IPS-1 plays an essential role in dsRNA-induced stress granule formation by interacting with PKR and promoting its activation

Eukaryotic cells form macromolecular aggregates called stress granules in response to different stress conditions, such as heat shock, oxidative stress, endoplasmic reticulum (ER) stress, UV irradiation and viral infections (Kedersha and Anderson, 2002). Stress granule formation might initiate from the phosphorylation of eukaryotic translation initiation factor 2α (eIF2α) by four kinases – protein kinase regulated by dsRNA (PKR, also known as EIF2AK2), heme-regulated kinase (HRI, also known as EIF2AK1), general control nondepressible 2 kinase (GCN2, also known as EIF2AK4) and PKR-like ER kinase (PERK, also known as EIF2AK3) (Anderson and Kedersha, 2002). In some circumstances, stress granule formation might occur independently of eIF2α phosphorylation, such as during hippuristanol-induced stress (Anderson and Kedersha, 2008; Grousl et al., 2013; Mazroui et al., 2006; Qin et al., 2011).

The components of stress granules include stalled 48S initiation translation complexes and a group of RNA-binding proteins, such as T-cell-restricted intracellular antigen-1 (TIA-1), TIA-1-related protein (TIAR, also known as TIAL1) and Ras GAP SH3-domain-binding protein 1 (G3BP-1) (Anderson and Kedersha, 2008; Ohn and Anderson, 2010). In addition, other cellular proteins could be packed into stress granules, including Tudor Staphylococcal Nuclease (Tudor-SN, also known as SND1), adenosine deaminase acting on RNA 1 (ADAR1) (Weissbach and Scadden, 2012), DAZL (Kim et al., 2012), p90 ribosomal S6 kinase 2 (RSK2, also known as RPS6KA3), Ras homolog gene family member A and Rho-associated coiled-coil forming kinase 1 (RhoA and ROCK1) (Tsai and Wei, 2010), 2-oxoglutarate and Fe(II)-dependent oxygenase domain containing 1 (OGFOD1) (Wehner et al., 2010), melanoma differentiation-associated gene 5 (MDA5, also known as IFIH1) and retinoic-acid-inducible gene I (RIG-I, also known as DDX58) </emph>(Langereis et al., 2013; Onomoto et al., 2012). When conditions that are suitable for translation are restored, the components of stress granules are dynamically released so that translation is rapidly resumed. Therefore, stress granules are believed to serve as a temporary repository for translationally silent mRNA, thereby promoting cellular survival (Anderson and Kedersha, 2002; Arimoto et al., 2008; Tsai and Wei, 2010).

Innate immunity is another cellular response against external stress, particularly that induced by microbial invasion (Takeuchi and Akira, 2010). During viral infection, double-stranded (ds)RNA, a viral replication intermediate (Alexopoulou et al., 2001), is detected as a pathogen-associated molecular pattern (PAMP) by cellular receptors, including membrane-bound Toll-like receptor 3 (TLR3) (Akira and Takeda, 2004), cytoplasmic RIG-I and MDA5 (Kang et al., 2002; Yoneyama et al., 2004). RIG-I and MDA5 act as sensors that detect intracellular viral dsRNA, and they initiate signaling through IFN-β promoter stimulator (IPS-1, also known as Cardif, MAVS or VISA), which is expressed on the mitochondrial membrane (Kawai et al., 2005). Upon binding to RIG-I or MDA5, IPS-1 undergoes a conformational change, switching to prion-like aggregates and transmitting the innate signaling through the TBK-1 and IKK kinases, thus leading to the induction of proinflammatory cytokines and IFNs (Hou et al., 2011; Meylan et al., 2006; Seth et al., 2005). Recent studies have shown that several RNA-binding proteins involved in innate immune signaling, including RIG-I, MDA5, PKR and ADAR1, are packed into influenza A virus (IAV)-induced antiviral stress granules (avSGs) (Onomoto et al., 2012) and mengovirus- and measles-virus-induced stress granules (Langereis et al., 2013; Okonski and Samuel, 2013). Nevertheless, it is unclear whether these proteins have a direct impact on stress granule formation.

The aim of the current study was to examine the relationship between stress granule formation and the innate immune response. We investigated the role of four key molecules associated with innate immunity (the intracellular sensors PKR, RIG-I and MDA5, and the adaptor IPS-1) in stress granule formation induced by different stresses. We found that IPS-1 and PKR play a crucial role in the process of stress granule formation induced by dsRNA, a common viral byproduct, but not by other stresses. We found that IPS-1 is physically associated with PKR and promotes PKR activation. For the first time, we demonstrate an unexpected role for IPS-1 in mediating stress granule formation, which it performs in addition to its well-defined role in innate immunity.

To examine whether innate immune proteins, including the dsRNA sensors RIG-I, MDA5, PKR and an adaptor protein IPS-1, have a direct impact on the formation of stress granules, we silenced these proteins by RNAi and compared the extent of stress granule formation induced by two eIF2α-dependent stimuli (dsRNA and heat shock) and one eIF2α-independent stimulus (hippuristanol). The efficacy of each siRNA was monitored by western blotting or real-time PCR. Western blotting data showed that >95% of RIG-I, IPS-1 and PKR protein was depleted in cells transfected with the respective siRNA, validating a high efficacy of siRNA-mediated knockdown (Fig. 1A–C). To assess the extent of MDA5 knockdown, we measured its mRNA level instead of the protein level, owing to its low endogenous abundance in A549 cells (Fig. 1D).

