As one of the earliest discovered and most broadly studied SG types, SGs induced by acute high-dose exposure to sodium arsenite have become the canon by which SGs are defined (Kedersha et al., 1999; Kedersha et al., 2002; Aulas and vande Velde, 2015; Aulas et al., 2017; Advani and Ivanov, 2020; Riggs et al., 2020). So-called “canonical” or arsenite-like SGs—caused by many stresses including heat shock (Kedersha et al., 1999), thapsigargin (Aulas et al., 2017), and bisphenol A (Fay et al., 2021) among others—are associated with the inhibition of global protein synthesis and the preferential translation of stress-induced transcripts (Harding et al., 2000a). Bulk translation is inhibited through two main pathways (Advani and Ivanov, 2019): 1) the phosphorylation of the alpha subunit of eukaryotic initiation factor 2 alpha (P-eIF2α), which occurs in mammalian cells via one of four serine/threonine kinases: (heme-regulated eIF2α kinase (HRI) (McEwen et al., 2005), protein kinase R (PKR) (Srivastava et al., 1998), PKR-like ER kinase (PERK) (Harding et al., 2000b) and general control nonderepressible 2 (GCN2) (Wek et al., 1995); and 2) the inhibition of the mammalian target of rapamycin (mTOR) (Proud, 2019). The former mechanism inhibits recycling of the eIF2/tRNA_i ^Met/GTP ternary complex required for translation initiation (Wek, 2018), while the latter mechanism results in hypophosphorylated eIF4E binding protein (4E-BP) which binds eIF4E and inhibits translation (Proud, 2019). When translation is inhibited, 48S translation pre-initiation complexes and untranslated mRNAs accumulate and aggregate in the cytoplasm, favoring a liquid-liquid phase separation (LLPS) event (Guillén-Boixet et al., 2020; Sanders et al., 2020; Yang et al., 2020) and assembling into SGs. Canonical SGs are in dynamic equilibrium with polysomes, as drugs that inhibit polysome disassembly (e.g., emetine, cycloheximide) inhibit SG formation, while drugs that induce premature polysome disassembly (e.g., puromycin) promote SG assembly (Kedersha et al., 2000; Kedersha et al., 2002).

The first protein markers associated with SGs were the RNA-binding proteins TIA-1 and TIAR, as well as the cytoplasmic poly(A)-binding protein (PABPC1) (Kedersha et al., 1999). Soon thereafter, the Ras GTPase-activating protein-binding protein 1 (G3BP1) was identified as a key modulator of SG assembly (Tourrière et al., 2003). It is now firmly established that G3BP1 is a master regulator of SG assembly, and most known SG subtypes require G3BP1 or its close homolog G3BP2 for their assembly (Kedersha et al., 2016; Guillén-Boixet et al., 2020; Sanders et al., 2020; Yang et al., 2020). Related to their relationship with eIF2α-mediated translational control, canonical SGs contain poly(A)+ mRNAs as well as stalled translation 48S-preinitiation complexes including the eIF3 complex, eIF4G, and the small ribosomal subunit (Kedersha et al., 2002; Aulas et al., 2017) (Table 1).

While the composition of canonical SGs have been intensively catalogued through affinity labeling and biochemical purification studies (Jain et al., 2016; Markmiller et al., 2018; Youn et al., 2019; Marmor-Kollet et al., 2020; An et al., 2022), the functional consequences of SG formation remain unclear. One proposed function for SGs is the modulation of apoptosis^12,13 (Arimoto et al., 2008; Arimoto-Matsuzaki et al., 2016; Park et al., 2020). In a now classic report, Arimoto et al. showed that SGs caused by overexpression of G3BP1 conferred resistance to GADD45-mediated apoptosis (Arimoto et al., 2008). They further demonstrated that hypoxia, which triggers SGs, protected cells from etoposide-mediated apoptosis. In both cases, they confirmed that sequestration of the signaling scaffold protein RACK1 at the SG was essential for protection from apoptosis (Arimoto et al., 2008). Other putative functions for SGs include mRNA triage (Kedersha et al., 2005) and stress-induced translational control (Dey et al., 2010); while SGs likely play a role, the idea that SG formation is absolutely essential or even largely responsible for such processes has largely been discredited (Mateju et al., 2020; Glauninger et al., 2022; Mateju and Chao, 2022).

