Dance with the Devil: Stress Granules and Signaling in Antiviral Responses

Cells have evolved highly specialized sentinels that detect viral infection and elicit an antiviral response. Among these, the stress-sensing protein kinase R, which is activated by double-stranded RNA, mediates suppression of the host translation machinery as a strategy to limit viral replication. Non-translating mRNAs rapidly condensate by phase separation into cytosolic stress granules, together with numerous RNA-binding proteins and components of signal transduction pathways. Growing evidence suggests that the integrated stress response, and stress granules in particular, contribute to antiviral defense. This review summarizes the current understanding of how stress and innate immune signaling act in concert to mount an effective response against virus infection, with a particular focus on the potential role of stress granules in the coordination of antiviral signaling cascades.


Introduction
Viruses depend on the host translation apparatus to express viral proteins. By hijacking and redirecting ribosomes, translation factors, and RNA binding proteins (RBPs), viruses modulate the cellular translatome in favor of their needs and virus progeny production. In rare cases, viruses such as hantaviruses, myoviruses, haloarchaeal viruses, and giant viruses encode viral proteins that substitute for translation initiation, elongation, and termination factors, transfer RNAs (tRNAs), aminoacyl-tRNA synthetases [1][2][3][4][5][6], and-as lately identified via metagenome analysis-also ribosomal proteins [7]. Although the use of the host-encoded translation machinery saves coding capacity, the ensuing dependency on host factors to accomplish this critical step of the viral life cycle makes viruses vulnerable. To counteract virus reproduction, cells have evolved highly specialized stress sensors that detect viral products and actively suppress both host and viral translation. Hence, limiting the access, availability, or activity of the translation apparatus can be considered an effective defense mechanism and integral component of the antiviral response.
Immune sensors detect viral nucleic acids as non-self through specific features, termed pathogen-associated molecular patterns (PAMPs), among these incoming viral DNA and RNA genomes and viral replication intermediates. The activation of immune sensors elicits an antiviral

General Control Nonderepressible 2 (EIF2AK4)
GCN2 is usually activated under conditions of amino acid starvation through its histidyl-tRNA synthetase-related domain that binds uncharged tRNAs [155]. This domain also binds to viral RNA genomes in vitro, notably those of sindbis virus (SINV), poliovirus, and HIV-1, and in vivo studies confirmed that infection with SINV and HIV-1 leads to activation of GCN2, eIF2α phosphorylation, and translational inhibition of viral RNA [156,157]. Furthermore, HIV-1 induces protease-mediated cleavage of GCN2, indicating that GCN2 has an antiviral function that the virus tries to suppress [157]. This notion is supported by the observation that GCN2-deficient or GCN2-mutant mice are more susceptible to infection with SINV [156], murine cytomegalovirus, and human adenovirus [158].
In the future, it will be important to further characterize the mechanism behind GCN2 activation in virus-infected cells since this might also be an indirect consequence of virus-or IFN-γ-induced amino acid deprivation.

Heme-Regulated Inhibitor (EIF2AK1)
A direct implication of HRI in virus infection was reported only recently for its fish homologues. In flounder cells (Paralichthys olivaceus), an upregulation of HRI was observed upon infection with Scophthalmus maximus rhabdovirus and poly(I:C) treatment, at both the mRNA and protein levels [159]. Furthermore, overexpression of HRI in orange-spotted grouper (Epinephelus coioides) resulted in an increase in eIF2α phosphorylation and inhibition of red-spotted-grouper nervous necrosis virus (RGNNV) replication [160]. Both studies suggested that HRI might have a similar function to PKR in fish.
Normally, HRI is activated by high intracellular levels of reactive oxygen species (ROS) and other imbalances of redox homeostasis, e.g., those induced by arsenite [161]. Several viruses interfere with Viruses 2020, 12, 984 7 of 47 mitochondrial, peroxisomal, and ER functions and can thereby lead to ROS production, as observed during infection with influenza virus [162], flaviviruses such as DENV, WNV, and JEV [163], as well as chronic viruses such as HCV and hepatitis B virus [164][165][166][167], HIV [168], and poxviruses [169]. In many cases, ROS production represents a necessary event for viral replication or other processes such as capping [170,171]. At the same time, viruses manipulate the antioxidative defense system to maintain ROS levels in a range that is optimal for their purpose, without inducing cell death [172]. However, in contrast to fish, there are no reports, to our knowledge, showing an activation and implication of HRI upon virus infections in mammalian cells.

Stress Kinases-Mediators of Innate Immune Signaling
Beside their function in translational control, stress kinases were reported to be involved in a multitude of immune modulatory processes. In the following section, we summarize the currently known connections between the immune and stress sensing pathways.
Notably, all these signaling pathways converge on the activation of the transcription factors IFN-regulatory factor (IRF) 3 and 7 by the TANK-binding kinase 1 (TBK1) and nuclear factor-κ-B (NF-κB) by the IκB kinase (IKK) complex [9,10,178,179]. Nuclear translocation of IRF3/7 phosphorylated forms leads to the induction of type I IFN genes. Upon secretion, IFNs act in an autocrine and paracrine manner, engage with IFN receptors and induce Janus kinase/signal transducers and activators of transcription (JAK/STAT) signaling [180], leading to the transcriptional activation of hundreds of IFN-stimulated genes (ISGs) with antiviral activity [181], including PKR. Thereby, type I IFNs establish an antiviral state that is critical for containment of viral infections. In parallel, phosphorylation of IκB by the IKK complex results in its dissociation from NF-κB and its degradation. In turn, NF-κB translocates into the nucleus where it initiates transcription of numerous pro-inflammatory cytokines and chemokines (e.g., interleukin (IL)-1, IL-6, IL-8, tumor necrosis factor α (TNF-α)), as well as anti-apoptotic proteins (e.g., Bcl-2, cFLIP, Fas) that promote cell survival [182,183].
The expression of cytokines and IFNs is further controlled at the posttranscriptional level via mitogen-activated kinases Erk1/2 and p38 MAPK signaling, through the downstream kinases Mnk1/2 and MK2. The phosphorylation of eIF4E by Mnk1/2, for instance, upregulates translation of IκB and IRF-1 [184,185]. Moreover, MK2-dependent phosphorylation of Tristetraprolin (TTP), an RBP that binds to AU-rich elements (AREs) located in the 3 UTR of many cytokine and IFN mRNAs, prevents rapid degradation of these mRNAs and further promotes their translation [186][187][188].