Subsequently, these siRNA-transfected cells were mock-treated or stimulated with the immunostimulant polyinosinic:polycytidylic acid [poly(I:C)], a synthetic analog of dsRNA. The high molecular mass (hereafter referred to as high molecular weight, HMW) version of poly(I:C) is sensed by MDA5, whereas the low molecular mass (low molecular weight, LMW) version is sensed by RIG-I. The percentages of stress-granule-positive cells were determined in an immunofluorescence staining assay using an antibody against TIA-1, a classical hallmark of stress granules (Anderson and Kedersha, 2002). In HMW-poly(I:C)-transfected cells, knockdown of PKR significantly reduced the percentage of TIA-1-positive cells (Fig. 1E,F). Interestingly, the formation of stress granules was also significantly reduced in IPS-1-depleted cells (22% contained stress granules, versus 42% of cells treated with a control siRNA, P<0.001). The knockdown of MDA5 led to a smaller but still significant effect on stress granule formation (P<0.05). By contrast, the percentage of stress-granule-positive cells was comparable between RIG-I-knockdown and control cells (P>0.05). Moreover, the depletion of both RIG-I and MDA5 also led to a significant reduction in the amount of stress granule formation (P<0.01). Similarly, in LMW-poly(I:C)-transfected cells (Fig. 1G,H), a significant reduction in the percentage of stress-granule-positive cells was seen in cells that were depleted of PKR, IPS-1, MDA5 or RIG-I and MDA5, but not in cells that were depleted of RIG-I alone (Fig. 1G,H). Moreover, the formation of stress granules in response to heat shock and hippuristanol treatment was examined. Both heat shock and hippuristanol treatment led to stress granule formation in most cells (70% and 98% of cells, respectively). However, the depletion of none of the innate immune molecules had any significant effect on stress granule formation in response to heat shock or hippuristanol (Fig. 1I–K). These data suggested that role of PKR and IPS-1 in the formation of stress granules was dsRNA specific, but was independent of dsRNA length.

As PKR–eIF2α is the dominant signaling pathway involved in dsRNA- and virus-induced formation of stress granules, we hypothesized that the effects of IPS-1 and MDA5 on stress granule formation might be exerted through PKR and eIF2α. To test this hypothesis, we compared the activation levels of PKR and eIF2α in cells depleted of RIG-I, MDA5, IPS-1 or PKR. Cells were stimulated by treatment with poly(I:C), and the phosphorylated (active) forms of PKR and eIF2α were detected by western blotting with specific antibodies. In HMW-poly(I:C)-treated cells, phosphorylation of PKR and eIF2α was detected more readily than in control cells, as expected (Fig. 2A, lane 2 versus lane 1). Interestingly, the phosphorylated forms of PKR and eIF2α were dramatically reduced in IPS-1- or PKR-knockdown cells upon poly(I:C) transfection (Fig. 2A, lanes 4 and 10 versus lane 2), suggesting that IPS-1 influences PKR activation. MDA5 depletion impaired the phosphorylation of PKR and eIF2α to a more moderate extent (lane 8), whereas knockdown of RIG-I did not affect phosphorylation of PKR and eIF2α in response to poly(I:C) (lane 6 versus lane 2). Likewise, knockdown of IPS-1 and PKR, but not of RIG-I, significantly reduced the activation of PKR and eIF2α in LMW-poly(I:C)-treated cells (Fig. 2B). However, MDA5 deficiency did not decrease the amount of phosphorylated PKR and eIF2α in response to treatment with LMW poly(I:C) (lane 6 versus lane 2), which might be due to the lack of responsiveness of MDA5 to short-length dsRNA. To further examine the redundancy of RIG-I and MDA5 in dsRNA sensing and stress granule formation, cells were simultaneously transfected with siRIG-I and siMDA5, followed by challenge with HMW and LMW poly(I:C). Western blotting data showed that the PKR activation triggered by HMW poly(I:C), but not by LMW- poly(I:C), was reduced in cells depleted of both RIG-I and MDA5, as seen in MDA5-depleted cells (Fig. 2C,D), suggesting that MDA5 but not RIG-I was involved in the formation of stress granules. Of note, the total PKR protein amounts in all poly(I:C)-treated cells were lower than in control cells, which is most likely caused by translational inhibition or a caspase-dependent cleavage of PKR protein (Saelens et al., 2001), consistent with our previous observations (Zhang and Samuel, 2007). Total protein levels of eIF2α were comparable in each lane.

To further validate the role of IPS-1in PKR autophosphorylation, purified IPS-1 and PKR proteins were used in a PKR kinase assay. The reaction mixtures were subjected to western blotting using the monoclonal antibody against phospho-PKR (Thr446). As shown in Fig. 2E, the IPS-1 protein alone clearly increased the phosphorylation of PKR on Thr446 (lane 3) and further enhanced PKR phosphorylation in the presence of poly(I:C) (lane 4). Surprisingly, poly(I:C) alone did not induce the phosphorylation of PKR on Thr446 (lane 2). These data suggested that the PKR autophosphorylation on Thr446 was dependent on IPS-1.

