Stress granule sequestration of OXPHOS gene transcripts exacerbates glycolytic restriction

22 Stress granules (SGs) are dynamic organelles formed under cellular stress and they 23 are generally regarded as protective entities. Meanwhile, their role in pathogenesis is 24 becoming increasingly recognized, but the underlying mechanisms remain elusive due 25 to the diverse nature of both stress types and biological contexts. Here we investigate 26 SG dynamics and temporal changes in bulk and SG-associated transcriptomes under 27 different regimens that inhibit glycolysis. We subject cells to either single assaults of 28 glucose depletion (GD) or 2-deoxy-D-glucose addition (2DG) or a combined 29 treatment (GD+2DG). We find that SGs formed under these conditions exhibit 30 distinct properties, including eIF2α phosphorylation dependency, mRNA composition, 31 and capacity to disassembly. Our results show that SGs induced by GD+2DG 32 uniquely trap oxidative phosphorylation (OXPHOS) gene transcripts, leading to 33 mitochondrial dysfunction. We provide evidence suggesting that the persistency of 34 SGs formed under GD+2DG treatment is interwoven with mitochondrial dysfunction 35 resulting in heightened apoptosis, effects that can also be recreated under single 36 assaults when combined with mitochondrial inhibition. Our findings suggest that SG 37 formation induced by inhibiting a single metabolic pathway can widen its impact in 38 intensifying cellular metabolic stress under specific conditions, providing mechanistic 39 insights into the paradoxical dual nature of SGs in stress response and pathology. 40 41

42 Eukaryotic cells respond to stress by compartmentalizing translationally aberrant 43 ribonucleoproteins into stress granules (SGs), which are dynamically assembled and 44 disassembled to regulate mRNA and protein availability (Van Leeuwen et al. 2019; 45 Campos-Melo et al. 2021). This process enables cells to conserve energy, survive 46 stress, and recover efficiently. However, dysregulation in SG assembly or 47 disassembly can render cells vulnerable to stress, contributing to the pathogenesis of 48 various diseases, including amyotrophic lateral sclerosis (ALS) and frontotemporal 49 dementia (FTD) (Zhang et al. 2019; Parameswaran et al. 2023; Buchan et al. 2013; 50 Cui et al. 2023; Wolozin et al. 2019; Cui et al. 2024). Understanding the molecular 51 mechanisms governing SG dynamics is therefore critical to elucidation of the 52 underlying causes of these disorders. 53 54 SG formation as a cellular process is mediated by diverse signaling pathways 55 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint molecularly, depending on the precise nature and context of the stress. For instance, 56 under arsenite-induced oxidative stress or heat shock, SG assembly is primarily 57 triggered by the phosphorylation of the translation initiation factor eIF2α 58 (Frydrýšková et al. 2020; Aulas et al. 2017; Szczerba et al. 2023). Beyond this 59 canonical pathway, alternative mechanisms, such as the activation of 4EBP1 and 60 disruption of the eIF4F complex, have also been implicated in SG formation under 61 conditions such as selenite exposure, hydrogen peroxide treatment, or complete 62 glycolysis inhibition (Tauber et al. 2020; Emara et al. 2012; Wang et al. 2022; 63 Fujimura et al. 2012). These findings highlight both versatility and specificity of SG 64 responses to cellular stress. 65 66 The characteristics of SGs can also be influenced by the intensity and duration of the 67 stress. Time-course analyses reveal that under arsenite exposure, the size and 68 dynamics of the SG core, marked by the G3BP1 protein, remain stable over 2 hours, 69 suggesting minimal changes in their biochemical state (Wheeler et al. 2016). In 70 contrast, heat shock-induced SGs exhibit significant changes in their core size over 71 time, accompanied by a proteomic transition as the stress prolongs (Mateju et al. 72 2017; Hu et al. 2023). Furthermore, the dynamics of SG dissolution upon alleviation 73 of the stress or cellular adaptation to the stress may also rely on context-specific 74 mechanisms, which can involve distinct signaling pathways, molecular chaperons, 75 and macromolecule degradation systems (Jia et al. 2024; Hofmann et al. 2021). 76 However, most of the existing studies have focused on acute stress-induced SGs, and 77 a comprehensive knowledge is currently lacking with regard to the dynamics and 78 composition of SGs formed under chronic stress (Hofmann et al. 2012; Cherkasov et 79 al. 2013; Huang et al. 2020; Reineke et al. 2018, 2019; Youn et al. 2018, 2019; Khong 80 et al. 2017; Namkoong et al. 2018; Somasekharan et al. 2020; Frydrýšková et al. 81 2020). 82 83 Nutrient starvation and mitochondrial inhibition have emerged as powerful paradigms 84 for studying chronic stress-induced SGs (Reineke et al. 2018; Fu et al. 2016; 85 Eiermann et al. 2022; Wang et al. 2022; Aguilera-Gomez et al. 2017; Amen et al. 86 2021; Sfakianos et al. 2018; Pernin et al. 2024). For example, when glucose, serum, 87 glutamine and pyruvate were deprived all together, SGs assembled slowly, reached 88 peak formation at 8 hours, persisted at 16 hours, and disassembled upon addition of 89 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint glucose (Amen et al. 2021; Reineke et al. 2018). However, existing models are 90 complicated by the fact that energy depletion—due to either nutrient withdrawal or 91 mitochondrial dysfunction—not only represents a specific type of stress but also 92 impairs the cell’s ability to assemble and disassemble SGs in response to other 93 stressors (Jain et al. 2016; Pernin et al. 2024; Wang et al. 2022; Eum et al. 2020). 94 Here we investigate mechanisms underlying SG dynamics under glycolytic inhibition, 95 aimed at disentangling these complexities to provide insights into chronic stress 96 responses. 97 98 In this study, we compare SG dynamics under different long-term glycolytic 99 inhibition regimens. We show that while treatments of glucose depletion (GD) or 2-100 deoxy-D-glucose (2DG) addition induce SGs that dissipate as stress prolongs from 8 101 to 24 hours, the combined GD+2DG treatment results in a persistent presence of SGs 102 and an increase in apoptotic cells over time. Time-resolved bulk RNA-seq reveals a 103 unique upregulation of oxidative phosphorylation (OXPHOS) pathway activity under 104 GD+2DG. However, this upregulation seems to be a futile response since transcripts 105 of many OXPHOS genes are specifically sequestered within SGs, as evidenced by 106 G3BP1-APEX2-enriched RNA-seq. We verify a reduced expression of OXPHOS 107 proteins and mitochondrial dysfunction, properties that differentiate cells that are 108 treated with GD+2DG from those under single treatments. We suggest that 109 mitochondrial defects in these cells were responsible for SG persistence, and such SG 110 persistency can be recapitulated in cells under single glycolytic inhibition with 111 mitochondrial disruption. Our findings support a feedback model in which SG 112 formation itself has a role in widening the impact of glycolytic restriction through 113 limiting OXPHOS gene expression, creating a scenario of a deepened, perpetuating 114 cellular stress. 115 116

117 GD, 2DG and GD+2DG induce SGs with distinct formation kinetics 118 The aim of this study was to compare how cells may form stress granules (SGs) in 119 response to different assaults that inhibit a common metabolic pathway, glycolysis. 120 Here we investigated this question through subjecting 143B cells to either single or 121 combined assaults. For single assaults, cells were treated with glucose depletion 122 (GD) or with addition of 2-deoxy-D-glucose (2DG), a glucose analog that directly 123 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint blocks glycolysis. For the double assault, 2DG was supplemented while glucose was 124 withdrawn from the medium (GD+2DG). We examined the characteristics of SG core 125 during their formation using G3BP1 as the marker (Figure 1A). ~20% of the cells 126 under the treatment of GD or 2DG became SG-positive (SG+) at the peak time of 8 h 127 and 1 h, respectively, while 75% cells were SG+ under the treatment of GD+2DG at 128 the peak time of 1 h (Figure 1B). The number of SGs per cell and the average size of 129 SGs were comparable between the two single-assault groups but distinct from the 130 GD+2DG group (Figure 1C-D). Based on these SG characteristics and those detailed 131 in the following sections, we regard GD as chronic-mild stress, 2DG as acute-mild 132 stress, and GD+2DG as acute-severe stress. As explained further below, we propose 133 that the previously reported eSG (Wang et al. 2022) is a result of a further exacerbated 134 stress state that goes beyond a mere glycolysis restriction. 