The integrated stress response remodels the microtubule-organizing center to clear unfolded proteins following proteotoxic stress

Cells encountering stressful situations activate the integrated stress response (ISR) pathway to limit protein synthesis and redirect translation to better cope. The ISR has also been implicated in cancers, but redundancies in the stress-sensing kinases that trigger the ISR have posed hurdles to dissecting physiological relevance. To overcome this challenge, we targeted the regulatory node of these kinases, namely, the S51 phosphorylation site of eukaryotic translation initiation factor eIF2α and genetically replaced eIF2α with eIF2α-S51A in mouse squamous cell carcinoma (SCC) stem cells of skin. While inconsequential under normal growth conditions, the vulnerability of this ISR-null state was unveiled when SCC stem cells experienced proteotoxic stress. Seeking mechanistic insights into the protective roles of the ISR, we combined ribosome profiling and functional approaches to identify and probe the functional importance of translational differences between ISR-competent and ISR-null SCC stem cells when exposed to proteotoxic stress. In doing so, we learned that the ISR redirects translation to centrosomal proteins that orchestrate the microtubule dynamics needed to efficiently concentrate unfolded proteins at the microtubule-organizing center so that they can be cleared by the perinuclear degradation machinery. Thus, rather than merely maintaining survival during proteotoxic stress, the ISR also functions in promoting cellular recovery once the stress has subsided. Remarkably, this molecular program is unique to transformed skin stem cells, hence exposing a vulnerability in cancer that could be exploited therapeutically.

. This selective repurposing of the translational machinery forms the 50 foundation of a cellular stress response aimed at restoring homeostasis. 51 52 The best studied translational target of the ISR is Atf4, encoding activating 53 transcription factor 4 (ATF4) whose translation is repressed in homeostatic conditions 54 by the presence of inhibitory uORFs in its 5' UTR. Upon stress, Atf4 translation is 55 unleashed, and the newly synthesized protein mediates transcriptional upregulation of 56 stress response genes, thus acting as a crucial regulator in the balance of cell death 57 and survival. In fact, known ATF4 targets include both cytoprotective and apoptotic 58 factors. That a single ISR target can have such diverging and context-specific roles 59 highlights the complexity of the cellular stress response and indicates the possibility 60 that the ISR might translationally regulate additional cellular programs to coordinate 61 survival, adaptation and recovery from stress. 62 63 At the interfaces of cellular proliferation, apoptosis, survival, and protein synthesis, the 64 ISR has naturally emerged as an area of interest in cancer research. Although studies 65 on the role of the ISR in cancer have been conflicting, the notion that the ISR is 66 protective for cancer cells has gained traction in recent years (Ghaddar et al., 2021;67 Koromilas, 2015). This is in line with the observation that cancer cells experience 68 hostile microenvironments characterized by nutrient deprivation and low oxygen 69 availability. Additionally, the proliferative stress and elevated metabolic demands of 70 cancer cells create an increased reliance on the cellular mechanisms that maintain 71 proteostasis, a central function of the ISR (Cubillos-Ruiz et al., 2017). Indeed, 72 proteasome inhibition as an anticancer therapy, is aimed at exploiting this vulnerability. 73 That said, although proteasome inhibitors have become the standard of care for 74 multiple myeloma and mantle-cell lymphoma (Manasanch & Orlowski, 2017), this 75 strategy has been less effective in solid tumors, suggesting that these tumors may 76 utilize mechanisms that enable them to cope, survive and recover in the face of 77 proteotoxic stress, thereby evading therapeutics (Tian et al., 2021).

79
In the current study, we focus on the most common and life-threatening solid tumors, 80 squamous cell carcinomas (SCCs), which affect stratified squamous epithelia of the 81 skin, esophagus, lung and head and neck. To investigate the roles of the ISR in this 82 cancer, we took advantage of our ability to culture the tumor-initiating stem cells from 83 mouse skin SCCs and generated primary, clonal cell lines in which the endogenous 84 eIF2α alleles were replaced by myc-epitope tagged but otherwise fully functional 85 versions of either wild-type eIF2α or eIF2α-S51A (ISR-null). Although unable to mount 86 an ISR in the face of stress, ISR-null SCC stem cells formed tumors that were similar 87 in morphology and proliferation characteristics to controls. However, both in vivo and 88 in vitro, when challenged with proteotoxic stress, ISR-null SCC cells fared 89 considerably more poorly than their ISR-competent counterparts. 90 91 While the HRI and PERK kinases have emerged as major players in sensing misfolded 92 proteins (Abdel-Nour et al., 2019;Harding et al., 1999), a detailed understanding of 93 how the ISR promotes proteostasis is lacking. To probe into the mechanisms that link 94 the ISR to proteostasis, we coupled genetics, cell biology, pharmacological inhibitors, 95 ribosomal profiling and finally functional analyses. We traced the connection to a group 96 of centrosomal proteins that become selectively translated in response to prototeoxic 97 stress and which act by strengthening the organizing center for the microtubule 98 dynamics that are needed to efficiently amass unfolded proteins at the 99 pericentrosomal locale, where they can be efficiently targeted for destruction and 100 clearance. Our findings add a new dimension-microtubule dynamics--to the role of 101 the ISR not only in stress, but also in the recovery of cells to stress. In so doing, our 102 findings also expose a hitherto unappreciated vulnerability of cancer cells when they 103 are unable to mount an ISR in the face of proteotoxic stress. 104 105 Results 106 107 Generation of ISR-null SCC cells 108 109 In order to directly test the role of the integrated stress response (ISR) in SCC cells, 110 we generated eIF2α-S51A, or "ISR-null" cancer cells using a knockout and 111 reconstitution strategy. For this purpose, we used an aggressive HRas G12V murine, 112 primary-derived skin SCC line expressing an eGFP reporter . We 113 transduced these cells with lentiviral constructs harboring a PGK promoter-driven 114 cDNA encoding either the wild-type (S51) or phospho-dead (S51A) eIF2α protein 115 (encoded by the eif2s1 locus) ( Figure 1B). The eif2s1 transgenes each contained a 116 synonymous mutation in a protospacer adjacent motif (PAM) site that rendered it 117 resistant to a small guide RNA (sgRNA) that could be used to specifically target the 118 endogenous eif2s1 gene for CRISPR/Cas9 deletion. 119 120 After cells were transduced with the myc-tagged eif2s1 constructs, cells were 121 transfected with liposomes harboring CRIPSR-Cas9/sgRNA ribonucleoproteins, which 122 targeted the ablation of the endogenous eif2s1 alleles, and thus left the myc-tagged 123 S51 or S51A transgenes as the sole source of eIF2α expression. Following ablation 124 and reconstitution, single cells were isolated by fluorescence activated cell sorting 125 (FACS) and used to generate stable SCC clones. Successful targeting of the two 126 endogenous eif2s1 alleles was verified by genomic sequencing ( Figure 1B). 127 128 Two of each S51 and S51A eIF2α replacement clones were chosen for further study. 129 In all assays presented in Figures 1 and 2, the clones of the same eIF2α status grew 130 and behaved similarly. Hence for the purposes of the current study, we show the 131 results on pooled clones displaying common genotypes. Immunoblot analyses 132 revealed that total eIF2α levels were comparable to the parental eIF2α clone, and the 133 replacement eIF2α proteins exhibited the expected increase in size due to the epitope 134 tag ( Figure 1C). Importantly, when we treated these SCC lines with sodium arsenite 135 to induce oxidative stress and activate the HRI kinase (Sendoel et al., 2017), we 136 observed that like the parental clone, the stressed eIF2α S51 cells displayed phospho-137 S51 immunolabeling, while the eIF2α S51A clones were refractory to phosphorylation. 138 Taken together, these results underscored the efficacy of our knockout and 139 replacement strategy, and verified the dramatic difference in stress-induced abilities 140 of our two clones to target eIF2α phosphorylation at the heart of the ISR.

