Stress granules are non-membrane bound RNA-protein (RNP) assemblies that form when translation initiation is limited and contain a biphasic structure with stable core structures surrounded by a less concentrated shell. The order of assembly and disassembly of these two structures remains unknown. Time course analysis of granule assembly suggests that core formation is an early event in granule assembly. Stress granule disassembly is also a stepwise process with shell dissipation followed by core clearance. Perturbations that alter liquid-liquid phase separations (LLPS) driven by intrinsically disordered protein regions (IDR) of RNA binding proteins in vitro have the opposite effect on stress granule assembly in vivo. Taken together, these observations argue that stress granules assemble through a multistep process initiated by stable assembly of untranslated mRNPs into core structures, which could provide sufficient high local concentrations to allow for a localized LLPS driven by IDRs on RNA binding proteins.https://doi.org/10.7554/eLife.18413.001
Stress granules are non-membranous assemblies of mRNA and protein (mRNP) that form when translation initiation is limiting, which occurs during many stress responses. Stress granules are thought to influence mRNA function, localization, and to affect signaling pathways (Buchan, 2014; Buchan and Parker, 2009; Kedersha et al., 2013). Normally, stress granule formation is a dynamic, reversible process. However, pathological mutations in proteins that either increase the formation, or decrease the clearance of stress granules, can lead to abnormal accumulation of aggregates that share components with stress granules (Buchan et al., 2013; Dormann et al., 2010; Izumi et al., 2015; Kim et al., 2013). Abnormal accumulation of stress granule-like aggregates is associated with neurodegenerative disease (Li et al., 2013; Ramaswami et al., 2013). The molecular interactions and mechanisms that regulate stress granule assembly and how these may become altered in disease remain unknown.
Stress granules are members of an emerging class of non-membrane bound organelles and are thought to represent multicomponent viscous liquid droplets that from spontaneously by liquid-liquid phase separation (LLPS) (Brangwynne, 2013; Brangwynne et al., 2009, 2015; Elbaum-Garfinkle et al., 2015; Hyman and Simons, 2012; Hyman et al., 2014). In vitro LLPS droplets display many properties of in vivo stress granules including fusion, wetting, shearing, and dynamicity (Lin et al., 2015; Molliex et al., 2015; Murakami et al., 2015; Patel et al., 2015; Zhang et al., 2015). LLPS in vitro can be driven by high local concentrations of proteins containing intrinsically disordered regions (IDRs) (Lin et al., 2015; Molliex et al., 2015; Nott et al., 2015; Wright and Dyson, 2015). Since IDR-containing proteins are enriched in stress granules and other mRNP granules (Decker et al., 2007; Gilks et al., 2004; Jain et al., 2016; Kato et al., 2012; King et al., 2012; Reijns et al., 2008), this has led to the suggestion that IDRs on mRNA binding proteins form hetero- and homotypic interactions that drive initial formation of stress granules and P-bodies (Gilks et al., 2004; Decker et al., 2007; Lin et al., 2015; Molliex et al., 2015; Patel et al., 2015; Toretsky and Wright, 2014; Weber and Brangwynne, 2012; Zhang et al., 2015). This model is supported by observations in vitro wherein high concentrations of IDRs are sufficient to spontaneously form LLPS possibly through the interplay of weak electrostatic and hydrophobic homo- and heterotypic protein-protein interactions (Lin et al., 2015; Molliex et al., 2015; Murakami et al., 2015; Nott et al., 2015; Pak et al., 2016; Patel et al., 2015).
The interactions amongst IDRs driving phase-separation are enhanced by altering the molecular milieu either through lower salt concentration, molecular crowding, addition of RNA, or lower temperature (Lin et al., 2015; Molliex et al., 2015; Nott et al., 2015; Patel et al., 2015). In contrast, small aliphatic molecules, such as 1,6-Hexanediol, are hypothesized to perturb weak hydrophobic interactions, and thus disassemble assemblies that exhibit liquid-like properties in vitro (Patel et al., 2007; Ribbeck and Görlich, 2002). These in vitro observations suggest that manipulation of these properties might be useful for discerning the molecular interactions and mechanisms that drive assembly and regulate stress granules in vivo. Indeed, since mammalian stress granules have been reported to disassemble when cells are treated with 1,6-Hexanediol, they have been inferred to be LLPSs (Kroschwald et al., 2015).
Analysis of mammalian stress granules has revealed they are comprised of two phases: a less concentrated shell, which disassembles upon cell lysis and thus behaves like a LLPS formed by weak interactions; and more stable, internal 'core' structures (Jain et al., 2016). Interestingly, formation of in vitro liquid droplets by IDRs have been observed to mature into a second, less dynamic phase comprised in part of stable amyloid-like assemblies (Han et al., 2012; Kato et al., 2012; Kwon et al., 2013; Lin et al., 2015; Molliex et al., 2015; Murakami et al., 2015; Patel et al., 2015). A parsimonious model therefore is that stress granules may initially assemble to first form a LLPS comprised of multivalent, weak and dynamic interactions, and over time a less dynamic, more viscous second phase matures from the continued supersaturation of IDRs, thereby generating stress granule cores (Jain et al., 2016). Here, we set out to test several predictions of this model: (i) that lower temperature should enhance granule assembly, as LLPS driven by IDRs on RNA binding proteins have been shown to be enhanced at lower temperatures (Molliex et al., 2015; Nott et al., 2015); (ii) that stress granules are sensitive to drugs, such as 1,6-Hexanediol, that are predicted to disrupt weak hydrophobic interactions; (iii) that stress granule dynamics decrease and core size should increase over time as stress granule structure matures.
In contrast to the expected results, we observed that stress granule assembly is inhibited at lower temperatures, despite the extent of translation repression being temperature independent. Moreover, we observe that 1,6-Hexanediol triggers stress granule formation in both yeast and mammalian cells, as well as altering many cellular structures including the cytoskeleton and the nuclear pore complex. Finally, we observe stress granule dynamics and core size is unchanged during prolonged stress. Given these observations we suggest an alternative model where untranslating mRNPs initially oligomerize into stable cores, thereby nucleating an initial shell layer and providing a platform for the growth of a more dynamic shell around these cores, followed by merger of these individual core/shell assemblies into large mature stress granules. Stress granule disassembly is also a stepwise process where shell dissipation occurs first followed by clearance of cores.