135 136 The pathway to SG formation can be either dependent or independent of 137 phosphorylation of the translation initiation factor eIF2α (p-eIF2α). Western blot 138 analysis revealed a significant increase of p-eIF2α/eIF2α ratio in both single-assault 139 groups but not in GD+2DG (Figure 1E). We verified this phenomenon at the single-140 cell level through immunofluorescence staining. By comparing the signals of anti-p-141 eIF2α and O-propargyl-puromycin (OPP) between SG+ and SG- cells, we observed a 142 correlation of SG presence to both p-eIF2α elevation and reduction in protein 143 synthesis under single but not double assaults (Figure 1F-G). Consistently, GADD34, 144 a factor involved in p-eIF2α dephosphorylation, was reduced under single but not 145 double assaults (Figure S1). To verify the distinct dependences of p-eIF2α in SG 146 formation in our system, we treated cells with ISRIB, an antagonist of ISR that acts 147 downstream to all eIF2α kinases and specifically reverses the cellular effects of p-148 eIF2α (Sidrauski et al. 2015). Our results show that ISRIB effectively reduced both 149 the fraction of SG+ cells and the level of p-eIF2α under single assaults, an effect that 150 was diminished under GD+2DG (Figure 1H-J). Together, these results suggest that 151 cells respond differently to a set of assaults that share the common target of 152 glycolysis. For convenience, we refer to SGs formed under single and double assaults 153 in our system as Type I and Type II SGs, respectively. 154 155 OXPHOS gene transcripts are compartmentalized in Type II SGs 156 The protective role of SGs has been attributed to the compartmentalization of RNPs 157 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint that can protect them from degradation and preserve their availability for future use if 158 and when cells recover (Das et al. 2022). While the protein composition of SGs has 159 been extensively studied, the specific mRNA species associated with SGs may also be 160 informative about the specificity or diversity of responses to the inducing stress. To 161 investigate the mRNA composition of SGs induced under our conditions, we 162 generated an G3BP1-APEX2 fusion protein in a biotin-based proximity labeling 163 experiment for enriching SG-associated mRNAs. Immunofluorescence staining 164 showed subcellular colocalization between the fusion protein and the biotinylation 165 signal (Figure 2A). We sequenced cDNA libraries prepared from bulk and G3BP1-166 associated mRNAs under no treatment (UT), GD, 2DG or GD+2DG at the peaking 167 time for SG+ cells. Pairwise correlation between independent replicates (Pearson’s 168 correlation ρ > 0.995) indicated an excellent data reproducibility. Importantly, only a 169 minimal level of correlation (ρ = 0.172~0.371) was observed between the bulk-seq 170 and APEX-seq results within each pair (Figure 2B). These results support a selective 171 enrichment of G3BP1-associated mRNAs under our experimental conditions. 172 173 Based on a hierarchical clustering analysis, we found that, among all G3BP1-174 associated transcriptomes, the GD+2DG groups were most distant from the other 175 groups (Figure 2B). To further evaluate the difference in the mRNA contents, we 176 enriched G3BP1-associated transcriptomes by comparing each of the three stress 177 conditions against those from UT. Such an analysis led to the identification of 1540, 178 5172 and 3798 stress-dependent, G3BP1-associated genes under GD, 2DG and 179 GD+2DG, respectively (Figure 2C). We referred to these stress-dependent genes as 180 SDGs. Among them, 458 genes were shared among all three conditions. Functional 181 annotation analysis revealed that these common SDGs were enriched in “cell cycle”, 182 “TGF-β signaling pathway”, “RNA degradation”, “valine, leucine and isoleucine 183 degradation” and “human immunodeficiency virus 1 infection” (Figure 2D). These 184 results, which confirm our recently reported effect of long-term glucose depletion on 185 cell cycle progression (Zheng et al. 2023; see also Figure S4 for cell cycle analysis of 186 the current study), are supportive of the hypothesis that SGs store away transcripts 187 which encode proteins with “unwanted” functions at the time of stress. 188 189 Transcript length is one of the features that determine the selectivity of mRNAs 190 sorting into SGs. Here we evaluated the transcript length distributions of SDGs 191 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint identified from the three conditions of glycolytic stress. We found that, similar to 192 endoplasmic reticulum (ER) stress or arsenite exposure (Namkoong et al. 2018; 193 Khong et al. 2017; Ren et al. 2023; Van Treeck et al. 2018), all three glycolytic 194 assaults resulted in SDGs whose transcripts were much longer than those of their non-195 SDG counterparts (Figure 2E). However, SDGs induced by GD+2DG had 196 significantly shorter transcript lengths than those induced by GD or 2DG alone 197 (Student’s t-test p = 0.004 and 0.03, respectively). This finer selection of SDG 198 contents likely reflects a tailored response strategy to the double assault. Indeed, 199 functional annotation analysis showed that SDGs identified from GD+2DG were 200 significantly involved in neurodegenerative diseases such as amyotrophic lateral 201 sclerosis (ALS), Huntington disease (HD) and Parkinson disease (PD; Figure 2F). 202 Among the specifically-enriched pathways, genes involved in oxidative 203 phosphorylation (OXPHOS) have the shortest transcript lengths, and many of these 204 genes are shared by the pathways of ALS, HD, PD, and reactive oxygen species 205 (ROS; Figure 2G). Together, our results suggest a qualitative difference in mRNA 206 composition of SGs between single and double glycolytic assaults and, more 207 importantly, a selective sequestration of OXPHOS gene transcripts by SGs induced by 208 the double assault. 209 210 Mitochondrial function is defective in cells that form Type II SGs 211 The preferential enrichment of OXPHOS genes in SGs formed under the double 212 assault suggests a possibility of mitochondrial dysfunction. To test this, we first 213 performed Western blot analysis on two selected proteins that are encoded by SDGs: 214 NDUFB4 serves as an accessory subunit of NADH dehydrogenase (mitochondrial 215 complex I), and COX6B1 constitutes a critical subunit of cytochrome c oxidase 216 (complex IV). The levels of both proteins were significantly reduced under GD+2DG 217 but not GD or 2DG alone (Figure 3A-B). To investigate whether the selective 218 sequestration of the transcripts of mitochondrial SDGs led to a general mitochondrial 219 dysfunction, we determined the levels of three other representative mitochondrial 220 proteins not encoded by SDGs: TOM20 is a subunit of the outer mitochondrial 221 membrane translocase (TOM complex), ND6 is a mitochondrial-encoded subunit of 222 complex I, and ATP6 is a mitochondrial-encoded subunit of ATP synthase (complex 223 V). Our results showed a significant reduction in the level of these mitochondrial 224 proteins under GD+2DG but not GD or 2DG alone (Figure 3C-D). Consistently, 225 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint GD+2DG treatment yielded a most prominent reduction in both TOM20 226 immunofluorescence staining signal and mitochondrial area size in SG+ cells relative 227 to SG- cells (Figure 3E-F). Interestingly, we also detected a higher mitochondrial 228 proximity of Type II SGs than those of SGs (Figure 3G), a phenomenon that has been 229 reported to link SGs to metabolic remodeling (Amen et al. 2021). As expected of a 230 mitochondrial dysfunction, the levels of intracellular ATP and mitochondrial 231 respiration were significantly reduced under GD+2DG relative to GD or 2DG alone 232 (Figure 3H-J). Together, these results document a unique feature of Type II SGs, a 233 preferential retention of mitochondria-related mRNAs, and the accompanying 234 mitochondrial defects. 235 236 Type II SGs differ from Type I SGs in their failure to dissolve during prolonged 237 treatment 238 Previous studies suggest that intracellular ATP deficiency can prevent conventional 239 SGs from disassembling (Eum et al. 2020; Jain et al. 2016; Wang et al. 2022). Based 240 on our finding that mitochondrial defects were linked only to SGs formed under 241 GD+2DG, we predicted that SGs formed under either GD or 2DG alone would 242 engage in disassembly without the release of glycolytic insult, i.e., during prolonged 243 treatment. Therefore, we analyzed the fate of SGs by extending the duration of 244 treatments to 24 h. Under GD alone, the number of SG+ cells, the average SG 245 number and the average SG size were all significantly reduced after 8 h (Figure 4A). 246 At 24h, SG+ cells approached a minimum level (~0.95%) that was basically a 247