SGs are dynamic. Canonical SGs rapidly exchange factors with the surrounding cytoplasm, though some factors exchange more rapidly than others (Kedersha et al., 2000). The extent to which factors move in and out of SGs is related to the structure of the granule, which appears to have a less dynamic core surrounded by a more dynamic shell (Wheeler et al., 2016). SGs that are more liquid-like also tend to be more dynamic, whereas those associated with gel or solid-like states (including some non-canonical and pathogenic subtypes, described below) are less dynamic (Kroschwald et al., 2015; Zhang et al., 2019).

The so-called “non-canonical” SG is a category used to collect SGs that do not mirror the formation, composition or function of canonical SGs as primarily understood from studies of arsenite and heat shock SGs (Advani and Ivanov, 2020). This SG subtype was originally dubbed non-canonical based on the lack of the canonical SG protein eukaryotic initiation factor 3 (eIF3) (Advani and Ivanov, 2020; Hofmann et al., 2021) in SGs triggered by sodium selenite (Fujimura et al., 2012), which suggested that the 48S preinitiation complex was not recruited to these SGs. Like canonical SGs, non-canonical SGs form acutely (in 4 h or less of stress), require G3BP1 or G3BP2 for their formation (Yang et al., 2020), and are associated with translational arrest, though they may or may not be associated with eIF2α phosphorylation depending upon the stress (Aulas et al., 2017; Advani and Ivanov, 2020; Hofmann et al., 2021). Non-canonical SGs are reported to form in response to many stresses including sodium selenite (Fujimura et al., 2012), nitric oxide (NO) (Aulas et al., 2018), the eIF4A inhibitor Rocaglamide A (RocA) (Aulas et al., 2017), hydrogen peroxide (Emara et al., 2012), and UV radiation (Kedersha et al., 1999; Pothof et al., 2009). There is broad variation within this non-canonical category as to which components do and do not localize at the SG, examples of which are described in Table 1. Others have attempted to further subdivide canonical and non-canonical SGs into designations including Types I, II, and III (Hofmann et al., 2021), however there is no consensus and some conflicting characterizations among these classifications (Advani and Ivanov, 2020; Hofmann et al., 2021), and thus we will address all acute non-canonical SG subtypes holistically in this Perspective.

Non-canonical SGs are reported to be functionally distinct from canonical SGs in that they are believed to be cytotoxic and less dynamic (Advani and Ivanov, 2020). However, there are few published studies that experimentally validate these characterizations. Relative to cell death, the most thoroughly characterized non-canonical SGs are those caused by nitric oxide (NO). NO SGs were determined to be triggering a non-apoptotic and potentially necrotic cell death by assaying a combination of propidium iodide and trypan blue staining, ATP levels and Caspase-3 cleavage (Aulas et al., 2018). For other non-canonical stresses like selenite, RocA, and UV, a decrease in cell viability was associated with SGs, but a mechanism of death was not elucidated (Fujimura et al., 2012; Aulas et al., 2018). Interestingly, while canonical arsenite-induced SGs are believed to be anti-apoptotic (Arimoto et al., 2008), the inverse relationship does not appear to be a universal feature of non-canonical SG subtypes, in that the lack of sequestration of pro-apoptotic factors like RACK1 at non-canonical SGs cannot simply be interpreted to mean that those pro-apoptotic factors are therefore active and inducing apoptotic cell death associated with non-canonical SG formation. Relative to their dynamics, the only report of which we are aware that directly measured dynamic behavior of a protein component of a non-canonical SG is in the context of NO-induced SGs (Aulas et al., 2018). G3BP1 recovery to photobleached NO SGs was approximately 10% lower than in arsenite SGs (Aulas et al., 2018). To our knowledge, this summarizes the extent to which the cytotoxic and dynamic properties of acute non-canonical SGs have been directly assessed in the literature.

The canonical and non-canonical labels are most frequently associated with acute SG formation. Chronic stresses and disease states are associated long-term SG formation. Chronic starvation induced SGs (stSGs) appear after prolonged (8–16 h) starvation of glucose, serum, glutamine and pyruvate (Reineke et al., 2018). These stSGs require both G3BP1 and eIF2α phosphorylation, and are in dynamic equilibrium with polysomes. StGSs appear to be pro-apoptotic, as assessed by Annexin V staining (Reineke et al., 2018), but the dynamic properties of these stSGs remain unknown. Acute metabolic stress, such as complete glucose starvation, glycolysis inhibitors (2-deoxy-D-glucose) and inhibitors of mitochondrial respiration and ATP synthesis (CCCP, oligomycin) are also known to cause stress granules (Kedersha et al., 2002; Wang et al., 2019; Amen and Kaganovich 2020). It is unclear whether or how prolonged metabolic stress relates to acute metabolic stress with respect to SG composition and function, though future investigation in this area could provide a better understanding of the evolution of SGs over prolonged stress conditions.