We have previously described that mammalian stress granules formed after 60 min of sodium arsenite (NaAsO2) treatment contain substructures (referred to as cores) that are stable in lysates, as well as a less concentrated surrounding structure, referred to as the shell (Jain et al., 2016). The shell region of stress granules does not persist in our lysis conditions, which suggests it is sensitive to dilution, one of the properties of an LLPS.
The biphasic stress granules seen after 60 min in mammalian cells could form by one of two models (Figure 1A). First, the increased pool of untranslated mRNAs bound by proteins containing IDRs, could lead to the formation of a LLPS based on IDR-IDR interactions that would be very fluid and dynamic. In this view, cores would assemble within this initial LLPS in a second step, possibly due to the increased local concentration of core components (Jain et al., 2016). In support of this view, LLPS derived from RNA binding protein IDRs in vitro mature to include more stable, possibly amyloid-like, substructure (Lin et al., 2015; Molliex et al., 2015; Murakami et al., 2015; Patel et al., 2015).
In a second model, stress granule formation could be initiated by the oligomerization of mRNPs into stable core structures. Because these stable core structures would concentrate mRNA binding proteins with IDRs, stable cores could function as nucleation sites for LLPS by creating a high local concentration of IDRs to drive LLPS formation. Over time, cores may then form into larger LLPS assemblies with multiple core structures within each stress granule. In this latter model the formation of the 'shell' structure of the stress granule would be analogous to the formation of the nuclear pore, wherein a high local concentration of FG-repeat containing proteins is imposed by the structure of the nuclear pore, which then facilitates the formation of phase transition in the pore region (Frey and Görlich, 2007; Hülsmann et al., 2012; Schmidt and Görlich, 2016).
To distinguish between these two models for stress granule assembly we examined the time course of stress granule assembly, as well as the biochemical, ultrastructural, and dynamic properties of stress granules at different stages of their assembly.
To examine the time course of stress granule assembly, we exposed human osteosarcoma cells (U-2 OS) expressing a GFP-G3BP1 fusion protein (Kedersha et al., 2008; Ohn et al., 2008) to NaAsO2 and collected images over time using a spinning disc confocal microscope. We observed that stress granules began to form 13–15 min after stress induction, which correlated with an increase in eIF2α phosphorylation (Figure 1B,C, Video 1). The initial stress granules then grew in size both by accretion of material and by fusion with other stress granules (Figure 1B). The area of GFP-G3BP1 stress granules continued to increase with the majority of stress granule assembly completed by 40 min. Stress granules induced with thapsigargin or osmotic stress showed a similar time course of stress granule growth albeit the rates of granule induction were slightly different (Figure 1C). Our data is consistent with the idea that assembly of a mature stress granule occurs via a multistep process: an initial nucleation event that may be influenced by the stress stimulus and a consistent growth phase that requires an addition of substrate and fusion of the small initial stress granules into larger assemblies.
If a stress granule shell forms first, then stress granules formed at early time points should be comprised of only a shell, no core structures, and would not be stable in lysates. To test this, we examined if the earliest detectable stress granules contain compartments that are stable in lysates (Jain et al., 2016).
Lysis of cells at 5 min, 10 min, and 15 min after NaAsO2 induction revealed that by 15 min, when stress granules are first visible in cells, we observed an increase in the number of stable stress granule cores detected in lysates (Figure 2A). We identify these GFP-G3BP1 foci in lysates as stress granule cores since they contain PABP1, poly(A)+ RNA, and eIF4G (Figure 2B). Similarly, yeast stress granules are stable in lysates as soon as they are microscopically detectible in vivo (data not shown). We interpret this observation to indicate that as soon as stress granules are observed in cells by microscopy they contain core assemblies that are stable in lysates.
As an additional test to see if stress granules formed at early times contain core substructures, we examined the ultrastructure of stress granules by super-resolution microscopy (Structured Illumination Microscopy [SIM]) at different times. Using SIM for G3BP1, as assessed by mapping pixel intensity for individual granules, we observed that stress granules contained substructure as soon as they were large enough to be examined by SIM (20 min) (Figure 3).
Based on these two observations, we conclude that core assembly represents an early event in stress granule formation.
Previous in vitro experiments demonstrate that IDR-containing stress granule proteins can undergo a maturation event leading to the formation of more stable complexes and possibly amyloid-like assemblies (Kato et al., 2012; Lin et al., 2015; Molliex et al., 2015; Murakami et al., 2015; Nott et al., 2015; Patel et al., 2015; Zhang et al., 2015). To test if stress granules show a similar 'maturation' in cells, we examined both the dynamics and core dimensions at different times during stress granule formation.
We examined the properties of the stable components of stress granules in lysates to see if there was an increase in size of the core structures over time, which would imply a continued growth of these stable core structures and might reveal maturation of the core structure within a larger stress granule. By Nanosight Nanoparticle Tracking analysis of lysates from U-2 OS cells at different times after stress, we observed that GFP-G3BP1 granule core size does not significantly change between 15 min (median size, 259 ± 74 nm), 30 min (median size, 258 ± 24 nm), 60 min (median size, 273 ± 12 nm), and 120 min (median size, 247 ± 4 nm) of NaAsO2 stress (one-way ANOVA, p-value = 0.94) (Figure 4). We interpret these results to demonstrate that GFP-G3BP1 granule core size does not appreciably mature into larger stable assemblies within the time periods assessed.
In a second experiment, we examined if the dynamics of stress granules change over time as assessed by FRAP (Fluorescence Recovery After Photobleaching) analysis of GFP-G3BP1. This experiment is analogous to experiments done in vitro where the recovery rates of LLPS driven by IDRs of RNA binding components decrease over time revealing a maturation process within these reconstituted assemblies (Lin et al., 2015; Molliex et al., 2015; Murakami et al., 2015; Patel et al., 2015; Zhang et al., 2015). Thus, we examined if the dynamics of normal stress granules change over time, which might reveal an analogous maturation process in cells. We observed that the recovery rates of GFP-G3BP1 were essentially identical after 30 min, 60 min, or 120 min of exposure to NaAsO2 (Figure 5), although the half-time of recovery was slightly faster at 30 min (18.5 s) as compared to 60 min (35 s) and 120 min (31 s). This slightly faster recovery rate is likely to simply reflect that at 30 min stress granules are still undergoing growth, which will increase the apparent rate of recovery. We interpret this observation to argue there is not a global transition of stress granule components to a more stable state within the time periods assessed. However, we cannot rule out that specific components of stress granules become less dynamic over time.