level seen in untreated cells (~1%). Similarly, under 2DG alone, these 248 SG features continuously declined after the peaking time at 1 h (Figure 4B). In 249 contrast, under GD+2DG, SGs were maintained even at 24 h in all examined cells (N 250 = 1,091; Figure 4C). These differences in SG dynamics between GD or 2DG alone 251 and GD+2DG were also observed in HeLa cells (Figure S5). 252 253 Given our result that eIF2α phosphorylation and translation repression are 254 differentially impacted by double or single assaults, we sought to evaluate the 255 dynamic properties of these processes during prolonged treatments. We detected a 256 decline in p-eIF2α under GD or 2DG alone, which was correlated with SG 257 dissociation dynamics (Figure 4D-E), whereas p-eIF2α exhibited a continuous 258 increase under GD+2DG despite its insignificance at the time of SG formation 259 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint (Figure 4F). Puromycin incorporation assays documented consistently continuous 260 declines in global protein synthesis in all three conditions (Figure 4G-I), supporting 261 that translation repression can be achieved through both p-eIF2α-dependent and -262 independent pathways. Together, these results document a dynamic disassembly 263 process for Type I SGs formed under sustained single glycolytic assaults, a process 264 lacking in cells that form Type II SGs. 265 266 Cells forming Type II SGs upregulate the bulk mRNA expression of OXPHOS 267 genes 268 To analyze transcriptional responses to different glycolytic assaults as a function of 269 time, we generated bulk mRNA-seq datasets from cells under GD, 2DG or GD+2DG 270 from different durations. We were particularly interested in comparing the two 271 distinct phases of SG lifecycle: formation (0~8, 0~1 and 0~1 h under GD, 2DG and 272 GD+2DG, respectively) and dissolution/persistence (8~16, 1~4 and 1~4 h under GD, 273 2DG and GD+2DG, respectively). We performed a time-series clustering analysis to 274 identify gene ontologies (GOs) with specific dynamic patterns: rise-falling, fall-rising, 275 monotonically rising, and monotonically falling (Figures 5A; see Materials and 276 Methods). A total of 9 pathways shared the same dynamics in cells under all three 277 treatments (Figure 5B), the majority of which (8/9) were monotonically rising such as 278 “IRES dependent viral translational initiation” and “eukaryotic translation initiation 279 factor 3 complex Eif3m” (Figure 5C). This result supports the known mechanism of 280 IRES-mediated translation upon stress (Yang et al. 2019; Lacerda et al. 2019) as a 281 common cellular response to different glycolytic assaults. 282 283 There were 26 pathways that shared the same dynamics between GD alone and 2DG 284 alone but not with GD+2DG. Among these, 3 monotonically-rising pathways were 285 related to ER stress (Figure 5D-E), indicative of a transcription-level response to 286 elevated p-eIF2α and reduced protein synthesis under either single assault. In 287 contrast, among the pathways with a unique temporal pattern in cells under GD+2DG, 288 the top significant GOs were related to mitochondrial functions, including 289 “mitochondrial protein containing complex”, “aerobic respiration” and “oxidative 290 phosphorylation” (Figure 5F). The majority of genes in these pathways exhibited an 291 overall increasing trend as a function of treatment duration under GD+2DG but not 292 under GD or 2DG alone (Figure 5G). It is particularly worth noting that this 293 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint expression increase in OXPHOS genes coincided with the acute trapping of their 294 transcripts by SGs (Figure 2F-G). Given the observation of mitochondrial defects 295 including the reduced expression of OXPHOS genes, we postulate that cells were 296 making a compensatory response to the double glycolytic assault, a response that 297 would only end up being futile. 298 299 Mitochondrial inhibition prevents Type I SG dissolution during prolonged 2DG 300 treatment 301 To test whether mitochondrial defects under the double assault may underlie the 302 inability of cells to reverse SGs, we analyzed cells under treatments of 2DG along 303 with inhibitors that target different mitochondrial respiratory complexes. If our 304 hypothesis is correct, we would expect mitochondrial inhibition to render SGs formed 305 under Type I condition non-dissolvable. Our results show that, while each 306 mitochondrial inhibitor alone showed little inducement of SGs, combining a 307 mitochondrial inhibitor with 2DG led to a dramatic induction of SGs within 1 h 308 (Figure 6A). In addition, both the fraction of SG+ cells and the p-eIF2α level were 309 persistent during the prolonged treatments (Figure 6A-B). Furthermore, in ρ0 cells, 310 which are depleted of mitochondrial DNA, GD alone induced SGs that shared similar 311 dynamics with those formed in 143B cells under GD+2DG (Figure 6C-D). 312 313 To test whether a lack of any specific OXPHOS gene activity may lead to a similar 314 phenotype, we combined siRNA knockdown with 2DG treatment. Here we tested 315 two OXPHOS SDGs (NDUFB4 and COX6B1) and two OXPHOS non-SDGs 316 (NDUFB6 and COX6B2). Figures 6E-I show that all of these siRNAs significantly 317 decreased the protein level of TOM20 and, importantly, SGs formed under these 318 conditions showed Type II hallmarks, including persistence at 24 h and proximal 319 mitochondrial association (see also Figure 6E-F for a detectable difference in 320 characteristics of SGs between SDG knockdown and non-SDG knockdown at 1 h). 321 Together, these results support an involvement of mitochondrial activity in SG 322 dissolution under prolonged glycolytic assaults, an engagement that, under the double 323 assault, is diminished by SG sequestration of OXPHOS transcripts. 324 325 Dissolution of Type I SGs requires HSP70 activity whose inhibition alters 326 OXPHOS gene expression 327 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint Our results described thus far document a defining difference between single and 328 double glycolytic assaults, i.e., the ability of SGs to dissociate under prolonged 329 treatments (Figure 4A-C). To gain insights into the dissolution process of SGs formed 330 under these assaults, we examined the potential dependence of SG clearance on 331 autophagy and the HSP70 chaperone, respectively. We found that, while wortmannin, 332 a potent autophagy inhibitor, affected neither SG formation nor dissolution (Figure 333 7A), siRNA against HSP A1A and VER-155008 targeting HSP70 both effectively 334 preserved the fraction of SG+ cells during prolonged treatment of GD or 2DG alone 335 (Figure 7B-C). Importantly, the expression level of HSP70 was significantly 336 increased in cells undergoing SG dissolution but not in cells during prolonged 337 GD+2DG treatment (Figure 7D). These results suggest a crucial role of HSP70 338 activity in SG dissolution during prolonged glycolytic stress. 339 340 To explore HSP70-mediated regulation of SG dissolution during prolonged glycolytic 341 stress, we performed bulk RNA-seq of cells that were preconditioned with HSP A1A 342 siRNA and then co-treated with GD for different durations (GD+siHSP). Differential 343 expression analysis between GD and GD+siHSP at each corresponding time point 344 showed a significant enrichment of downregulated genes in OXPHOS at 0 and 16 h 345 but not at 8 h (Figure 7E), suggesting a requirement of HSP70 activity for OXPHOS 346 gene expression before the onset of glycolytic stress and during SG dissolution when 347 the stress prolonged. In addition, by intersecting SDGs identified from G3BP1-348 APEX2 experiments and HSP70-dependent SG-dissolution genes (upregulated at 16 h 349 relative to 8 h under GD but not under GD+siHSP), we obtained a list of 19 genes, 350 which might be sequestered by SGs under the control of HSP70 (Figure 7F). Among 351 these genes, SP ATA18 encodes the protein MIEAP, which promotes accumulation of 352 lysosomal proteins in mitochondrial matrix and elimination of damaged proteins 353 inside mitochondria (Ikari et al. 2024; Gaowa et al. 2018). RT-qPCR experiments 354 confirmed that SP ATA18 was significantly increased from 8 to 16 h under GD only 355 when HSP70 activity was intact (Figure 7G). Therefore, HSP70 may mediate the 356 expression of mitochondrial quality control genes such as SP ATA18 to control SG 357 dissolution. 358 359 Cell survival under prolonged glycolytic assaults is correlated with SG 360 dissolution 361 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint To test whether SG dissolution is an adaptive function that supports cell survival 362 during prolonged glycolytic assaults, we performed flow cytometry with propidium 363 iodide and Annexin V staining. We detected a pronounced fraction of apoptotic cells 364 at 24 h under GD+2DG, but not under GD or 2DG alone (Figure 8A). Consistently, 365 genes involved in apoptosis were enriched in upregulated genes under prolonged 366 GD+2DG but not under prolonged GD or 2DG alone according to bulk mRNA-seq 367 and RT-qPCR experiments (Figure 8B-C). Furthermore, when the dissolution of Type 368 I SGs was blocked by HSP70 inhibition (GD+VER or 2DG+VER), the fraction of 369 apoptotic cells was significantly increased (Figure 8D). Together these results show 370 that cell survival is correlated with SG dissociation under prolonged glycolytic 371 assaults, and that persistent SGs represent a sign for cells reaching an impasse. 372 373