Pathological SGs (pSGs) are chronic SGs that form in diseased cells and are hypothesized to seed the accumulation of irreversible and toxic disease aggregates. pSGs have been most closely associated to date with ALS and Alzheimer’s Disease (Vanderweyde et al., 2012; Vanderweyde et al., 2016; Ash et al., 2014; Aulas and vande Velde, 2015; Apicco et al., 2018; Zhang et al., 2019; Marmor-Kollet et al., 2020; An et al., 2022). Several excellent reviews have been written on the subject (Wolozin, 2014; Aulas and vande Velde, 2015; Wolozin and Ivanov, 2019), including one within this Research Topic (Rhine et al., 2022). Interestingly, stSGs and pSGs can contain many of the components of canonical granules including poly(A)+ RNA, eIF3, and eIF4G (Table 1), though unlike canonical SGs they are clearly associated with cell death (Liu-Yesucevitz et al., 2010; Vanderweyde et al., 2012; Reineke et al., 2018). This observation raises the question of whether the categorization of SGs into canonical and non-canonical groups based primary on composition data will remain a useful proxy for the breadth of SG function as we continue to discover SGs in new contexts.

Within the past several years, several research groups have noted specific relationships between SGs and membranous organelles (Nicchitta 2022). SGs caused by activation of the unfolded protein response (UPR) can form directly on the endoplasmic reticulum (ER) and incorporate ER-associated mRNAs (Child et al., 2021), which are underrepresented in the transcriptome of SGs caused by arsenite (Khong et al., 2017). These ER-associated SGs contained G3BP1 and were inhibited by cycloheximide, suggesting they are bona fide SGs. SGs were also recently reported to associate with the ER in an autophagy-independent disassembly process that was specific for SGs caused by heat shock (Gwon et al., 2021). It is unknown whether SGs caused by other stresses can associate with the ER, however it has been observed that the ER acts as a facilitator of fission of both SGs and processing bodies (PBs) caused by arsenite stress (Lee et al., 2020). Relationships to other membranous organelles have also been noted. SG nucleation localized to the plasma membrane has been observed in yeast in response to starvation, and was linked to protein kinase C signaling under stress (Amen and Kaganovich 2020). Lysosomal damage was found to trigger SG formation, and a subset of those granules were physically associated with the damaged lysosome (Jia et al., 2022). Similarly, significant SG association with the mitochondrial membrane and with peroxisomes has been reported (Amen and Kaganovich 2021). The composition and function of these membrane-associated SGs remains unclear, but their observation adds an interesting dimension to our consideration of the role of SGs in cellular physiology.

UV SGs were first reported in 1999 by Kedersha et al. (1999) (as data not shown) as containing the SG marker proteins TIA-1, TIAR, and PABPC1. The first published evidence of UV-induced cytoplasmic aggregation was in 2005, when Teixeira et al. were investigating processing bodies [PBs, a cytoplasmic mRNP complex associated with mRNA decay (Luo et al., 2018)], in the budding yeast S. cerevisiae (Teixeira et al., 2005). Then in 2008, UV-induced poly(A)+ RNA-containing bodies were reported in S. cerevisiae by Gaillard and Aguilera (2008), but were not thought to be bona fide SGs due to the lack of co-localization of putative SG homologs. At that time however, SGs had not yet been discovered in S. cerevisiae, and thus core SG markers had not been established in that species.