Our observations suggest that core formation is an early event in stress granule assembly; however, we cannot distinguish by light microscopy if cores precede shell formation or not. To distinguish between these early events, we first tested whether stress granule assembly is enhanced at lower temperatures. The logic of this experiment is based on multiple experiments in vitro showing that IDRs from RNA binding proteins can form either LLPS (Lin et al., 2015; Molliex et al., 2015; Murakami et al., 2015; Nott et al., 2015; Zhang et al., 2015) or hydrogels, which in this context are a meshwork assembly of protein filaments (Han et al., 2012; Kato et al., 2012). In these experiments, the LLPS or hydrogel formation is enhanced at lower temperatures, which would be expected to lower the critical concentration required for assembly (Molliex et al., 2015; Murakami et al., 2015; Nott et al., 2015; Kato et al., 2012). If these types of interactions drive stress granule assembly in vivo, then stress granule assembly should be more efficient at lower temperatures. To test this possibility, we examined stress granule formation at different temperatures in GFP-G3BP1 U-2 OS cells, in response to NaAsO2 treatment.
We observed that stress granules formed less efficiently at lower temperatures with a continuing decline in the rate, overall number and size of stress granule formed as the temperature declined from 37°C to 27°C (Figure 6A,B). Two observations suggest that lower temperatures do not perturb translational inhibition. First, we observe a stress-specific induction of eiF2α-phosphorylation to similar levels amongst all the temperatures assessed (Figure 6C). Second, via polysome profiling, we observed a similar extent of translation repression after 20 min of NaAsO2 stress at 30°C as compared to 37°C. (Figure 6E). Taken together, we conclude that despite a sufficient extent of translation repression and an environment shown to be conducive for phase separation in vitro (i.e. lower temperatures), we observe a delay in stress granule formation at lower temperatures.
In principle, stress granule formation could be dependent on higher temperatures either because the interactions holding stress granules together are disrupted at lower temperatures, or because the assembly of stress granules is an ordered process with some step in the process dependent on higher temperatures. If the interactions holding stress granules together are dependent on higher temperatures, then stress granules formed at higher temperatures should disassemble at lower temperatures. Conversely, if stress granules are assemblies that have a temperature dependent step in their assembly, then once formed they should persist after a drop in the temperature. To test these possibilities, we induced stress granule formation at 37°C for 45 min then imaged the cells at a lower temperature (30°C) for 15 min (Figure 6D). We observed that once formed, stress granules persist at a lower temperature and do not appreciably decrease in size. It should be noted that shifting cells to 30°C (without NaAsO2) does not induce stress granule formation (Figure 6D). This is inconsistent with stress granules being held together by interactions that are enhanced at higher temperatures, and instead argues stress granules have a step in their assembly that is inhibited at low temperatures, but once formed stress granules are stable at both high and low temperatures. This suggests that the primary driving forces of stress granule assembly are not weak interactions between IDR domains that are enhanced at low temperatures.
Our observations suggest that promoting LLPS by lower temperature is not sufficient to enhance granule formation. To test if an LLPS shell precedes and is necessary for granule core, we decided to test the effect of inhibiting or disrupting LLPS formation by the addition of 1,6-Hexanediol.
1,6-Hexanediol has previously been used to disrupt various structures that are expected to represent LLPS, in vitro and in vivo (Kroschwald et al., 2015; Molliex et al., 2015; Patel et al., 2007; Ribbeck and Görlich, 2002; Updike et al., 2011). As the nuclear pore is proposed to be a LLPS (Patel et al., 2007; Ribbeck and Görlich, 2002; Updike et al., 2011), a prediction is that 1,6-Hexanediol should disrupt this structure in vivo. Consistent with this prediction, the addition of 10% 1,6-Hexanediol to yeast cells (Kroschwald et al., 2015), rapidly disrupted the nuclear pore, as adjudged by the loss of punctate, peri-nuclear localization of the nuclear pore protein, Nsp1 (Figure 7A). However, we noticed that the effect of 1,6-Hexanediol was not limited to cellular structures that are known to represent a LLPS. We observed that 1,6-Hexanediol (but not 1, 4, 6-Hexanetriol) caused a significant disruption of actin and tubulin organization, as adjudged by the localization of Sac6-GFP (an actin binding protein) and Tub1-GFP (Alpha-Tubulin) respectively (Figure 7A and data not shown). This disruption of cytoskeleton organization occurred within two minutes of 1,6-Hexanediol treatment. Interestingly, Sac6 changed localization twice – once rapidly to go from punctate to diffuse; and later from diffuse to into rod-like structures in the cytoplasm (Figure 7A). Membrane bound cellular structures such as the Endoplasmic Reticulum (ER, observed using Get1-GFP) and mitochondria (observed using Atp1-GFP) were not affected by 1,6-Hexanediol. These observations suggest that 1,6-Hexanediol will disrupt a variety of cellular structures.
We also examined how 1,6-Hexanediol affected mammalian cells. Two observations suggest 1,6-Hexanediol affects cellular viability and gross cellular morphology of mammalian cells. First, HeLa cells demonstrated a 40% reduction in cellular viability as assessed by RealTime-Glo MT Cell Viability Assay following exposure to 3.5% 1,6-Hexanediol as compared to untreated cells. Second, HeLa cells exposed to 1,6-Hexanediol demonstrated rapid widespread morphological changes within seconds including membrane blebbing (Figure 7B; Video 2). Thus, owing to the adverse effects of 1,6-Hexanediol on cell viability and morphology, it’s use in mammalian cells in culture may be complicated.