142
To further interrogate the functionality of our SCC lines in mounting an ISR, we 143 monitored their protein synthesis rates in response to stress. Using puromycin 144 incorporation as a gauge to measure nascent protein synthesis, we observed that 145 translation within eIF2α S51 cells was comparable to the native eIF2α parental clone, 146 and markedly dampened upon stress as expected ( Figure 1D). Consistent with the 147 inability of eIF2α-S51A to become phosphorylated and decommissioned, the eIF2α-148 S51A SCC cells sustained higher levels of protein synthesis during oxidative stress 149 than their control counterparts ( Figure 1D). Similar results were seen upon exposure 150 of our SCC cell lines to tunicamycin, which induces ER-stress and eIF2α-S51 151 phosphorylation through PERK kinase.

156
Typical of a classical ISR, global protein synthesis was reduced in our SCC cells 157 exposed to bortezomib, consistent with pathway activation ( Figure 1D and Figure 1-158 figure supplement 1A). In contrast, however, there did not seem to be a marked 159 difference in protein synthesis rates between bortezomib treated eIF2α-S51 and 160 eIF2α-S51A cells. In this regard, SCC cells appeared to face proteotoxic stress in a 161 manner distinct from a classical integrated stress response.

163
Probing deeper, we found that under the serum-rich culture conditions used here, S51 164 and S51A cells proliferated at comparable rates and showed no differences in viability 165 ( upon exposure to stress, eIF2α-S51A cells were significantly more sensitive than SCC 168 cells with an intact ISR. Particularly, after 24 hrs of bortezomib treatment, the 169 percentage of cells positive for the apoptotic marker Annexin-V was appreciably higher 170 in S51A over S51 cultures ( Figure 1E; Figure 1-figure supplement 1E). Taken together, 171 our collective data thus far pointed to the view that when faced with proteotoxic stress, 172 SCC cells harboring a functional ISR may gain a fitness advantage over their ISR null 173 counterparts not by selectively dampening down global protein synthesis, but rather 174 by shifting to a translational landscape that is more favorable for survival. 175 176 177 The ISR promotes efficient clearance of protein aggregates 178 179 When epidermal stem cells acquire a single oncogenic mutation and will eventually 180 progress to SCC, they activate alternative translational pathways suggestive of an ISR 181 (Blanco et al., 2016;Sendoel et al., 2017). Intriguingly, however, when we injected our 182 SCC cells into the skins of immunocompromised, athymic (Nude) mice, S51A and S51 183 SCC cells both formed tumors that grew similarly and showed similar morphologies 184 ( . This was not attributable to 'escapers' that 185 had somehow circumvented the S51A mutation in vivo, as FACS sorted SCC stem 186 cells from our S51A tumors still displayed insensitivity to sodium arsenite-induced 187 eIF2α phosphorylation in contrast to their S51 counterparts (Figure 2- figure  188 supplement 1B).

189
Most strikingly, when we treated our mice with bortezomib, the in vivo S51A SCC 190 tumors exhibited significantly greater sensitivity to proteotoxic stress and apoptosis 191 than their counterparts ( Figure 2B). These findings were particularly relevant given 192 that despite intense interest in proteasome inhibitors as a potential new line of cancer 193 therapeutics, solid tumors have shown resistance to these drugs (Fournier et al., 2010;194 Manasanch & Orlowski, 2017). Taken together, our results raised the tantalizing 195 possibility that if the ISR is first crippled in solid tumors, their tumor-propagating stem 196 cells may be increasingly sensitive to added stress.

198
While the ISR is known to be activated by the types of unfolded proteins that 199 accumulate upon proteasome inhibition (Figure 1-figure supplement 1A) (Pakos-200 Zebrucka et al., 2016), it is incompletely understood how the pathway promotes cell 201 recovery and survival in the face of this stress. Intrigued that ISR-null cells and tumors 202 had an intrinsic liability in coping with protein aggregate stress, we established a two-203 step in vitro model of first triggering SCC cells to accumulate an excess of unfolded 204 proteins and then allowing them to recover from the stress. To this end, we treated 205 cultured control or ISR-null SCC cells in vitro with a saturating dose of bortezomib for 206 6 hrs, at which point we washed the cells and switched to fresh media to allow cells to 207 recover ( Figure 2C). Following this bortezomib "pulse", control SCC cells with a 208 competent ISR recovered and began to proliferate within 24 to 48 hrs. In striking 209 contrast, ISR-null SCC cells took a full 24 hrs longer than their counterparts before 210 they began to proliferate again ( Figure 2D). Intriguingly, this could not be imputed to 211 differences in viability, since following this treatment regime, ISR-null and ISR-intact 212 SCC stem cells displayed no difference in Annexin-V positivity either at 6 hrs of 213 bortezomib treatment or at 24 hrs after recovery (Figure 2-figure supplement 2A-C).

215
The first step in targeting unfolded proteins for degradation is their ubiquitination 216 (Smith et al., 2011). We therefore examined the clearance of ubiquitinated proteins in 217 SCC cells recovering from proteotoxic stress. Following treatment with bortezomib, 218 cells were lysed in Radio-Immunoprecipitation Assay (RIPA) buffer, and soluble and 219 insoluble proteins were then fractionated by centrifugation. Each fraction was 220 normalized based on the protein concentration of the soluble fraction and then 221 subjected to polyacrylamide gel electrophoresis and analyzed by anti-ubiquitin 222 immunoblots.

224
ISR-null and control SCC cells responded similarly to bortezomib, displaying a rapid 225 jump in ubiquitinated proteins within 6 hrs of treatment ( Figure 2E-F). After withdrawing 226 bortezomib, however, the rate at which ubiquitinated proteins were cleared from the 227 insoluble fraction was remarkably reduced in the ISR-null cells compared to controls 228 SCC cell lysates. As lysine 48-linked polyubiquitin is the specific mark of proteins 229 targeted for proteasomal degradation (Thrower et al., 2000), we performed anti-K48-230 polyubiquitin immunoblot analyses ( Figure 2E). These data confirmed that SCC cells 231 that are unable to mount an ISR are not defective in their E3-ubiquitin-ligase system 232 per se, but rather are impaired in their ability to efficiently clear proteins that are 233 marked for proteosomal destruction.

235
We also used this assay to interrogate the source of the insoluble protein aggregates 236 induced by bortezomib. To do so, we treated cells concurrently with cycloheximide, to 237 block new protein synthesis and with bortezomib, to inhibit proteosome-mediated 238 aggregated protein clearance. Inhibiting translation elongation with cyclohexamide or 239 translation initiation with harringtonine nearly quantitatively blocked the buildup of 240 protein aggregates following 6 hrs of bortezomib treatment. These data suggested that 241 if an acute block in proteasome function occurs and SCC cells cannot respond quickly 242 by slowing new protein synthesis, an imbalance in proteostasis arises, resulting in an 243 accumulation of newly synthesized, unfolded proteins (Figure 2-figure supplement 244 3A).