Impact of Stress Kinases on Innate Immune Signaling Pathways
IFN and cytokine production is not only regulated by the abovementioned immune sensing pathways. Several lines of evidence indicate that eIF2α-kinases modulate immune signaling pathways at several levels ( Figure 1).
A first mechanism is a direct consequence of translational repression by the eIF2α-kinases, which lowers the levels of key regulatory proteins, especially labile proteins such as IκB, A20, and SHIP-1, leading to elevated activity of NF-κB and IRF3 [189][190][191][192]. A second mechanism involves CHOP, a central uORF-regulated transcription factor induced during the ISR. CHOP causes transcriptional inhibition of peroxisome proliferator-activated receptor γ, which, in turn, is a negative regulator of NF-κB transcriptional activity [193]. Thereby, CHOP indirectly augments NF-κB activity under stress conditions. Furthermore, PKR-mediated translation inhibition was found to be required for full activation of the stress-activated JNK upon poly(I:C) transfection [194].
Independently of its function in translational control, PKR orchestrates a variety of immune and survival pathways, thereby influencing cell fate decisions. For instance, PKR has been implicated in NF-κB activation by directly phosphorylating the NF-κB inhibitor IκB [195]. However, later reports indicated that PKR activates NF-κB signaling indirectly, either in a translation-dependent manner as described above, or by providing a signaling platform via its dsRBD, which recruits various signaling molecules and allows PKR to function as a scaffold independently of its kinase activity. Accordingly, PKR was shown to interact with the β subunit of the IKK complex [196][197][198] as well as with TNF receptor-associated factor (TRAF) 2, TRAF5, and TRAF6 [199]. Additionally, PKR was found to directly interact with components of the RIG-I/MDA5 signaling pathway. For instance, during the very early response to HCV infection, PKR interacts with MAVS and TRAF3, thereby inducing ISG15. Subsequent ISGylation of RIG-I interferes with RIG-I activation and thus, limits IFN induction [200]. In contrast, DHX36-mediated activation of PKR in response to IAV and NDV infection promotes RIG-I activation [201]. PKR was also reported to associate with MDA5 and to stimulate IFN-β production via the MAVS-IRF3/7 signaling cascade in response to VACV infection [202]. In addition, the interaction of PKR with MAVS is suggested to promote PKR activation in response to dsRNA [202,203]. cGAS together with G3BP1 was shown to form a complex with PKR, which is important for cGAS activation and IFN production upon dsDNA exposure. Vice versa, PKR was also activated within the complex [204]. Interactions of PKR with immune-related transcription factors, including STAT1 [205] and STAT3 [206], have also been reported. Finally, PKR coordinates IKK, JNK, and p38 MAPK activity in response to pro-inflammatory stimuli such as TNF-α and IL-1 [207,208].
Another mode by which PKR affects immune signaling is through binding to cis-acting regulatory elements in cytokine mRNAs. Binding of PKR to an element in the TNF-α 3 UTR was shown to promote TNF-α pre-mRNA splicing [209] and binding to the IFN-γ 5 UTR appears to suppress IFN-γ translation beyond the effect of PKR on global protein synthesis [210]. PKR has also been implicated in sustaining IFN-α/β mRNA integrity in response to a subset of RNA viruses that activate MDA5 specifically, but the exact mechanism still needs to be clarified [211].
Unlike those identified for PKR, links between the other stress kinases, namely GCN2, PERK, and HRI, and innate immune signaling pathways, are less well understood and are often not studied in the context of virus infections. In vitro, HRI mediates NF-κB activation by phosphorylating IκB [212]. Further to this, HRI was shown to be important in fish for the NF-κB-mediated immune response to RGNNV [160]. Reports about GCN2 focus on its ability to negatively regulate inflammatory responses or positively influence antigen-presentation [213][214][215]. Direct interactions with immune signaling components remain to be investigated. PERK acts as an activator of the JAK1/STAT3 signaling pathway in the context of neuroinflammation [216] and was shown to be important for poly(I:C)-induced TLR inflammatory signaling [217]. On the other hand, PERK activation was reported to induce the degradation of the type I IFN receptor IFNAR1, and thereby inhibits JAK/STAT-mediated production of many ISGs. Viruses such as HCV or vesicular stomatitis virus (VSV) take advantage of that and actively trigger IFNAR1 degradation via PERK activation [218,219]. Further investigations about a possible direct implication of HRI, GCN2, and PERK in innate immune signaling are needed.
Downstream of the eIF2α-kinases, signal transducers also link the ISR to the IFN signaling. ATF4, which is translationally activated upon eIF2α phosphorylation, directly interacts with IRF7, and affects IFN-α/β induction in response to viral infection. This interaction is bidirectional, since IRF7 upregulates ATF4 expression and activity, while ATF4 inhibits IRF7 activation [220]. Moreover, GADD34 expression was proposed to be induced by the MAVS-IRF3/7 pathway in response to poly (I:C) or infection with Viruses 2020, 12, 984 9 of 47 VSV [192]. Additionally, GADD34 is involved in negative feedback upon TNF-α activation: it recruits CUE domain-containing 2 (CUEDC2) to PP1 and thereby leads to the dephosphorylation of IKKα and β and a decrease in NF-κB activity [221]. cytokines. TBK1, on the other hand, phosphorylates IRF3/7, whose nuclear translocation mediates the transcriptional activation of IFN-α/β (left side). Cytosolic dsRNA that accumulates during viral replication is sensed by the stress kinase PKR. As a consequence, phosphorylation of eIF2α strongly represses the translation of most cellular mRNAs, while the translation of factors related to the stress response (ATF4, CHOP, and GADD34) or cell survival (XIAP and Bcl-2) is selectively favored (right side). The other eIF2α-kinases, GCN2, PERK, and HRI, contribute to translation suppression by detecting virus-induced changes in cellular homeostasis such as nutrient deprivation and accumulation of reactive oxygen species (ROS) or unfolded proteins in the ER. In the cytosol, untranslated mRNAs condense together with numerous RBPs and form SGs (1). Upon infection with certain viruses, innate immune sensors, PKR, regulators of stress and immune sensors, and interferon (IFN)-induced effectors localize in stress granules (SGs) together with viral RNA or DNA, forming a signaling platform (2) that coordinates and potentiates the antiviral response (3,4,5). PKR is at the crossroads of stress and innate immune signaling pathways: PKR interacts with RIG-I, promotes its activation, and amplifies the downstream signaling cascade (6). A similar interaction exists with cGAS (7). PKR promotes MDA5 filament formation, is activated by MDA5 to enhance downstream MAVS signaling (8), and, in turn, can also be activated by MAVS (9). PKR affects pro-inflammatory responses by interacting with TRAFs to activate the NF-κB signaling and potentially the JNK/p38 MAPK pathway leading to AP-1 activation (10), or directly regulates NF-κB activity via phosphorylation of IκB and IKK (11). Finally, PKR is involved in controlling the stability and translation of IFN mRNAs (12). Black arrows indicate signaling pathways; dashed red arrows indicate crossroads between stress response pathways, innate immune signaling, and SGs.
Altogether, the translational inhibition of unstable immune-regulatory proteins, as well as interactions of stress kinases or downstream signal transducers with components of the immune signaling cascade, promote the transcription of IFNs and pro-inflammatory cytokines. Hence, induction of the stress response, especially the activation of PKR, upon viral infections strongly contributes to establishing a pro-inflammatory and antiviral state. At the same time, PKR, PERK, ATF4, and GADD34, however, also initiate distinct negative feedback loops in order to fine-tune and possibly locally and temporally restrict IFN production to prevent an overactivation of the innate immune system.
An unsolved question is how antiviral proteins are synthesized when global translation is repressed. While uORFs facilitate translation of mRNAs under such conditions, as exemplified by ATF4 [222], most mRNAs encoding antiviral proteins do not appear to harbor uORFs. GADD34 is induced as a negative feedback regulator upon translational inhibition, leading to dephosphorylation of eIF2α and resumption of translation. Mathematic modeling in combination with flow cytometry experiments suggest that this negative feedback loop contributes to the stochastic expression of IFN in individual cells upon poly(I:C) treatment and VSV infection, whereby transcriptional induction of IFNs during translation-off states is followed by synthesis of IFNs during translation-on states [192,223]. Recently, two different laboratories made another important contribution to the understanding of how antiviral factors are preferentially produced in response to viral infections. They discovered that ribonuclease L (RNase L), which is activated in response to dsRNA, strongly depletes cellular mRNA pools through wide-spread mRNA degradation, but specifically leaves mRNAs encoding IFNs, cytokines, and other defense proteins intact [224][225][226]. The mechanism by which these mRNAs are excluded from RNase L-mediated cleavage remains unclear, but probably involves regulatory sequences, RNA secondary structures, and RBPs that protect individual mRNAs from degradation. Hence, it appears that the combination of increased transcription and selective mRNA stabilization and translation permits the synthesis of antiviral factors under conditions of repressed global translation.
An early link between the ISR, in particular the formation of SGs, and the temporal control of cytokine production comes from studies on the adaptive immune system. In naïve T helper cells, IL-4 mRNA was found to accumulate in SG-like foci in the cytoplasm during T cell priming, concomitant with elevated phosphorylation of eIF2α. The release of these mRNAs then allowed for the rapid production of cytokines during T cell restimulation [227], suggesting that SGs can exert storage and regulatory functions outside of classical stress conditions.