To test the effect of 1,6-Hexanediol on already assembled stress granules and P-bodies in yeast, we glucose starved yeast cells for 15 min, followed by incubation with 10% 1,6-Hexanediol as previously described (Kroschwald et al. 2015). Markers of stress granules and P-bodies changed localization twice (similar to Sac6-GFP above). First, both assemblies quickly disassembled within two minutes, however, after 10 min, they reappeared, often completely co-localized (Figure 8A). This is in contrast to a previous report that suggested that yeast P-bodies, but not stress granules, are sensitive to 1,6-Hexanediol treatment under glucose starvation conditions (Kroschwald et al. 2015). We observed that P-bodies formed upon 1,6-Hexanediol treatment are less intense than the P-bodies formed under glucose starvation conditions and this may have limited their earlier detection.
By using strains with tagged components of stress granules and P-bodies, we observed that in the absence of any other stress, 1,6-Hexanediol (and not a Digitonin control) also induced robust Pab1-GFP (a stress granule marker) granule formation within 10 min of treatment (Figure 8B). Interestingly, these Pab1-GFP foci also overlapped with Edc3-mCherry – a P-body protein (Figure 8B). Thus, exposure of yeast cells to 1,6-Hexanediol is an effective inducer of stress granule-like assemblies.
To determine how pre-existing mammalian stress granules react to 1,6-Hexanediol, we induced stress granules with NaAsO2 in HeLa cells and then treated cells with 1,6-Hexanediol. Similar to yeast stress granules, we observed an initial reduction in total stress granule numbers, which coincided with alterations in cellular morphology; however, after 10 min, stress granules re-appeared (Figure 9A). These effects of 1,6-Hexanediol were also observed in U-2 OS cells, suggesting this is not a cell specific phenotype (data not shown).
In the absence of any other stress, 1,6-Hexanediol treatment resulted in the formation of stress granule-like assemblies. These assemblies stained positive for stress granule markers G3BP1 and PABP1 (Figure 9B). Thus, similar to yeast cells, 1,6-Hexanediol treatment of mammalian cells in culture can induce stress granule-like assemblies.
Several observations suggest the assemblies induced by 1,6-Hexanediol are bona fide stress granules. First, 1,6-Hexanediol induces the formation of assemblies that contain known stress granule proteins in both yeast and mammalian cells (HeLa and U-2 OS cells) (Figure 10A). Second, 1,6-Hexanediol induced assemblies are sensitive to cycloheximide treatment, a drug known to block stress granule assembly, in both yeast and mammalian cells (Figure 10A). Third, 1,6-Hexanediol-induced assemblies are stable in lysates suggesting that the material properties of these assemblies contain a relatively stable set of interactions similar to normal stress granules (Figure 10B). Fourth, eiF2α phosphorylation is induced to a similar extent by 1,6-Hexanediol compared to NaAsO2 (Figure 10C). Finally, FRAP analysis of 1,6-Hexanediol-induced granules in mammalian cells indicates that 1,6-Hexanediol induced stress granules behave similar to normal stress granules (Figure 10D).
Our data suggests a model for stress granule assembly wherein mRNPs oligomerize to form stable cores and over time individual cores then dock with one another through a more dynamic shell. We reasoned stress granule disassembly may occur in a reverse process where a less stable shell may dissipate initially followed by core disassembly or clearance. To test this prediction, we examined stress granule disassembly using live cell imaging. Several observations suggest that, like stress granule assembly, stress granule disassembly occurs in a multi-step process.
First, live cell imaging of stress granule disassembly shows that stress granules start out as large complexes and disassemble within a narrow time window. Within the limitations of our ability to detect stress granule disassembly, we observe for large stress granules, disassembly appears to occur in two steps: first, larger stress granules break into smaller foci followed by disassembly and/or clearance of these smaller foci. During disassembly, we observe the appearance of filamentous structures emerging at the edges of the disassembling stress granule; however, we should point out that due to the resolution of our microscope we are unable to characterize the nature of these structures (Figure 11BC). Interestingly, once stress granules disassemble into smaller assemblies, these small assemblies become microscopically undetectable within a similar time scale as required for their initial detection during assembly (Figure 11A,B, Video 3).
Second, examination of stress granule disassembly reveals non-uniform distribution within stress granules undergoing disassembly wherein areas of relative concentration appear to persist once the surrounding shell structures have dissolved (Figure 11B). Finally, we observe stress granules are stable in lysates at the same time points that disassembly is occurring (Figure 11C).
Taken together, stress granule disassembly may occur through multiple steps wherein RNA is titrated out of stress granule into translation leading to structural instability and subsequent disassembly of a larger stress granule complex into smaller core structures that are then disassembled or cleared by autophagy.
In this manuscript, we set out to distinguish between two different models to explain the processes of stress granule formation, maturation, and disassembly. In the first model, stress granules form by a loose condensation of IDR containing proteins to form an initial phase-separated, dynamic structure and over time stable cores mature within this dynamic structure. This model is suggested by recent findings in vitro, demonstrating LLPS can be driven by IDRs and these LLPS subsequently mature to form a second, less dynamic phase (Lin et al., 2015; Molliex et al., 2015; Murakami et al., 2015; Patel et al., 2015; Zhang et al., 2015). In the second model, we considered that stress granules initially condense to form stable cores, possibly with a nascent shell layer, and these cores merge into larger structures through interactions between the shell structures.
Several lines of evidence argue that stress granule assembly involves the early formation of stable core structures that then assemble into larger stress granules, each of which can contain multiple cores, surrounded by a less dense shell layer (Figure 12). First, as soon as stress granules are observed in cells, they are stable in lysates (Figure 2). Second, the size distribution of the stable components of stress granules in lysates is similar from 15 to 120 min of stress, arguing that stress granule core formation is not a late step in stress granule assembly (Figure 4). Third, as soon as technically possible to assess, the FRAP behavior, as observed by the GFP-G3BP1 mobile fraction, of stress granules remains unchanged from 30 min to 120 min of stress (Figure 5). This also argues that stress granules are not undergoing a dramatic change in biochemical state, and is inconsistent with models where stress granules initially consist of a uniform and dynamic LLPS, that then matures into a mature biphasic stress granule. Finally, when examined using super resolution microscopy, we observe granules exhibit a heterogeneous distribution of protein even at early time points (Figure 3). Taken together, these observations are consistent with granules assembling from the oligomerization of individual mRNPs into stable core complexes. The oligomerization of individual mRNPs into core structures would be analogous to a liquid-liquid phase separation driven by multivalent interactions, but because the core structures are stable to dilution, could be considered a liquid-solid demixing phase separation. It should be noted we cannot rule out the formal caveat that stress granules initially condense into a weak dynamic assembly, that immediately transitions to a stable assembly, or that stress granule shells and cores form independently as two immiscible liquid phases which later fuse as has been suggested to explain the assembly of nucleolus (Feric et al., 2016). We hypothesize that these stable core assemblies could provide a structural platform, due to the high concentration of IDRs on stress granule components to then rapidly phase-separate a dynamic shell that grows as a result of fusion with other small granules and surface exchange with an increasing pool of untranslating mRNPs.
