Oxidative Stress Increases the Number of Stress Granules in Senescent Cells and Triggers a Rapid Decrease in p21waf1/cip1 Translation*

Very little is known as to how the accumulation of senescent cells during aging may affect our ability to cope with various stresses. Here we show that the assembly of stress granules (SGs) is part of the early events used by senescent cells to respond to certain stresses. Although SGs can form in response to stress during senescence activation, their number significantly increases once the cells are fully senescent. This increase correlates with a rapid decrease in the expression levels of the cyclin kinase inhibitor p21, an important activator of senescence. Throughout stress, p21 mRNA is stabilized and localizes to SGs, but only during late senescence does this localization interferes with its translation. Additionally, we observed that when the stress is relieved, senescent cells produce lower levels of p21 protein, which correlates with a small delay in SG disassembly. Therefore, our data suggest that SG formation and the reduction in p21 protein levels represent two main events by which senescent cells respond to stress.

Senescence exists as a stress response whereby cells experience irreversible cell cycle arrest. Because of this antiproliferative effect, senescence is considered a possible tumor suppressor mechanism (1,2). These observations have led to the design of many chemotherapeutics that act by inflicting severe DNA damage to tumor cells, triggering cellular senescence (3). Recent studies have indicated that although aging is characterized by a significant increase in senescent cells, old organisms are more cancer-prone than their younger counterparts (2,4). In fact, even though senescent cells reach a state where they appear incapable of further division, they remain metabolically active (5,6), suggesting that they can still react and adapt to extracellular assaults which, with time, may affect their general behavior.
In young organisms senescent cells (at least, those induced by acute p53 activation) are efficiently eliminated by the immune system (7,8). However, with aging, some of these cells acquire the ability to escape elimination and continue to survive for years undetected (1,2). Many studies have indicated that the age-dependent accumulation of senescent cells leads not only to a decrease in tissue regeneration and repair (9) but also to cancer development (5,6). This results mainly from the fact that senescent cells up-regulate the expression and the secretion of toxic factors (6), such as matrix metalloproteinases (e.g. MMP-3), epithelial growth factors (e.g. vascular endothelial growth factor), and inflammatory cytokines (e.g. interleukin-6) (6, 9 -11). These substances disrupt tissue integrity and function, providing a better milieu and stromal support for cancer cells. Surprisingly, very little is known about the affect stress can have on the expression of many genes in senescent cells which might explain how and why the senescence phenotype switches from being a means to stop cancer to becoming an activator of malignancy.
The induction of the senescence phenotype requires specific conditions that lead to the up-regulation and activation of many pro-senescence factors. It is well established that the cyclin-dependent kinase p21 waf1/cip1 (referred to hereafter as p21) is required for the initial steps leading to the senescence phenotype (2,12). Mounting evidence indicates that p21 expression is regulated at the transcriptional and posttranscriptional levels in response to different stimuli including those known to trigger senescence (13)(14)(15)(16). Indeed, during the transition from the proliferative to the senescence state, the translation of p21 mRNA is regulated by the RNA-binding proteins calreticulin (proliferative) and CUGBP1 (senescence). At the proliferative state, calreticulin binds the 5Ј-untranslated region of p21 mRNA, blocking its translation; however, in senescent cells CUGBP1 (an activator of p21 mRNA translation) relieves this inhibition by displacing calreticulin and binding to the same element in the 5Ј-untranslated region (17). Likewise, it has been shown that under some stresses the expression of p21 mRNA may be regulated at the level of stability. For example, in response to ultraviolet radiation (UV), the half-life of p21 mRNA increases by a mechanism involving the RNA-binding protein HuR (13). However, it is not known whether stress could affect the expression levels as well as the function of p21 protein in senescent cells.