SGs as Immune Signaling Platforms in Antiviral Defense
While SGs assemble in response to translational shut-off, they are not necessary for translation suppression under stress conditions. Rather, SGs were proposed to function as (i) hubs for modulating local protein and mRNA concentrations; (ii) timers for the stress response, marking a "window of opportunity" during which the stress can be resolved or an apoptotic program will be initiated; (iii) joint assemblies of unfolded proteins and translation complexes that coordinate the activities of the protein synthesis and folding machineries; (iv) storage sites for pre-assembled initiation complexes, allowing for rapid resumption of protein synthesis when cells recover from stress; (v) signaling platforms that connect stress sensors with effectors of immune responses, especially in the context of viral infection [15,16,18,228,229].
Viruses deploy many different strategies to inhibit the formation of SGs, suggesting that SGs serve an antiviral function. In particular, flaviviruses have evolved numerous mechanisms to interfere with SG assembly. The core protein of JEV, for instance, interacts with Caprin1 to relocalize G3BP1 and USP10 to the perinuclear region, thereby preventing SG formation [230]. Other SG components seem to be sequestered by viral RNAs, as reported for TIA1 and TIAR, which bind to the genomic RNA 3 end of DENV, WNV, and tick-borne encephalitis virus [231,232]. At later stages of infection, flavivirus genomes are degraded by the cellular exoribonuclease XRN1 from the 5 end up to a compact pseudoknot structure in the 3 UTR, leading to the cytosolic accumulation of the remaining subgenomic flaviviral RNA (sfRNA) [233,234]. G3BP1, G3BP2, Caprin1, and USP10 are sequestered by sfRNAs, which leads to inhibition of ISG mRNA translation and thus, attenuation of the antiviral response [235]. A similar sequestration strategy was observed for SeV infection, during which trailer RNAs, i.e., small abortive products of viral genome replication, bind TIAR, and inhibit SG formation [236][237][238]. Our own work showed that flaviviruses uncouple the stress response from translation control by blocking both eIF2α phosphorylation and SG assembly, while protein synthesis is still suppressed [58]. In line with this observation, other laboratories reported that the expression of single viral proteins such as Zika virus (ZIKV) capsid, NS3, NS2B-3, or NS4A is sufficient to inhibit SG assembly [239][240][241]. The precise mechanism behind this inhibition remains to be uncovered. In the following, we will focus on the role of SGs within the complex signaling network that controls cellular reprogramming towards an antiviral state (Table 1 and Figure 2).

G3BP1 at the Interface between SGs and the IFN Response
G3BP1 turns out to be a target of particular importance for viruses, which aim to interfere with its function as a SG nucleator and regulator of immune responses. Moreover, G3BP1 appears to facilitate replication of many viruses. Several viral proteins were shown to directly recruit G3BP1 to sites of viral replication, e.g., HCV polymerase NS5B [271,[296][297][298], Chikungunya virus nsP3 [299,300], Semliki Forest virus and SINV Nsp3 [301,302], and Junín virus N protein [303]. Murine norovirus inhibits the formation of canonical SGs not only by redistributing G3BP1 together with the NS3 protein to sites of viral replication, but also by modifying the interactome of G3BP1 [304]. Notably, G3BP1 is directly involved in translation of the norovirus genome by associating with the VPg viral cap complex to help ribosome recruitment [305].  . Certain RBPs and cellular mRNAs (e.g., cytokine mRNAs) preferentially localize in SGs (C). Viral components and ITAFs that control viral gene expression are also detected in SGs (D). SG composition is highly stressand cell type-specific. Notably, many components have been detected in SGs under metabolic or environmental stress conditions, while an investigation in the context of viral infection is still missing. The localization and function of some SG components is dependent on the interaction with specific SG core proteins, indicated in turquoise.

G3BP1 at the Interface between SGs and the IFN Response
G3BP1 turns out to be a target of particular importance for viruses, which aim to interfere with its function as a SG nucleator and regulator of immune responses. Moreover, G3BP1 appears to facilitate replication of many viruses. Several viral proteins were shown to directly recruit G3BP1 to sites of viral replication, e.g., HCV polymerase NS5B [271,[296][297][298], Chikungunya virus nsP3 [299,300], Semliki Forest virus and SINV Nsp3 [301,302], and Junín virus N protein [303]. Murine norovirus inhibits the formation of canonical SGs not only by redistributing G3BP1 together with the NS3 protein to sites of viral replication, but also by modifying the interactome of G3BP1 [304]. Notably, G3BP1 is directly involved in translation of the norovirus genome by associating with the VPg viral cap complex to help ribosome recruitment [305].
In recent years, G3BP1 was also found to promote both the inflammatory and the IFN responses. Overexpression of G3BP1 was shown to induce the formation of SGs to which innate immune factors such as PKR, oligoadenylate synthetase (OAS) 2, and RNase L are recruited, which, in turn, promote the production of certain cytokines such as IL-17, MIP-3a/b, and MCP-5 via activation of the NF-κB and JNK pathways [256]. Independently of its role in SG assembly, G3BP1 was also shown to interact with the RNA sensor RIG-I and positively regulates its downstream signaling [306,307], as well as with the DNA sensor cGAS, thereby enhancing IFN-β production [308]. This might be the reason why several members of the Picornaviridae and related viruses have evolved G3BP1-cleaving proteases, as reported for the Leader protein of foot-and-mouth disease . Certain RBPs and cellular mRNAs (e.g., cytokine mRNAs) preferentially localize in SGs (C). Viral components and ITAFs that control viral gene expression are also detected in SGs (D). SG composition is highly stress-and cell type-specific. Notably, many components have been detected in SGs under metabolic or environmental stress conditions, while an investigation in the context of viral infection is still missing. The localization and function of some SG components is dependent on the interaction with specific SG core proteins, indicated in turquoise.
In recent years, G3BP1 was also found to promote both the inflammatory and the IFN responses. Overexpression of G3BP1 was shown to induce the formation of SGs to which innate immune factors such as PKR, oligoadenylate synthetase (OAS) 2, and RNase L are recruited, which, in turn, promote the production of certain cytokines such as IL-17, MIP-3a/b, and MCP-5 via activation of the NF-κB and JNK pathways [256]. Independently of its role in SG assembly, G3BP1 was also shown to interact with the RNA sensor RIG-I and positively regulates its downstream signaling [306,307], as well as with the DNA sensor cGAS, thereby enhancing IFN-β production [308]. This might be the reason why several members of the Picornaviridae and related viruses have evolved G3BP1-cleaving proteases, as reported for the Leader protein of foot-and-mouth disease virus, Theiler's murine encephalomyelitis virus (TMEV) and mengovirus [280,309,310], the proteinase 3C of poliovirus, coxsackievirus B3 and encephalomyocarditis virus (EMCV) [245,[311][312][313], as well as the 3C-like proteinase NS6 of feline calicivirus [314]. Hence, G3BP1 and SGs seem to be at the nexus of the stress response and innate immune signaling. The role of SGs as a signaling platform of the antiviral response will be described in the paragraph below.