374 In this study, we document a divergent formation of two distinct types of SGs under 375 glycolytic inhibition. Type I SGs, which are induced by single assaults (GD or 2DG), 376 resemble arsenite-induced SGs in their dependence on eIF2α phosphorylation for 377 assembly and HSP70 activity for disassembly. In contrast, Type II SGs, which are 378 induced by GD+2DG and previously referred to as energy deficiency-induced SGs 379 (eSGs), exhibit a distinct biochemical state characterized by the sequestration of 380 OXPHOS gene transcripts and mitochondrial dysfunction. Our comparative analysis 381 of these two types of SGs suggests a feedback loop between the formation of Type II 382 SGs and mitochondrial dysfunction, a regulatory loop that can render Type II SGs 383 non-dissociable under sustained stress (see Figure 9 for a graphic model). Effectively 384 the sequestration of OXPHOS transcripts within SGs under double assault widens the 385 impact of glycolytic inhibition to further exacerbate energy deficit and prevent SG 386 disassembly. This feedback loop highlights the dual roles of SGs in stress adaptation 387 and pathogenesis, providing insights into the molecular mechanisms underlying SG-388 associated diseases. 389 390 The dynamic and reversible nature of physiological SGs enables cells to adapt to and 391 recover from stressful conditions. However, the persistence of SGs containing 392 pathological contents can lead to cell dysfunction and death (Ivanov et al. 2019; 393 Zhang et al. 2019; Mahboubi et al. 2017; Sato et al. 2024). Such a paradoxical 394 divergence highlights a delicate control of the quality and dynamics of SGs. In fact, 395 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint disassembly upon the release of the inducing stressors such as sodium arsenite and 396 heat shock has been defined as a key feature of acute SGs (Marmor-Kollet et al. 2020; 397 Hofmann et al. 2021; Buchan et al. 2009). Interestingly, SGs that are induced by the 398 proteasome inhibitor MG132 can also disassemble during a long-term treatment 399 without eliminating the stressor (Ganassi et al. 2016; Mazroui et al. 2007; Wang et al. 400 2022). Similarly, our Type I SGs, but not Type II SGs, are dissolved within 24 h 401 when glycolytic restriction remains in effect. Therefore, different regimens of 402 glycolytic stress induce distinct SG responses, resembling a physiological-to-403 pathological transition that accompanies disease progression. In this context, it is 404 worth noting that, while the initiating differences between the single and double 405 assaults on glycolytic inhibition might not be major on their own, the positive 406 feedback loop that involves the sequestration of OXPHOS mRNAs and the persistent 407 nature of type II SGs likely have contributed significantly to the bifurcation in the SG 408 types. 409 410 It has been shown that severe energy deficiency can prevent arsenite-induced SGs 411 from disassembling (Jain et al. 2016; Wang et al. 2022). Our time-resolved 412 transcriptomic analysis shows that the bulk mRNA level of OXPHOS genes exhibits 413 an increase along the GD+2DG treatment time, but this increase in mRNA expression 414 is ultimately unsuccessful because their proteins remain at a reduced level and 415 mitochondrial functions remain impaired. We suggest that the preferential 416 sequestration of OXPHOS gene transcripts by Type II SGs is responsible for 417 undercutting the effect of the compensatory transcriptional upregulation, leading to a 418 perpetual mitochondrial dysfunction and energy stress, accompanied by a persistence 419 of SGs and increased cell death. 420 421 There is a long-standing hypothesis that SGs serve as temporal storages and silent 422 sites of untranslated mRNAs and unused RNA-binding proteins (Kedersha et al. 423 2002; Ivanov et al. 2019). Our APEX2-based transcriptomic analysis uncovers 424 distinct profiles of G3BP1-associated transcripts between Type I and Type II SGs. 425 Notably, Type II SGs preferentially sequester shorter transcripts, including OXPHOS 426 genes, despite the conventional preference for long transcripts due to enhanced RNA-427 RNA interactions (Campos-Melo et al. 2021; Ren et al. 2023; Lee et al. 2019; Khong 428 et al. 2017). This unique mRNA composition reflects a specific cellular response to 429 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint complete glycolytic inhibition and further underscores the role of Type II SGs in 430 modulating mitochondrial function. 431 432 In conclusion, our work suggests a mechanism through which SG formation can 433 transition between their protective and deleterious roles. Such a transition can be 434 achieved through the operation of a feedback loop between SG dynamics and 435 mitochondrial function. Our study provides a fresh perspective for understanding the 436 pathogenesis of SG-associated diseases and potential therapeutic targets. 437 438

439 Cell culture and treatment 440 143B and HeLa cells were cultured in DMEM (Gibco) with 10% FBS, 100 U/ml 441 penicillin and 100 U/ml streptomycin under 5% CO₂ at 37°C. The mitochondrial 442 DNA-less ρ⁰ 206 cells, derived from 143B cells, were cultured under the same 443 condition except an addition of 50 μg/ml uridine. For glycolysis inhibition, cells were 444 rinsed in PBS and then transferred to glucose-free DMEM, DMEM with 25 mM 2-445 deoxy-D-glucose (2DG), or glucose-free DMEM with 2DG for various durations as 446 described in main text. For mitochondrial inhibition, cells were rinsed in PBS and 447 then transferred to DMEM with 1 μM rotenone, 5 μM antimycin A or 1.5 μM 448 oligomycin. For p-eIF2α inhibition, cells were rinsed in PBS and then transferred to 449 DMEM with 500 nM ISRIB (MCE, HY-12495A) for 1 h. For autophagy inhibition, 450 cells were rinsed in PBS and then transferred to DMEM with 1 μM wortmannin 451 (MCE, HY-10197) for 8 h. For HSP70 inhibition, cells were rinsed in PBS and then 452 transferred to DMEM with 50 μM VER-155008 (MCE, HY-10941). 453 454 siRNA transfection 455 All siRNAs were designed by DSIR (http://biodev.extra.cea.fr/DSIR/DSIR.html) and 456 synthesized by GenePharma. For transfection, cells were treated with an siRNA at a 457 final concentration of 50 nM using jetPRIME (Polyplus) for 36 h before other 458 treatments. The oligo sequences are listed in Table S1. 459 460 Quantitative RT-PCR 461 Cells were seeded in 6-well plates to reach approximately 80~90% confluence, and 462 total RNA was extracted using TRIzol (TaKaRa, 9109). The purified RNA was 463 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint quantified by NanoDrop Spectrophotometer, and 1 μg RNA was subjected to reverse 464 transcription using ABScript III RT Master Mix with gDNA Remover (ABclonal, 465 RK20429). Quantitative PCR was performed using 2× Universal SYBR Green Fast 466 qPCR Mix (ABclonal, RK21203). For each experimental group, three independent 467 reverse transcription experiments were conducted and β-actin was used as an internal 468 control. The primers are listed in Table S1. 469 470 Western blot 471 Cells were washed with PBS and then lysed in RIPA buffer (FUDE, FD009) with 472 Benzonase Nuclease (Beyotime, D7121) and complete Protease and phosphatase 473 inhibitor cocktail (Beyotime, P1048). Lysates were loaded onto 10% SDS-PAGE and 474 proteins were transferred to PVDF membranes (Millipore, IPVH00010). Membranes 475 were incubated with rocking first in TBST with 5% milk at room temperature for 1 h, 476 then in TBST with 5% milk and primary antibody at 4°C overnight. After three 477 washes, membranes were incubated in TBST with secondary antibody and 5% milk at 478 room temperature for 1 h. After another three washes, ECL Western Blotting Substrate 479 (Vazyme Biotech, E412-01) were used for detection. 