246
We next addressed the extent to which the proteosome versus autophagy is 247 responsible for the clearance of these ubiquitin-marked protein aggregates during the 248 recovery phase following bortezomib treatment. To do so, we treated with bortezomib 249 for 6 hrs, and then after washing out the drug, we either allowed both pathways to 250 participate in clearance or added bafilomycin A1 to block the autophagy pathway. As 251 shown in Figure 2-figure supplement 3B, BafA1 delayed recovery partially but not fully. 252 Taken together, these data indicated that both autophagy and the proteasome 253 cooperate in clearing these ubiquitin-marked protein aggregates during the recovery 254 phase following proteotoxic stress. 255 256 257 The ISR responds to proteotoxic stress by upregulating translation of 258 centrosomal proteins 259 260 Our findings pointed to the view that the ISR is a major player in the process of cellular 261 recovery from proteotoxic stress. The delayed rate in protein aggregate clearance 262 seen in ISR-null cells was not attributable to a higher rate of global protein synthesis, 263 as this was comparable to the ISR-competent control cells ( Figure 1D). We therefore 264 asked whether the ISR might be required to drive translation of select mRNAs upon 265 proteotoxic stress, and if so, whether the proteins produced under such circumstances 266 might give us clues into how the ISR functions in recovery. To this end, we performed 267 ribosome profiling to landscape the ISR-mediated impact on translation during 268 proteotoxic stress ( Figure 3A) (Ingolia et al., 2009;McGlincy & Ingolia, 2017).

270
Briefly, we subjected control and ISR-null SCC cells to proteotoxic stress (bortezomib) 271 versus vehicle control. Quadruplicate samples of each condition were then lysed in 272 the presence of cycloheximide in order to preserve the ribosome location along 273 transcripts. Total mRNAs were saved, and the remaining lysates were treated with 274 RNase-I to digest away all mRNAs that were not protected by ribosomes. Total 275 mRNAs and the ribosome protected fragments (RPFs) were then prepared for deep 276 sequencing, and the translational efficiency was analyzed by assessing the ratio of 277 RPFs to total mRNA reads, genome wide.

279
Reads were aligned to the mouse reference genome, and quality control was 280 performed to confirm that we had successfully purified and sequenced RNA fragments 281 that were protected by actively translating ribosomes. Technical replicates were found 282 to co-vary within samples by subjecting them to principal component analysis ( Figure  283 3-figure supplement 1A). As a second quality control, metagene analysis was 284 performed, which demonstrated that the majority of RPFs were found within coding 285 segments, with relative peaks at the translation start and stop sites, as would be 286 expected from a high quality ribosome profiling dataset ( Having verified the efficacy of our data, we next focused on the translational response 292 to proteotoxic stress in eIF2α-S51 control SCC cells bearing an intact ISR. To this end, 293 we sought to identify mRNAs that displayed increased ribosome occupancy (RPF FC 294 >1.5 and padj(RPF) <0.05) in bortezomib compared to vehicle treated cells. To 295 eliminate possible translational variances arising from transcriptional differences, we 296 also normalized the RPF reads of each transcript according to total mRNA levels 297 (Ingolia et al., 2009;McGlincy & Ingolia, 2017). This allowed us to identify genes with 298 translational efficiencies (TE) (RPF/mRNA) that are sensitive to bortezomib, and 299 whose translational changes are at the heart of ISR-mediated differences ( Figure 3B). 300 Interestingly, in our S51-SCC cells with an intact ISR, proteotoxic stress provoked the 301 translational upregulation of 199 mRNAs (RPF FC >1.5, padj (RPF) < 0.05 and TE FC 302 > 1.5) ( Figure 3C). 303 304 Next, we performed the same analysis to compare ISR-competent and ISR-null cells 305 upon proteotoxic stress. This revealed 167 mRNAs whose translational upregulation 306 was dependent on the presence of an intact ISR (S51-bort vs S51A-bort (RPF FC 307 >1.5, padj (RPF) < 0.05 and TE FC > 1.5)) ( Figure 3D). These findings clearly showed 308 that when SCC cells are refractory to proteotoxic stress-induced phosphorylation of 309 eIF2α, their translational program is selectively perturbed. 310 311 To curate a list of specific ISR-targeted mRNAs, we identified genes for which the 312 translation changed in response to bortezomib and only in cells with an intact ISR. We used gene ontology enrichment analysis (GO-term) to ask if the ISR targeted sets 321 of mRNAs corresponded to common biological processes or cellular components. The term analysis of the 24 upregulated ISR-targets was especially interesting ( Figure 3G).

326
Regulation of stress induced transcription featured prominently in the top GO-terms 327 for biological processes and was exemplified by the presence of ATF4, a well-328 established stress induced transcriptional regulator that is activated in premalignant 329 SCC cells by non-canonical translation when eIF2α is phosphorylated (Sendoel et al., 330 2017). Most intriguing, however, were the top three GO-terms for cellular components: 331 Centrosomal proteins, Microtubule organizing center (MTOC) proteins and 332 Microtubule cytoskeletal proteins (Figures 3G and genes highlighted in green in Figure  333 3F).

335
The centrosome is the major cytoplasmic MTOC within interphase cells (Caviston & 336 Holzbaur, 2006;Sanchez & Feldman, 2017;Woodruff et al., 2017). In SCC cells, like 337 most other mammalian cells, the MTOC is located near the nucleus, where it is 338 surrounded by pericentriolar proteins important for plus-ended microtubule growth 339 towards the cell periphery. The polarity of microtubules establishes polarized 340 transport, which depending upon the motor protein involved, can occur either towards 341 the minus ends at the MTOC, or along the plus ends of the microtubules towards the 342 focal adhesions.

344
Although the ISR had not been previously found to regulate the microtubule 345 cytoskeleton, we were intrigued by the possibility that the ISR was influencing 346 microtubule dynamics in response to proteotoxic stress, which in turn might affect how 347 efficiently proteins are cleared during the recovery phase. This possibility was all the 348 more compelling because protein aggregate formation and clearance is known to 349 depend on microtubule-dependent intracellular transport (Kopito, 2000). Of further 350 intrigue was the mRNA for ATF5, which like that for ATF4, displayed dramatic 351 increases in translation specifically in ISR-competent SCC cells exposed to 352 proteotoxic stress and not in the ISR-null SCC counterparts ( Figures 3H and 3I). ATF5 353 has been previously demonstrated to play a non-transcriptional role at the 354 centrosome (Madarampalli et al., 2015), strengthening the hypothesis that ISR 355 activation was driving a subset of centrosomal proteins to preserve the response to 356 proteotoxic stress. We therefore set out to evaluate how the centrosome and 357 microtubule dynamics were changing in response to stress in SCC cells with or without 358 an intact ISR. 359 360 361 The ISR protects centrosomal microtubule dynamics 362 363 To evaluate the status of the centrosome during stress, we performed 364 immunofluorescence for centrosomal markers pericentrin (PCNT) and γ-tubulin 365 (TUBG1). Fluorescence intensities of both markers were significantly increased in 366 control, but not ISR-null, cells and specifically in response to proteotoxic stress ( Figure  367 4A and Figure 4-figure supplement 1). Quantifications further revealed that the overall 368 size of the MTOC was also enlarged in response to this stress, but only when the ISR 369 was intact ( Figure 4B). This was especially intriguing because the pericentriolar region 370 that surrounds the centrioles of the microtubule organizing centers is a special site of 371 protein catabolism within the cell, concentrating both autophagosome and lysosomes 372 as well as proteasomes and proteins marked for degradation (Freed et al., 1999;Liu 373 et al., 2016;Wigley et al., 1999) . Moreover, although the highly transformed, long-374 passaged HeLa cell line lacks physiological relevance, it was notable that proteosome 375 inhibition in these cells resulted in the accumulation of pericentriolar material, including 376 pericentrin and γ-tubulin (Didier et al., 2008). When we placed these data in the context 377 of our new findings, we became all the more interested that the ISR may be at the crux 378 of reinforcing microtubule dynamics following proteotoxic stress.

380
If our premise was valid, then when cells cannot trigger an ISR in the face of stressful 381 situations, microtubule-dependent processes should be vulnerable. To evaluate 382 microtubule dynamics during the recovery phase of our cells following proteotoxic 383 stress exposure, we briefly interrupted existing microtubule assembly/disassembly 384 dynamics with nocodazole and then examined nascent microtubule nucleation 385 initiated from the centrosome ( Figures 4C and 4D).