SGs, a Platform to Initiate IFN Signaling?
In cells infected with IAV lacking the NS1 protein (IAV∆NS1) and thus, unable to counteract PKR, viral RNA was found to co-localize in SGs together with the immune sensors RIG-I and MDA5 as well as other antiviral effectors such as OAS, RNase L, and PKR. A similar redistribution of RIG-I was also observed upon infection with EMCV, adenovirus, and SINV. In these infection models, the inhibition of SG assembly by depletion of essential SG components resulted in a strong attenuation of IFN production and an increase in viral replication. Based on these results, Onomoto et al. proposed the term "antiviral SGs" (avSGs) to indicate that SGs may serve as a platform for the activation of innate immune sensors by viral RNA and the subsequent initiation of the IFN response [242]. While most RNA viruses are sensed by RIG-I alone or by RIG-I and MDA5 together [315], MDA5 is essential for IFN-α/β induction in response to Picornaviridae infection [316,317]. In agreement with the idea of avSGs, MDA5 and dsRNA were found to locate in EMCV-induced SGs at early times post infection. At later stages of EMCV infection, cleavage of G3BP1 by the EMCV protease 3C resulted in SG dissolution concomitant with a reduced IFN-β and cytokine response, which could be rescued expressing a cleavage-resistant G3BP1 mutant [245]. The characterization of avSGs formed upon infection with SeV, again revealed the presence of many antiviral components and SG formation was shown to be important for IFN-β production and viral restriction [263]. Furthermore, late during NDV infection, uncapped viral RNA(+) derived from read-through transcription was found to accumulate together with RIG-I in SGs and trigger RIG-I activation. Consistent with the idea of avSGs, IFN-β induction could be reduced by the disruption of SGs in this model [243].
In contrast to these findings, some studies suggest that SG formation may not be necessary for IFN production upon virus infection. During infection with a mutant mengovirus lacking the Leader protein, for instance, localization of MDA5 to SGs was not a prerequisite for IFN induction. However, it may be important to note that viral dsRNA could not be detected within SGs in this infection model, which may indicate that SGs adopt an antiviral function as platforms for RIG-I or MDA5 activation only when they accumulate viral RNA [95]. Another study showed that disruption of SGs in IAV∆NS1and SeV-infected cells did not reduce but actually increased RIG-I-mediated IFN-β production [202], which is in direct contradiction with previous observations [242,263]. It is therefore still unclear whether the changes observed in the IFN response ensue from the disruption of SGs as a coordinating platform or are due to the immunomodulatory function of single SG components. Additional experiments based on genetic ablation of other SG components, interference with SG assembly by cycloheximide or ISRIB treatments, or the use of eIF2α S51A mutant cells might be necessary to answer this question Finally, MDA5 and RIG-I were also detected in SGs formed under other stress conditions such as heat shock and arsenite treatment [95,242], which raises the possibility that MDA5/RIG-I can also assemble with endogenous (e.g., damaged) RNAs in SGs, or that these sensors passively phase separate in SGs.
Whether SGs contribute to the antiviral response upon infection with DNA viruses is even less clear. Upon DNA virus infection, viral DNA is sensed by specialized DNA sensors, among those cGAS. In addition, viral transcripts, cytosolic noncoding RNAs transcribed by the RNA polymerase III, as well as Alu-derived host RNAs can activate MDA5 or RIG-I [318]. In the case of VACV∆E3L, MDA5-mediated IFN-β production was unchanged upon SG dissolution, and rather dependent on the immunomodulatory function of PKR [202]. HSV-1 infection induces formation of SGs, in which dsRNA (whose origin is unclear) was detected. In this infection model, IFN production was induced by the DNA sensing cGAS-STING pathway rather than by MDA5/RIG-I, and preceded PKR activation and SG formation, indicating that SGs are dispensable for cGAS-mediated IFN production upon HSV-1 infection [121]. Interestingly, cGAS was reported to associate with dsDNA, G3BP1, and PKR in a complex and one study claimed the presence in cytoplasmic foci upon dsDNA transfection [204,308]. While it is not clear if these foci represent canonical SGs, they were proposed to be essential for cGAS activation and downstream IFN-β production [204]. Of note, polyglutamine binding protein 1 (PQBP1), identified as a co-sensor of cGAS in the context of HIV infection [319], was shown to localize to SGs [70,255]. Hence, SGs might support cGAS-mediated IFN production under specific conditions or for particular virus types. However, this needs to be further investigated.
Given these discrepancies, the link between SGs and innate immune signaling appears to be complex. Besides the co-localization of innate immune sensors with viral RNAs or DNAs within SGs, the type of virus and the presence or absence of regulatory factors are likely to affect the potential of SGs to serve as platforms initiating the IFN response. It should further be noted that studies so far have mainly focused on the impact of SGs on the production of IFN-β or cytokines such as IL-6, RANTES, and CXCL10. Hence, it would be interesting to investigate how SGs might influence the expression of other IFN types and subtypes, and how this could affect virus replication.

Regulators of the Innate Immune Sensors
RLR activity is tightly controlled by multiple posttranslational modifications and protein interactions. Tripartite motif protein 25 (TRIM25) and Riplet are two examples of E3 ubiquitin ligases, which, according to a current model, mediate consecutive K63-linked ubiquitination of RIG-I. The first ubiquitination event mediated by Riplet leads to the exposure of the RIG-I CARD domains and their subsequent ubiquitination by TRIM25. Thereby RIG-I gets activated, oligomerizes, and initiates downstream MAVS signaling [320]. Both E3 ubiquitin ligases were found to be recruited together with RIG-I to SGs when cells were exposed to poly(I:C) [249]. It is not clear, however, whether co-localization of these proteins within SGs is important for the sequential ubiquitination of RIG-I and downstream signaling events. Using a bimolecular fluorescence complementation assay and super-resolution microscopy in SeV-infected cells, RIG-I was identified in two distinct complexes-RIG-I/TRIM25 and RIG-I/MAVS-with different cellular localization [248]. While TRIM25/RIG-I complexes localized in SGs, RIG-I/MAVS complexes remained attached to mitochondrial membranes. An interaction between MAVS and TRIM25 was barely observed [248], contrasting previous studies in which TRIM25 was proposed to act as an ubiquitin ligase of MAVS [321,322]. RIG-I/TRIM25 complexes further appeared to be in close proximity with mitochondria upon virus stimulation [248], consistent with previous reports about physical contacts between avSGs and MAVS [242,243]. These findings suggest that RIG-I first needs to interact with TRIM25 within SGs to become ubiquitinated and activated before it is released from the complex to activate MAVS at mitochondria. It should be noted that TRIM25 also has a second function: when bound to FAT10, it stabilizes the protein, which acts as a negative regulator of RIG-I and prevents the formation of avSGs under conditions of inflammation [323]. Recent in vivo studies suggest that Riplet alone is sufficient for RIG-I activation [324,325], and confirmed a role of Riplet in modulating the kinetics of RIG-I recruitment into SGs. However, the ability of Riplet to activate RIG-I signaling was not dependent on SG formation [324]. Apart from Riplet and TRIM25, other E3 ubiquitin ligases can mediate RIG-I ubiquitination including MEX3C, an RNA-binding E3 ligase that was found to localize in NDV-induced avSGs and plays an essential role in RIG-I-mediated IFN signaling [244]. Moreover, RIG-I activity is regulated by the activity of several kinases, phosphatases, and acetyl transferases, some of which localize to SGs such as histone deacetylase 6 (HDAC6) [252], PKC-α [251], and the α-subunits of CKII [250].
MDA5 activity is similarly regulated by posttranslational modifications [326]. RIO kinase 3 (RIOK3) is involved in maintaining MDA5 inactive under normal conditions by phosphorylating its C-terminal domain, thereby interfering with MDA5 filament formation on dsRNA and downstream IFN signaling. Upon poly(I:C) exposure, RIOK3 was shown to be partially recruited to SGs, together with MDA5 [246]. DNAJ heat shock protein family (Hsp40) member B1 (DNAJB1), a member of the Hsp40 family, was identified as an interactor of MDA5 that negatively regulates MDA5-MAVS signaling. Upon poly(I:C) treatment, DNAJB1 in conjunction with Hsp70 was found to co-localize with SG markers and the mitochondrial membrane, negatively affecting MDA5 oligomerization and MAVS aggregation [247]. If and how the co-localization of MDA5 together with the DNAJB1-Hsp70 complex or RIOK3 within SGs affects MDA5-MAVS signaling still needs to be investigated.
Laboratory of genetics and physiology 2 (LGP2, DHX58) belongs to the RLR family and acts as a modulatory co-receptor of RIG-I and MDA5 [315]. LGP2 has been shown to localize in SGs upon IAV∆NS1 infection [242]. Moreover, the positive activator of PKR (PACT, PRKRA) was identified as an LGP2 interactor essential for RLR regulation [327], and also shown to localize in SGs upon treatment with hippuristanol [257], a steroid compound that interferes with cap-dependent translation [270]. Interestingly, PACT was also found in a complex with RIG-I and MDA5, positively regulating their downstream signaling [328,329]. In addition, two members of the pumilio RBP family, PUM1 and PUM2, localized to SGs and were found to associate with LGP2 upon NDV infection, enhancing its dsRNA binding ability through conformational changes and thereby increasing the IFN response [253]. Whether co-localization of LGP2 with PACT, PUM1, and PUM2 in SGs is necessary for the regulatory effects on RIG-I and MDA5 activity is currently not clear.