480 481 The following primary antibodies were used: HRP-conjugated β-Actin Rabbit mAb 482 (1:5000, ABclonal, AC028), HRP-conjugated β-Tubulin Mouse mAb (1:5000, 483 ABclonal, AC030), rabbit anti LC3B (1:1000, ABclonal, A19665), mouse anti 484 HSP70/HSPA1 (1:2000, ABclonal, A1507), rabbit anti HSP70 (1:2000, ABclonal, 485 A23457), rabbit anti phospho-eIF2α (S51; 1:1000, Cell Signaling, 3398), mouse anti 486 phospho-EIF2S1 (Ser51) (1:1000, proteintech, 68023-1-Ig), rabbit anti EIF2S1 487 (1:1000, proteintech, 82936-1-RR), rabbit anti EIF2S1/ EIF2A (1:1000, proteintech, 488 11170-1-AP), rabbit anti GADD34 (1:1000, proteintech, 10449-1-AP), mouse anti 489 TOM20 (1:1000, proteintech, 66777-1-Ig), rabbit anti TOM20 (1:1000, proteintech, 490 11802-1-AP), rabbit anti ATP6 (1:1000, ABclonal, A23150), rabbit anti COX6B1 491 (1:1000, proteintech, 11425-1-AP), rabbit anti MT-ND6 (1:1000, ABclonal, A17991), 492 rabbit anti NDUFB4 (1:1000, proteintech, 27931-1-AP), rabbit anti COX6B2 (1:1000, 493 proteintech, 11437-1-AP), rabbit anti NDUFB6 (1:1000, proteintech, 16037-1-AP). 494 495 Immunofluorescence staining 496 Cells were washed in PBS, fixed in 4% paraformaldehyde at room temperature for 20 497 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint min, and rinsed in PBS buffer containing 0.1% Triton X-100 (PBST) for 10 min. 498 Then cells were blocked in PBST with 3% BSA for 1 h and incubated at 4°C 499 overnight with primary antibody. After three washes, cells were incubated in PBST 500 with secondary antibody and 3% BSA at room temperature for 1 h. After another 501 three washes, cells were mounted in Antifade Mounting Medium with DAPI 502 (Beyotime, P0131). 503 504 The following primary antibodies were used: rabbit anti G3BP1 (1:200, ABclonal, 505 A3968), mouse anti TOM20 (1:200, proteintech, 66777-1-Ig), rabbit anti TOM20 506 (1:200, proteintech, 11802-1-AP), 488-conjugated G3BP1 pAb (1:200, proteintech, 507 CL488-13057), mouse anti p-EIF2S1 (Ser51) (1:200, proteintech, 68023-1-Ig), rabbit 508 anti p-eIF2α (1:200, Cell Signaling, 3398), mouse anti Myc tag (1:500, proteintech, 509 60003-2-Ig). 510 511 OPP staining 512 O-propargyl puromycin (OPP) staining was performed using Click-iT™ Plus Alexa 513 Fluor™ 555 Picolyl Azide Toolkit (Thermo Fisher, C10642) according to the 514 manufacturer’s instruction. After three washes, cells were mounted in Antifade 515 Mounting Medium with DAPI (Beyotime, P0131). 516 517 Image analysis and quantification 518 For each glass slide, >= 5 different fields were imaged by Olympus FV1000 Confocal 519 Microscope or Nikon Instruments A1 Confocal Laser Microscope. ImageJ tools were 520 used to identify individual cells (based on DAPI signals) and quantify SG 521 characteristics (based on G3BP1 signals). Cells with at least five G3BP1-positive 522 spots detected in the cytoplasm were considered as SG+. In each SG+ cell, the 523 number of SGs was measured using Analyze Particles and the aggregated area size of 524 SGs was measured using ROI Manager. 525 526 For quantification of p-eIF2α and OPP, the fluorescence intensities within each 527 identified cell were summed, background-subtracted and normalized to the cell area 528 size using ImageJ. For quantification of TOM20, the total area size of fluorescent 529 signals was measured by ROI Manager and normalized to the cell area. 530 531 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint RNA sequencing 532 Total RNA was extracted using RNAiso Plus (TaKaRa, 9109). The libraries were 533 generated using V AHTS Universal V8 RNA-seq Library Prep Kit for Illumina 534 (Vazyme, NR605). Sequencing was performed on Novaseq 6000 (Nanjing Jiangbei 535 New Area Biopharmaceutical Public Service Platform). Read quality was assessed 536 using fastqc and adaptor sequences were removed using trim_galore v0.6.10. Then 537 reads were aligned to GRCh38 using hisat2 v2.2.1, and summarized using 538 featureCounts v2.0.1. Quantification of transcript isoforms was performed using 539 StringTie v2.2.1 for assembly and analysis. Differential expression analysis was 540 conducted using edgeR v4.4.0 with a cutoff fold change > 2 and adjusted p-value < 541 0.05. Functional annotation, including Gene Ontology (GO) and Kyoto Encyclopedia 542 of Genes and Genomes (KEGG) enrichment analyses, was carried out using 543 clusterProfiler v4.14.3 with adjusted p-value < 0.05. 544 545 APEX-based proximity labeling and RNA sequencing analysis 546 G3BP1-APEX2-Myc was generated by an in-frame fusion of G3BP1, APEX2 and 547 Myc tag in pAcGFP1-N1 vector. After transfection, cells were cultured for 48 h to 548 have adequate expression, and then subjected to different experimental treatments. 549 The resulting samples were biotin-labeled as previously described (Somasekharan et 550 al. 2020). For immunofluorescence staining analysis, rabbit anti G3BP1 (1:200, 551 ABclonal, A3968), mouse anti Myc tag (1:500, proteintech, 60003-2-Ig) and Alexa 552 Fluor™ 647-conjugated streptavidin (1:400, Thermo Fisher, S21374) antibodies was 553 used. For RNA-seq analysis, biotinylated RNAs were pulled down using C1 554 Streptavidin beads (Thermo Fisher) according to the manufacturer’s instruction. To 555 identify stress-dependent G3BP1-associated genes (SDGs), we performed differential 556 expression analysis by comparing APEX-seq under a given stress condition over 557 APEX-seq under no treatment in edgeR v4.4.0 with a cutoff fold change > 2 and 558 adjusted p-value < 0.05. 559 560 Time-series RNA sequencing analysis 561 We generated bulk mRNA-seq datasets from cells before, at and after the SG peaking 562 time under GD (0, 8 and 16 h), 2DG (0, 1 and 4 h) or GD+2DG (0, 1 and 4 h). For 563 each dataset, we used GSV A v2.0.1 to calculate Geneset Activity Score (GAS) of all 564 GO terms, limma v3.62.1 to identify GO terms with differential GASs (fold change > 565 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint 2 and adjusted p-value < 0.05), and Mfuzz v2.66.0 to cluster differential GOs with the 566 same dynamic patterns of GAS. 567 568 FACS analysis of ROS, apoptosis and cell cycle 569 Cells were treated with Reactive Oxygen Species Assay Kit (Beyotime, S0035M), 570 Annexin V-FITC assay kit (Beyotime, C1062) or Cell Cycle and Apoptosis Analysis 571 Kit (Beyotime, C1052) according to the manufacturer’s instructions, respectively, and 572 then fluorescence was quantified by flow cytometry. 573 574 Measurement of ATP 575 Cells were assayed by Enhanced ATP Assay Kit (Beyotime, S0027) according to the 576 manufacturer’s instruction. 577 578 Measurement of Oxygen Consumption rate 579 Cells were plated onto a Seahorse XF96 Cell Culture Microplate (Agilent) at a density 580 of 1.0 × 104 cells/well. After an overnight incubation with 5% CO₂ at 37 °C, the 581 culture medium was replaced by the assay medium containing 1 mM pyruvate, 4 mM 582 glutamine, and with or without 25 mM glucose (as the control and the GD group 583 respectively). Then sequential treatments with 1 µM oligomycin, 2 µM FCCP, and 1 584 µM rotenone+antimycin A allowed for generating the full profile of OCR. 585 586 Statistics 587 Each sequencing data has at least two replicate samples. Each quantitative 588 experiment has at least three independent samples. Unless otherwise stated in the 589 legend, all quantitative results were presented as mean ± standard error of the mean 590 (SEM), and analyses of the mean were presented as unpaired two-tailed Student's t-591 test. 592 593 Data availability 594 All raw RNA-seq data generated in this study have been submitted to the NCBI 595 BioProject database under accession number PRJNA1289020. 596 597

598 This study was supported by the National Natural Science Foundation of China 599 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint (32470584) and the National Key R&D Program of China (2021YFC2700403). We 600 acknowledge support of Zhejiang University School of Medicine affiliated Women’s 601 Hospital. 602 603 Author contributions 604 W.Z., J.M. and F.H. conceived the study and designed the experiments; W.Z., M.X. 605 and X.L. performed experiments and generated data; W.Z. and R.X. analyzed the data 606 and generated all figures; Y .G. and M.G. provided technical and managerial support; 607 J.M. and F.H. acquired funding; W.Z., R.X., J.M. and F.H. wrote the paper and all 608 approved the paper. 609 610 Competing interests 611 The authors declare no competing interests. 612 Figures 613 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint 614 Figure 1. SGs are differentially formed under different assaults against 615 glycolysis. 616 (A) Representative images of SG formation (blue: DAPI, red: G3BP1) in 143B cells 617 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint under no treatment (UT), glucose deprivation for 8 h (GD), treatment with 2DG for 1 618 h (2DG), and the combined treatment for 1 h (GD+2DG). Scale bar: 20 μm. Inset 619 box: a zoom-in view of SGs. 620 (B) Percentages of SG+ cells. Mean ± SEM (standard error of the mean) was 621 calculated from three independent replicate experiments under each condition. 