Stress granules are shock absorbers that prevent excessive innate immune responses to dsRNA

collaborative

Stress granules restrain rather than amplify double-stranded RNA sensing, and cells lacking the granule nucleators G3BP1 and G3BP2, UBAP2L or PKR respond to double-stranded RNA with excessive RIG-I-like receptor, PKR and OAS activation and MAVS-dependent apoptosis.

Max Paget; Cristhian Cadena; Sadeem Ahmad; Hai-Tao Wang; Tristan X. Jordan; Ehyun Kim; Beechui Koo; Shawn M. Lyons; Pavel Ivanov; Benjamin tenOever; Xin Mu; Sun Hur

2023 · Molecular Cell · primary research

Senior authors

Sun Hur

Correspondence

Sun Hur

Research areas & themes

Citation

Paget M, Cadena C, Ahmad S, Wang HT, Jordan TX, Kim E, Koo B, Lyons SM, Ivanov P, tenOever B, Mu X, Hur S. Stress granules are shock absorbers that prevent excessive innate immune responses to dsRNA. Molecular Cell. 2023. Volume 83, Issue 7, pages 1180-1196.e8.

DOI 10.1016/j.molcel.2023.03.010. PMID 37028415. PMCID PMC10170497.

Max Paget and Cristhian Cadena are designated as having contributed equally. Sun Hur is the lead contact and sole corresponding author.

One-sentence contribution

Stress granules restrain rather than amplify double-stranded RNA sensing, and cells lacking the granule nucleators G3BP1 and G3BP2, UBAP2L or PKR respond to double-stranded RNA with excessive RIG-I-like receptor, PKR and OAS activation and MAVS-dependent apoptosis.

Executive summary

Double-stranded RNA is detected by RIG-I-like receptors, by protein kinase R and by the oligoadenylate synthetases, and it also triggers the assembly of stress granules, cytoplasmic condensates nucleated by G3BP1 and G3BP2 together with UBAP2L. Stress granules had been proposed to serve as signaling platforms for RIG-I-like receptors, based on the concentration of those receptors in granules together with viral RNA and on reduced interferon induction when G3BPs were knocked down. Other observations sat uneasily with that model, including the exclusion of double-stranded RNA from granules and the failure of granule-disrupting agents to impair signaling.

This study approaches the question with three independent genetic routes to granule deficiency, deleting G3BP1 and G3BP2, deleting UBAP2L, or deleting PKR, and with stimuli that decouple granule formation from double-stranded RNA sensing.

All three granule-deficient backgrounds showed stronger signaling in response to defined 162 base pair 5-prime triphosphate double-stranded RNA, measured by transcriptome, cytokine messenger RNA and protein, IRF3 phosphorylation and nuclear translocation, and MAVS activity in a cell-free assay. PKR and the OAS-RNase L arm were also hyperactive. Granule-deficient cells underwent pronounced caspase-dependent apoptosis, which was largely rescued by deleting MAVS but not by deleting IRF3, and which was relieved by blocking tumor necrosis factor alpha. The same protection extended to endogenous double-stranded RNA accumulating after ADAR1 knockdown. During infection, granule-deficient cells showed both stronger signaling and, independently of MAVS, higher viral protein per infected cell.

Scientific context

Double-stranded RNA was long regarded as a purely viral product, generated by RNA-dependent RNA polymerization or by convergent bidirectional transcription of DNA genomes, but the paper notes that dysregulated cellular processes can also produce it and that responses to double-stranded RNA underlie pathologies from autoimmunity to neurodegeneration and metabolic disease. RIG-I and MDA5 recognize double-stranded RNA, multimerize, and induce MAVS multimerization, leading to IRF3 and NF-kappaB activation and induction of type I interferons and other antiviral genes.

Double-stranded RNA also triggers stress granules, a conserved condensate response also induced by heat shock and oxidative stress. Kinases including PKR phosphorylate eIF2 alpha and suppress global translation, and stalled ribosome-messenger RNA complexes aggregate with cytoplasmic proteins including the nucleators G3BP1, G3BP2 and UBAP2L.