The assembly of stress granule cores as an initial step in assembly solves a conundrum. The issue is that IDRs are thought to interact with each other to drive LLPS often by weak dynamic interactions, proposed to involve Arg-Aromatic interactions in some cases (Nott et al., 2015; Pak et al., 2016). Since these interactions are limited to few amino acids, one anticipates that any given interaction would be expected to have low specificity, and as such would be expected to interact with many proteins when dispersed in the cytosol. Thus, for such weak and promiscuous interactions to drive a priori the assembly of a stress granule would be difficult since the interactions between stress granule components would always be in competition with other cytosolic factors. When the initial step in stress granule assembly is formation of a stable core, which by definition involves relatively stable interactions, the core structure could then provide a high local concentration of IDRs on stress granule components, which could then be envisioned to trigger a local phase separation, possibly what we refer to as the shell phase of a stress granule. This general principle is also seen in the formation of the nuclear pore, wherein the creation of a high local concentration of FG repeat proteins by the structure of the nuclear pore, allows for a phase transition within the nuclear pore itself, even if the interactions between the FG repeats are relatively weak and non-specific (Frey and Görlich, 2007; Hülsmann et al., 2012; Schmidt and Görlich, 2016).
This work provides two observations that suggest a difference between the process of in vitro LLPS driven by IDRs on RNA binding proteins and the formation of stress granules in cells. First, although low temperatures promote LLPS driven by IDRs of RNA binding proteins (Molliex et al., 2015; Nott et al., 2015), we observed that stress granules formed less efficiently at lower temperatures, despite efficient polysome disassembly (Figure 6). Second, although 1,6-Hexanediol can prevent LLPS by IDRs from RNA binding proteins (Molliex et al, 2015), we observed that 1,6-Hexanediol effectively induces stress granules in both yeast and mammals (Figures 8, 9, and 10). These observations suggest that the process of stress granule assembly in cells is not primarily driven by weak, dynamic homo- and heterotypic interactions between IDRs on RNA binding proteins analogous to the LLPS driven by these protein domains in vitro. However, we hypothesize that weak, dynamic interactions between IDRs could be important in forming the shell structure of stress granules, once the core is nucleated by more stable interactions.
In contrast to our results, 1,6-Hexanediol has been reported to dissolve yeast P-bodies and mammalian stress granules (Kroschwald et al., 2015). We do not understand the basis for the difference between our results and those published earlier, although it could be due to differences in cell handling, growth conditions, or analysis. In any case, the sensitivity of a cellular structure to 1,6-Hexanediol does not appear sufficient to conclude the structure forms by LLPS. We suggest this cautionary note for three reasons. First, we observed that some, but not all, cellular structures were sensitive to 1,6-Hexanediol, including well described cytoskeleton elements (Figure 7). Second, one should anticipate that, depending on the chemical nature of interactions driving assembly, many different types of assemblies could be sensitive to 1, 6-Hexanediol. Finally, one should also expect LLPS assemblies can form by many types of promiscuous or specific interactions, some of which would be expected to be resistant to 1,6-Hexanediol.
Previous work has suggested that small stress granules that form early in a stress response require microtubules and their associated motors to form larger stress granules (Chernov et al., 2009; Fujimura et al., 2009; Ivanov et al., 2003; Loschi et al., 2009). In those experiments, treatment of cells with nocadazole, which depolymerizes microtubules, prevents the fusion of the small initial stress granules into larger assemblies. Based on these results, a reasonable model is that cores are brought together to form larger assemblies by movement on microtubules. Disassembly may involve a reverse process where the microtubule network may facilitate retrograde transport of mRNPs away from a disassembling stress granule.
We observe stress granule disassembly occurs in a reverse process where a less stable shell dissolves initially followed by core disassembly and or clearance by autophagy. This observation has several corollaries. First, upon re-establishment of translation, mRNA is thought to be in rapid equilibrium between the cytosol and stress granules. This exchange appears to influence stress granule structural integrity and may account for titration of select RNAs into translation. The lag in the clearance of granule cores may reflect the requirement of a myriad of ATP-dependent remodeling complexes (e.g. heat shock 70 or p97/VCP AAA-ATPase complexes [Buchan et al., 2013; Walters et al., 2015]) and could serve as a cytoprotective mechanism to acutely re-nucleate stress granules if the cell re-encounters stress.
In mammalian cells, a biphasic stress granule architecture provides multiple layers of functionality within the larger tunable stress-specific RNA transcriptome. First, a shell in mammalian cells provides a scaffold for dynamic exchange of select RNAs out of the translational pool. This could represent a thermostatic mechanism where select RNAs (possibly important housekeeping genes) could be still translated under stress without overtly outcompeting the translation of stress-specific messages essential for survival under stress. These shell-specific RNAs would be expected to be primed for reentry into translation during recovery and may also represent a metastable structural scaffold that fractures during stress granule disassembly. Meanwhile, further compartmentalization into less dynamic cores could provide an additional layer of sequestration of select messages. This could possibly serve as a biochemical sorting mechanism where 'fit' mRNPs could be remodeled under stress by concentrating similar RNPs, while 'unfit' mRNPs could be sorted for decay or not remodeled to reenter the translational pool.
In degenerative disease, this multistep process might become altered leading to aberrant formation of granules. Hyper-nucleation or inefficient clearance of stress granule cores may facilitate a pathological transition by lowering the free energy landscape to seed an β-amyloid aggregate by providing a continued platform for nucleation.