Stress Kinase PKR and Its Regulators
Localization of the dsRNA-sensor PKR in SGs is a prime example of their functional relevance [121,256,330]. PKR is recruited to SGs through its direct interaction with the NTF2 and PXXP domains of G3BP1 [330]. By complex formation with G3BP1 and Caprin1, PKR can be activated in a dsRNA-independent manner. The high local concentrations of all three components in SGs are likely what drive formation of the tripartite G3BP1/Caprin-1/PKR complex. Activated PKR is subsequently released from SGs into the cytosol, where it phosphorylates eIF2α and thereby promotes both translation repression and SG persistence [330]. Hence, SGs are part of a feed-forward loop that amplifies PKR activity and restricts viral replication. This amplification mechanism might explain why so many viruses simultaneously target PKR and SG assembly. One may speculate that the ubiquitin-specific peptidase USP10, which binds to G3BP1 and inhibits SG formation [331], could affect PKR activation and other antiviral signaling events in SGs.
In contrast, activation of PKR by MAVS was reported to occur outside of SGs prior to its recruitment into SGs [203]. This finding corroborates the idea that PKR is activated in sequential steps and that SGs help to maintain PKR activity at later stages of the infection. Interestingly, PKR was also reported to localize both inside and at the outer membrane of mitochondria [99]. Since SGs and mitochondria are often found in close proximity [248], it is tempting to speculate that PKR shuttles between the two compartments. Accordingly, recruitment of PKR to SGs could affect not only the activation status of PKR within protein complexes inside SGs, but also within protein complexes that specifically form inside or at mitochondria. This scenario is well in line with the tight collaboration between SGs and mitochondria in the RIG-I/MAVS activation pathway described above (Figure 3).
Besides G3BP1 and Caprin1, PACT, transactivation response element RNA-binding protein (TRBP), and P58 IPK modulate PKR activity by directly targeting its kinase domain. It is generally thought that PACT stimulates PKR whereas TRBP and P58 IPK inhibit PKR activity, although exceptions have been reported and may be cell-context specific [332][333][334][335][336]. The interaction between PKR, PACT, and TRBP is especially important for cellular physiology during viral infections and has implications in stress recovery and the RNA-induced silencing pathway [335,337,338]. However, it is currently unclear whether the recruitment of PKR into SGs affects its interaction with PACT, TRBP, and P58 IPK . Upon hippuristanol treatment, PACT localizes in SGs where it was proposed to form a complex with argonaute 2 (AGO2) [257]. Furthermore, the SG proteins TIA1 and TIAR were found to suppress PKR activity, although they seem to exert this effect by controlling PACT pre-mRNA splicing, independently of their role in SG assembly [336]. In addition to PACT [257], several modulators of PKR activity, including NFAR1/2 [258,259], adenosine deaminase RNA-specific 1 (ADAR1) [77,262] and Staufen [260], were found to localize in SGs or modulate SGs formation. Further studies are needed to assess whether these PKR regulators relocalize into, or are specifically excluded from, SGs upon virus infection. The differential activation of PKR within different complexes and cellular compartments is likely to dictate the spectrum of PKR targets and thereby decides whether PKR has pro-apoptotic, anti-apoptotic, inflammatory, or translational effects.
(ADAR1) [77,262] and Staufen [260], were found to localize in SGs or modulate SGs formation. Further studies are needed to assess whether these PKR regulators relocalize into, or are specifically excluded from, SGs upon virus infection. The differential activation of PKR within different complexes and cellular compartments is likely to dictate the spectrum of PKR targets and thereby decides whether PKR has pro-apoptotic, anti-apoptotic, inflammatory, or translational effects.

Oligoadenylate Synthase and RNase L
Proteins of the OAS family are IFN-induced effectors that are activated by dsRNA binding [340]. OAS proteins polymerize ATPs into 2′-5′-linked oligoadenylates (2-5A), whose lengths vary from dimers to 30-mers [341]. Trimeric or longer 2-5A molecules serve as unique secondary messengers to activate RNase L, which induces RNA degradation [342]. Studies indicate the presence of several proteins of the OAS family within SGs, including OAS1, OAS2, and OAS-like protein (OASL), as well as its mouse homolog OASL1 [159,242,256,267]. OASL, in contrast to the other members of the OAS family, lacks 2-5A synthetase activity but instead, contains ubiquitin-like (UBL) domains [343,344]. OASL was found to co-localize with RIG-I in SGs upon SeV infection and enhances RIG-I-signaling via its UBL domains, mimicking K63-linked ubiquitination. A similar function was suggested for one of the mouse homologues, OASL2 [266]. Interestingly, in the case of DNA virus infections, OASL/OASL2 had the opposite effect, negatively regulating cGAS signaling [345]. OASL1, the second mouse OASL homologue, was similarly shown to be recruited to IAV-induced SGs and to interact with many SG components, including PKR and MDA5 [267]. OASL1 was proposed to act both as a positive and negative regulator of immune signaling. By binding to viral RNA, OASL1 supports MDA5-mediated IFN production early in infection, while its binding to IRF7 mRNA prevents IRF7 synthesis and thereby dampens transcriptional activation of IFN at later time points. However, while SGs are supposed to be important for the activation of the MDA5-MAVS signaling, translational control of IRF7 was not affected by the absence of SGs [267]. Whether the localization of the active 2-5A synthetases OAS1 and OAS2 within SGs has an impact on their function still needs to be clarified.
RNase L activation upon binding of 2-5A results in the cleavage of viral and cellular ssRNAs into small fragments with 5′-hydroxyl and 3′-monophosphate ends [346]. These fragments, which

Oligoadenylate Synthase and RNase L
Proteins of the OAS family are IFN-induced effectors that are activated by dsRNA binding [340]. OAS proteins polymerize ATPs into 2 -5 -linked oligoadenylates (2-5A), whose lengths vary from dimers to 30-mers [341]. Trimeric or longer 2-5A molecules serve as unique secondary messengers to activate RNase L, which induces RNA degradation [342]. Studies indicate the presence of several proteins of the OAS family within SGs, including OAS1, OAS2, and OAS-like protein (OASL), as well as its mouse homolog OASL1 [159,242,256,267]. OASL, in contrast to the other members of the OAS family, lacks 2-5A synthetase activity but instead, contains ubiquitin-like (UBL) domains [343,344]. OASL was found to co-localize with RIG-I in SGs upon SeV infection and enhances RIG-I-signaling via its UBL domains, mimicking K63-linked ubiquitination. A similar function was suggested for one of the mouse homologues, OASL2 [266]. Interestingly, in the case of DNA virus infections, OASL/OASL2 had the opposite effect, negatively regulating cGAS signaling [345]. OASL1, the second mouse OASL homologue, was similarly shown to be recruited to IAV-induced SGs and to interact with many SG components, including PKR and MDA5 [267]. OASL1 was proposed to act both as a positive and negative regulator of immune signaling. By binding to viral RNA, OASL1 supports MDA5-mediated IFN production early in infection, while its binding to IRF7 mRNA prevents IRF7 synthesis and thereby dampens transcriptional activation of IFN at later time points. However, while SGs are supposed to be important for the activation of the MDA5-MAVS signaling, translational control of IRF7 was not affected by the absence of SGs [267]. Whether the localization of the active 2-5A synthetases OAS1 and OAS2 within SGs has an impact on their function still needs to be clarified.
RNase L activation upon binding of 2-5A results in the cleavage of viral and cellular ssRNAs into small fragments with 5 -hydroxyl and 3 -monophosphate ends [346]. These fragments, which tend to form duplex structures, can activate RIG-I/MDA5-MAVS signaling pathways and enhance the response to virus infection [347,348]. Recently, RNase L was reported to localize in SGs upon transfection of 2-5A. The resulting RNase L cleavage products were found to activate PKR, hence triggering SG formation and subsequent induction of IRF3-mediated IFN response [263]. Contradictory to these findings, earlier reports indicate that RNase L activity reduces PKR protein levels and eIF2α phosphorylation via destabilizing PKR mRNA [349]. In addition, RNase L was shown to inhibit SG assembly or reduce the size of SGs, likely through the global decay of cellular mRNAs, while antiviral mRNAs (e.g., IFN-β mRNA) escape degradation [224][225][226]. Hence, depending on the cell type and the stage of infection, RNase L can apparently enhance or restrict SG formation and the SG-associated antiviral response. In turn, SGs may potentially serve as platforms for interactions between RNase L and viral RNAs, thus enhancing the antiviral activity of RNase L.