The paper states that the physiological functions of granule formation remain unclear. Granules were initially thought to be sites of translational suppression, a view recent work has argued against. In innate immunity they were proposed to be signaling scaffolds for RIG-I-like receptors, supported by receptor and viral RNA concentration in granules and by reduced interferon induction on G3BP knockdown, and consistent with the frequent targeting of granules by viruses. The paper then lists the contrary evidence. Double-stranded RNA is excluded from granules even though some viral RNAs are enriched. Granule-disrupting agents such as cycloheximide do not impair signaling. Stressors such as arsenite and heat shock produce granules with receptor colocalization but without receptor activation. And granules have been reported to suppress other innate pathways including the NLRP3 inflammasome and MAPK signaling.

Central question

Do stress granules serve as platforms that promote RIG-I-like receptor signaling in response to double-stranded RNA, or do they instead constrain the magnitude of that response, and what are the consequences for cell survival and for viral replication?

Experimental strategy

The design addresses two problems at once, the ambiguity of correlative colocalization data and the confounding of granule loss with loss of other functions of the nucleator proteins.

The first is handled by measuring signaling output at several levels rather than inferring it from localization, including whole transcriptome, cytokine messenger RNA and secreted protein, IRF3 phosphorylation and nuclear translocation, and MAVS activity assayed cell-free by mixing isolated mitochondrial fractions with a common pool of resting cytosolic extract and radiolabeled IRF3. That last assay is important because it reads the signaling potential of MAVS itself against a fixed downstream background.

The second is handled by using three genetically distinct routes to granule deficiency. G3BP1 and G3BP2 and UBAP2L are nucleators, PKR is not a nucleator but supplies the translational arrest that granule formation requires. The three have different additional functions, so a phenotype shared by all three is more plausibly attributable to granules. The paper is explicit that PKR alone would be a poor model, since it affects both translation and granule formation and is itself subject to granule-mediated feedback.

A further decoupling exploits the fact that PKR-deficient cells fail to make granules in response to double-stranded RNA but still make them in response to thapsigargin, which acts through PERK, or to nutrient starvation. Adding those stressors to PKR-deficient cells restores granules without changing the double-stranded RNA stimulus, which tests whether granules as such suppress signaling rather than whether the specific route to them matters.

The stimulus itself is a defined in vitro transcribed 162 base pair double-stranded RNA bearing 5-prime triphosphate, chosen so that receptors are stimulated potently without the confounding effect of viral antagonists. Complementation of G3BPs back into knockout cells provides the gain-of-function control.

Downstream, epistasis is used to identify what kills the cells. Knockouts of MAVS, IRF3, PKR and RNase L, plus inhibitors of TBK1, JAK, NF-kappaB and caspases and blocking antibody to tumor necrosis factor alpha, partition the death phenotype among the arms of the response.

Finally, the question is asked in two more settings, during infection with four viruses selected to engage RIG-I or MDA5, and with endogenous double-stranded RNA produced by ADAR1 knockdown with interferon beta priming, which tests whether the protective role extends beyond exogenous stimulation.

Key findings

1. Transfected 162 base pair 5-prime triphosphate double-stranded RNA induced granules in U2OS cells that depended on G3BPs and were cycloheximide sensitive, and that were enriched for RIG-I, MAVS, TRAF proteins and TBK1, along with the mitochondrial proteins COXIV and NIX (Figure 1A, Figure S1B, Figure S1C). Cy5-labeled double-stranded RNA showed minimal granule localization whether delivered by lipid transfection or electroporation (Figure 1A). The observation is that signaling machinery concentrates in granules while the ligand does not, which the authors read as inconsistent with granules being sites of receptor activation.

2. G3BP-deficient cells showed enhanced antiviral signaling at 6 hours by transcriptome (Figure 1B, Figure S2A), by IFNB, IL6 and RANTES messenger RNA (Figure 1C) and secreted protein (Figure 1D), across all doses tested (Figure 1E). Complementing G3BPs restored granules and suppressed signaling (Figure S2C).