One of the main mechanisms by which the cell responds to a variety of extracellular assaults is the formation of cytoplasmic foci known as stress granules (SGs) 2 (18 -20). These entities are considered a part of the survival pathways that are activated with the onset of many stresses to protect key cellular elements that allow a quick recovery if the stress is rapidly removed (20 -22). SGs are also known to interfere with general translation (18 -20) and to recruit and prevent the decay of many mRNAs (22,23). In response to several stresses, some key posttranscriptional regulators such as the RNA-binding proteins HuR (24), tristetraprolin (25), and Ras-GAP SH3-binding protein (G3BP) (26) localize to SGs. Recently it has been demonstrated that although some stresses require the phosphorylation of the initiation translation factor eIF2␣ to assemble SGs (22,23,27), others do not (28,29). The importance of SG-mediated modulation of gene expression during stress is underscored by the effect that the assembly and disassembly of SGs have on the translation of p21 mRNA in cells treated with the proteasome inhibitor MG132. Indeed, during the first 6 h of MG132 treatment, SGs assemble and recruit p21 mRNA, leading to its stabilization as well as to the inhibition of its translation. These SGs subsequently disassemble in an Hsp70-dependent mechanism, allowing the recovery of p21 translation that correlates with the later activation of the apoptotic pathway and cell death (22). The role of Hsp70 protein in cell recovery from stress has been previously demonstrated (30,31). Hence, the fact that in senescent cells exposed to heat shock the expression of Hsp70 protein is reduced (32,33) could be an indication of a decrease in their ability to recover from stress.
The features described above argue that under stress conditions SGs are one of the key posttranscriptional regulators of gene expression. At this time, however, we do not know whether SG formation is part of the mechanisms by which senescent cells modulate the expression of many genes in response to a variety of stresses. In this study we addressed these issues and found that the number of arsenite-induced SGs significantly increases in senescent cells when compared with their proliferative counterparts. We showed that this increase coincides with the reduction of p21 protein expression in senescent cells exposed to oxidative stress. We discuss the molecular mechanism leading to this reduction and its functional consequences on the ability of senescent cells to recover from stress.

EXPERIMENTAL PROCEDURES
Cell Lines and Cultures-IDH4 cells were generously provided by Dr. Myriam Gorospe, NIH, Aging Institute, Baltimore, MD. WI38 cells were generously provided by Dr. Chantal Autexier, Lady Davis Institute, McGill University, QC, Canada. All cell lines were maintained in Dulbecco's modified Eagle's medium (Sigma) supplemented with 10% fetal bovine serum (Sigma) and penicillin/streptomycin (Sigma). Media supplemented with 1 M dexamethasone (dex) (34) (needed for the induction of the SV40 large T antigen) was used to ensure proliferation of IDH4 cells. To induce the senescence of IDH4 cells, dex was removed from the media, and fetal bovine serum (FBS) was replaced by charcoal-stripped FBS (Sigma). Serial passage of cells was performed to induce the senescence of WI38 cells. Cells were treated with 0.5 mM of sodium arsenite (AS) for 30 min to induce oxidative stress and SG formation.
␤-Galactosidase Staining-The Senescent Cell Staining kit (obtained from Sigma) was used to detect senescent cells as previously described (35) following the manufacturer's guidelines.
Immunofluorescence-Cells were grown to confluency on coverslips. The cells were exposed to stress as described under "Results" then washed with PBS and fixed with 3% paraformaldehyde. The cells were subsequently permeabilized with 0.1% Triton X-100, PBS, and washed three times with cold 1% normal goat serum, PBS. Coverslips were then incubated with primary antibodies against HuR (1/1500), G3BP (1/1000), Fragile X retardation factor 1 (FXR1; 1/1000), and CUGBP1 (1/1000), diluted in 1% normal goat serum/PBS for 1 h at room temperature. Coverslips were then washed twice with 1% normal goat serum, PBS and incubated with secondary antibody coupled to Alexa Fluor488/594 (Molecular Probes) for 1 h at room temperature. 4Ј,6-Diamidino-2-phenylindole (1:20,000) was added 5-10 min before the end of the incubation with the secondary antibodies. Fluorescence microscopy was performed using the Zeiss Axiovision 4.3 microscope.