Finally in 2009, Pothof et al. (2009) investigated the DNA damage response (DDR) to UV in HeLa cells, and discovered that the cytoplasmic condensates they observed in response to UV are SGs, based on the co-localization of TIA-1 and Ago2, and that UV SGs may be involved in microRNA-mediated silencing in response to DNA damage. Several more recent publications characterize UV SGs as non-canonical SGs lacking eIF3 and eIF4G (Table 1) (Moutaoufik et al., 2014; Aulas et al., 2017; Ying and Khaperskyy, 2020). We confirm here that UV SGs also lack the pro-apoptotic scaffolding protein RACK1, 4 h after UV treatment of U2OS human osteosarcoma cells (Figure 1A). Like other non-canonical subtypes, UV SGs are G3BP-dependent, eIF2α-independent, and cytotoxic (Pothof et al., 2009; Moutaoufik et al., 2014; Aulas et al., 2017; Ying and Khaperskyy, 2020). Surprisingly however, there are conflicting reports about whether UV SGs contain poly(A)+ RNA (Moutaoufik et al., 2014; Aulas et al., 2017).

It was widely assumed in the early years of their discovery that all SGs contained mRNA. SGs are defined as mRNA-containing bodies or mRNP aggregates in many articles and reviews (Anderson and Kedersha, 2006; Riggs et al., 2020; Hofmann et al., 2021). In 2014, Moutaoufik et al. (2014) observed the colocalization of poly(A)+ RNA at UV SGs by co-localizing an oligo (dT) probe with the Fragile X Mental Retardation Protein (FMRP) as a SG marker in NIH-3T3 cells 18 h post-UV exposure. This report is, to our knowledge, the first published account of RNA localization to UV SGs in a mammalian cell type. However, a study published in 2017 by Aulas et al. (2017) found no poly(A)+ RNA localization at UV SGs, using G3BP1 as a SG marker in Hap1 human haploid cells, 2 h post-UV.

To independently validate the observation of poly(A)+ RNA at UV SGs, we treated U2OS cells with 15 J/m² UVC, and used fluorescence in situ hybridization (FISH) to examine poly(A)+ RNA localization to UV SGs using G3BP1 as a SG marker. We find, as did Aulas et al. (2017), that poly(A)+ RNA is not enriched at UV SGs at 4 h post-UV exposure, as detectable by FISH (Figure 1B). It is tempting to speculate that perhaps UV light degrades cellular pools of mRNA, which could explain why poly(A)+ RNA FISH signal does not accumulate at UV SGs. However, the literature actually suggests that UV stabilizes and inhibits the degradation of mRNAs (Bollig et al., 2002; Gowrishankar et at., 2005; Gaillard and Aguilera, 2008) Click or tap here to enter text. mRNAs are sequestered in the nucleus upon UV treatment (Burgess et al., 2011) which reduces the cytoplasmic mRNA pool and may contribute to the loss of signal at UV SGs. PABP1 and PABP4 also relocalize to the nucleus, however the poly(A)+ RNA nuclear retention and PABP relocalization under UV stress appear to be independent (Burgess et al., 2011), which may support a model by which PABPs dissociate from their RNA targets under UV stress. Although it is reported UV can stabilize mRNAs, it is also known that UV induces reactive oxygen species and that oxidized RNAs can be targeted for degradation (Wurtmann and Wolin, 2009), meaning that some mRNA degradation may also be possible. Arsenite SGs have been shown to contain reactive oxygen species (ROS), which may further enhance RNA damage and subsequent degradation (Hu et al., 2021), however whether UV SGs contain ROS remains unknown.

To determine whether it was likely that mass degradation of mRNA was responsible for the lack of poly(A)+ RNA we observed at UV SGs, we treated U2OS with UV, then used arsenite to induce SGs at 3 h post-UV exposure. We observe that arsenite SGs induced in UV irradiated cells contain poly(A)+ RNA, though the intensity of localization is decreased compared to arsenite SGs in non-UV treated cells (Figure 1C). Therefore, we submit that mass degradation of mRNA cannot fully explain the lack of poly(A)+ RNA signal at UV SGs, as we show it is possible to assemble poly(A)+-containing SGs after UV treatment. We conclude that poly(A)+ RNAs are either not enriched or not detectable by FISH in UV SGs, though the mechanism of this phenomenonis unknown. As UV SGs are generally smaller than other canonical SGs, it may be the case that the RNA they contain cannot be detected above the cytoplasmic background using FISH. The results do not preclude the possibility that deadenylated mRNAs, non-coding mRNAs and/or small RNAs may still be present at UV SGs, and this possibility warrants further investigation. UV-induced damage may also affect the detection of poly(A)+ RNA using this method. If, however, it is the case that UV SGs are not RNA granules, it would mean that UV SGs do not play any role in the RNA-associated putative functions of SGs such as mRNA triage and translational control.