3. IRF3 phosphorylation and nuclear IRF3 were higher in G3BP-deficient cells at 6 hours but declined sharply by 24 hours, mirroring IFNB messenger RNA, while RANTES and IL6 remained elevated at both times (Figure 1C, Figure 1F, Figure 1G). The divergent time course of the IRF3 arm is what the authors later attribute to caspase feedback.

4. In the cell-free assay, mitochondrial fractions from stimulated cells activated IRF3 dimerization, this required MAVS, and fractions from G3BP-deficient cells were more potent than wild type (Figure 1H, Figure 1I). This places the difference at the level of MAVS signaling potential rather than only at downstream readouts.

5. The hyperactivation generalized to A549, HeLa and human bronchial epithelial cells, to both delivery methods, and to double-stranded RNAs of different lengths and sequences (Figure S3A to Figure S3G). Signaling driven by gain-of-function MDA5, gain-of-function RIG-I or STING, which occurs without granule formation, was unaffected by G3BP loss (Figure S3H, Figure S3I), which the authors read as indicating the effect is specific to granule-forming conditions. Basal transcriptomes showed no consistent pattern of baseline inflammation in G3BP-deficient cells (Figure S3E).

6. PKR-deficient and UBAP2L-deficient cells formed G3BP1 foci, but these were smaller, less frequent and lacked MAVS or TIAR enrichment, distinguishing them from conventional granules (Figure 2A, Figure 2B). Both backgrounds nonetheless showed hyperactive signaling at 6 hours (Figure 2C, Figure 2D, Figure 2E). At 24 hours the PKR-deficient cells diverged from the other two, with IL6 and RANTES falling below wild type, which the authors attribute to PKR affecting translation as well as granules, noting that translation itself positively influences immune signaling (Figure 2F).

7. In PKR-deficient cells, thapsigargin or nutrient starvation restored granules containing RIG-I, MAVS and TIAR and reduced signaling at 6 hours, while neither stressor suppressed signaling in wild-type cells where double-stranded RNA alone already makes granules (Figure 2G, Figure 2H, Figure 2I). The suppression therefore tracks with granule presence rather than with the route to granule formation.

8. PKR, OAS3 and RNase L were enriched in granules (Figure 3B, Figure 3C). PKR phosphorylation and ATF4 were higher in G3BP-deficient cells (Figure 3D) and ribosomal RNA integrity was more degraded, indicating higher RNase L activity (Figure 3E). Deleting MAVS partly rescued RNA integrity, so part of the PKR and OAS hyperactivation is secondary to enhanced interferon-driven induction of those sensors. Granules therefore suppress all three arms.

9. G3BP-deficient cells underwent pronounced cell death after double-stranded RNA at all doses, by detachment, Sytox uptake, caspase-3 and caspase-7 activity and live-dead staining (Figure 4A to Figure 4C, Figure S4A), while etoposide and staurosporine killed both genotypes equally (Figure 4D), so the hypersensitivity is stimulus specific. UBAP2L-deficient and PKR-deficient cells behaved similarly (Figure 4E), and the phenotype held in A549, HeLa and bronchial epithelial cells (Figure S4B to Figure S4H).

10. Death showed apoptotic morphology, PARP and caspase-3 cleavage, and activation of caspase-8 and caspase-9 but not caspase-1 (Figure 4F to Figure 4H, Figure S4I). A pan-caspase inhibitor blocked death and PARP cleavage completely, while pyroptosis and necroptosis inhibitors did not (Figure 4I, Figure 4J, Figure S4J).

11. The pan-caspase inhibitor restored IRF3 phosphorylation and IFNB messenger RNA at 24 hours without affecting them at 6 hours, in both U2OS and A549 G3BP-deficient cells, and had no effect in wild-type cells (Figure 4K to Figure 4M). Clear caspase-dependent cleavage of RIG-I, MAVS and IRF3 was not observed (Figure S4K). The interpretation offered is that caspase activation, rather than cleavage of any one identified substrate, produces the late decline in the IRF3 arm.