Editing of dsRNA by Adenosine Deaminase
ADAR1 is an IFN-induced RNA editing enzyme that catalyzes the C6 deamination of adenosine (A) to inosine (I) within dsRNAs [350]. In cellular RNAs, A-to-I editing weakens RNA duplex structures [351], thereby avoiding the activation of PKR, RIG-I, and MDA5 by self-dsRNAs, most importantly Alu-derived dsRNAs [352]. ADAR1 is recruited to SGs in response to different stresses, IFN treatment, poly(I:C) transfection, and MV and HCV infection [77,245,262,353,354]. Considerable evidence supports the notion that ADAR1 acts as an immune suppressor by promoting the survival and replication of RNA viruses. Wild type MV infection causes induction of IFN-β and subsequent activation of ADAR1. Infection with C-deficient MV, which is not able to counteract PKR, induces SGs in ADAR1 knockout but not in ADAR1-sufficient cells, indicating that ADAR1 antagonizes SG formation [353][354][355]. Thus, ADAR1 sustains viral replication through inhibition of both SG formation and IFN production. However, this proviral effect of ADAR1 activity appears to be virus-specific, since ADAR1 was also found to work as an antiviral immune modulator against HCV infection [356]. Therefore, ADAR1 seems to be a regulator for both SG assembly and antiviral immune responses, yet it remains to be determined whether the recruitment of ADAR1 to SGs influences the decision of whether ADAR1 acts in a pro-or antiviral manner.

Zinc-Finger Antiviral Protein
Zinc-finger antiviral protein (ZAP or PARP13) is an IFN-induced effector protein, which is expressed in two isoforms (ZAP-S and ZAP-L), both lacking poly(ADP-ribose) polymerase (PARP) activity [357]. ZAP is recruited to SGs and participates together with other family members in maintaining SG integrity [264]. In addition, it has a range of different antiviral functions, depending on the factors it interacts with. Acting as an RNA sensor, ZAP was shown to bind to murine leukemia virus (MLV) transcripts, mediating their degradation through recruitment of the RNA exosome machinery. Interestingly, exosome components such as EXOSC5 were found to localize within RNA granules (SGs and PBs) in a ZAP-dependent manner. Similarly, MLV transcripts were recruited [358]. However, it is not clear if the co-localization of these factors into granules is important for viral restriction. Another study found that ZAP was similarly recruited to SINV-induced SGs together with viral RNA. ZAP mutants revealed that recruitment to SGs correlates with ZAP antiviral activity. However, this study also indicated that ZAP interaction partners are important for antiviral function [265]. TRIM25, as an example, was shown to be an essential co-factor for the ability of ZAP to block SINV translation [359]. Beside the exosome, ZAP was further shown to interact with RNA-induced silencing complex (RISC) components such as AGO2, repressing the miRNA ability against antiviral factors, hence mediating an increase in many ISGs [360]. In addition, ZAP-S was reported to interact with RIG-I to strengthen IFN signaling upon IAV and NDV infections [361]. Contradictory to this study, upon knockdown of ZAP-S and exposure with poly(U/UC) RNA, others reported an increase in IFN-β, IFN-λ2, and IFN-λ3 mRNA levels, indicating that ZAP-S has a function in the degradation of those mRNAs, therefore rather in the resolution of the antiviral response. In addition, they claimed that the different cellular localization of both isoforms (with ZAP-S excluded from sites of viral replication) has an impact on their binding preference for host or viral RNA [362]. It would be interesting to investigate how the co-localization of ZAP-S/ZAP-L target RNAs and interactors within SGs shapes ZAP-dependent antiviral functions.

Other DEAD/H-Box Proteins
DEAD/H-box proteins (DDX, DHX) are a large family of helicases involved in multiple cellular processes, which cover nearly all aspects of RNA metabolism [363,364] and include the RNA sensors RIG-I (DDX58) and MDA5 (RH116). DEAD/H-box proteins can be engaged in cytosolic viral RNA sensing, promote or inhibit IFN and cytokine signaling, function as co-factors for other immune sensors and regulators, and control viral replication by protein/protein and protein/RNA interactions. Whether these proteins promote or inhibit viral infections seems to be virus-specific and depend on the cellular context. The helicases DDX1, DDX3, and DDX21, for instance, are frequently required for efficient cytokine and IFN production, whereas DDX19 has been described as a negative regulator of type I IFN production in the context of viral infections [20]. DHX36 (RHAU) was shown to interact with PKR in an RNA-dependent manner, leading to activation of PKR, subsequent SG assembly, and enhanced RIG-I signaling upon IAV∆NS1 and NDV infection [201]. DHX30 was shown to exert an antiviral role by directly interacting with and stimulating ZAP activity against MLV [365]. Proteome analyses of purified SGs induced by oxidative stress confirmed the presence of multiple DEAD/H-box proteins inside SGs, including DDX1, DDX2 (eIF4A), DDX3, DDX6 (Rck), DDX19 (Dbp5), DDX21, DDX47, and DDX50 as well as DHX30 and DHX36 [70]. Localization in SGs could also be confirmed by immunofluorescence microscopy for DDX1, DDX2, DDX3, DDX6, DDX19, and DHX36 [201,[268][269][270][271][272][273][274][275][276][277]. Currently, it is not well understood if the immune regulatory functions of these proteins are affected by their localization in SGs, and localization in SGs during viral infections has so far only been addressed for DDX3, DDX6, and DHX36 [201,271,274,366].
For the RNA helicase DDX3, the connection to viral infections and SG assembly has been studied extensively [367]. Furthermore, HDAC6-mediated deacetylation of DDX3 is needed for the maturation of SGs upon stress, a process by which SGs fuse and grow in size [368]. The ability of DDX3 to induce SG assembly was demonstrated to depend on DDX3 s interaction with eIF4E. A DDX3 mutant with impaired eIF4E binding instead did not only show reduced SG assembly but also reduced cell survival upon stress [272]. In the context of HCV infection, recruitment of DDX3 into SGs has a proviral effect. By recruiting DDX3 into SGs, HCV can prevent DDX3 from exerting its positive effect on the translation of PACT mRNA [366]. Moreover, DDX3 was shown to recruit IKKα into SGs concomitant with activation of IKKα. This activation, however, did not induce pro-inflammatory NF-κB signaling, but enhanced proviral lipogenic gene expression [271]. Future studies will have to show how DDX3 mediates IKKα activation in SGs, if this mechanism operates also upon infection with other viruses, and whether this mode of IKKα activation can also exert antiviral effects via the canonical NF-κB signaling.
Finally, DDX6 was shown to co-localize with RIG-I in SGs upon infection with an NS1-deficient Influenza B virus, and promote IFN-β induction [274]. However, the effect of DDX6 on IFN induction seems to be independent of SGs since it was also observed in cells infected with wild type influenza B virus, which inhibits SG formation [274]. Taken together, experimental data for modulation of DEAD/H-box protein activities via recruitment to SGs, especially in the context of viral infections, are scarce and await further investigation.