The activation of PKR involves binding to dsRNA, dsRNA-binding domain (dsRBD)-mediated dimerization and autophosphorylation (Lemaire et al., 2006; Sadler, 2010). To address the mechanism by which IPS-1 facilitates PKR activation, we examined the involvement of IPS-1 in PKR dimerization through co-immunoprecipitation (co-IP) assays. First, as the PKR N-terminus has been suggested to mediate PKR dimerization (Patel and Sen, 1998), we overexpressed an HA-tagged version of the PKR N-terminus (PKRN–HA) comprising two dsRBDs in IPS-1-sufficient or IPS-1-deficient cells, followed by poly(I:C) stimulation (to mimic exposure to dsRNA). Protein was pulled down from whole-cell extracts using an anti-HA antibody, and protein complexes were detected by western blotting using an anti-PKR antibody. The endogenous full-length PKR was precipitated with PKRN–HA regardless of the absence or presence of dsRNA (Fig. 3A, upper panel, lanes 1–4). Upon stimulation with dsRNA, more endogenous co-immunoprecipitated PKR was detected in control (siNC)-transfected cells (upper panel, lane 2 versus lane 1), and this dsRNA-mediated association between PKR and PKRN–HA was significantly reduced in IPS-1-depleted cells (upper panel, lane 4 versus lane 3).

Next, we generated a construct expressing full-length PKR tagged with HA at the N-terminus and FLAG at the C-terminus (HPKRF) to further probe the role of IPS-1 in PKR activation. IPS-1-sufficient or IPS-1-deficient cells were transfected with pSG5-HPKRF plasmid and harvested for co-IP using an anti-FLAG antibody. Without dsRNA stimulation (Fig. 3B), comparable amounts of endogenous PKR (lower band) were precipitated with HPKRF (upper band) in IPS-1-sufficient and IPS-1-deficient cells (lanes 1 and 3). Stimulation with dsRNA significantly increased the PKR–HPKRF association (upper panel, lane 2 versus lane 1), which was greatly diminished by IPS-1 depletion (upper panel, lane 4). These data implied that IPS-1 might be essential for the dsRNA-dependent dimerization of PKR, but not for the dsRNA-independent dimerization.

To determine the underlying mechanism by which IPS-1 promotes PKR activation, we tested the possibility that IPS-1 interacts with PKR. A549 cells were co-transfected with plasmids expressing IPS-1–FLAG and PKR–HA fusion proteins, and were harvested for the co-IP assay using antibodies against HA or FLAG. The results showed that the anti-HA antibody was able to co-immunoprecipitate PKR or IPS-1 when they were expressed together in the absence or presence of poly(I:C) (Fig. 4A). Similarly, anti-FLAG antibody co-immunoprecipitated both PKR and IPS-1 in the mock- or poly(I:C)-transfected cells (Fig. 4A). To validate the interaction between IPS-1 and PKR, purified IPS-1–FLAG and PKR–HA were incubated in vitro, followed by a pulldown assay using antibodies against FLAG or HA. Our data revealed that the anti-FLAG antibody was able to pull down both IPS-1–FLAG and the PKR–HA (Fig. 4B). Reciprocally, the anti-HA antibody also pulled down both proteins (Fig. 4C). Next, IPS-1–His or PKR–GST fusion proteins were expressed in Escherichia coli and purified by using antibodies against GST or His respectively. IPS-1–His and PKR–GST were incubated in vitro, followed by a pulldown assay using GST resin. The western blot clearly showed that IPS-1 was co-precipitated with PKR in the absence or presence of poly(I:C) (Fig. 4D). These co-IP and GST pulldown data indicate that IPS-1 has a direct interaction with PKR that is independent of dsRNA.

To map which domains of PKR and IPS-1 are responsible for their interaction, we generated a series of constructs expressing different forms of PKR or IPS-1. The PKR protein forms included full-length PKR (PKR–HA), the N-terminus of PKR comprising two dsRNA binding domains (dsRBDs) (PKRN–HA) and the C-terminus of PKR comprising the kinase domain (PKRC–HA) (Fig. 5A). The IPS-1 protein forms included full-length IPS-1 (IPS-1–FLAG), IPS-1 with CARD deleted (ΔCARD–FLAG), IPS-1 with deletion of the proline-rich domain (ΔPro–FLAG) and IPS-1 with deletion of the transmembrane domain (ΔTM–FLAG) (Fig. 5B). Cells were transfected with the plasmid encoding IPS-1–FLAG, along with the plasmid encoding PKR–HA, PKRN–HA or PKRC–HA, followed by co-IP assay using anti-FLAG antibody. The western blotting data showed that the anti-FLAG antibody immunoprecipitated IPS-1–FLAG with PKR–HA or PKRN–HA but not with PKRC–HA, indicating that the dsRBDs of PKR mediate its association with IPS-1 (Fig. 5C). Subsequently, cells were transfected with PKR–HA-expressing plasmid, along with the plasmid encoding IPS-1–FLAG, ΔCARD–FLAG, ΔPro–FLAG or ΔTM–FLAG, followed by co-IP assay using an anti-FLAG antibody. PKR was immunoprecipitated with IPS-1–FLAG, ΔPro–FLAG or ΔTM–FLAG but not with ΔCARD–FLAG (Fig. 5D). Furthermore, an in vitro PKR kinase assay was performed using purified full-length and deletion forms of IPS-1. Western blotting using an antibody that detected p-PKR (Thr446) showed that the presence of IPS-1–FLAG, ΔPro–FLAG or ΔTM–FLAG clearly increased PKR phosphorylation at Thr446, whereas ΔCARD–FLAG only weakly promoted PKR activation (Fig. 5E). These data suggested that CARD is indispensable for mediating the IPS-1–PKR interaction and PKR activation.