Northern and Slot Blots-Northern and slot blots were performed as previously reported (37,38). Total RNA was extracted from cells using the TRIzol reagent (Invitrogen). The RNA was transferred to a Hybond-N membrane, hybridized with [ 32 P]dCTP-labeled human p21, p53, hsp70, and glyceraldehyde-3-phosphate dehydrogenase probes, washed, and then exposed to Biomax films. The mRNA stability experiments were performed as described (39).
Polysome Fractionation-Subconfluent IDH4 cells were switched to dex-free media for 10 days, then exposed or not 0.5 mM AS for 30 min. These cells were then prepared for polysome fractionation as described (40). The sucrose gradients were prepared using sucrose solutions of 10 and 50% sucrose in 20 mM HEPES, pH 7.6, 100 mM KCl, and 5 mM MgCl 2 .
All cells to be used for polysome fractionation were washed twice with 5 ml of cold PBS containing 100 g/ml cycloheximide. The cells were then scraped with a rubber policeman in 1 ml of cold PBS containing 100 g/ml cycloheximide, and cells from the same treatment conditions were combined in one tube. Cells were then centrifuged at 1000 rpm for 10 min. The cell pellets were each resuspended in 425 l of hypotonic lysis buffer composed of 5 mM Tris, pH 7.5, 2.5 mM MgCl 2 , 1.5 mM KCl. The resuspended cells were transferred to a prechilled Eppendorf tube and then the following were added: cycloheximide (final concentration, 100 g/ml), dithiothreitol (final concentration, 2 mM), and 11.8 l of RNAguard. This mix was vortexed, and then 25 l of 10% Triton X-100 and 25 l of 10% sodium deoxycholate were added. After vortexing again, the cell solutions were centrifuged for 2 min at 4°C. The superna-  Arsenite triggers the assembly of high number of G3BP1-containing stress granules in fully senescent IDH4 cells. IDH4 induced for senescence and exposed to AS treatment as described in Fig. 1 were also grown in the presence or absence of cycloheximide (CHX). Cells were subsequently fixed, permeabilized, and were analyzed by immunofluorescence with antibodies against HuR and G3BP proteins. Bars, 20 M. The graph illustrates the number of SGs formed (visualized by G3BP staining) on the different days of senescence. The number of SGs in each experiment was calculated using three random fields each containing 10 cells. Data are represented as means of three experiments Ϯ S.E. (error bars). DAPI, 4Ј,6-diamidino-2-phenylindole. tants were loaded on the sucrose gradient columns and then centrifuged at 35,000 rpm for 2 h at 4°C. The fractions were collected with the fraction collector, 24 fractions per lysate, with a sensitivity of 0.2, a chart speed of 30, and a pump speed of 40. RNA was isolated from all fractions as described (41).
Fluorescence in Situ Hybridization-The fluorescence in situ hybridization experiments were performed as described (22). A DNA fragment of ϳ500 bp corresponding to the coding region of human p21 was amplified by PCR using the following primers fused to either a T7 or T3 minimal promoter sequence: T7-p21 forward, 5Ј-TAA TAC GAC TCA CTA TAG GGG GAA GTA GCT GGC ATG AAG CC-3Ј, and T3-p21 reverse, 5Ј-AAT TAA CCC TCA CTA AAG GGG AAG ACC ATG TGG ACC TGT CA-3Ј. The PCR product was used as the template for in vitro transcription of the p21 probe needed for fluorescence in situ hybridization. The antisense (T3) and sense (T7) probes were prepared using digoxigenin-RNA labeling mix from Roche Diagnostics. The RNA probes were quantified, denatured, and incubated with permeabilized cells at 37°C overnight in the hybridization buffer (50% formamide, 5ϫ SSC (1ϫ SSC ϭ 0.15 M NaCl and 0.015 M sodium citrate), 50 mM phosphate buffer, pH 7.4, 5ϫ Denhardt's solution, 1 mM EDTA, and 250 ng/l salmon sperm DNA). After the hybridization, the cells were incubated with anti-G3BP antibody for 1 h at room temperature. Finally, the cells were incubated with secondary goat anti-rabbit antibody and anti-digoxigenin antibody for immunofluorescence.