Antiviral SG Functions Beyond
In addition to providing an immune signaling platform, SGs have been connected to other antiviral and pro-survival effects (Figure 2). SGs were suggested to contribute to an antiviral state by (i) sequestration of viral RNAs, (ii) clearing viral components through granulophagy, (iii) sequestering RBPs together with viral and cellular mRNAs to modulate gene expression programs, (iv) acting as signaling hubs to coordinate a general stress response, and (v) controlling the initiation of apoptotic programs.

Sequestration of Viral RNAs in SGs or SG-Like Structures
In many cases, SGs appear to sequester viral mRNAs and/or viral genomes to prevent viral replication. Evidence comes from studies using mutant viruses that are no longer able to hinder SG assembly or counteract PKR, as well as from viruses that per se cannot prevent SG assembly. Examples of mutant viruses, for which viral RNAs were found to relocalize in SGs, are VHS-1-deficient HSV-1 and NS1-deficient IAV [121,242,369]. Furthermore, HIV-1 nef mRNA was observed to localize in SGs in a Sam68 mutant cell background [370].
The formation of "antiviral granules" (AVGs) that form in close proximity to viral replication factories and silence viral mRNAs was observed in cells infected with VACV and rabies virus. AVGs form upon PKR activation by viral dsRNA and contain several canonical SG markers, including G3BP1, TIA1, Caprin1, and USP10. Unlike SGs, however, AVGs do not dissolve upon cycloheximide treatment [371][372][373], suggesting that their components are not in exchange with polysomes. Enrichment of viral mRNAs in AVGs is associated with their silencing and the inhibition of viral protein translation. The inhibition of AVG formation by knockdown or knockout of TIA1 instead resulted in increased viral replication [372,373]. Similar cytosolic granules were also observed during infection with TGEV and were found to accumulate both TGEV genomic and subgenomic RNAs together with TIA1 and the polypyrimidine tract-binding protein (PTB). The occurrence of these granules was supposed to either spatiotemporally control viral gene expression or to restrict viral translation and hence, infection [281].
HPIV3 is an example where canonical SGs are induced early during infection, leading to sequestration and silencing of viral mRNAs. However, SG assembly is suppressed at later stages of infection by the HPIV3 proteins N and P, and viral mRNAs are shielded in inclusion bodies for efficient replication [120]. EMCV infection is another example where SGs are induced transiently and sequester EMCV RNA [245]. While sequestration of viral RNAs in SGs is generally thought to represent an antiviral strategy, it is also possible that in some cases, viral replication might benefit from the concentration of viral RNAs in SGs. Given that many viruses including poliovirus, infectious bronchitis virus, mengovirus, and TMEV efficiently prevent the recruitment of viral RNAs into SGs [280,[374][375][376], it is reasonable to postulate that SGs primarily serve an antiviral function.
In conclusion, sequestration of viral RNAs within SGs might be a mechanism that applies only to certain viruses to restrict viral replication and spreading. Whether the recruitment of these viral RNAs is an active sorting process or just a consequence of random inclusion is unknown. Generally, whether the presence of viral RNAs in granules or SGs is an antiviral mechanism or a spatiotemporally controlled way to ensure viral replication is unclear.

Granulophagy
SGs are known to be cleared by autophagy, also termed granulophagy [377,378]. Given that LINEand SINE-derived retroelement RNAs are specifically sorted into SGs and PBs and subsequently cleared by granulophagy [379], this process could help cells to clear viruses and trapped viral RNAs. First indications for such a model came from reports on the role of autophagy in clearing poly(I:C)-induced SGs [267]. Recent work on coxsackievirus A16 revealed that SGs induced early during infection were later cleared by autophagy as a means to suppress antiviral immune signaling [380]. Some viruses hijack the autophagic degradation pathway or use autophagosome-derived compartments to support their replication [381,382]. Infectious bursal disease virus is an example of a virus that subverts the autolysosome for virus assembly and maturation in an acidic environment [383]. It is tempting to speculate that in these cases, granulophagy may also serve a proviral function. In conclusion, there is emerging evidence that viral sequestration and granulophagy might be an effective host mechanism to suppress viral spreading. Furthermore, as it is often the case, some viruses appear to subvert this mechanism to their own advantage.

Recruitment of RBPs and Associated RNAs
During stress, numerous RBPs are known to relocalize into SGs. In the context of viral infections, the recruitment of RBPs that function as ITAFs, controlling the activity of viral and cellular IRESs, and the sequestration of RBPs regulating the expression of cytokine and stress response mRNAs may have a direct impact on the antiviral response [384]. Among canonical and non-canonical ITAFs (see for reviews [384,385]), G3BP1 [69,386], heterogeneous nuclear ribonucleoprotein (hnRNP) A1 [278,387], poly(rC) binding protein 2 (PCBP2), [279,388,389], receptor for activated C kinase 1 (RACK1) [292,390], PTB [280,391], RNA-binding motif protein 4 (RBM4) [282], Hu Antigen R (HuR) [392,393], and hnRNPK [283,394] localize in SGs. Interestingly, the anti-apoptotic Bcl-xL mRNA was found to be sequestered together with its inhibitory ITAFs in SGs, which prevents its translation during osmotic stress [395]. However, there is no experimental evidence that relocalization of ITAFs in SGs would affect the translation of viral IRES-containing mRNAs.
Since SGs are considered to be separated biochemical compartments within the cytoplasm, they might contribute to mounting a stress-responsive gene expression profile by affecting mRNA translation and stability rates. An interesting observation is that cells lacking the ability to form SGs tend to over-react to stress. In such cells, stress-induced mRNAs are induced to higher levels, while repressed mRNAs are more strongly repressed. This led to the proposition that SGs, by recruiting RBPs alone or together with their target mRNAs, may provide a buffering system for changes in gene expression [396]. In addition, the recruitment of mRNAs together with translation initiation factors into SGs is thought to help maintain these mRNAs in a translationally silent state during stress and allow for rapid reinitiation of their translation upon resolution of stress. Moreover, the recruitment of certain mRNAs into SGs was found to correlate with enhanced mRNA stability [397]. It should be noted, however, that this view was challenged by a recent single molecule analysis, which found that mRNAs originating from SGs are translated and degraded at similar rates to their cytosolic counterparts, at least during a harsh condition such as incubation with arsenite [398].
The RBPs TIA1, TIAR, HuR, and TTP are involved in regulating the stability and translation of many mRNAs encoding pro-and anti-inflammatory cytokines. At the same time, these RBPs belong to a group of well-characterized SG proteins [399]. Of particular importance to viral infections, IFN-ß and IFN-γ mRNAs contain AREs, which mediate control of mRNA stability through binding of HuR and TTP [400,401]. Stress-dependent phosphorylation events are known to control both the activity and SG localization of several of these RBPs [186,402]. TTP, for instance, is phosphorylated via the stress-activated p38 MAPK/MK2 cascade, which prevents TTP from binding to AREs, mediating degradation of ARE-containing mRNAs, and associating with SGs [186,188]. While it is intriguing that protein activity and SG localization are controlled by the same phosphorylation event, it is still not clear whether inclusion in or exclusion from SGs per se is relevant for the effector functions as well as the RNA and protein interactions of these posttranscriptional regulators of cytokine expression.
In addition to the sequence-specific RBPs, microRNA (miR) binding and subsequent assembly of the RISC are known to suppress translation and induce the decay of cytokine mRNAs, either in the cytoplasm or after recruitment into PBs. Localization of AGO2 in SGs was found to counteract RISC activity in the cytosol, thus contributing to enhanced mRNA stability [284]. Relocalization of AGO2 from PBs and SGs to lipid droplets is further required for miR-122-dependent HCV replication [297]. In addition, components of the nonsense-mediated decay (NMD) machinery such as UPF1 and SMG1 often localize in SGs. It will be important to determine whether the relocalization of these factors into SGs affects the efficiency of NMD, not least because viral genomes are generally prone to activate NMD due to their multicistronic gene arrangement [403].
Finally, two different sets of transcripts are preferentially recruited to SGs-m 6 A-modified mRNAs [286] and mRNAs containing a 5 -terminal polypyrimidine tract (5 TOP) motif [398]. m 6 A modification of cellular mRNAs was found to change upon infection by DENV, ZIKV, WNV, and HCV, and influence the expression of hundreds of mRNAs that regulate flavivirus infection [404]. Notably, m 6 A modification controls the production and stability of the IFN-β mRNA [405,406] and the translation of the transcription factor Foxo3, an antagonist of ISG transcription [407]. YTHDF proteins, which specifically recognize m 6 A, were found to interact with m 6 A-containing mRNAs at the shell of SGs [286]. Given the role of m 6 A in regulating the expression of innate immune factors, the enrichment of m 6 A-containing mRNAs in SGs may contribute to the control of viral replication. The preferential recruitment of 5 TOP mRNAs in SGs is also interesting. Since these mRNAs encode all ribosomal proteins and several translation factors, their sequestration could exert a sustained suppressive effect on virus replication by reducing the long-term availability of the translation apparatus [26].