As PKR has been reported to be a component of stress granules that are induced by several viruses (Okonski and Samuel, 2013; Onomoto et al., 2012), we next used immunofluorescence microscopy to examine whether PKR is recruited to stress granules upon stimulation of cells with dsRNA. In A549 and HeLa cells, PKR was diffusely distributed in the cytoplasm, whereas TIA-1 was predominantly located in the nuclei. In both dsRNA-stimulated A549 and HeLa cells, nuclear TIA-1 translocated to cytoplasmic granules as expected. Interestingly, PKR protein was localized in the granules that overlapped with regions of TIA-1 staining (Fig. 6A,B), suggesting that PKR was recruited by dsRNA-induced stress granules. As the compositions of stress granules vary in response to different stresses (Kedersha et al., 2005), we also examined whether PKR is a component of stress granules that are induced by other stresses. After heat shock or hippuristanol treatment, A549 and HeLa cells were stained with antibodies against PKR and TIA-1. In response to heat shock or hippuristanol, TIA-1 translocated from nuclei to form cytoplasmic stress granules (Fig. 6A,B), in agreement with previous reports (Anderson and Kedersha, 2008; Gilks et al., 2004). By contrast, PKR remained evenly distributed in the cytoplasm after heat shock or hippuristanol stimulation (Fig. 6A,B). The colocalization analysis showed that the overlap coefficient between PKR and TIA-1 was significantly higher in dsRNA-stimulated cells, but not in heat-shock- or hippuristanol-stressed cells, than in the mock-treated cells (0.86 versus 0.53 in A549, P<0.01, Fig. 6C; 0.8 versus 0.56 in HeLa, P<0.01, Fig. 6D), indicating that the recruitment of PKR to stress granules is stress specific.

To test whether IPS-1 is recruited to stress granules together with PKR, IPS-1 subcellular localization was visualized by confocal immunofluorescence staining microscopy using a mitochondrial indicator, MitoTracker Red, and antibodies against IPS-1 and TIA-1. The staining pattern of IPS-1 overlapped with that of MitoTracker Red in control cells, whereas, in dsRNA-stimulated cells, IPS-1 protein formed clusters that partially overlapped with MitoTracker Red (Fig. 7A). In addition, cells were double-stained with antibodies against TIA-1 and IPS-1 (Fig. 7B), or TIA-1 antibody and MitoTracker Red (Fig. 7C). A small amount of TIA-1 aggregates were observed to overlap with IPS-1-positive clusters, whereas TIA-1-positive stress granules did not colocalize with MitoTracker Red. The colocalization analysis revealed that the overlap coefficient between TIA-1 and IPS-1 was slightly, but not significantly, increased in poly(I:C)-treated cells (P>0.05, Fig. 7D), and the overlap coefficients between TIA-1 and MitoTracker Red were comparable in mock and poly(I:C)-treated cells (Fig. 7D).

To determine whether stress granules or innate immune responses play a role in cell physiology or viral replication, we infected the cells with vesicular stomatitis virus carrying the gene encoding GFP (VSV-GFP). As IPS-1 and PKR function in both signaling pathways, we silenced two additional proteins – RIG-I, the receptor sensing VSV, and G3BP-1, a key component of stress granules. As predicted, the depletion of RIG-I, IPS-1 and PKR significantly decreased the level of dsRNA-induced IFN-β expression (Fig. 8A). Although G3BP-1 knockdown largely abolished the stress granule formation induced by dsRNA (data not shown), its depletion barely affected the induction of IFN-β. Moreover, the cell viabilities were comparable among all the mock-treated cells (around 85%, Fig. 8B) or among the dsRNA-treated cells (around 75%, Fig. 8B), suggesting that either stress granule formation or innate immune response do not have a pronounced effect on cell survival upon VSV infection. In the mock-treated cells, the percentages of VSV-GFP-positive cells were comparable (approximately 75%) in most tested cells except those with RIG-I knockdown, agreeing with our previous finding that PKR is not a major anti-VSV protein (Zhang and Samuel, 2007). In dsRNA-stressed cells, the percentage of cells that were positive for GFP was largely reduced in siNC-transfected cells (5.6%), as a result of IFN induction and stress granule formation induced by dsRNA. Moreover, the percentage of cells with GFP staining was significantly restored by depletion of PKR, IPS-1 or G3BP-1 (around 15.9%, 16.2% and 9.0% of cells were GFP positive, respectively, Fig. 8C,D). These data suggested that stress granules and innate immune responses confer an advantage in protecting cells against VSV infection. Surprisingly, RIG-I knockdown led to decreased viral replication in mock-treated cells (39% GFP-positive cells, P<0.01, Fig. 8C,D) and a comparable viral replication in dsRNA-stressed cells (3.8% GFP-positive cells, P>0.05, Fig. 8C,D), implying that RIG-I might play some proviral role in VSV replication.