622 Student’s t-test: **** p < 0.0001, *** p < 0.001, ** p < 0.01, * p 0.05 623 (same throughout the manuscript). 624 (C) Number of SGs per SG+ cell. For each condition, ≥ 49 individual cells were used 625 for quantification. 626 (D) Aggregated area proportion of SGs per SG+ cell. For each condition, ≥ 49 627 individual cells were used for quantification. 628 (E) Western blot analysis of p-eIF2α (serine 51) and total eIF2α. Each condition has 3 629 independent replicate experiments. Quantification was performed as p-eIF2α/eIF2α 630 ratio. 631 (F) Representative images of fluorescent immunostaining (blue: DAPI, red: G3BP1, 632 green: p-eIF2α) in SG- and SG+ cells under the four conditions, respectively. For 633 each bar, average background-subtracted p-eIF2α intensity per unit cell area size was 634 quantified from ≥ 10 individual cells. 635 (G) Representative images of fluorescent immunostaining (blue: DAPI, red: G3BP1, 636 green: OPP) in SG- and SG+ cells under the four conditions, respectively. For each 637 bar, average background-subtracted OPP intensity per unit cell area size was 638 quantified from ≥ 10 individual cells. 639 (H) Representative images of SG formation (blue: DAPI, red: G3BP1) in cells under 640 different glycolytic assaults combined with ISRIB. Each condition has 3 independent 641 replicate experiments. 642 (I-J) Western blot analysis and quantification of p-eIF2α and total eIF2α in cells under 643 different glycolytic assaults combined with ISRIB. Each condition has 3 independent 644 replicate experiments. 645 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint 646 Figure 2. Different glycolytic assaults result in different SG-associated 647 transcripts. 648 (A) Colocalization of G3BP1-APEX2-Myc fusion proteins with biotinylation signals 649 (blue: DAPI, red: G3BP1, gray: MYC, green: biotin) in cytoplasm and SGs under UT, 650 GD, 2DG, and GD+2DG. Scale bar: 10 μm. 651 (B) Hierarchical cluster analysis on pair-wise dissimilarities (calculated as 1 - 652 Pearson’s correlation coefficient) among all bulk and G3BP1-APEX2-Myc-enriched 653 RNA-seq datasets. Each condition has 2 independent replicates. 654 (C) Venn diagram shows the numbers of stress-dependent G3BP1-associated genes 655 (SDGs) resulted from GD, 2DG, and GD+2DG. 656 (D) Gene pathways enriched with common SDGs among GD, 2DG, and GD+2DG. 657 Blue dots represent enriched pathways, with darker colors indicating higher 658 significance, and gray dots denote genes within specific pathway. 659 (E) Transcript lengths of SDGs identified from GD, 2DG and GD+2DG and the other 660 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint expressing genes (non-SDGs). Shown are the results based on the lengths of all 661 quantified transcript isoforms. The trend remains consistent whether using transcript 662 isoforms with the longest total length or the longest CDS (Figure S2). 663 (F) Gene pathways enriched with SDGs identified from GD, 2DG and GD+2DG, 664 respectively. 665 (G) Transcript length distributions of SDGs in the GD+2DG-enriched pathways of 666 OXPHOS, PD, HD, ALS, and ROS. 667 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint 668 Figure 3. Mitochondrial dysfunction under double assault against glycolysis. 669 (A-B) Western blot analysis of NDUFB4 and COX6B1, which were identified as 670 SDGs under GD+2DG. 671 (C-D) Western blot analysis of TOM20, ND6 and ATP6, which were identified as 672 non-SDGs. 673 (E-F) Representative images of immunofluorescence staining against G3BP1 (gray) 674 and TOM20 (red and binary). Quantification was performed by normalizing the 675 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint aggregated area size of identified mitochondria using TOM20 with the cell size. N ≥ 7 676 for each data point. 677 (G) Pearson’s correlation coefficients show pixel intensity correlation between the 678 two channels within single cells of panel E. The analysis was performed by 679 “Colocalization Finder” in ImageJ. Each dot represents one single cell. 680 (H) Relative levels of intracellular ATP under GD, 2DG and GD+2DG at the 681 corresponding SG peaking times. 682 (I) Oxygen consumption rate (OCR) was measured in cells sequentially treated with 683 oligomycin, trifluoromethoxy carbonylcyanide phenylhydrazone (FCCP), and 684 rotenone+antimycin A. Errorbars represent SEM computed from three independent 685 measurements. 686 (J) Relative rates of basal respiration, maximal respiration, ATP production and proton 687 leak, normalized to the corresponding measurements in the side-by-side UT samples. 688 689 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint 690 Figure 4. Distinct stress granule dynamics of cells under different assaults 691 against glycolysis. 692 (A-C) Representative images and quantifications of SG formation (blue: DAPI, red: 693 G3BP1) in 143B cells treated by GD (A), 2DG (B), and GD+2DG (C) for different 694 durations. 695 (D-F) Western blot analysis of p-eIF2α (serine 51) and total eIF2α in cells treated by 696 GD, 2DG, GD+2DG for different durations. Quantification shown on the right was 697 performed by normalizing p-eIF2α to eIF2α. Errorbars represent SEM computed 698 from three independent replicates. 699 (G-I) Western blot analysis of puromycin incorporation assay in cells treated by GD, 700 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint 2DG, GD+2DG for different durations. Quantification shown on the right was 701 performed by normalizing each set of experiments to the control group at 0 h. 702 Errorbars represent SEM computed from three independent replicates. 703 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint 704 Figure 5. Transcriptomic dynamics identifies upregulation of OXPHOS genes as 705 a specific response to double glycolytic assaults. 706 (A) Geneset Activity Score (GAS) analysis identifies four clusters of gene ontologies 707 (GOs) with distinct temporal patterns during the prolonged glycolytic assaults: rise-708 falling (I, dark brown), fall-rising (II, dark green), monotonically rising (III, dark 709 blue), monotonically falling (IV , dark purple). X axis represents time points in 710 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint relation to the SG peaking time; y axis represents changes in GAS, defined as the 711 temporal variation in GSV A-derived scores of GO terms; each line represents one GO 712 term. Shown is the result from GD; see Figure S3 for the results from 2DG and 713 GD+2DG. 714 (B) Venn diagrams illustrate the numbers of GOs with specific temporal patterns that 715 are common or unique in cells under GD, 2DG and GD+2DG. 716 (C) Gene pathways that share the same temporal pattern under all three conditions. 717 Color codes are the same as in (A). 718 (D) RT-qPCR analysis of DNAJC3, DNAJB11 and PPIB, three genes identified in (E). 719 Errorbars represent one standard deviation computed from 3 independent replicate 720 experiments. 721 (E) Gene pathways that share the same temporal pattern between GD alone and 2DG 722 alone but not GD+2DG. Color codes are the same as in (A). 723 (F) Gene pathways that exhibit a specific temporal pattern under GD+2DG but 724 behave differently under GD or 2DG alone. Color codes are the same as in (A). 725 (G) Heat maps showing scaled expression of OXPHOS genes during the prolonged 726 treatments under GD, 2DG and GD+2DG. 727 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint 728 Figure 6. Mitochondrial inhibition renders SGs formed under 2DG treatment 729 non-dissociable. 730 (A) Representative images and quantifications of SG formation (blue: DAPI, red: 731 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint G3BP1) in 143B cells treated with 2DG and mitochondrial inhibition (1 μM rotenone, 732 5 μM antimycin A, or 1.5 μM oligomycin). Scale bar: 20 μm 733 (B) Western blot analysis of p-eIF2α (S51) and eIF2α in 143B cells treated with 2DG 734 and mitochondrial inhibition. 735 (C-D) Representative images and quantification of SG formation (blue: DAPI, red: 736 G3BP1) and p-eIF2α in ρ0 cells were treated with GD for different durations. Scale 737 bar: 20 μm. 738 (E-F) Representative images and quantification of immunofluorescence staining 739 against G3BP1 (red) and TOM20 (gray) in 143B cells treated with 2DG and a siRNA 740 targeting negative control (NC), NDUFB4, COX6B1, NDUFB6, or COX6B2. Scale 741 bar: 20 μm. 742 (G) Pearson’s correlation coefficients quantify the spatial overlap between SGs 743 (G3BP1 intensity) and mitochondria (TOM20 intensity) across treatments. Each dot 744 represents measurements from one single cell. 745 (H) Western blot analysis of TOM20, NDUFB4, NDUFB6, COX6B1, and COX6B2 746 proteins in 143B cells treated with or without 2DG and a siRNA. 