Senescent Cells Respond to Stress by Increasing the Number of SGs Compared with Their Proliferative
Counterpart-To investigate the formation and the role of SGs during senescence, we first followed their assembly in senescent cells exposed to a variety of stresses. We . During senescence, AS treatment triggers a significant reduction in the levels of p21 proteins only in late senescent cells. A, the effect of AS on the expression levels of eIF2␣ and phosphorylated eIF2␣ (p-eIF2␣) during senescence. Western blots using antibodies specific to phosphorylated eIF2␣, eIF2␣, and G3BP were performed with total cell extracts and harvested at different time points during the induction of senescence in the presence or absence of AS. Shown are representatives of two independent experiments. B, the bar graphs represent the expression level of phosphorylated eIF2␣ protein in each time point normalized to the expression levels of eIF2␣. The intensity of the signal in each lane was measured using ImageQuant software. Each bar graph represents the ratio of phosphorylated eIF2␣ over eIF2␣ that was normalized to the intensity of G3BP for each time point. The histogram presents the results from A as a mean Ϯ S.E. (error bars) from three independent experiments. C and D, AS-induced oxidative stress affects the expression of p21 protein only in late senescent cells. C, Western blots were performed as described in A and probed with antibodies to p21, p53, CUGBP1, and G3BP. G3BP levels were used as a loading control. D, bar graphs represent the expression levels of p21 is each time point in the presence (ϩ) or absence (Ϫ) of AS. The expression levels of p21 are represented as ratios that were determined as described above. The histogram presents the results from C as the mean Ϯ S.E. (error bars) from three independent experiments. used two model systems, serial passage of WI-38 human diploid fibroblasts and IDH4 human fibroblasts (16). It is well established that senescence can be induced in IDH4 cells by growing them in a dex-free media (dex induces the expression of the SV40 large T antigen that maintains the proliferative state of IDH4 cells) for a few days and in WI-38 by serial passages until they reach the population doubling number of 39 or more (42,43). One of the well used biomarkers for replicative senescence is senescence-associated ␤-galactosidase activity (35). While testing senescence activation in our cell systems, we observed that Ͼ95% of IDH4 cells presented a senescence-associated ␤-galactosidase staining 8 days after the removal of dex (supplemental Fig. 1A). As expected (16), the same pattern of ␤-galactosidase staining was also obtained in WI-38 fibroblasts as early as population doubling number 36 and was more pronounced at population doubling number 39 and later (supplemental Fig. 1B).