Stress granule formation and innate immunity are two distinct mechanisms by which cells defend themselves from stresses derived from viral infection. Herein, we have investigated their relationship by studying the role of four innate immune proteins in the stress granule response, and our data provide several novel findings and implicate IPS-1 in the formation of stress granules.

In the present study, dsRNA was used as a major stimulator to induce stress granule formation because it is a known activator of innate immunity and represents a common stress encountered during viral infection. We demonstrated that one of the dsRNA sensors, PKR, is a dominant factor controlling eIF2α activation and the process of stress granule formation under conditions of dsRNA-induced stress, agreeing with our previous observation that PKR is the only kinase mediating dsRNA-induced phosphorylation of eIF2α (Zhang and Samuel, 2007). This observation further extends the notion that PKR plays a key role in viral-infection-induced stress granule formation, as reported for RNA viruses, including measles virus, hepatitis C virus, IAV and respiratory syncytial virus (Dabo and Meurs, 2012; Lindquist et al., 2011; Okonski and Samuel, 2013; Onomoto et al., 2012; Ruggieri et al., 2012). It is conceivable that PKR most likely is the major eIF2α kinase involved in the stress granule formation that is induced by other dsRNA-producing viruses, either DNA or RNA viruses.

Importantly, the innate immune adaptor IPS-1 was found to have a novel function in regulating stress granule formation, which is mediated through PKR. To date, IPS-1 has been implicated as a common adaptor in a variety of signaling pathways, owing to its ability to interplay with different signaling molecules. Initially, IPS-1 was identified as a common adaptor of RIG-I and MDA5 in the innate signaling pathway triggered by viral infection or dsRNA (McWhirter et al., 2005). Subsequently, IPS-1 was found to interact with a sensor of DNA and dsRNA, DHX9, in myeloid dendritic cells (Zhang et al., 2011). Most recently, IPS-1 was found to recruit the inflammasome protein NLRP3 to mitochondria (Arnaud et al., 2011). In the present study, we demonstrate that IPS-1 acts as an adaptor in stress granule formation, through an association with the key stress granule regulator, PKR. We show that the interaction between IPS-1 and PKR is mediated by the dsRBDs of PKR and the CARD of IPS-1. Interestingly, the deletion of the IPS-1 proline domain resulted in an increased interaction with PKR, implying that the proline domain might be an inhibitory domain or a conformational change might be required for IPS-1 to interact with PKR. By contrast, deletion of the transmembrane domain of IPS-1 led to a decreased interaction with PKR, suggesting that IPS-1 needs to be localized on the mitochondrial membrane for an optimal interaction.

Furthermore, our data indicated that the interaction between IPS-1 and PKR might directly affect the dsRNA-dependent dimerization of PKR. Our co-IP data showed that the endogenous PKR can associate with N-terminal dsRBDs or full-length PKR in the absence of dsRNA and IPS-1, agreeing with a previously established notion that PKR might form a weak dimer in latent cells (Langland and Jacobs, 1992; Patel et al., 1995). The IPS-1 depletion abrogated the dsRNA-mediated PKR–dsRBDs and PKR–PKR associations, suggesting that dsRNA-mediated PKR dimerization requires the participation of IPS-1. Consistently, we observed that the in vivo phosphorylation of PKR and eIF2α triggered by both HMW and LMW poly(I:C) was markedly reduced by IPS-1 depletion, and the presence of IPS-1 significantly increased the in vitro phosphorylation of PKR on Thr446, which is independent of, but enhanced by, dsRNA. Interestingly, in an in vitro assay, dsRNA alone did not lead to PKR phosphorylation on Thr446, although dsRNA has been shown to induce PKR phosphorylation (Li et al., 2006). This difference could be caused by different detection assays used – we used a specific antibody against phosphorylated Thr446, whereas the autoradiography protocol used in other studies enables the detection of phosphorylation of several sites, including Thr258, Ser242, Thr255 or Thr451, in addition to Thr446 (Romano et al., 1998; Taylor et al., 1996). As phosphorylation in the activation loop (Thr446 or Thr451) functions to facilitate substrate binding and catalysis (Williams, 1999), the observation that IPS-1 assists PKR autophosphorylation on Thr446 might illustrate a potential role of IPS-1 in PKR signaling. In this respect, our finding that IPS-1 interacts with PKR and facilitates its activation could help us to uncover more mechanisms of their common functions. For example, the pro-apoptotic effect of IPS-1 in response to viral infection could possibly be exerted through PKR-mediated signaling.