387
In the absence of stress, the ISR-null state did not appreciably affect MTOC-initiated 388 microtubule dynamics ( Figure 4D). In striking contrast, the vulnerability of microtubule 389 dynamics in the ISR-null state became clear when we exposed our cells to the 390 proteotoxic stress and monitored the recovery process. Even after 4 hrs of bortezomib 391 recovery, microtubule growth from the MTOC was still markedly impaired in the ISR-392 null compared to control SCC cells ( Figure 4D). Since microtubule dynamics were 393 independent of the ISR in the unstressed state, these results suggested that the ISR 394 is required to reinforce the microtubule dynamics involved in cellular recovery following 395 proteotoxic stress.

397 398
The ISR is required for aggresome formation 399 400 Our evidence that protecting microtubule dynamics during proteotoxic stress is a major 401 function of the ISR was all the more compelling because of known role for microtubules 402 in assembling ubiquitinated protein aggregates into larger membraneless structures 403 called aggresomes (Johnston et al., 1998;Wigley et al., 1999). Through their ability to 404 associate with proteasomes and autophagosomes, aggresomes appear to be a key 405 intermediate in the clearance of improperly folded proteins.

407
With these insights in mind, and the general view that aggresomes function beneficially 408 in cellular recovery from proteotoxic stress, we evaluated aggresome formation in our 409 SCC cells following withdrawal of bortezomib ( Figure 5A). To this end, we used 410 immunofluorescence for p62/SQSTM1, a protein involved in shuttling ubiquitinated 411 protein aggregates to the aggresome (Christian et al., 2010). As judged by 412 convergence of intense anti-p62 immunofluorescence to a single large perinuclear 413 spot within each S51 cell, misfolded protein aggregates began to coalesce into the 414 aggresome soon after bortezomib withdrawal, peaking at approximately 8 hrs into the 415 recovery phase ( Figure 5B, top panels). By 24 hrs, the recovery phase appeared to 416 be complete, as the aggresome was no longer present (see quantifications at right). 417 These findings were in good agreement with the clearance of ubiquitinated misfolded 418 proteins that had accumulated during the proteosome block ( Figure 2E).

420
In striking contrast to S51 cells, the S51A cells lacking an intact ISR showed 421 pronounced defects in their ability to clear misfolded proteins that had accumulated 422 during bortezomib treatment. The rise in cytoplasmic p62 immunofluorescence during 423 the proteosomal block indicated that small protein aggregates had formed in the S51A 424 cells exposed to proteotoxic stress ( Figure 5B, bottom panels). However, even at 8 425 hrs after bortezomib withdrawal, only a few S51A cells displayed the bright perinuclear 426 spot reflective of the aggresome. Moreover, while S51 cells had cleared their protein 427 by 24 hrs after culturing in normal media, S51A cells still exhibited anti-p62 428 coalescence, indicative of a marked delay in clearance of unfolded proteins.

430
Further signs of cellular defects during the recovery process were evident at the 431 ultrastructural level. After 8 hrs of recovery, electron dense protein aggregates 432 pushing into the nucleus were observed in the perinuclear regions of S51 but not S51A Since these structures did not accumulate in S51 cells, it seemed likely that the root 443 of this phenotype was the inability of ISR-null cells to cope with de novo protein 444 synthesis in the face of acute proteotoxic stress. However, this did not explain why 445 clearance of cytoplasmic protein aggregates was markedly diminished in S51A cells.

446
Rather, the surprising dependence upon an intact ISR for SCC cells to form 447 aggresomes during proteotoxic stress recovery led us to posit that the ISR functions 448 critically in sustaining the necessary microtubule dynamics to enable efficient transport 449 of misfolded proteins to the MTOC, where they can form aggresomes and be targeted 450 for perinuclear proteosomal and autophagosomal clearance.

452
To further challenge the relation between microtubule dynamics and aggresome 453 formation, we used the drug, paclitaxel, a chemotherapy that stabilizes microtubules 454 while at the same time disrupting their dynamics. Strengthening the link between the 455 ISR, microtubule dynamics, and aggregate clearance, treating cells with paclitaxel 456 following the induction of protein aggregates potently blocked aggresome formation, 457 phenocopying the ISR-null cells during this process ( The integrated stress response is required for migration and focal adhesion 464 homeostasis following protein aggregate stress 465 466 During the course of our previous experiments, we made several observations 467 regarding the cellular morphology, which supported the conclusion that the ISR was 468 required to maintain proper microtubule dynamics in the face of protein aggregate 469 stress. First, we noticed that control, but not ISR-null cells, went through a dramatic 470 change in cell shape while recovering from aggregate stress. Specifically, control cells 471 "unspread", becoming more compact and tall between 4 and 8 hrs after washing out 472 bortezomib, the timepoints correlating with peak aggresome formation. ISR-null cells 473 on the other hand, did not similarly round, and instead they maintained a flattened 474 shape with some cells also forming elongated processes. Examples of these 475 differences are shown in Figure 6-figure supplement 1.

477
The long cellular extensions were reminiscent of that seen when keratinocytes 478 displayed defects in the turnover of focal adhesions, a process that we had previously 479 shown depends not only upon focal adhesion proteins (Schober et al., 2007), but also 480 on the ability of microtubules to deliver turnover cargo to the focal adhesions (Wu et 481 al., 2008). Specifically, microtubule-mediated cargo transport is essential for focal 482 adhesion disassembly and cell migration (Ezratty et al., 2005;Yue et al., 2014). While 483 the transport of protein aggregates to the centrosome involves minus end-directed 484 molecular motors, transport to the focal adhesions requires plus end-directed motors. 485 Reasoning that a defect in MTOC-mediated microtubule dynamics might affect both 486 processes, we examined the focal adhesions in control and ISR-null cells following 487 proteotoxic stress. As judged by immunofluorescence for vinculin, upon proteotoxic 488 stress the ISR promoted a remodeling in cell morphology evident by diminished focal 489 adhesion and F-actin staining. By contrast, ISR-null cells retained large focal 490 adhesions suggestive of impaired cytoskeletal dynamics. ( Figure 6A). This ISR-491 mediated difference was specific to stress and was not seen when control and ISR-492 null cells were cultured in normal media under unstressed conditions. Compellingly, 493 further analysis of our ribosome profiling dataset revealed a significant enrichment of 494 focal adhesion genes within mRNAs translationally downregulated upon ISR induction 495 ( Figure 3-figure supplement 1D), suggesting that ISR-mediated translational control 496 stands at the crux of these dramatic morphological changes by selectively 497 upregulating microtubule dynamics and, at the same time, downregulating focal 498 adhesion components. 499 500 To directly evaluate focal adhesion dynamics in our cells we generated a zyxin-iRFP 501 construct that would allow for live imaging of cells. Indeed, ISR-null SCC cells 502 recovering from proteotoxic stress possessed large, stable focal adhesions that did 503 not disassemble, rendering the cell immobile over the time frame whereas control SCC 504 cells disassembled their large focal adhesions and displayed considerable dynamics 505 ( Figure 6B and Supplemental Video 1). To confirm these aberrant dynamics, we 506 performed a scratch wound assay to induce cell migration following bortezomib 507 treatment. In agreement with our previous data showing large and stable focal 508 adhesion, ISR-null cells showed markedly impaired mobility relative to control cells 509 and failed to migrate into the wound site ( Figure 6C and Supplemental Video 2-3). 510 Taken together, these results provided further evidence that in the absence of an ISR, 511 SCC cells exposed to proteotoxic stress are slow to ignite the microtubule dynamics 512 necessary for recovery. 513 514 ATF5 acts downstream of the ISR to promote aggresome formation and the 515 recovery from proteotoxic stress 516 517 Our collective data strongly implicated the ISR in regulating microtubule dynamics with 518 the purpose of promoting aggresome assembly and the efficient clearance of protein 519 aggregates that accumulate during proteotoxic stress. Additionally, our analysis of 520 ISR-dependent translational changes, revealed a selective upregulation of 521 centrosomal proteins upon proteotoxic stress ( Figure 3G). In particular, we were 522 intrigued by ATF5, which has been described as an integral component of the MTOC 523 (Madarampalli et al., 2015). In agreement with these previous findings, confocal 524 microscopy revealed that upon recovery from proteotoxic stress, ATF5 localized at the 525 centrosome in ISR-competent SCC cells ( Figure 7A). Hence, we reasoned that 526 translational upregulation of ATF5 might be needed to regulate microtubule dynamics 527 in the face of proteotoxic stress. To test this hypothesis, we engineered ISR-null cells 528 to overexpress doxycycline-inducible, GFP-tagged ATF5 ( Figures 7B and 7C).