Several laboratories have shown that treatment of cells with oxidative stress-inducing drugs such as AS triggers the formation of SGs (22,23). It is also well established that proliferative and senescent cells differentially respond to oxidative stress (44 -46). Thus, it was of high interest to determine whether SGs could assemble in response to oxidative stress. IDH4 cells grown in proliferative (day 0) or in senescent media (day 4 and 10 post-dex removal) were exposed to 0.5 mM AS for 30 min (Fig. 1). These cells were then fixed, and immunofluorescence experiment was performed using antibodies against three well established SGs markers: FXR1 (22, 47) (Fig. 1), HuR (24), and G3BP (26) (Fig. 2). We observed that AS-induced SGs were visible in both proliferative and senescent cells (Fig. 1, compare  panels 1, 7, and 13 to 4, 10, and 16 and Fig. 2, compare panels 1,2,13,14,25, and 26 to panels 5, 6, 17, 18, 29, and 30). The number of SGs was 5-fold higher in late senescent cells (day 10) when compared with their proliferative counterpart (Figs. 1 and 2, graphs, compare day 0 to day 10). In early senescent cells (day 4), however, the number of SGs increased by only 2-fold (Figs. 1 and 2, graph, compare day 0 to day 4).Because, as expected (16), the levels of HuR protein significantly decreased during senescence (supplemental Fig. 2), we defined the number of SGs using FXR1 (Fig. 1) and G3BP (Fig. 2) as a marker. The same results (SG assembly and the increase in their number during senescence) was also obtained using the WI-38 cell system (supplemental Fig. 3). To characterize these foci as bona fide SGs, we assessed their assembly in the presence or absence of cycloheximide (CHX) (Fig. 2). Cycloheximide is an inhibitor of translation elongation that has previously been shown to trap mRNAs on polysomes, causing the disassembly of SGs (48). We observed that the addition of cycloheximide to AS-treated proliferative and senescent IDH4 cells prevented SG formation (Fig. 2, panels 9, 10, 21, 22, 33, and 34). Additionally, we observed that SGs can also form in senescent cells when exposed to other stresses such as heat shock (supplemental Fig.  4) or MG132 (22) (data not shown). Because AS treatment did not affect both the general distribution of the ␤-actin protein (data not shown) and the elongated shape of senescent cells (Fig. 2 compare panels 25 and 26 to panels 29 and 30), we concluded that the increase in the number of SGs was not due to a disruption of the cytoskeleton-based cellular structure. More-

. AS treatment increases the half-life of p21 mRNA in both proliferative and senescent cells.
A, the increase in the steady state levels of p21 mRNA during senescence is not affected by AS. Total mRNA was prepared from IDH4 cells induced for senescence and exposed to 0.5 mM AS for 30 min. The expression level of p21 mRNAs was determined using 10 g of total mRNA in a Northern blot analysis. Endogenous glyceraldehyde-3-phosphate dehydrogenase (GAPDH) mRNA was assessed as a loading control. Shown are representatives of three independent experiments. B, C, and D, AS stabilizes p21 mRNA in both proliferative and senescent IDH4 cells. B and C, proliferative (day 0) and senescent (day 8) IDH4 cells were treated with or without AS and then incubated with 5 g/ml actinomycin D (ActD) for the indicated times. 25 g (day 0) and 10 g (day 8) of total RNA was prepared, and p21 mRNA levels were subsequently verified by Northern blot analysis. The expression of the glyceraldehyde-3-phosphate dehydrogenase mRNA was also assessed and used as a loading control. D, p21 mRNA was quantified using the ImageQuant software program, standardized against glyceraldehyde-3-phosphate dehydrogenase message, and plotted as the percentage of remaining mRNA compared with message levels at the 0 time point (where there is 100% of the maximum mRNA level). Error bars, S.D. of three independent experiments. MARCH 27, 2009 • VOLUME 284 • NUMBER 13 JOURNAL OF BIOLOGICAL CHEMISTRY 8881 over, our recovery experiments clearly showed that these SGs disassemble completely 2-4 h upon the removal of stress albeit with a small but reproducible delay in fully senescent cells (supplemental Fig. 5). Therefore, together our data suggested that forming a high number of SGs is one of the early responses used by senescent cells to cope with a variety of stresses.