Our immunofluorescence studies indicated that PKR and IPS-1 might disassociate upon PKR activation. PKR protein was recruited into stress granules that were induced by dsRNA stress, but not those induced by heat shock or hippuristanol. dsRNA alone, independent of other viral gene products, is sufficient to induce the translocation of PKR into stress granules, as seen in the case of IAV-induced avSGs and measles-virus-induced stress granules (Okonski and Samuel, 2013; Onomoto et al., 2012). By contrast, IPS-1 protein forms some mitochondria-associated clusters upon dsRNA stimulation, consistent with recent reports that IPS-1 undergoes a conformational switch to produce prion-like aggregates upon viral infection or dsRNA stimulation (Hou et al., 2011; Meylan et al., 2006; Seth et al., 2005). Notably, some of these IPS-1-positive clusters overlapped with stress granules, so the possibility that IPS-1 colocalizes with stress granules could not be ruled out, as stress granule formation is a highly dynamic process.

Intriguingly, the impacts of another two dsRNA sensors, RIG-I and MDA5, on dsRNA-induced stress granule formation are distinct. RIG-I and MDA5 are responsible for recognizing different lengths of dsRNA molecules or viruses to initiate innate immune signaling (Kato et al., 2008). RIG-I was dispensable for PKR activation and stress granule formation, regardless of dsRNA lengths, which we expected as previous work has shown that the interaction between PKR and IPS-1 takes place prior to and was independent of interaction between RIG-I and IPS-1 (Arnaud et al., 2011). As for MDA5, it plays a moderate role in stress granule formation induced by HMW or LMW poly(I:C); in addition, it has a moderate effect on PKR and eIF2α phosphorylation induced by HMW dsRNA, but not by LMW dsRNA. In combination with the report that MDA is a component of stress granules (Langereis et al., 2013), and that MDA5 displays differential effects on PKR phosphorylation and stress granule formation, we speculated that the major mechanism by which MDA5 affects stress granule formation is through mediating the assembly of stress granule components downstream of eIF2α phosphorylation, rather than through directly regulating the activation of PKR or eIF2α.

Based on the data presented here, we propose that the role of IPS-1 in PKR activation is to provide a physical platform for PKR to form dsRNA-dependent dimers. In unstressed cells, PKR protein resides in the cytoplasm as monomer or weak dimer, and dynamically interacts with IPS-1. After binding to dsRNA, PKR molecules form more dimers in association with IPS-1, thereby undergoing autophosphorylation. Activated PKR is released from IPS-1 and phosphorylates downstream eIF2α, resulting in inhibition of translation. Finally, a variety of components, including PKR, stalled initiated translation complexes and TIA-1, etc., assemble to form stress granules. After dissociation from the PKR dimer, IPS-1 interacts with RIG-I and becomes activated, forming prion-like aggregates on mitochondrial membranes (Hou et al., 2011).

We found that both stress granules and innate immunity protect cells from viral infection. Depletion of IPS-1 or PKR abolished both IFN production and stress granule formation, leading to a higher rate of VSV replication than depletion of G3BP-1 alone, which only eliminated the stress granule formation. These data suggest that innate immunity and the formation of stress granules might act cooperatively to combat VSV. In addition, as mitochondrial IPS-1 has been suggested to participate in the apoptosis induced by VSV infection (Guan et al., 2013), blockade of apoptosis by IPS-1 knockdown might be another contributing factor to the recovery of viral replication. Unexpectedly, the cell viabilities in response to VSV infection were not correspondingly decreased by depletion of IPS-1, PKR or G3BP-1, suggesting that other responses, rather than stress granule formation and innate immunity, play a dominant role in cell death caused by VSV infection.

The fact that IPS-1 and PKR participate in stress granule formation and innate signaling indicates that these two distinct responses have a close connection. To overcome stress-granule-dependent and innate immunity defenses, viruses must have developed their own strategies to counteract IPS-1 or PKR (Lifland et al., 2012; Onomoto et al., 2012; Ruggieri et al., 2012; Tu et al., 2012). However, the storage of cellular components in stress granules could potentially pose a negative effect on the innate immune response, because stress granules not only recruit innate immune signaling molecules, including PKR, RIG-I and MDA5, but also effector molecules, such as OAS, RNase L and ADAR1 (Onomoto et al., 2012; Weissbach and Scadden, 2012). Our data show that blockade of stress granule formation by G3BP-1 knockdown did not have a significant effect on dsRNA-induced IFNs, consistent with a recent study showing that the recruitment of MDA5 by stress granules did not affect the induction of type-I IFNs resulting from mengovirus infection (Langereis et al., 2013). These observations suggest that the processes of stress granule formation and innate immune signaling do not engage in direct competition for these cellular components. Nonetheless, this observation needs to be further validated in other models, because the abundances of the relevant proteins in different cells might vary.

In summary, our work demonstrates for the first time that IPS-1 is involved in the process of stress granule formation through its interaction with PKR, providing evidence that IPS-1 acts as a novel adaptor in an additional anti-stress response. Further investigations into the biological significance of the interaction between IPS-1 and PKR will provide us with more in-depth knowledge of the roles of IPS-1 and PKR in these cellular signaling pathways.