12. Deleting MAVS largely rescued viability in G3BP-deficient U2OS and A549 cells and reduced caspase cleavage (Figure 5A, Figure 5B, Figure S5A), whereas deleting IRF3 did not rescue viability despite abolishing IFNB induction, and TBK1 or JAK inhibitors did not help (Figure 5A, Figure 5C, Figure S5B). Death is therefore MAVS dependent but IRF3 and interferon independent.

13. Pro-apoptotic genes including TNF, FAS and TNFRSF10B were hyperinduced in a MAVS-dependent but IRF3-independent way (Figure 5C, Figure 5D), tumor necrosis factor alpha secretion was markedly elevated (Figure 5E), and blocking antibody to it significantly relieved death (Figure 5F). NF-kappaB inhibitors reduced both TNF expression and death (Figure S5C, Figure S5D). Tumor necrosis factor alpha alone did not kill cells (Figure S5E), so the authors conclude it cooperates with other factors rather than acting as a sufficient death signal.

14. Deleting PKR or RNase L partially relieved death in G3BP-deficient U2OS cells, less completely than deleting MAVS (Figure 5G, Figure 5H). In A549 cells, deleting RNase L rescued significantly while deleting PKR did not (Figure S5F, Figure S5G), which the authors present as cell-type variation. They also note that deleting PKR increases death in a wild-type background while decreasing it in a G3BP-deficient background, and interpret this opposite behavior as reflecting granule loss in the first case and relief of translational inhibition in the second.

15. Across Sendai virus, influenza A virus lacking NS1, the M51R variant of vesicular stomatitis virus and encephalomyocarditis virus, granule-deficient cells mounted stronger signaling and underwent more death, both predominantly MAVS dependent (Figure 6A to Figure 6G, Figure S6).

16. Cell-to-cell spread of the three viruses with available antibodies was more restricted in G3BP-deficient cells, and this effect required MAVS, consistent with hyperactive signaling limiting spread. Unexpectedly, viral protein per infected cell was higher in G3BP-deficient cells and this was independent of MAVS (Figure 6H to Figure 6J, Figure S7A to Figure S7C), matching higher overall viral messenger RNA (Figure S7E to Figure S7H). The authors read this as two separable functions, receptor-dependent suppression of spread and receptor-independent restriction of replication.

17. ADAR1 knockdown with interferon beta priming produced granules and signaling in wild-type cells, and stronger signaling in G3BP-deficient cells with or without priming (Figure 7A, Figure 7B). Death was greater in G3BP-deficient cells, predominantly MAVS dependent and partly PKR and RNase L dependent, and was relieved by blocking tumor necrosis factor alpha or by pan-caspase inhibition (Figure 7C, Figure 7D). UBAP2L-deficient cells were likewise hypersensitive (Figure 7E). The protective role therefore extends to self-derived double-stranded RNA.

Mechanistic model

The authors state in their own limitations section that it is currently unclear precisely how stress granules regulate either the immune response or viral replication, so the molecular mechanism is not established.

What the data support is a functional model. Stress granules act as a buffer or shock absorber that slows the rate at which double-stranded RNA sensing ramps up and keeps its magnitude below a threshold at which the cell kills itself. Without granules, RIG-I-like receptors, PKR and the oligoadenylate synthetases are all more strongly activated, the MAVS arm drives NF-kappaB-dependent induction and secretion of tumor necrosis factor alpha along with other pro-apoptotic genes, and caspase-dependent apoptosis follows. The interferon arm through IRF3 does not contribute to that death and instead shows a spike followed by a sharp decline caused by caspase-dependent negative feedback.

Directly demonstrated are the hyperactivation of all three sensing arms in granule-deficient cells across three genetic backgrounds and several cell types, the increased MAVS signaling potential measured cell-free, the MAVS dependence and IRF3 independence of the death phenotype, the role of tumor necrosis factor alpha established by blocking antibody and by NF-kappaB inhibition, the caspase dependence of death established by inhibitor and cleavage markers, the caspase dependence of the late IRF3 decline, and the extension of all of this to infection and to ADAR1 deficiency.