Stress Granules and Senescence
AS Treatment Specifically Affected the Expression of p21 Protein Only in Senescent Cells-It is well established that in proliferative cells the formation of AS-induced SGs correlates with the phosphorylation of the translation initiation factor eIF2␣ (23). To assess whether this is also the case during senescence, IDH4 cells grown in the presence (proliferative) or absence of dex (senescent) and exposed or not to 0.5 mM AS for 30 min were harvested at different days during the senescence process. Total extracts prepared from these cells were subjected to Western blot analysis using antibodies against the phosphorylated (p-eIF2␣) and non-phosphorylated (eIF2␣) isoforms of eIF2␣ protein. As expected, the phosphorylated eIF2␣ isoform was detected at all stages of the senescence process, only in AS-treated cells (Fig. 3A, compare lanes 1-5 to 6 -10). Our experiments showed that despite the general decrease in the expression levels of eIF2␣ protein in late senescence cells, we still detect a significant increase in the ratio of phosphorylated eIF2␣ when compared with total eIF2␣ in fully senescent cells exposed to AS for 30 min (Figs. 3, A, compare lanes 9 and 10 to  lanes 4 and 5, and B). Together these results indicated that the assembly of AS-induced SGs correlated with the phosphorylation of eIF2␣ in both proliferating and senescent cells.
Both the development and the maintenance of the senescence phenotype require the expression of several factors such as p53, CUGBP1, and p21 (1,17). To examine the expression of these proteins in senescent cells exposed to AS, we prepared total cell extracts as described for Fig. 3A and performed a Western blot analysis using specific antibodies against p21, p53, and CUGBP1 proteins (Fig. 3C). Antibodies against G3BP and Hsp70 (supplemental Fig. 6) proteins were used to assess equal loading. We observed that at early senescence (days 2-4), AS did not affect the expression of any of these factors. At late senescence, however, AS triggered a significant reduction (Ͼϳ75%) in the levels of p21 but not in the levels of p53 and CUGBP1 proteins (Figs. 3, C and D, compare days 0 -4 to 8 -10). These data indicated that despite the high expression levels of p53, which is known as one of the main activators of p21 transcription during senescence (49,50), AS stress specifically reduced the expression levels of p21 protein only in late senescent cells.

AS Treatment Prevents the Translation of p21 mRNA in Late Senescent Cells, and This Correlates with Its Rapid Recruitment
to SGs-The fact that the level of p53 protein was not affected in AS-treated senescent cells raised the possibility that under these conditions, the expression of p21 protein is regulated in a p53-independent manner. This effect could be due to a reduction in the transcription, the stability, or the translation of the p21 mRNA. To assess these possibilities we first performed Northern blot analysis on total mRNA isolated from proliferative and senescent IDH4 cells exposed or not to 0.5 mM AS for 30 min. We observed that the steady state levels of p21 mRNA increased by 20-fold in fully senescent cells in the presence or absence of AS (Fig. 4A). This result ruled out the possibility that AS affected the transcription levels of p21 gene.
Next, we performed actinomycin D (a known inhibitor of polymerase II transcription (37,39)) pulse-chase experiments to examine the effect of AS treatment on the stability of p21 mRNA in IDH4 cells grown in the presence (day 0) or absence of dex (day 8). We observed that AS treatment increased the half-life of p21 mRNA by Ͼ4 h in proliferative and by Ͼ2 h in late senescent cells (Figs. 4, B-D). Collectively, these data showed that the pronounced decrease in the expression of p21 protein (Fig. 3C) is not due to changes in the stability or the steady state levels of the p21 mRNAs. Therefore, it is possible that exposing senescent cells to AS stress triggers cellular mechanisms that interfere with the translation of p21 message. This possibility was confirmed by following the distribution of p21 mRNA in polysome gradients in proliferative and during the senescence of IDH4 cells exposed or not to 0.5 mM of AS for 30 min. Slot blot analysis (38) showed that during proliferation (day 0) and early senescence (day 4) AS treatment did not affect the association of p21, p53, and hsp70 mRNAs with heavy polysomes. However, in fully senescent cells (day 10), AS treatment triggered the rapid dissociation of p21 mRNA (but not p53 and hsp70 mRNAs) from heavy polysomes (Fig. 5). Because RNA association with heavy polysomes is a strong indication of active translation (51), our observations argue that oxidative stress prevented p21 translation only in late senescence cells. Additionally, these observations are consistent with our data showing that although AS treatment did not affect the expression levels of p53 (Fig. 3C) and Hsp70 proteins (supplemental Fig. 6), AS significantly reduced the level of p21 protein only in fully senescent IDH4 cells (Fig. 3C).