Human lung carcinoma epithelial cells (A549), human cervical cancer cells (HeLa) and human embryonic kidney cells (293T) were maintained in Dulbecco's modified Eagle's medium (DMEM) supplemented with 5% (v/v) fetal bovine serum (Gibco, CA), 1% sodium pyruvate, 100 µg/ml penicillin and 100 units/ml streptomycin (Invitrogen, CA) at 37°C.

Cells (3×10⁴) were seeded onto coverslips in 24-well plates. Cells were transfected with 300 ng of poly(I:C) (tlrl-pic, tlrl-picw, InvivoGen, MO) by using Lipofectamine 2000 (Invitrogen, CA), or were incubated at 43°C for 1 h or incubated with 1 µM hippuristanol [kindly provided by Junichi Tanaka (Mazroui et al., 2006)] for 1 h. Cells were incubated with 50 nM MitoTracker Red (Molecular Probes) for 20 min, washed in phosphate-buffered saline (PBS) and fixed in 4% (v/v) paraformaldehyde. Cells were permeabilized in 0.2% Triton X-100 for 15 min and blocked in blocking buffer (5% bovine serum albumin in PBS) for 1 h, followed by incubation with primary antibodies and secondary antibodies. The primary antibodies used were against TIA-1 (Santa Cruz, CA), PKR (Santa Cruz, CA) and IPS-1 (BETHYL, TX). The secondary antibodies used were Cy3-conjugated goat anti-rabbit-IgG (Millipore, MA) and Alexa-Fluor-488-conjugated anti-goat-IgG (R&D, MN). Cells were then stained with 0.2 mg/ml DAPI (Invitrogen, CA) and visualized using a Zeiss fluorescence microscope or a Zeiss LSM710 confocal microscope. Positively stained cells were quantified using the PicCnt 100× software. For colocalization assays, images were typically acquired with the same optical slice thickness in every channel and analyzed by using the Carl Zeiss software ZEN. Mander's coefficient was used to measure colocalization.

The sequences of primers used in PCR to amplify tagged full-length PKR, IPS-1, truncated fragments of PKR and IPS-1, or PKR and IPS-1 are listed in supplementary material Table S1. The PCR templates were plasmid pEF-BOS IPS-1-FLAG (Addgene, MA) and pSG5-PKR^RSC (a kind gift from Stefan Rothenburg, NIH, MD) (Zhang et al., 2009). PCR fragments were cloned into the pSG5 vector for eukaryotic expression, or cloned into the pGEX-6P-1 (PKR) or pET28a (IPS-1) vector for prokaryotic expression after digestion by restriction enzymes.

The sequences of siRNAs prepared by Invitrogen with dTdT overhangs were as follows: 5′-CCUCAUCUCUUUGUUCUAA-3′ and 5′-GCAUGGGCCAGAAGGAUUUCA-3′ for PKR (siPKR), 5′-GGAAGAGGUGCAGUAUAUU-3′ for RIG-I (siRIG-I), 5′-GGUGAAGGAGCAGAUUCAG-3′ for MDA5 (siMDA5), 5′-UAGUUGAUCUCGCGGACGA-3′ and 5′-CCACCUUGAUGCCUGUGAA-3′ for IPS-1 (siIPS-1) and 5′-GAGCCAGUAUUAGAAGAAA-3′ for G3BP-1 (siG3BP-1). Dharmacon's negative siRNA with a scrambled sequence were used as a negative control (siNC). siRNAs were transfected into A549 cells by using Lipofectamine 2000 (Invitrogen, CA) according to the manufacturer's instructions.

Total RNA was prepared from A549 cells using TRIzol reagent (Invitrogen, CA) following the manufacturer's instructions. Then, 1 µg of total RNA was reverse-transcribed to produce cDNAs and then amplified using SYBR Green Master Mix (Bio-Rad, CA) following the manufacturer's protocol. The primer pairs used were as follows: glyceraldehyde-3-phosphate dehydrogenase (GAPDH), forward primer 5′-GCCTTCCGTGTCCCCACTG-3′ and reverse primer 5′-CGCCTGCTTCACCACCTTC-3′; MDA5, forward primer 5′-TGGACATAACAGCAACATGG-3′ and reverse primer 5′-CACTCTGGTTTTTCCACTCC-3′; IFN-β, forward primer 5′-AAACTCATGAGCAGTCTGCA-3′ and reverse primer 5′-AGGAGATCTTCAGTTTCGGAGG-3′. Quantitative real-time PCR was performed using the CFX96 Real-Time PCR System (Bio-Rad, CA). Relative mRNA levels were calculated after normalization to GAPDH.

