the pEF–BOS–FLAG–RIG-I plasmid, and Abs to RIG-I, ECMV, and NDV; and laboratory members for discussion.

The authors have no financial conflicts of interest.