530
To understand whether ATF5 is needed to clear protein aggregates, we first examined 531 the clearance of ubiquitinated proteins during the recovery phase following a pulse of 532 bortezomib. Remarkably, expression of ATF5 significantly improved the ability of ISR-533 null cells to clear ubiquitinated proteins, as assessed by immunoblot of ubiquitinated 534 proteins in the insoluble fraction of cells after 24 hrs recovery from bortezomib ( Figure  535 7D and 7E). Strikingly, this corresponded with a significant increase in aggresome 536 formation, indicating that ISR-mediated ATF5 up-regulation is needed to promote the 537 retrograde transport of misfolded proteins and the subsequent accumulation of protein 538 aggregates at the MTOC ( Figure 7F). This effect was mediated via protection of 539 microtubule dynamics as evidenced by increased pericentrin mean fluorescence 540 intensity at the centrosome in ISR-null cells expressing ATF5 after recovery from 541 bortezomib treatment ( Figure 7G). 542 543 Finally, we asked whether ISR-mediated ATF5 induction promotes cell survival upon 544 proteotoxic stress of SCC cells in vivo. To this end, we turned to our grafting model 545 where ISR-null cells displayed increased sensitivity to proteasome inhibition. 546 Consistent with our hypothesis, ATF5 expression rescued the selective sensitivity to 547 bortezomib shown by ISR-null cells. This occurred in the absence of changes in 548 viability in tumors at steady state ( Figure 7H). Altogether, these data implicate ATF5 549 as a critical ISR target, which orchestrates microtubule dynamics to facilitate the 550 clearance of protein aggregates and promote cell recovery upon proteotoxic stress. 551 552 553 Discussion 554 555 Although a handful of reports have linked the actin cytoskeleton to eIF2a 556 dephophorylation (Chambers et al., 2015;Chen et al., 2015), cytoskeletal regulation 557 has not been viewed as a primary function of the ISR. In the current study, we 558 unearthed a novel and essential role for the ISR in regulating the cytoskeleton, 559 specifically through preserving microtubule dynamics in stressful situations. We 560 showed that by doing so, proteotoxically stressed cells can accomplish the necessary 561 intracellular trafficking to efficiently clear misfolded proteins and prevent them from 562 accumulating and overtaxing the ER. 563 564 Our findings led us to the remarkable conclusion that the ISR not only is required to 565 mount a stress response, but also to maintain cell function during recovery from stress. 566 Although the existence of negative feedback loops that downregulate the ISR upon 567 termination of stress has long been appreciated (Novoa et al., 2001), an active role for 568 the ISR in cellular recovery following a stressful experience has hitherto gone largely 569 unrecognized.

571
By temporally profiling the translational differences that arise when proteotoxically 572 stressed SCC cells are unable to mount an ISR, we gained insights into the 573 mechanisms underlying the ISR's importance. Specifically, we learned that when ISR-574 competent cells are exposed to proteotoxic stress, they redirect their translational 575 machinery to a cohort of mRNAs encoding centrosomal proteins. We show that these 576 proteins function in bolstering the MTOC and reinforcing its microtubule dynamics. 577 This then facilitates efficient microtubule-mediated transport of misfolded proteins to 578 the perinuclear space, where they can be assembled into aggresomes and targeted 579 for ubiquitin-mediated degradation during the stress recovery phase. Indeed, as we 580 showed, when eIF2α cannot be phosphorylated and the ISR core is thereby crippled 581 in proteotoxically stressed cells, the centrosomal proteins are not translated, and 582 microtubule dynamics emanating from the MTOC are disrupted. As judged by 583 deficiencies in the aggresome assembly and in focal adhesion turnover, both 584 retrograde and anterior grade transport of microtubule cargo are slow to recover 585 following stress, thereby impairing cell fitness. Compellingly, the ISR not only 586 upregulates a subset of centrosomal proteins, but also downregulates focal adhesion 587 components, surfacing a network of ISR-regulated cytoskeletal dynamics that directs 588 morphological changes critical for cell recovery.

590
Microtubule trafficking is important to bring misfolded proteins to the perinuclear 591 space, and our results provided compelling evidence that the ISR functions in this 592 process. However, our findings also pointed to a hitherto unappreciated role of the ISR 593 in aggresome assembly specifically. Our studies revealed that this function is in part 594 mediated by ATF5, a translational target of the ISR in proteotoxically stressed SCC 595 cells and a previously documented structural component of the centrosomal MTOC 596 (Madarampalli et al., 2015). We showed that during recovery from protein aggregate 597 stress, the MTOC size increases concomitantly with aggresome formation, and both 598 of these events depend upon an intact ISR. Our data support a model in which ISR-599 mediated translational upregulation of centrosomal proteins, including ATF5, is 600 required to remodel the MTOC and concentrate protein aggregates in the perinuclear 601 space so that they can be degraded and cleared by proteosomes and 602 autophagosomes.

604
Several studies have provided tantalizing evidence that some cells may be able to 605 asymmetrically partition their centrosomal aggregates when division resumes 606 following proteotoxic stress, such that one daughter (e.g. a stem cell) remains healthy, 607 while the other daughter inherits the aggregates and becomes slated to differentiate 608 (Morrow et al., 2020;Rujano et al., 2006;Singhvi & Garriga, 2009). We did not see 609 evidence for this in our SCC cells. Rather, there was a lag in cell cycle reentry such 610 that it coincided with the timing at which accumulated ubiquitinated misfolded proteins 611 were cleared (Figures 2D and 5B). This coupling was especially striking in comparing 612 the behaviors of ISR-null versus ISR-competent cells. Thus, during the recovery 613 phase, ISR-null cells were delayed by > 24 hrs in both aggregate clearance and cell 614 cycle re-entry. That said, when the ISR was crippled either in vivo or in vitro, SCC cells 615 were jeopardized in their overall ability to survive proteotoxic stress. In this scenario, 616 as a major regulator of cell survival in response to proteotoxic stress, our findings 617 suggest that by coupling pharmacologic inhibition of the ISR (Sidrauski et al., 2015) 618 with proteosomal inhibitors such as bortezomib, such a regimen may find an Achilles 619 heel for this family of difficult to treat cancers. 620 621 Ideas and speculation 622 Our findings are also likely to have relevance for other diseases, including 623 neurodegenerative diseases, which are driven by or associated with protein 624 aggregates (Soto & Pritzkow, 2018). To date, the function of the ISR in 625 neurodegenerative diseases has been attributed largely to the emergence of 626 alternative translation pathways and to ATF4 transcriptional activity that drives 627 expression of cytoplasmic chaperones and balances cell survival and cell death (Bond 628 et al., 2020;Costa-Mattioli & Walter, 2020). In SCC cells, although it is possible that 629 chaperones may be upregulated indirectly through ISR-driven transcription factors, 630 these proteins did not emerge as major translational targets of the ISR in our ribosome 631 profiling analysis. Moreover, since cells have parallel pathways, most notably the heat 632 shock response, that can upregulate chaperones when needed (San Gil et al., 2017), 633 it seems unlikely that this would be ISR's sole function in maintaining proteostasis.