747 (I) Quantification of TOM20 protein levels in panel H. 748 749 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint 750 Figure 7. HSP70 is responsible for SG dissociation during prolonged treatments. 751 (A-C) Representative images and quantifications of SG formation (blue: DAPI, red: 752 G3BP1) in 143B cells treated with GD+ wortmannin (A), GD+HSP A1A siRNA (B) 753 and 2DG+VER-155008 (C). Each experimental condition has at least 2 independent 754 replicate experiments. 755 (D) Western blot analysis of HSP70 in cells treated by GD, 2DG, GD+2DG for 756 different durations. 757 (E) Functional enrichment analysis of differential expression genes between GD and 758 GD+siHSP at each corresponding time point. 759 (F) Venn diagram shows the numbers of genes from SDGs under GD, upregulated at 760 16 h relative to 8 h under GD (GD-rising) and upregulated at 16 h relative to 8 h 761 under GD+siHSP (GD+siHSP-rising). Left panel represents the top 5 genes from the 762 intersection of SDGs under GD and HSP70-dependent SG-dissolution genes 763 (upregulated at 16 h relative to 8 h under GD but not under GD+siHSP), ranked by 764 the differences in upregulation levels between GD and GD+siHSP. 765 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint (G) RT-qPCR analysis of SP ATA18 identified in (F). Errorbars represent the standard 766 deviation calculated from 3 independent replicate experiments. 767 768 769 770 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint 771 Figure 8. Cell survival under prolonged glycolytic assaults. 772 (A) Flow cytometry analysis of apoptosis using Annexin V and propidium iodide 773 staining in cells under UT, GD, 2DG, and GD+2DG for 8 or 24h. Errorbars represent 774 SEM computed from three independent replicates. 775 (B) Gene pathways enriched with upregulated genes under prolonged GD (24h), 2DG 776 (8h) and GD+2DG (8h), respectively. 777 (C) Heatmap for fold changes of genes involved in apoptosis (resulted from (B)) 778 under prolonged GD, 2DG and GD+2DG, compared to the control. RT-qPCR 779 confirmed the increase in the mRNA levels of BCL2L11, FOS and ITPR1, three genes 780 in the apoptosis pathway, under GD+2DG but not under GD or 2DG alone. 781 (D) Flow cytometry analysis of apoptosis in cells treated with VER-155008, 782 GD+VER, and 2DG+VER. Errorbars represent SEM computed from three 783 independent replicates. 784 785 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint 786 Figure 9. A model showing the formation of two distinct types of SGs under 787 glycolytic inhibition. 788 The double assault (glucose depletion + 2DG treatment) leads to the generation of 789 Type II SGs, which specifically sequester mRNA molecules related to oxidative 790 phosphorylation and intensify cellular energy deficits. This energy crisis drives Type 791 II SGs into a non-dissolvable state. By contrast, single glycolytic assaults lead to the 792 formation of Type I SGs, which can naturally dissolve during prolonged treatment. 793 Type I SGs may be transformed into the non-dissolvable state by additional 794 mitochondrial inhibition (or HSP70 dysfunction). This model depicts the scenario 795 where SGs serve as both a stress responder and a driver of metabolic collapse, 796 offering mechanistic insight into the association of pathological SGs. 797 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint 798 Figure S1. GADD34 is reduced and negatively correlated with p-eIF2α under 799 single glycolytic assaults but not under double assault. 800 (A-B) Western blot analysis of GADD34, p-eIF2α, and eIF2α in cells under GD, 2DG 801 or GD+2DG. Quantification was performed by normalizing p-eIF2α to eIF2α and 802 normalizing GADD34 to its corresponding internal control. Mean ± SEM was 803 calculated from three independent replicate experiments. 804 805 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint 806 Figure S2. Comparison of transcript lengths under different conditions. 807 (A, C): Transcript lengths of SDGs resulting from GD, 2DG, GD+2DG and non-808 SDGs, measured using the longest total length (A), or the longest CDS (C), 809 respectively. 810 (B, D): Distribution of transcript lengths for SDGs enriched in specific pathways 811 under GD+2DG, including OXPHOS, PD, HD, ALS, and ROS, measured using the 812 longest total length (B) or the longest CDS (D), respectively. 813 814 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint 815 Figure S3. Time-course Geneset Activity Score (GAS) analysis under different 816 conditions. 817 (A-B) Same as Figure 5A but results from 2DG (A) and GD+2DG (B), respectively. 818 819 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint 820 Figure S4. Cell cycle is similarly disturbed by different glycolytic assaults. 821 (A-B) Flow cytometry analysis of cell cycle using PI staining in cells under GD for 8h 822 (A) and GD+2DG for 1h (B). Both treatments present a moderate increase in the 823 proportion of G1 cells: from 28.63 ± 7.72% to 32.61 ± 6.29% under GD and from 824 27.12 ± 6.21% to 32.32 ± 4.41% under GD+2DG. 825 826 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint 827 Figure S5. Distinct stress granule responses of HeLa cells under different 828 glycolytic assaults. 829 (A-C) Same as Figure 2A-C but results from HeLa cells. 830 831 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint Table S1. Primer and oligo sequences. 832 RT-qPCR primers Gene Forward primer 5’-3’ Reverse primer 5’-3’ DNAJC3 CTGCAGTACGAAGGTGCTGA ACGGCAGCATGAAACTGAGA DNAJB11 ATCAAAGTTGTCAAGCACCC A GGCCTGGTGATCTTATCCCG PPIB GCGGCCGATGAGAAGAAGA CGTAGATGCTCTTTCCTCCTGT BCL2L11 GCTACCAGATCCCCGCTTTT CCTGCCTCATGGAAGCCATTG FOS TGGCGTTGTGAAGACCATGA AGTTGGTCTGTCTCCGCTTG ITPR1 GAGTTTCAGCCCTCAGTGGA GCAGAGTGGTGGGATCTAGC SPATA18 GAAGAGAACACCCTTCCCGC TGATCACACGTGTTTGTGTTG T siRNAs Gene Sequence 5’-3’ NC UUCUCCGAACGUGUCACGUTT HSP70/HSPA1A CGUCCAUGGUGCUGACCAAGA COX6B1 GCGAUAUCUCUGUGUGCGAAU COX6B2 CGUGGAUGUUGGAUGUGGAAG NDUFB4 AGAUGUCGUUCCCAAAGUAUA NDUFB6 GAAAGAAUUUCCUGAUCAACA .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint

833 Aguilera-Gomez A, Zacharogianni M, Van Oorschot MM, Genau H, Grond R, 834 Veenendaal T, Sinsimer KS, Gavis ER, Behrends C, Rabouille C. 2017. Phospho -835 Rasputin Stabilization by Sec16 Is Required for Stress Granule Formation upon 836 Amino Acid Starvation. Cell Reports 20: 935–948. 837 Amen T, Kaganovich D. 2021. Stress granules inhibit fatty acid oxidation by 838 modulating mitochondrial permeability. Cell Reports 35: 109237. 839 doi:10.1016/j.celrep.2021.109237. 840 Aulas A, Fay MM, Lyons SM, Achorn CA, Kedersha N, Anderson P, Ivanov P. 2017. 841 Stress-specific differences in assembly and composition of stress granules and related 842 foci. Journal of Cell Science 130: 927–937. 843 Buchan JR, Kolaitis R-M, Taylor JP, Parker R. 2013. Eukaryotic Stress Granules Are 844 Cleared by Autophagy and Cdc48/VCP Function. Cell 153: 1461–1474. 845 Buchan JR, Parker R. 2009. Eukaryotic Stress Granules: The Ins and Outs of 846 Translation. Molecular Cell 36: 932–941. 847 Campos-Melo D, Hawley ZCE, Droppelmann CA, Strong MJ. 2021. The Integral 848 Role of RNA in Stress Granule Formation and Function. Front Cell Dev Biol 9: 621 -849 779. 850 Cui Q, Bi H, Lv Z, Wu Q, Hua J, Gu B, Huo C, Tang M, Chen Y, Chen C, et al. 2023. 851 Diverse CMT2 neuropathies are linked to aberrant G3BP interactions in stress 852 granules. Cell 186: 803-820.e25. 853 Cui Q, Liu Z, Bai G. 2024. Friend or foe: The role of stress granule in 854 neurodegenerative disease. Neuron 112: 2464–2485. 855 Das S, Santos L, Failla AV, Ignatova Z. 2022. mRNAs sequestered in stress granules 856 recover nearly completely for translation. RNA Biology 19: 877–884. 857 Eiermann N, Stoecklin G, Jovanovic B. 2022. Mitochondrial Inhibition by Sodium 858 Azide Induces Assembly of eIF2α Phosphorylation-Independent Stress Granules in 859 Mammalian Cells. IJMS 23: 5600. doi:10.3390/ijms23105600. 860 Emara MM, Fujimura K, Sciaranghella D, Ivanova V, Ivanov P, Anderson P. 2012. 861 Hydrogen peroxide induces stress granule formation independent of eIF2α 862 phosphorylation. Biochemical and Biophysical Research Communications 423: 763 –863 769. 864 Eum H, Shin Y, Song Y, Kim Y, Kang S-W. 2020. ATP-driven reactions are required 865 for the assembly of large stress granules. Biochemical and Biophysical Research 866 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint Communications 521: 238–244. 867 Frydrýšková K, Mašek T, Pospíšek M. 2020. Changing faces of stress: Impact of heat 868 and arsenite treatment on the composition of stress granules. WIREs RNA 11: e1596. 869 doi:10.1002/wrna.1596. 870 Fu X, Gao X, Ge L, Cui X, Su C, Yang W, Sun X, Zhang W, Yao Z, Yang X, et al. 871 2016. Malonate induces the assembly of cytoplasmic stress granules. FEBS Letters 872 590: 22–33. 