A549 cells were seeded in 12-well plates (8×10⁴ per well). After transfection with siRNAs for 48 h, cells were transfected with 1 µg poly(I:C) by using Lipofectamine 2000 (Invitrogen, CA) for 6 h. Whole-cell extracts were prepared in the presence of 1 mM phenylmethylsulfonyl fluoride (PMSF) and 1% (v/v) protease inhibitor cocktail (Sigma, MO) as described previously (Li et al., 2013). Proteins were fractionated by electrophoresis on sodium dodecyl sulphate–10% polyacrylamide gels, transferred to nitrocellulose membranes, blocked and then probed with an appropriate dilution of primary antibody in PBS containing 3% (w/v) skimmed milk. Rabbit polyclonal antibodies were used to detect PKR (Santa Cruz, CA), phospho-PKR (Epitomics, CA), phospho-eIF2α (Cell Signaling Technology, MA), RIG-I (Cell Signaling Technology, MA), IPS-1 (Bethyl, TX), HA (Sigma, MO) and FLAG (Beverly, MA); goat polyclonal antibody was used to detect eIF2α (Santa Cruz, CA), mouse monoclonal antibody was used to detect β-actin (Sigma, MO). Western blot visualization was performed with IRDye-800-CW-conjugated anti-rabbit-IgG or IRDye-680-CW-conjugated anti-mouse-IgG secondary antibodies according to the manufacturer's protocols (LI-COR, NE). Immunoreactive bands were visualized using an Odyssey infrared imaging system.

For the analysis of interactions between endogenous PKR and exogenously expressed PKRN–HA or full-length PKR, HeLa cells were transfected with siNC or siIPS-1 and pSG5-PKRN-HA or pSG5-HPKRF for 48 h. For analysis of the interaction between PKR and IPS-1, A549 cells (7×10⁵ per well, six-well plate) were transfected with plasmids expressing IPS-1–FLAG and PKR–HA fusion proteins. Cells were then transfected with 3 µg of poly(I:C) for 1 h and collected in RIPA lysis buffer (50 mM Tris-HCl, 0.5% NP-40, 1% Triton X-100, 1 mM EDTA, 150 mM NaCl, 1 mM PMSF and 1% v/v protease inhibitor cocktail). The protein concentration was determined by using the Bradford assay. Subsequently, 1 mg of cell lysate was incubated with 20 µl of monoclonal anti-HA or anti-FLAG agarose (Sigma, MO) at 4°C overnight. Beads were collected by centrifugation and washed extensively with NET-RIPA wash buffer (50 mM Tris-HCl, 0.5% NP-40, 1 mM EDTA, 150 mM NaCl), and proteins were eluted in SDS loading buffer.

pSG5-IPS-1-FLAG and pSG5-PKR-HA were expressed in HEK293T cells by transient transfection. At 48 h after transfection, cells were lysed in RIPA lysis buffer and centrifuged at 12,000 g for 15 min. The supernatants were incubated with anti-FLAG or anti-HA beads (Sigma, MO) and eluted with the FLAG peptide or HA peptide (Sigma, MO). pET28A-IPS-1 and pGEX-6P-PKR were transformed into Rosetta 2(DE3). Expression of IPS-1–His and PKR–GST fusion proteins was induced by the addition of 500 µM isopropyl-β-d-thiogalactopyranoside (IPTG) at 20°C for 16 h, and purified using His- (Merck, NJ) and GST-binding resin (Merck, NJ) according to the manufacturer's instructions.

Purified full-length or truncated IPS-1 proteins were individually incubated with full-length or truncated PKR proteins for 1 h, followed by incubation with anti-FLAG or anti-HA beads for 1 h. 100 nM IPS-1–His and PKR–GST fusion proteins purified from E. coli were incubated in the absence or presence of 5 µg/ml poly(I:C), followed by overnight incubation with GST-binding resin. The beads were collected by centrifugation and proteins were eluted in SDS loading buffer for western blot analysis.

Autophosphorylation assay of PKR was performed in kinase buffer (20 mM Tris-HCl pH 7.6, 50 mM KCl, 25 mM MgCl₂, 1% protease inhibitor cocktail and 1 mM PMSF) containing purified PKR, full-length or deletion forms of IPS-1, dsRNA (2 mg/ml) and ATP (0.1 mM) at 30°C for 30 min. Reaction mixtures were analyzed by SDS-PAGE and western blotting using the anti-phospho-PKR antibody (Thr 446) (Epitomics, CA).

Vesicular stomatitis virus (VSV-GFP) was kindly provided by Dongyan Jin (Kok et al., 2011). A549 cells were transfected with siRNAs against PKR, IPS-1, RIG-I or G3BP-1 for 48 h. Cells was transfected with poly(I:C) for 3 h and then infected with VSV-GFP (multiplicity of infection = 1). Total cellular RNAs were extracted before viral infection and analyzed by real-time PCR to determine the IFN-β mRNA levels. Cell viability was assessed by using the MTT [3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyl-2H tetrazolium bromide] assay. Fluorescence microscopy images were taken at 8-h post-infection to visualize the viral replication.

Statistical significance was determined by using unpaired two-tailed Student's t-tests. P<0.05 was considered to be statistically significant.

We thank Junichi Tanaka (University of the Ryukyus, Okinawa, Japan) for generously providing hippuristanol, and we thank Dongyan Jin (University of Hong Kong, Pokfulam, Hong Kong) for generously providing VSV–GFP. We thank Chia-Yi Yu (National Defense Medical Center, Taipei, Taiwan.) and Yi-Ling Lin (National Defense Medical Center, Taipei, Taiwan.) for their helpful suggestion on IPS-1 expression, and we thank Meili Wei (Sun Yat-sen University, GuangZhou, China.) for assistance on IPS-1 purification.