635
The necessity of an ISR-driven pathway to preserve centrosome dynamics during 636 proteotoxic stress raises an important question regarding centrosome function in 637 neurodegenerative disorders. If ISR-driven translation of centrosomal proteins is 638 required to protect microtubule dynamics during stress, it follows that the accumulation 639 of unfolded proteins should negatively impact the MTOC and/or its microtubule-640 associated dynamics. In Alzheimer's disease and other tauopathies, the microtubule 641 associated protein Tau is the main driver of the aggregates of neurofibrillary tangles 642 that ensue, and this alone is likely to negatively impact microtubule-mediated 643 trafficking in neurons (Ballatore et al., 2007). In addition, misfolded protein aggregates 644 often inadvertently sequester properly folded cellular proteins, such as p62 and the 645 disaggregase p97/VCP, which can exacerbate their toxic effects on cells (Donaldson 646 et al., 2003;Olzscha et al., 2011;Yang & Hu, 2016). It seems plausible that misfolded 647 protein aggregates might either sequester low-complexity centrosomal or 648 pericentrosomal proteins (Woodruff et al., 2017), or structurally interfere with MTOC 649 function. In fact, one study found that proteotoxic stress does indeed inhibit 650 centrosome function in neurons (Didier et al., 2008), raising the question as to whether 651 the ISR might also be involved in maintaining the MTOC of long-lived, non-dividing 652 neurons, which face challenging microtubule-mediated cellular trafficking dynamics of 653 the likes that few if any other cell type of the body does.

655
In summary, our work supports a model in which most if not all cells, but likely cancer 656 cells and neurons in particular, rely upon the ISR to restore proteostasis following 657 protein aggregate stress. By redirecting the translational machinery towards 658 synthesizing proteins involved in enlarging the MTOC and bolstering microtubule 659 dynamics, the ISR aids in the intracellular microtubule-mediated trafficking necessary 660 to assemble aggresomes at the MTOC, where they can be efficiently targeted to the 661 perinuclear protein degradation machinery. 662 663 664

CRISPR cloning 666
Lentiviral particles containing the gene replacement construct, pLKO-PGK-eIF2α-667 mycTag-P2A-NeoR were prepared by transfecting the lentiviral plasmid along with 668 packaging plasmids into HEK293T cells, and viral supernatant was collected 48 hours 669 post transfection. This construct was integrated into the genome of a clonal, parental 670 primary SCC mouse line by incubating 100 μL with 0.1 mg/mL polybrene for 8 hours. 671 48 hours later cells with the integrated construct were selected with 0.5 mg/mL 672 Neomycin selection for 2 days. Following selection the endogenous allele was 673 targeted for deletion using CRISPR-Cas9 RNP particles. The replacement allele had 674 a synonymous mutation in the PAM site rendering it resistant to this CRIPSR 675 construct. CRISPR-Cas9 RNP particles targeting the endogenous allele were 676 prepared as follows: eIF2α gRNA (target sequence: ATATTCCAACAAGCTGACAT) 677 was designed using Guidescan software (Perez et al., 2017) and complexed with 678 ATTO550-tracrRNA and Cas9. All reagents were acquired from IDTdna's "AltR 679 system". Duplexed gRNA:tracrRNA was prepared by mixing 1 μM of each component 680 in IDTdna duplex buffer, heated to 95° C in a thermocycler and annealed by gradually 681 lowering the temperature to 25° C at a rate of 0.1° C/second. Duplexed 682 gRNA:tracrRNA was complexed with Cas9 by mixing 1 μM of the duplex with 1 μM 683 Cas9 in OptiMEM (ThermoFisher) and incubating at room temperature for 5 minutes.

684
RNP complexes then were transfected into 60% confluent 12 well plates using 685 RNAiMAX as follows: 30 μL of 1 μM RNP complexes were mixed with 4.8 μL RNAiMAX 686 in 335 μL of optiMEM and RNP-lipid complexes were allowed to form for 15 minutes 687 at room temperature. At the end of the incubation 400 μL of complexes were added 688 dropwise to cells, and media was changed 16 hours later. 48 hours after transfection 689 single cells were isolated using fluorescence activated cell sorting (FACS) as follows: 690 Cells were dissociated with trypsin (Gibco) and resuspended in 500 μL of FACS buffer 691 (PBS supplemented with 5% FBS and 5 uM EDTA), and single, ATTO-550 positive 692 cells were sorted using BD FACSAria cell sorter into wells of 96-well plates containing 693 100 μL of 50-50 mixture of fresh media and conditioned media. Clones that grew to 694 confluency were transferred to 12 well plates, and following growth to confluency in 12 695 well plates, cells were dissociated with trypsin and frozen in freezing media 696 supplemented with 10% FBS and 10% DMSO. At this stage a small aliquot of cells 697 (75% of the plate) was lysed in 200 μL of QuickExtract DNA Extraction solution 698 (Lucigen) and gDNA was prepared by heating to 65° C for 10 minutes followed by heat 699 inactivation at 95° C for 2 minutes. These gDNA samples were used for further 700 analysis. For HRI KO cells the protocol was exactly the same except that gRNA 701 targeting HRI (target seq: ATTTAAACACCTGTTTGGAG) was used. 702 703 NGS analysis of CRISPR outcomes 704 Knockout of the endogenous allele was evaluated using primers targeting a 300 705 basepair region of genomic DNA with the targeted locus in the middle of the amplicon. 706 Primers had 5' overhangs with sequences compatible with the Illumina Nextera XT 707 index primers (R: overhang: GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAG, L 708 overhang: TCGTCGGCAGCGTCAGATGTGTATAAGAGACAG). Amplicons were 709 generated with 1 μL of input gDNA and the NEB phusion kit according to 710 manufacturer's instructions, and amplicons were isolated using Agencout Ampure XP 711 beads (Beckman Coulter). A second barcoding PCR was performed using Nextera XT 712 index primers as follows: 2 μL of cleaned amplicons were used as input, primers were 713 added so that each isolated clone had a unique combination of left and right barcodes, 714 and barcodes were added using a 8 cycle PCR reaction with 55° C annealing temp, 715 again with the NEB Phusion kit, and the barcoded amplicons were cleaned and primer 716 dimers were removed using Ampure XP beads. Amplicons were normalized to the 717 same concentration, pooled, and sequenced using a single Illumina MiSeq Nano lane 718 using the 250 basepair, paired end kit. Demultiplexed reads were analyzed and 719 screened for indels using the RGEN Cas Analyzer (http://www.rgenome.net). A KO 720 clone was confirmed if the only reads detected in that sample were indels that would 721 create a frameshift.

723
Cell culture 724 Primary murine SCC cells were generated and cultured in E medium supplemented 725 with 15% FBS and 50 mM CaCl2 as previously described . Cells 726 were passaged 3 times per week and passage numbers were maintained counting 727 from the point of cell line generation. Frozen cell stocks were generated by freezing 728 cells in complete media supplemented with 10% additional FBS and 10% DMSO.

729
ATF5-GFP overexpressing cells were generated by infecting ISR-null cells with ATF5-730 GFP lentivirus in the presence of 5 µg/mL polybrene. Infected cells were selected 731 using 2 µg/mL puromycin for 5 days. Induction of ATF5 was induced using 1 µg/mL 732 doxycycline at the time of plating for eachexperiment.