How granules exert the suppression is proposed rather than shown. The authors offer that immune molecules appear to be recruited to granules independent of their activation state, which may exert a sequestration effect that retards activation, and they raise as an alternative that transient transit through granules may alter signaling activity through post-translational modification or association with inhibitory molecules. They also note that they do not know what fraction of signaling molecules is localized within granules at any time. The identification of tumor necrosis factor alpha as a contributor rather than a sufficient cause is explicitly stated, since the cytokine alone did not kill cells.

The reconciliation with earlier reports that granules amplify signaling is offered as a hypothesis. The authors propose that the IRF3 and interferon axis, commonly used as the single readout of receptor signaling, behaves non-monotonically in granule-deficient cells because of caspase feedback, and that this plus differences in how granules were disrupted may account for conflicting results in the literature.

The authors also flag, in their limitations, that other biological processes affected in common by all three genetic perturbations and all three chemical perturbations could in principle contribute to the observed effects.

Conceptual or technical advance

The study reverses the prevailing reading of a correlation. Concentration of sensing machinery inside a condensate had been taken as evidence of a signaling platform, and this work shows that the same colocalization is compatible with, and here accompanied by, suppression. The exclusion of the ligand from granules while the receptors are enriched makes that argument concrete.

Methodologically, the combination of three genetic routes to granule deficiency with stressors that restore granules without changing the double-stranded RNA stimulus is a way to isolate the contribution of a condensate from the other functions of the proteins that build it. The PKR-deficient background in particular becomes an experimental instrument, because it separates the double-stranded RNA stimulus from granule formation in a way that pharmacological disruption cannot.

The finding also reframes what a condensate can be for. Rather than assigning granules an antiviral or proviral label, the study reports two effects that both preserve the cell, dampening a potentially lethal immune response and restricting viral replication through a receptor-independent route, and proposes homeostasis as the common denominator. The extension to ADAR1 deficiency connects the same mechanism to autoinflammatory pathology driven by self-derived double-stranded RNA.

Relationship to the broader research program

This study is led by the Hur laboratory at Harvard, and the author contributions statement records that the tenOever contribution, together with three other authors, was provision of reagents. It is a collaborative paper in which the tenOever laboratory supplied material rather than directing the work.

Its subject matter nonetheless intersects several recurring interests in the tenOever corpus, including double-stranded RNA sensing during virus infection, the magnitude and control of the interferon response, and the use of engineered viruses such as influenza A virus lacking NS1 as tools that strip away viral antagonism. The question of what sets the ceiling on an antiviral response, as opposed to what initiates it, is a thread that appears elsewhere in the corpus, but drawing that connection formally is category 3 synthesis and would require the other records to support it.

Related publications

• Kedersha, Anderson, Ivanov and colleagues, on the molecular mechanisms of stress granule assembly and disassembly. Relationship methodological foundation from collaborating laboratories, cited as reference 14, with Ivanov and Lyons as co-authors here.
• Onomoto and colleagues, and related work proposing stress granules as antiviral signaling platforms for RIG-I-like receptors. Relationship predecessor from other laboratories, cited as references 25 and 26, and the model this study argues against.
• Yoo and colleagues, and related reports questioning whether granules are sites of receptor activation, including the exclusion of double-stranded RNA from granules and the failure of cycloheximide to impair signaling. Relationship predecessor from other laboratories, cited as references 30 and 31.
• Work on ADAR1 deficiency leading to endogenous double-stranded RNA accumulation and aberrant MDA5, PKR and OAS activation. Relationship predecessor from other laboratories, cited as references 60 to 66, and the basis for the self-RNA experiments here.
• Prior work from the Hur laboratory on cooperative MDA5 filament assembly and on MAVS multimerization. Relationship methodological foundation, cited as references 10 to 13, and the source of the cell-free IRF3 dimerization assay used here.

Limitations and boundaries

The authors state their own limitations plainly. It is unclear precisely how granules regulate either the immune response or viral replication, they do not know what fraction of signaling molecules resides in granules, and it remains possible that processes other than granules are commonly affected by all three genetic and all three chemical perturbations and contribute to the results.