All GFP-tagged strains shown in Figure 7 (Sac6-GFP, Tub1-GFP, Nsp1-GFP, Get1-GFP and Atp1-GFP) were taken from the yeast GFP collection. These strains were grown in minimal media supplemented with a complete set of amino acids and 2% Dextrose at 30°C. For the experiments presented in Figure 8, BY4741 yeast was transformed with a single plasmid expressing Pab1-GFP and Edc3-mCherry (pRP1657). These strains were grown in –Ura media with 2% Dextrose at 30°C. Experiments in Figure 10A were done with 4741 transformed with a plasmid expressing Pab1-GFP (pRP 1363). Human osteosarcoma U-2 OS (expressing GFP-G3BP1, mRFP-DCP1a) (Kedersha et al., 2008; Ohn et al., 2008) and HeLa cells, maintained in DMEM, High Glucose, GlutaMAX with 10% fetal bovine serum, 1% penicillin/streptomycin, and 1 mM sodium pyruvate at 37°C/5% CO2 were used for all biochemical experiments, stress granule purification, and imaging experiments.
Yeast: NaN3 stress: Cells were treated with 0.5% NaN3 for 30 min at 30°C (unless otherwise stated). Mammalian: NaAsO2 stress: Cells were treated with 0.5 mM NaAsO2 at 37°C/5% CO2 (unless otherwise stated). Thapsigargin stress: Cells were treated with 100 nM Thapsigargin in DMSO for 1 hr at 37°C/5% CO2. Osmotic stress: Cells were treated with 375 µM Sorbitol at 37°C/5% CO2 for 1 hr.
A ‘granule-enriched’ fraction was prepared from yeast and mammalian cells as previously described (Jain et al., 2016). For yeast granule isolation, 50 mL of yeast were grown. For mammalian granule isolation, stressed U-2 OS cells were collected from 15 cm plates, pelleted and snap frozen in liquid nitrogen. Cell pellets (re-suspended in lysis buffer) were used to prepare granule enriched fractions. In lysate immunofluorescence and oligo-(dT) staining of stress granule cores was performed as previously described (Jain et al., 2016). 1,6-Hexanediol granules were isolated from U-2 OS cells grown in the presence of 3.5% 1,6-Hexanediol for 15 min.
Stress granule core enriched fractions were collected at respective time points and diluted in lysis buffer (1:2). Particles were analyzed using the NanoSight Nanoparticle Tracking Analysis system NS300 (Malvern) with syringe pump, a 488 nm laser, and a sCMOS camera. Five videos of 60 s were collected for each sample using the 488 nm laser and analyzed by NTA 3.0 software. Maximal concentration of particles (particles/ nm) was used for normalization.
U-2 OS cells were cultured on glass bottom 35 mm dishes. FRAP experiments were performed at indicated times as previously described (Jain et al., 2016).
Cells were grown to an OD600 of 0.4–0.7 in minimal media supplemented with the appropriate amino acids, at 30°C. These cells were then imaged for the ‘No Drug’ or ‘Pre-treat’ controls. These cultures were then pelleted at 3220Xg for 1 min, washed once with medium containing 10 μg/ ml Digitonin ± 10% 1,6-Hexanediol (similar to the concentrations used in Kroschwald et al., 2015), then resuspended in the same medium and returned to a shaker at 30°C. Cells were then imaged after 2 min, 10 min, 20 min and 30 min. For glucose starvation experiments, cells at log phase were pelleted at 3220Xg for 1 min, washed once in media without Dextrose, then resuspended in media without Dextrose and returned to a shaker at 30°C. After 15 min, these cells were then pelleted again at 3220Xg for 1 min and washed once and then resuspended in dextrose free media containing 10 μg/ ml Digitonin ± 10% 1,6-Hexanediol. These cells were then returned to a shaker at 30°C, and imaged after 2, 10, 20 and 30 min. All cyclohexamide treatments were performed at a final concentration of 100 μg/mL added to media at the same time as Digitonin ± 1,6-Hexanediol.
For treatment with 1,6-Hexanediol, complete media (DMEM supplemented with 10% FBS, 1% penicillin/streptomycin) was prepared containing 3.5% 1,6-Hexanediol. Media was exchanged and replaced with 3.5% 1,6-Hexanediol or complete media for indicated time periods. HeLa cells were stressed with NaAsO2 (0.5 mM) for 1 hr. Media was exchanged and replaced with 3.5% 1,6-Hexanediol and NaAsO2 (0.5 mM) or NaAsO2 (0.5 mM) only containing media at indicated time points. For all experiments, HeLa cells were fixed (4% paraformaldehyde) and co-stained for PABP1 (detected by Cy5 labeled secondary) and G3BP1 (detected by FITC labeled secondary). The same concentrations for staining were used in all experiments for both primary (1:200) and secondary (1:400) antibodies. For U-2 OS experiments, all experiments were conducted with 3.5% 1,6-Hexanediol, cyclohexamide (10 μg/mL), and NaAsO2 (0.5 mM) prepared in complete media.
Cellular viability was measured using RealTime-Glo MT Cell Viability Assay (Promega) according to manufacture’s protocol. HeLa cells were exposed to complete media containing 3.5% 1,6-Hexanediol for indicated time periods. Following exposure, normal media was exchanged and cells were incubated in presence of Realtime-Glo substrates for 1 hr or 14 hrs at 37°C. Luminescence measurements were performed using a Victor3 plate reader (PerkinElmer; Waltham, MA).
All yeast and mammalian images were acquired using a DeltaVision Elite microscope with a 100X objective using a PCO Edge sCMOS camera unless otherwise stated. ≥3 images were taken for each experiment comprising of 8 Z-sections each. Stress granule cores were visualized using Deltavision. All images were analyzed using ImageJ. In Figure 2, three independent experiments were performed for each time point. Granule core numbers were calculated using Deltavision, maximal image projections were determined using ImageJ, and thresholding was kept constant across all replicates and time points. Plotted are mean GFP-G3BP1 postiive particles (± standard deviation) normalized to particles at time zero. The x-axis was zeroed to 50 to allow for easier comparison within the graph. In Figure 9, percent HeLa cells (N=3, 50 cells/ replicate) with stress granules (G3BP1 and PABP1 double positive foci) were calculated for each condition.