734
Proliferation and cell viability assays 735 For proliferation assays 2500-5000 cells were plated per well in clear-bottom, black 736 optical 96 well plates (Nunc) and allowed to attach overnight. A baseline plate was 737 collected the next day as a zero hour sample, and then plates were collected at 24, 738 48, and 72 hours after this timepoint. At the time of collection media was washed and 739 cells were fixed in 4% PFA in PBS for 10 minutes at room temperature. After fixation, 740 PFA was washed with PBS and cells were stored in PBS at 4° C until the end of the 741 experiment. Following collection of the final plate, nuclei were stained in all samples 742 using 1 ug/mL DAPI in PBS for 5 minutes at room temperature. DAPI was washed and 743 replaced with PBS, and nuclei were imaged on a Biotek Cytation 5 high content 744 imager, and cells were counted using Gen5 software. To measure cell death, single-745 cell-suspensions were washed once with PBS, stained with 3 μl of AnnexinV-PE 746 conjugates (Thermo Fisher) and 0.1 μg/ml DAPI (Thermo Fisher) in 100 μl of 1X 747 Annexin Binding Buffer (Thermo Fisher) for 15 min at room temperature. Data were 748 collected using a BD LSR Fortessa X20 flow cytometer and analyzed using FlowJo 749 software.

751
Measurement of translation rates 752 Cells were plated to 75% confluency in 6 well plates. The next day cells were treated 753 with 50 μM sodium arsenite, 100 ng/mL tunicamycin or 100 nM borteozomib, or a 754 vehicle control and incubated at 37° C for 6 hrs. At this time 20 μM puromycin was 755 supplemented to the media, and cells were incubated for 30 min. Cells were then lysed 756 in RIPA buffer and translation rates were evaluated by immunoblotting for 757 puromycilated peptides using an anti-puromycin antibody.

759
Microtubule nucleation assay 760 Cells were plated to 50% confluency in glass slides (Millicell EZ, Millipore Sigma) and 761 treated with 100 nM bortezomib or a vehicle control for 6 hrs followed by a PBS wash 762 and replacement with fresh media. 4 hrs later the media was supplemented with 13 763 μM nocodazole, and cells were incubated at 37° C for 20 mins. At the end of the 764 incubation slides were washed and media was replaced with fresh, nocodazole-free 765 media, and microtubules were allowed to recover during a 2 min incubation at room 766 temperature. At this time cells were fixed in 4% PFA for 10 min at room temperature. 767 Additional control slides without nocodazole or with nocodazole and no recovery were 768 fixed as controls. Slides were processed for immunofluorescence targeting α-tubulin, 769 pericentrin, and λ-tubulin, immunofluorescence signal was imaged by confocal 770 microscopy, and microtubule nucleation rates were evaluated by quantifying the 771 relative α-tubulin signal intensity within the centrosome region, which was defined 772 using 3D volumetric assessment (Imaris) of pericentrin-positive volumes.

774
Scratch assay 775 Cells were plated in plastic bottom optical slides (Ibidi 80826) and allowed to reach 776 confluency in in 48 hours. At this time cells were treated with 100 nM bortezomib or a 777 vehicle control for 6 hours followed by PBS wash and replacement with fresh, drug-778 free media. At the time of wash scratch wounds were manually created by gently 779 scaping cells using a rubber cell scraper. Movement of wound-edge SCC cells was 780 evaluated using time-lapse confocal microscopy of GFP signal (marking all cells) with 781 images acquired every 2 minutes for 8 hours. Scratch closure was evaluated using 782 Imaris software to measure the percentage of scratch closed by leading edge cells 783 over the course of the experiment. 784 785

FACS isolation of ex vivo tumor cells 786
To sort SCC cells out of tumors formed from ISR-WT and ISR-null cells lines, Day 35 787 tumors were dissected from the skin and finely minced in 0.25% of collagenase 788 (Sigma) in HBSS (Gibco) solution. The tissue pieces were incubated at 37º C for 20 789 minutes with gently shaking. After a single wash with ice-cold PBS and samples were 790 further digested into single cell suspension in 0.25% Trypsin/EDTA (Gibco) for 10 min 791 at 37º C. After neutralization with the FACS buffer (PBS supplemented with 4% FBS, 792 5 mM EDTA, and 1 mM HEPES), single-cell suspensions was then centrifuged, 793 resuspended, and strained before preparing for staining. A cocktail of Abs for surface 794 markers at the predetermined concentrations (CD31-APC 1:100, Biolegend; CD45-795 APC 1:200, Biolegend, CD117-APC 1:100, Biolegend; CD140a-APC 1:100, Thermo 796 Fisher; CD29-APCe780, 1:250, Thermo Fisher, Biolegend) was prepared in the FACS 797 buffer with 100 ng/ml DAPI. Sorting was performed using a BD FACSAria equipped 798 with FACSDiva software to isolate a population of cells that was GFP-positive (pan-799 SCC), DAPI-negative (live), and APC-negative (dump gate to exclude immune cells, 800 endothelial cells, and fibroblasts), and new cell lines were established from bulk 801 populations of sorted cells.

803
Tumor allografting 804 Squamous cell carcinoma allografts were generated by intradermally injecting 1x10 5 805 SCC cells suspended in a 50:50 mix of PBS and growth-factor reduced Matrigel 806 (Corning, 356231) in an injection volume of 50 μL. Grafts were generated in the flanks 807 of 6-8 week old female (Nude) mice. Tumor dimensions were measured every 5 days 808 using electronic calipers and tumor volume was calculated using the formula V = 0.5 809 × length × width2. For tumor growth experiments sample sizes of N=8 were calculated 810 to yield an 80% power to detect a significant (p<0.05) effect size of 50% assuming 811 standard deviation of 25%, conservate estimates based on past lab experience 812 suggesting targeting ISR-related genes in SCC which yielded larger effect sizes 813 (Sendoel et al., 2017). Experiment was performed twice ISR-WT and ISR-null using 814 two different clones for each, with each experiment powered independently to 80%, 815 yielding total sample size of N=16 when pooling data. Experiments evaluating 816 apoptosis in vivo were set up for 80% power to detect a significant (p<0.05) effect size 817 of 66% with 25% SD, which yielded minimum sample size of N=4.

819
Immunofluorescence/histology 820 For immunofluorescence of tumors, samples were fixed in 4% PFA for 1 hr at room 821 temperature , dehydrated in 30% sucrose overnight at 4°C, and mounted into OCT 822 blocks and frozen. 14 μm thick sections were cut using a Leica cryostat deposited onto 823 SuperFrost Plus slides (VWR). For immunofluorescence of cells in culture, cells were 824 plated on glass slides (Millicell EZ, Millipore Sigma) coated with human plasma 825 fibronectin (Millipore Sigma) diluted to 100 μg/mL in PBS. At the conclusion of 826 experiments samples were fixed with 4% PFA for 10 minutes at room temperature. 827 Samples were permeabilized with 0.3% Triton-X100 in PBS and blocked using 2.5% 828 normal donkey serum, 2.5% normal goat serum, 1% BSA, 2% fish gelatin, and 0.3% 829 Triton X-100 in PBS. Primary antibodies were applied in blocking buffer overnight at 830 4° C. Samples were washed with 0.1% Triton X-100 and secondary antibodies with 831 Alexa 488, Alexa 594, and Alexa 647 were applied for 1 hour at room temperature in 832 blocking buffer containing 1 μg/mL DAPI. Slides were washed with 0.1% triton and 833 mounted using Prolong Diamond Antifade Mountant with DAPI (ThermoFisher).