All experiments are in human immortalized or primary cell lines, principally the osteosarcoma line U2OS with confirmation in A549, HeLa and bronchial epithelial cells. There is no animal work, so the proposed relevance to systemic inflammation, immunopathology and autoinflammatory disease is an extrapolation from cell-intrinsic observations.

Granule deficiency is produced by complete genetic loss of nucleators or of PKR, which is a more absolute perturbation than any physiological modulation of granule assembly, and the knockout lines have been without these proteins before the stimulus is applied. The complementation experiment restores G3BPs and suppresses signaling, which supports the assignment, but the other backgrounds are not complemented.

The three granule-deficient backgrounds are not equivalent. PKR-deficient and UBAP2L-deficient cells still form G3BP1 foci, so the comparison is between granules of different composition and size rather than between presence and absence. The PKR background additionally alters translation and diverges from the others at 24 hours, and the authors caution against using it as the sole model.

Conclusions differ between cell types in ways the paper does not resolve. Deleting RNase L rescued viability in A549 cells while deleting PKR did not, the reverse of part of the U2OS result, and deleting PKR has opposite effects on death depending on whether G3BPs are present.

The apoptosis mechanism is only partly resolved. Tumor necrosis factor alpha contributes but is not sufficient, deleting MAVS rescues largely but not completely and reduces caspase cleavage but not to completion, and no caspase substrate accounting for the late IRF3 decline was identified, with cleavage of RIG-I, MAVS and IRF3 not observed.

The principal stimulus is a defined synthetic double-stranded RNA delivered by transfection or electroporation, which is chosen deliberately to avoid viral antagonists but does not reproduce the location, quantity or kinetics of double-stranded RNA generated during replication. The paper also notes that its poly(IC)-triggered granules behaved differently from published reports and attributes this to reported variability among commercial poly(IC) reagents, which bounds comparison with that literature.

Virological readouts rely on immunofluorescence for spread and protein per cell using antibodies available for three of the four viruses, with no antibody for encephalomyocarditis virus, and infectious titer measurements are not the primary readout.

The ADAR1 experiments use transient small interfering RNA knockdown with interferon beta priming rather than genetic loss, so residual enzyme and the priming condition are part of the system.

Audience summaries

25 words

Stress granules turn out to restrain double-stranded RNA sensing rather than promote it. Cells that cannot form them overreact and die by immune-triggered apoptosis.

75 words

Cells detect double-stranded RNA through several sensors and also build stress granules, condensates once thought to serve as signaling platforms for those sensors. Using three separate genetic routes to granule deficiency, this study finds the opposite. Without granules, sensing is hyperactive, cells secrete excess tumor necrosis factor alpha and die by caspase-dependent apoptosis that requires MAVS but not interferon. The same protection applies to self-derived double-stranded RNA arising from ADAR1 deficiency.

150 words

Stress granules had been proposed as platforms that promote RIG-I-like receptor signaling, though double-stranded RNA itself is excluded from them and granule-disrupting drugs do not impair signaling. Using cells lacking G3BP1 and G3BP2, UBAP2L or PKR, this study finds that granules instead restrain double-stranded RNA sensing. All three backgrounds showed stronger transcriptional and cytokine responses, higher IRF3 activation and greater MAVS signaling potential in a cell-free assay, along with hyperactive PKR and OAS-RNase L arms. Granule-deficient cells underwent caspase-dependent apoptosis that was largely rescued by deleting MAVS but not IRF3 and was relieved by blocking tumor necrosis factor alpha. Restoring granules in PKR-deficient cells with thapsigargin or starvation suppressed signaling, tying the effect to granules rather than to the route that makes them. During infection, granule loss both enhanced signaling and, independently of MAVS, raised viral protein per cell. The molecular mechanism is not established.

Discoveries supported by this paper

Documented publication relationships

No explicit publication relationship was recorded in the reviewed graph.

Pathogens

Technologies