Live cell imaging was performed using a Nikon Spinning Disk Confocal microscope outfitted with an environmental chamber with O2, CO2, temperature, and humidity control. All images were acquired using a 100X objective with a 2x Andor Ultra 888 EMCCD camera. For temperature experiments, U-2 OS cells grown at 37°C in 35 mm dishes were transferred to environmental chamber at respective temperatures and allowed to equilibrate to respective temperatures for 10 min prior to image acquisition. For disassembly experiments, cells were washed twice in normal media and permitted to recover during image acquisition. All images were acquired by exciting for 200 ms using a 488 nm laser at a gain of 200 (unless otherwise stated). In Figure 2, GFP-G3BP1 foci were determined from thresholded (Otsu) images acquired at 20 s time intervals and plotted as a percentage of peak foci detected for 9 total cells acquired from three independent experiments. In Figure 7, bright field movie was acquired using a Nikon spinning disk microscope using a 20X air objective (HBO arc lamp). Time and event stamps and scale bars were added using ImageJ.
SIM (Structured Illumination Microscopy) was performed using the Nikon N-SIM microscope system run in 3D SIM mode with a Nikon 100X objective at a 1.49NA. Images were acquired using an Andor iXon DU897 EM-CCD camera. GFP-G3BP1 was visualized by exciting the sample using a 488 nm laser at a gain of 300. Samples were mounted using ProLong Diamond Mounting media incubated overnight at room temperature. Sample was excited for 600 ms/ image. Image reconstruction was performed in Nikon NIS Elements.
Phase transitions and size scaling of membrane-less organellesThe Journal of Cell Biology 203:875–881.https://doi.org/10.1083/jcb.201308087
Polymer physics of intracellular phase transitionsNature Physics 11:899–904.https://doi.org/10.1038/nphys3532
Role of microtubules in stress granule assembly: microtubule dynamical instability favors the formation of micrometric stress granules in cellsThe Journal of Biological Chemistry 284:36569–36580.https://doi.org/10.1074/jbc.M109.042879
Edc3p and a glutamine/asparagine-rich domain of Lsm4p function in processing body assembly in Saccharomyces cerevisiaeThe Journal of Cell Biology 179:437–449.https://doi.org/10.1083/jcb.200704147
Microscopic dissection of the process of stress granule assemblyBiochimica Et Biophysica Acta (BBA) - Molecular Cell Research 1793:1728–1737.https://doi.org/10.1016/j.bbamcr.2009.08.010
Stress granule assembly is mediated by prion-like aggregation of TIA-1Molecular Biology of the Cell 15:5383–5398.https://doi.org/10.1091/mbc.E04-08-0715
Liquid-liquid phase separation in biologyAnnual Review of Cell and Developmental Biology 30:39–58.https://doi.org/10.1146/annurev-cellbio-100913-013325
Disruption of microtubules inhibits cytoplasmic ribonucleoprotein stress granule formationExperimental Cell Research 290:227–233.https://doi.org/10.1016/S0014-4827(03)00290-8
Stress granules and cell signaling: more than just a passing phase?Trends in Biochemical Sciences 38:494–506.https://doi.org/10.1016/j.tibs.2013.07.004
Real-time and quantitative imaging of mammalian stress granules and processing bodiesMethods in Enzymology 448:521–552.https://doi.org/10.1016/S0076-6879(08)02626-8
Dynein and kinesin regulate stress-granule and P-body dynamicsJournal of Cell Science 122:3973–3982.https://doi.org/10.1242/jcs.051383
A role for Q/N-rich aggregation-prone regions in P-body localizationJournal of Cell Science 121:2463–2472.https://doi.org/10.1242/jcs.024976
Transport Selectivity of Nuclear Pores, Phase Separation, and Membraneless OrganellesTrends in Biochemical Sciences 41:46–61.https://doi.org/10.1016/j.tibs.2015.11.001
Assemblages: functional units formed by cellular phase separationThe Journal of Cell Biology 206:579–588.https://doi.org/10.1083/jcb.201404124
P granules extend the nuclear pore complex environment in the C. elegans germ lineThe Journal of Cell Biology 192:939–948.https://doi.org/10.1083/jcb.201010104
Classification of intrinsically disordered regions and proteinsChemical Reviews 114:6589–6631.https://doi.org/10.1021/cr400525m
Intrinsically disordered proteins in cellular signalling and regulationNature Reviews Molecular Cell Biology 16:18–29.https://doi.org/10.1038/nrm3920
Timothy W NilsenReviewing Editor; Case Western Reserve University, United States
In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.
Thank you for submitting your article "Distinct stages in stress granule assembly and disassembly" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by Timothy Nilsen as the Reviewing Editor and James Manley as the Senior Editor. The reviewers have opted to remain anonymous.
The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.
We have now received comments on your manuscript from three expert referees. As you will see, they were generally positive about the work but several issues will need to be dealt with via revision. Specifically points 2, 3 and 4 raised by Reviewer 2 as well as his/her minor points should be addressed thoroughly. In addition all of the referees felt that some conclusions were overstated or not rigorously supported by experiments. Please soften these statements appropriately.
In this manuscript the authors investigated the steps/mechanisms modulating the assembly/disassembly of Stress Granules during arsenite-induced stress. The authors performed a series of experiments which included: (1) verifying the kinetic of SG assembly in cells exposed to arsenite and overexpressing GFP-G3BP1 (2) examining the ultrastructure of these SGs using super-resolution microscopy (3) assessing the dynamics and core size dimensions at different time during SG formation (4) assessing their assembly at lower temperature in order to determine how SGs form. By performing these experiments the authors concluded that the assembly of SGs begins with the formation of a core and the subsequent assembly of the shell.
The disassembly also proceeds in a stepwise logical manner starting with the removal of the shell and the clearance of the core.
This is a very interesting manuscript which attempts to understand the mechanisms behind the assembly of SGs. The data provided are convincing, appropriate and support the conclusions set forth by the authors. There are, as such, no major concern with the work.