835
Microscopy and image analysis 836 Microscopy of tumors and 40X images of aggresomes and spreading cells were 837 performed using an Axio Observer Z1 epifluorescence microscope equipped with a 838 Hamamatsu ORCA-ER camera (Hamamatsu Photonics), and with an ApoTome.2 839 (Carl Zeiss) slider using a 20X air, 40X oil, or 63X oil objective. 63X confocal 840 microscopy images were collected on an Andor Dragonfly spinning disk imaging 841 system with a Leica DMi8 Stand and cMOS Zyla camera. Images were analyzed in 842 FIJI or Imaris. For fluorescence intensity measurements of tumors GFP masks were 843 generated and signal was measured within the mask. Aggresomes were manually 844 counted as discrete p62-positive juxtanuclear puncta on maximum intensity Z-845 projections. Cell dimensions were calculated using the length measuring tool, and cell 846 spreading was evaluated manually by observing for spread morphology. RGB images 847 were generated with FIJI and saved as TIFF files. For 3-dimensional reconstructions 848 and volumetric analyses of microtubule organizing centers, Imaris was used to 849 generate 3D images, and volumes were generated to create 3D volumes 850 encompassing the discrete puncta of pericentrin staining. Volume as well as summed 851 pericentrin and γ-tubulin fluorescence intensity were measured within these volumes 852 853 Electron microscopy 854 Cells were fixed in a solution containing 4% PFA, 2% glutaraldehyde, and 2 mM CaCl2 855 in 0.1 M sodium cacadylate buffer (pH 7.2) for 1 hr at room temperature, and then 856 placed at 4° C. Cells were next postfixed in 1% osium tetroxide and processed for 857 Epon embedding; ultrathin sections (60-65 nm) were then counterstained with uranyl 858 acetate and lead citrate, and images were acquired using a Tacnai G2-12 transmission 859 electron microscope equipped with an AMT BioSprint29 digital camera. 860 861 Cell fractionation 862 Cells were lysed in RIPA buffer (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1 mM EDTA, 863 1 mM EGTA, 1% Triton X-100, 0.5% deoxycorate, 0.1% SDS) containing protease 864 inhibitors (Complete mini, Roche) and phosphatase inhibitors (PhosStop), and lysed 865 for 10 minutes at room temperature. Membrane fraction was pelleted by centrifuging 866 at 1000 g for 10 minutes at 4° C. The supernatant (cytosolic fraction) was transferred 867 to new tubes which were centrifuged at 20,000 g for 30 min at 4° C. The supernatant 868 (RIPA-soluble fraction was transferred to a new tube, and the pellets (RIPA-insoluble 869 fractions) were washed with 300 μL of RIPA buffer, centrifuged 20,000 g for 10 min at 870 4° C, and then resuspended in 30 μL of 1X LDS-βME by vortexing vigorously and 871 boiling at 98° C for 10 minutes. The protein concentration of the RIPA-soluble fraction 872 was measured with a BCA assay (Pierce). RIPA-soluble fractions were mixed into 1X 873 LDS-βME and protein concentration was normalized. The insoluble fractions were 874 normalized by adding 1X LDS-bMe so that the same volume corresponds to the same 875 volume of insoluble-fraction lysate (eg. Insoluble fraction from 100 µg of cell lysate).

876
Samples were run on immunoblots as previously described and probed for ubiquitin 877 signal.

896
Antibodies and counterstains 897 The following antibodies and dilutions were used for immunoblotting. Ribosome profiling and total RNA libraries were pooled and sequenced on a 940 Novaseq using the S1, 1x100 bp kit. Reads were demultiplexed, trimmed using FastX 941 trimmer, and aligned to the mm10 reference genome using bowtie2. Sequences were 942 counted in bins of 5' UTR, CDS, and 3' UTR as defined using plastid 943 (https://plastid.readthedocs.io/en/latest/). Count data was analyzed using DESeq2 944 (Love et al., 2014), an R package designed for statistical analysis of gene counts 945 generated from Illumina-based sequencing. For expression analysis of ribosome 946 profiling data only reads in the CDS were included and reads coming from the first 15 947 codons (45bp) and last 5 codons (15bp) were excluded. Additionally, only reads of 948 size between 20-23bp or 26-32bp were counted. Gene lists were generated as 949 described in the main text, first by filtering genes with significant differences in RPF-950 read counts, and then genes that changed specifically at the translation level were 951 identified as the subset of filtered genes with translational efficiency (TE=normalized 952 RPF/normalized total RNA) fold changes greater than 1.5.

954
Data availability 955 The total mRNA and ribosome protected fragments datasets generated during the 956 ribosome profiling experimetns performed in this study have been deposited to the 957 Gene interval, in N=2 independent experiments with n=3-5 technical replicates per condition.

1159
Bar graph shows percentage of scratch closure ± SD. *p<0.05 ns, no statistical 1160 significance (one-way ANOVA with multiple comparisons). 1161 1162 A. Bar graph shows tumor volume quantification ± SD at day 40 in mice transplanted 1222 with either control or ISR-null SCCs. A minimum of 7 tumors per condition were 1223 quantified. ns, no statistical difference (t test).

1224
B. Secondary ex vivo control and ISR-null SCC cell lines were generated by FACS 1225 isolation of day 35 tumors initiated and grown from transplantation of our primary 1226 control and ISR-null SCC lines. p-eIF2α and ATF4 immunoblot following 4 hrs 1227 treatment with 50 μM sodium arsenite shows that neither ISR-null clone (N=2) has 1228 escaped ISR-ablation. This experiment rules out the trivial explanation that ISR-null 1229 SCC growth is due to 'escaper' cells that have reverted to a wild-type ISR state. 1230 1231 A. Schematic illustrates experimental design for our bortezomib pulse and recovery 1235 system. Cells are treated for 6 hrs with 100 nM bortezomib and let recover for 24 hrs. 1236 Viability is assessed at two time points (0 hrs recovery and 24 hrs recovery) 1237 B. ISR-null cells show no difference in viability after a short (6 hrs) pulse of 100 nM 1238 bortezomib. Bar graph shows percentage of live cells ± SD quantified as DAPI 1239 negative cells by FACS in three independent experiments. ns, no statistical 1240 significance (t test).

1241
C. ISR-null cells show no difference in viability after a 24 hrs recovery from a 6 hrs 1242 pulse of 100 nM bortezomib. Bar graph shows percentage of live cells ± SD quantified 1243 as DAPI negative cells by FACS in three independent experiments. ns, no statistical 1244 significance (t test). 1245 1246 A. PCA plot shows variance of 4 independent replicates. Two major sources of 1261 variance are identified in our dataset: S51 vs S51A (x-axis) and vehicle vs bortezomib 1262 (y-axis).

1265
C. Histograms show fraction of RPF reads corresponding to their length in nucleotides.

1266
As expected the majority of reads are found between 29-30 nt. High magnification images show centrosomal MTOC by pericentrin and g-tubulin 1277 staining in S51 and S51A cells vehicle-treated or treated with a bortezomib pulse (100 1278 nM, 6 hrs) and let to recover for 4 hrs. White arrows indicate MTOCs. Scale bar 10 1279 µm. 1280 1281 A. Formation of aggresomes is dependent on microtubule-mediated transport. 1300 Blocking microtubule dynamics by using 200 nM of paclitaxel inhibits aggresome 1301 formation. Graph shows percentage of cells with aggresomes as quantified by 1302 immunofluorescence in two independent biological experiments.

1303
B. Aggresomes formation is dependent on microtubule-mediated transport. 1304 Microtubule polymerization was blocked adding 13 µM nocodazole at the time of 1305 bortezomib wash-out. Percentage of cells with aggresomes ± SD is quantified in two 1306 independent biological replicates. 1307 1308         Source Data 11 Numerical data for each experiment included in the manuscript.

Source Data 12
Read counts for ribosome profiling and total RNA sequencing.