The finding that UV SGs are depleted of mRNA may also have implications for the assembly of SGs in general. Numerous reports have documented the role of RNA in driving LLPS during SG formation (van Treeck et al., 2018; Guillén-Boixet et al., 2020; Yang et al., 2020; Matheny et al., 2021). If indeed UV SGs do not contain mRNA, then one concludes that the interactions that drive UV SG assembly, unlike most other SG subtypes, must be primarily protein-driven. The lack of facilitation by RNA in LLPS of UV SG formation may also explain why UV SGs have been reported to be smaller and less numerous than other SG subtypes (Moutaoufik et al., 2014). Antiviral ribonucleases generally suppress SGs and have been used to demonstrate the importance of RNAs in SG formation (Burke et al., 2019; Burke et al., 2020); a similar approach could be applied in the case of UV SGs to further confirm or reject a role for RNA in UV SG assembly. Another reasonable prediction is that UV SGs may be more solid or gel-like than canonical or mRNA-containing SGs, as RNA is known to increase the liquid-like properties of biomolecular condensates (Guillén-Boixet et al., 2020; Yang et al., 2020; Bevilacqua et al., 2022). If indeed UV SG assembly is primarily protein-driven, then UV SGs may be a useful model for the specific study of essential protein-protein interactions within SGs. The dynamic properties of UV SGs have not yet been measured, but will no doubt be the topic of future investigation.

Because SGs have been implicated in seeding pathological protein aggregation (Wolozin and Ivanov, 2019), mechanisms of SG suppression are of intense interest. For example, SG induced by arsenite, a strong oxidizer, can be suppressed by strong antioxidants such as N-acetylcysteine (Szaflarski et al., 2016; Aulas et al., 2018) as well as by preconditioning that likely enhances cellular resilience to oxidative stress (Glass and Wente, 2019; Fay et al., 2021). SGs are also known to be suppressed in mitotic cells, because stalled translation elongation prohibits polysome disassembly in M phase (Gilad et al., 2007). While assessing the DNA damage response, Pothof et al. noted that UV-induced SGs did not form in cells that were positive for cyclin A, a marker of S phase (Pothof et al., 2009). By synchronizing their cells prior to irradiation, it was determined that UV SGs only appeared in cells that were irradiated in G2, passed through mitosis, and were in G1 phase at the assay timepoint. This observation of cell cycle dependence likely explains why only ∼10%–30% of cells in an asynchronous culture form UV SGs (Pothof et al., 2009; Aulas et al., 2017). The mechanism for the G1 specificity of UV SGs is unknown, however the ATM and ATR DNA damage checkpoint kinases were not required for UV SG formation (Pothof et al., 2009).

The observation of cell cycle dependence of UV SGs was confirmed in 2014 by Moutaoufik et al. (2014), who discovered that a dose of 10 J/m² UV resulted in complete G1 arrest. These researchers used FACS profiling to assess the cell cycle, while concurrently assessing SG formation by immunofluorescence microscopy, and reported that G1 arrest correlated with the highest SG formation, and resumption of the cell cycle corresponded with SG clearance. Although this implies UV SGs may be dissolvable, it is also possible that the cells with UV SGs underwent cell death, leaving behind SG-free cells to resume the cell cycle. Thus, suppression of UV SGs in S and G2 phases is a unique feature of UV SGs.

Because skin is the organ most directly affected by UV, we decided to examine UV SG formation in an untransformed human keratinocyte cell line, HaCaT. To our great surprise, we were completely unable to induce UV SGs in HaCaT cells. We applied increasing amounts of UV, from our starting dose of 15 J/M² up to 3,000 J/M², and observed zero UV SG formation in HaCaTs by G3BP1 staining (Figure 1D). We do not yet know if this intriguing observation of UV SG suppression is a common property of all keratinocytes or specific to HaCaT, nor do we know whether it is related to the mechanisms of G2, S, and M phase suppression described above. We should note here that all studies of UV SGs reported to date, including the data reported herein, use UVC (254 nm) light sources. UVC is well known to cause damage to human tissues (Roy, 2017), but does not penetrate the Earth’s atmosphere. It remains unknown whether UVA (400–315 nm) and UVB (315–280 nm) rays can trigger UV SG formation in any cell type, including keratinocytes. However, UVA causes significantly more reactive oxygen species in human tissues than other UV wavelengths (Karran and Brem, 2016), which would cause an increase in cellular stress over other wavelengths.