873 Fujimura K, Sasaki AT, Anderson P. 2012. Selenite targets eIF4E-binding protein-1 874 to inhibit translation initiation and induce the assembly of non -canonical stress 875 granules. Nucleic Acids Res 40: 8099–8110. 876 Ganassi M, Mateju D, Bigi I, Mediani L, Poser I, Lee HO, Seguin SJ, Morelli FF, 877 Vinet J, Leo G, et al. 2016. A Surveillance Function of the HSPB8 -BAG3-HSP70 878 Chaperone Complex Ensures Stress Granule Integrity and Dynamism. Molecular Cell 879 63: 796–810. 880 Gaowa S, Futamura M, Tsuneki M, Kamino H, Tajima JY, Mori R, Arakawa H, 881 Yoshida K. 2018. Possible role of p53/Mieap‐regulated mitochondrial quality control 882 as a tumor suppressor in human breast cancer. Cancer Science 109: 3910–3920. 883 Hofmann S, Kedersha N, Anderson P, Ivanov P. 2021. Molecular mechanisms of 884 stress granule assembly and disassembly. Biochim Biophys Acta Mol Cell Res 1868: 885 118876. doi:10.1016/j.bbamcr.2020.118876. 886 Hu S, Zhang Y, Yi Q, Yang C, Liu Y, Bai Y. 2023. Time-resolved proteomic 887 profiling reveals compositional and functional transitions across the stress granule life 888 cycle. Nat Commun 14: 7782. doi:10.1038/s41467-023-43470-1. 889 Ikari N, Honjo K, Sagami Y, Nakamura Y, Arakawa H. 2024. Mieap forms 890 membrane-less organelles involved in cardiolipin metabolism. iScience 27: 108916. 891 doi:10.1016/j.isci.2024.108916. 892 Ivanov P, Kedersha N, Anderson P. 2019. Stress Granules and Processing Bodies in 893 Translational Control. Cold Spring Harb Perspect Biol 11: a032813. 894 doi:10.1101/cshperspect.a032813. 895 Jain S, Wheeler JR, Walters RW, Agrawal A, Barsic A, Parker R. 2016. ATPase-896 Modulated Stress Granules Contain a Diverse Proteome and Substructure. Cell 164: 897 487–498. 898 Jia Y, Jia R, Dai Z, Zhou J, Ruan J, Chng W, Cai Z, Zhang X. 2024. Stress granules 899 in cancer: Adaptive dynamics and therapeutic implications. iScience 27: 110359. 900 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint doi:10.1016/j.isci.2024.110359. 901 Kedersha N, Chen S, Gilks N, Li W, Miller IJ, Stahl J, Anderson P. 2002. Evidence 902 That Ternary Complex (eIF2-GTP-tRNA i Met ) –Deficient Preinitiation Complexes 903 Are Core Constituents of Mammalian Stress Granules. MBoC 13: 195–210. 904 Khong A, Matheny T, Jain S, Mitchell SF, Wheeler JR, Parker R. 2017. The Stress 905 Granule Transcriptome Reveals Principles of mRNA Accumulation in Stress 906 Granules. Molecular Cell 68: 808-820.e5. 907 Lacerda R, Menezes J, Candeias MM. 2019. Alternative Mechanisms of mRNA 908 Translation Initiation in Cellular Stress Response and Cancer. In The mRNA 909 Metabolism in Human Disease, Vol. 1157 of Advances in Experimental Medicine and 910 Biology, pp. 117–132, Springer International Publishing, 10.1007/978-3 -030-19966-911 1_6 10.1007/978-3-030-19966-1_6 (Accessed March 28, 2025). 912 Lee C-Y, Seydoux G. 2019. Dynamics of mRNA entry into stress granules. Nat Cell 913 Biol 21: 116–117. 914 Luo Y-Y, Wu J-J, Li Y-M. 2021. Regulation of liquid –liquid phase separation with 915 focus on post-translational modifications. Chem Commun 57: 13275–13287. 916 Mahboubi H, Stochaj U. 2017. Cytoplasmic stress granules: Dynamic modulators of 917 cell signaling and disease. Biochimica et Biophysica Acta (BBA) - Molecular Basis of 918 Disease 1863: 884–895. 919 Marmor-Kollet H, Siany A, Kedersha N, Knafo N, Rivkin N, Danino YM, Moens 920 TG, Olender T, Sheban D, Cohen N, et al. 2020. Spatiotemporal Proteomic Analysis 921 of Stress Granule Disassembly Using APEX Reveals Regulation by SUMOylation 922 and Links to ALS Pathogenesis. Molecular Cell 80: 876-891.e6. 923 Mateju D, Franzmann TM, Patel A, Kopach A, Boczek EE, Maharana S, Lee HO, 924 Carra S, Hyman AA, Alberti S. 2017. An aberrant phase transition of stress granules 925 triggered by misfolded protein and prevented by chaperone function. EMBO J 36: 926 1669–1687. 927 Mazroui R, Di Marco S, Kaufman RJ, Gallouzi I-E. 2007. Inhibition of the Ubiquitin-928 Proteasome System Induces Stress Granule Formation. MBoC 18: 2603–2618. 929 Namkoong S, Ho A, Woo YM, Kwak H, Lee JH. 2018. Systematic Characterization 930 of Stress-Induced RNA Granulation. Molecular Cell 70: 175-187.e8. 931 Parameswaran J, Zhang N, Braems E, Tilahun K, Pant DC, Yin K, Asress S, Heeren 932 K, Banerjee A, Davis E, et al. 2023. Antisense, but not sense, repeat expanded RNAs 933 activate PKR/eIF2α-dependent ISR in C9ORF72 FTD/ALS. eLife 12: e85902. 934 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint doi:10.7554/eLife.85902. 935 Pernin F, Cui Q- L, Mohammadnia A, Fernandes MGF, Hall JA, Srour M, Dudley 936 RWR, Zandee SEJ, Klement W, Prat A, et al. 2024. Regulation of stress g ranule 937 formation in human oligodendrocytes. Nat Commun 15: 1524. doi:10.1038/s41467-938 024-45746-6. 939 Reineke LC, Cheema SA, Dubrulle J, Neilson JR. 2018. Chronic starvation induces 940 non-canonical pro-death stress granules. Journal of Cell Science jcs.220244. 941 doi:10.1242/jcs.220244. 942 Ren Z, Tang W, Peng L, Zou P. 2023. Profiling stress-triggered RNA condensation 943 with photocatalytic proximity labeling. Nat Commun 14: 7390. doi:10.1038/s41467-944 023-43194-2. 945 Sato K, Takayama K- I, Inoue S. 2024. Stress granule-mediated RNA regulatory 946 mechanism in Alzheimer’s disease. Geriatr Gerontol Int 24 Suppl 1: 7–14. 947 Sfakianos AP, Mellor LE, Pang YF, Kritsiligkou P, Needs H, Abou -Hamdan H, 948 Désaubry L, Poulin GB, Ashe MP, Whitmarsh AJ. 2018. The mTOR- S6 kinase 949 pathway promotes stress granule assembly. Cell Death Differ 25: 1766–1780. 950 Sidrauski C, McGeachy AM, Ingolia NT, Walter P. 2015. The small molecule ISRIB 951 reverses the effects of eIF2α phosphorylation on translation and stress granule 952 assembly. eLife 4: e05033. doi:10.7554/eLife.05033. 953 Somasekharan SP, Zhang F, Saxena N, Huang JN, Kuo I- C, Low C, Bell R, Adomat 954 H, Stoynov N, Foster L, et al. 2020. G3BP1-linked mRNA partitioning supports 955 selective protein synthesis in response to oxidative stress. Nucleic Acids Research 48: 956 6855–6873. 957 Szczerba M, Johnson B, Acciai F, Gogerty C, McCaughan M, Williams J, Kibler KV, 958 Jacobs BL. 2023. Canonical cellular stress granules are required for arsenite-induced 959 necroptosis mediated by Z- DNA–binding protein 1. Sci Signal 16: eabq0837. 960 doi:10.1126/scisignal.abq0837. 961 Tauber D, Tauber G, Khong A, Van Treeck B, Pelletier J, Parker R. 2020. Modulation 962 of RNA Condensation by the DEAD-Box Protein eIF4A. Cell 180: 411-426.e16. 963 Van Leeuwen W, Rabouille C. 2019. Cellular stress leads to the formation of 964 membraneless stress assemblies in eukaryotic cells. Traffic 9: 623-638. 965 Van Treeck B, Protter DSW, Matheny T, Khong A, Link CD, Parker R. 2018. RNA 966 self-assembly contributes to stress granule formation and defining the stress granule 967 transcriptome. Proc Natl Acad Sci USA 115: 2734–2739. 968 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint Wang A, Abulaiti X, Zhang H, Su H, Liu G, Gao S, Li L. 2022. Cancer Cells Evade 969 Stress-Induced Apoptosis by Promoting HSP70-Dependent Clearance of Stress 970 Granules. Cancers 14: 4671. doi:10.3390/cancers14194671. 971 Wang T, Tian X, Kim HB, Jang Y, Huang Z, Na CH, Wang J. 2022. Intracellular 972 energy controls dynamics of stress-induced ribonucleoprotein granules. Nat Commun 973 13: 5584. doi:10.1038/s41467-022-33079-1. 974 Wheeler JR, Matheny T, Jain S, Abrisch R, Parker R. 2016. Distinct stages in stress 975 granule assembly and disassembly. eLife 5: e18413. doi:10.7554/eLife.18413. 976 Wolozin B, Ivanov P. 2019. Stress granules and neurodegeneration. Nat Rev Neurosci 977 20: 649–666. 978 Yang Y, Wang Z. 2019. IRES-mediated cap-independent translation, a path leading to 979 hidden proteome. Journal of Molecular Cell Biology 11: 911–919. 980 Youn J- Y, Dunham WH, Hong SJ, Knight JDR, Bashkurov M, Chen GI, Bagci H, 981 Rathod B, MacLeod G, Eng SWM, et al. 2018. High -Density Proximity Mapping 982 Reveals the Subcellular Organization of mRNA-Associated Granules and Bodies. 983 Molecular Cell 69: 517-532.e11. 984 Youn J-Y, Dyakov BJA, Zhang J, Knight JDR, Vernon RM, Forman-Kay JD, Gingras 985 A-C. 2019. Properties of Stress Granule and P-Body Proteomes. Molecular Cell 76: 986 286–294. 987 Zhang P, Fan B, Yang P, Temirov J, Messing J, Kim HJ, Taylor JP. 2019. Chronic 988 optogenetic induction of stress granules is cytotoxic and reveals the evolution of 989 ALS-FTD pathology. eLife 8: e39578. doi:10.7554/eLife.39578. 990 Zheng D, Mao Y, Gao Y, He F, Ma J. 2023. Daught er cell fate choice instructed 991 preemptively by mother cells facing nutrient limitation. iScience 26: 107198. 992 doi:10.1016/j.isci.2023.107198. 993 .CC-BY 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted September 2, 2025. ; https://doi.org/10.1101/2025.09.02.673658doi: bioRxiv preprint