In this manuscript by Wheeler et.al. "Distinct stages in stress granule assembly and disassembly", the nature of stress granule heterogeneity and assembly dynamics are interrogated. Previous work by these authors had revealed that stress granules have a structure defined by a more stable "core", surrounded by a more dynamic "shell". Here the authors utilize light microscopy approaches, combined with chemical and physical (i.e. temperature) perturbations, to test two different models: 1) the core is formed first followed by a shell, or 2) the whole structure assembles and then a more mature core nulceates within. The authors conclude that the first model is correct. The conclusion of initial core assembly followed by dynamic shell formation is based on some findings indicating stable cores are present in stress granule from the early stage of assembly, and that the assembly of a liquid-like shell is not required for stress granule assembly. Overall, this study has some interesting observations but the conclusions are not strongly supported. In particular:
1) No evidence has been presented to show that there is no shell in any conditions examined. In order for the claim that stress granule assembly proceeds with the initial core formation to be valid, showing the presence of core at the early time point of assembly is not enough but showing the absence of shell in the initial stress granule is necessary. FRAP data in Figure 5 clearly indicate that most of G3BP in stress granule shows high mobility, consistent with liquid-like shells, 30 min after NaAsO2 treatment, which is an early time point of assembly by their definition. If shells are present in stress granule from the earliest time point they can study, it is hard to justify their argument that stress granule assembly is initiated by stable core formation.
2) Based on the fact that treatment of yeast and mammalian cells with 1,6-Hexanediol lead to formation of stress granules, the authors draw a conclusion that "the assembly of a LLPS shell is not necessary for stress granule assembly" (Results section, subsection “Stress granule disassemble through multiple discrete steps”, Figure 8 and 9). However, no evidence has been provided for the material state of 1,6-Hexanediol induced stress granules. This claim would require FRAP measurements on these stress granules, to examine if they are indeed mostly stable cores lacking liquid-like shells.
Related to this, it is necessary to discuss the potential origin of discrepancies in experimental results for 1,6-Hexanediol treatments between this work and Kroschwald et al..
3) In Figure 2A, there are already foci in cell lysate at time 0. Does this indicate the presence of stable cores without any stress? How do these pre-existing foci affect the size distribution measured in Nanosight experiments (Figure 4)? How does the size distribution look in the earlier time point, 10 and 20 min? This is relevant since by 30 min after NaAsO2 treatment the growth phase of stress granule assembly is almost finished (based on Figure 1C).
4) Figure 6C, what time point after NaAsO2 treatment is used for this western blot? How about at lower temperatures, like 27C? It looks like there is higher phosphorylation at 37C. This data is important to fully rule out the possibility of slower eIF2α phosphorylation at lower temperatures as an origin for delayed SG assembly.
5) A recent paper by Feric et.al. Cell 165(7)2016 describes liquid immiscibility of a as an organizing principle for the nucleolus. The model presented in that paper is one of two liquid droplets that do not mix with one another, with the core liquid prone to gelation over time- this seems to be a third model not considered in the current manuscript. Do the data rule out a model like this for stress granules?
Stress granules are cytoplasmic assemblies consisting of stable cores surrounded by a dynamic shell. Some stress granule proteins are able to undergo liquid-liquid phase separation (LLPS) in vitro, raising the possibility that LLPS may play a role in stress granule assembly. This model has gained some support recently since it has been demonstrated that liquid droplets can mature into more stable structures over time. How LLPS initially provides the specificity needed to recruit stress granule components and exclude other cytoplasmic proteins, however, is not known. An alternative model is that stress granule assembly is initiated by the formation of stable complexes through conventional protein/protein interactions. These cores create localized foci of high concentration of RNA-binding proteins, which in turn template local phase separation to create the more dynamic shell. In the present study, the authors present evidence in support of core-first/shell-later model using a time course analysis of stress granule assembly. They demonstrate that stable cores appear in cell lysates within minutes of stress induction and that stress granules have a non-homogeneous appearance immediately upon assembly. The authors also show that exposure to low temperature, which should enhance LLPS, interferes with stress granule formation. Furthermore, they show that treatment with hexanediol, which has been used to dissolve weak assemblies in cells, affects too many cellular structures to be diagnostic and in fact induces stress granules.
Although skeptics may argue that these studies are too descriptive to provide insights into mechanism (formation of small liquid droplets may be hard to visualize), the observations presented are certainly provocative and provide a reasonable model alternative for stress granule assembly. My only concern is the description of granule disassembly, which the authors describe as "shell dissipation" and "core fracturing". This seems to be an over-interpretation of the images shown in Figure 10, given their low resolution. But I agree with the authors that these images do not appear to be consistent with a simple phase transition. It may be helpful to point out to the readers the features (filaments?) that appear at the edges of the stress granules during disassembly (Figure 10B, 60 min timepoint).
In the Discussion, I suggest that the authors distinguish which aspects of their model is supported by data and which are more speculative or only supported by in vitro experiments. For example, what is the evidence that the dynamic shell is formed by phase separation and causes granules to fuse with one another?
In summary, this study provides several interesting observations that challenge a popular model for granule assembly. This study will be of interest to cell biologists and to the growing field of biophysicists that study phase separation.https://doi.org/10.7554/eLife.18413.017
- Roy Parker
- Joshua R Wheeler
- Roy Parker
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
We thank Denise Muhlrad, Robert Walters, Carolyn Decker, Sarah Mitchell, David Protter, and the Parker Lab for helpful discussions and feedback on the manuscript. The imaging work was performed at the CU Light Microscopy Core Facility and the BioFrontiers Institute Advanced Light Microscopy Core. Laser scanning confocal microscopy was performed on a Nikon A1R microscope acquired by the generous support of the NIST-CU Cooperative Agreement award number 70NANB15H226. Spinning disc confocal microscopy was performed on a Nikon Ti-E microscope acquired by the generous support of Professor Tom Cech, Professor Roy Parker, and the Howard Hughes Medical Institute. SIM imaging was performed on a Nikon N-SIM structured illumination super-resolution made possible by equipment supplements to R01 GM79097 (D Xue) and P01 GM105537 (M Winey).
- Timothy W Nilsen, Reviewing Editor, Case Western Reserve University, United States
- Received: June 2, 2016
- Accepted: August 15, 2016
- Version of Record published: September 7, 2016 (version 1)
© 2016, Wheeler et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.