SGs as Signaling Hubs to Coordinate the General Stress Response
Approximately 50% of the proteins in SGs are not known to bind RNA but rather contribute to cell signaling during stress. Thus, not only RBPs, but also various enzymes that catalyze posttranslational modifications localize in SGs. These include for example components of mTORC1 complex, JNK and its scaffold protein WDR62, the ubiquitin peptidase USP10, the protein arginine methyltransferase (PRMT) 1 and PRMT5, the glycosyltransferase OGG1, and the poly-ADP-ribose polymerase PARP12 [70,408]. Accordingly, proteins within SGs were found to carry a plethora of posttranslational modifications including phosphorylation [250,409], GlcNac glycosylation [408], sumoylation [410], neddylation [411], ubiquitination [380], and ADP-ribosylation [264].
The localization of enzymes within SGs can have different consequences. First, enzymes localize in SGs to modulate the assembly, disassembly, or composition of SGs. They are needed for SG biology, but SGs, in turn, are unlikely to influence the activity of these enzymes. Second, enzymes are sequestered in SGs as a way of silencing their activity. This may be particularly relevant for low abundance proteins with low SG shuttling rates, leading to efficient separation of the enzyme from its cytosolic or nuclear targets. Third, enzymes are recruited in their active form into SGs, or become activated inside SGs, to participate in signaling pathways. This mechanism is likely to benefit from the high concentration of enzyme and substrate within SGs.
One interesting example in this context is the recruitment of PARP12 into SGs. While the various cellular functions of PARP12 are not fully understood, PARP12 expression is induced by IFN and has an antiviral role that can potentially be modulated in a SG-dependent manner. The antiviral functions of PARP12 have been linked to its ability to bind viral and cellular RNAs, and a possible role in translation control [412,413]. PARP12 might further be involved in ADP ribosylation of AGO proteins within SGs, which potentially results in reduced microRNA activity [264]. Relocalization of PARP12 from the Golgi into SGs was further shown to result in the disassembly of the Golgi compartment and in a block of the anterograde membrane trafficking [287]-a route that many viruses rely on for particle assembly. In contrast, inflammation-induced association between PARP12 and p62/SQSTM1 outside of SGs was found to increase NF-κB signaling [412]. Moreover, PARP12-mediated ADP-ribosylation of ZIKV proteins NS1 and NS3 in the cytosol leads to their degradation and restricts virus replication [414]. Taken together, these findings point to a potential role of SG localization for PARP12 substrate selection and its multiple effects on virus replication.

Apoptosis Control by SGs
Several reports indicate that SGs fulfill important anti-apoptotic functions. An attractive concept is that cells, upon damage or insult, assemble SGs while they attempt to resolve the stressful conditions. Cytosolic phase separation could herein mark this "window of opportunity", and only if the damage persists and stress cannot be resolved cells would undergo apoptosis [15]. For several types of stress, SGs were found to efficiently sequester RACK1, thereby inhibiting RACK1-dependent activation of the pro-apoptotic mitogen-activated protein kinase MTK1-p38/JNK signaling axis [292]. Similarly, the localization of the small GTPase RhoA and the Rho-associated coiled-coil containing protein kinase 1 (ROCK1) in SGs hinders the induction of JNK-mediated apoptosis [295]. NF-κB-mediated pro-inflammatory signaling-which can also lead to apoptosis-is partially suppressed by eIF4G-dependent recruitment of TRAF2 into SGs during heat stress and arsenite exposure [293]. Furthermore, Astrin-dependent recruitment of mTORC1 into SGs also prevents apoptosis, which may arise from mTOR hyperactivation, upon exposure to arsenite [289]. An anti-apoptotic role was also revealed for the co-localization of TIA1 and the p90 ribosomal S6 kinase 2 (RSK2) in SGs upon oxidative stress [294]. The Fas-activated serine/threonine kinase (FASTK) is a pro-survival factor that also localizes in SGs and controls protein synthesis of the anti-apoptotic factors cIAP-1 and XIAP. Like RSK2, FASTK interacts with and mediates its anti-apoptotic effects by binding to and regulating TIA1 [415], though it is not yet clear if the SG-resident pool of FASTK is critical to its anti-apoptotic function. Finally, DDX3-dependent assembly of SGs was found to negatively correlate with DDX3-mediated induction of the NLP3 inflammasome and subsequent pyroptosis in LPS-challenged macrophages [416]. Notably, the above examples assign an anti-apoptotic role to SGs, indicating that SGs indeed play an active role in opening a "window of opportunity" for cells in distress, by delaying the onset of apoptosis. Most lytic viruses induce premature death of the infected host cell. To prevent this from turning into a disadvantage by limiting viral spread, some have evolved molecular strategies to counteract or delay apoptosis. One example is SeV, in which encoded trailer RNA binds TIAR to limit apoptosis and SG formation [238]. On the other hand, viruses can also actively trigger apoptosis to release progenies during later stages of lytic infections [417]. Whether SGs also serve as a timer function for the life span of virus-infected cells remains, however, to be determined.

Conclusions
Knowing whether RBPs and cellular or viral RNAs are actively "recruited", "sequestered", or just passively "phase separated" into SGs is much more than a matter of semantics-it is key to a mechanistic understanding of how SGs shape host-virus interactions and coordinate signaling events. Evidence for the presence of a specific RNA or protein in SGs largely relies on microscopy analyses, which have recently been complemented by the development of labeling methods, purification protocols, and the use of high-resolution imaging. However, most SG studies are performed in cells exposed to oxidative stress or other harsh conditions, and many aspects of SG biology still need to be addressed in virus-infected cells.
The co-evolution of virus and host can be described as an arms race that needs to be carefully balanced with the consequences of collateral damage. The dynamic nature of SGs gives testimony to this contest: while some viruses effectively suppress SG assembly, others face an initial wave of SGs that they can only curb during later stages of infection. The appearance of oscillating SGs during chronic virus infections is maybe the most stunning strategy by which cells contain "the enemy within" without suffocating from the antiviral response.
It is tempting to speculate that the condensation state, composition, and function of SGs may change over time as viral infections progress. Further studies building on the recent technical advances will be required to mechanistically dissect the antiviral role of SGs and provide a time-resolved understanding of how stress response pathways and innate immune signaling intersect in the context of viral infections. This may eventually allow us to harness the SGs' "dance with the devil" for antiviral therapeutic approaches against important human pathogens.

Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.