We were surprised to observe that in fully senescent cells the exposure to 0.5 mM AS for only 30 min led to this rapid decrease in the translation of p21 mRNA (Figs. 3, C and D, and 5). Because SGs are known to prevent the translation of the messages they recruit (20, 52, 53), we assessed whether p21 mRNA is recruited to SGs in senescent cells which might explain the rapid decrease in its translation. We performed RNA fluorescence in situ hybridization experiments (22) on IDH4 cells induced or not for senescence in the presence or absence of AS. We observed that endogenous p21 mRNA was recruited to ASinduced SGs during the different steps of the senescence process (Figs. 6, A-D). Under these conditions and as expected, the number of p21 mRNA-containing SGs significantly increased FIGURE 5. AS treatment reduces the levels of mRNAs that associate with heavy polysomes for p21 but not for p53 and hsp70. A-C, sucrose gradient polysome fractionation was performed on IDH4 cells at different stages of senescence (days 0, 4, 10) that were treated for 30 min with (ϩ) or without (-) 0.5 mM AS. Absorbance at wavelength 540 was measured to observe the fractionation of the polysomes (left panels). RNA was isolated from all fractions, and slot blot analysis was performed using specific probes for p21, p53, and hsp70 mRNAs. Shown are representatives of two independent experiments. D-F, bar graphs represent the sum of the intensity of all p21, p53, or hsp70 bands that associate with heavy polysomes as indicated normalized to the sum of the 18 S band for the same fractions. The intensity of the signal in each lane was measured using ImageQuant software. Each bar graph represents the ratio of p21, p53, or hsp70 mRNAs over 18 S signals. The histogram presents the results from A as a mean Ϯ S.E. (error bars) from two independent experiments. as the cells enter senescence (Fig.  6E). We observed that only ϳ20% (day 0) and ϳ30% (day 4) of total SGs recruit p21 mRNA in proliferative and early senescent cells, respectively. In fully senescent cells, however, Ͼ90% of total SGs contain p21 mRNA in response to AS treatment (Fig. 6E). Additionally, we observed that in response to AS, the translation activator of p21, CUGBP1 is also recruited to SGs (supplemental Fig. 7). Because in fully senescent cells AS does not alter the levels of p21 mRNA (Fig.  4A), our data argue that the decrease in p21 translation under these conditions (Figs. 3C and 5) resulted from the rapid recruitment to SGs of p21 mRNA. Therefore, our data raised the possibility that the recruitment of p21 mRNA to SGs could be part of the cellular mechanisms that participate in the blockage of p21 translation in senescent cells exposed to oxidative stress.
The role of p21 protein in the induction of senescence is well demonstrated. Our data showed that upon the removal AS stress, the disassembly of SGs is delayed in fully senescent cells compared with their proliferative counterpart (Fig.  supplemental Fig. 5). Therefore, we assessed whether this delay in SG disassembly could affect the recovery of p21 translation. Western blot analyses were performed on total extracts from senescent cells harvested at different time points after the removal of AS from the media using antibodies against p21 and G3BP proteins. We observed that the level of p21 protein significantly increased 2 to 4 h after the removal of AS (Fig. 7, A and B). This increase coincided with the disappearance of SGs in the early and late senescent cells (supplemental Fig. 5). Interestingly, the expression levels of p21 protein 4 h post AS-removal is significantly lower in late senescent cells when compared with their early counterparts at the same time point (Fig. 7B, compare 4 h post-recovery for days 4 and 10). These observations indicated that when cells are fully senescent, their ability to reestablish the normal translation levels of p21 protein during cell recovery from stress is affected.

DISCUSSION
Despite many advances in our understanding of the molecular mechanisms responsible for the senescence phenotype, very little is known about the cellular processes by which fully senescent cells respond to stress. In this study we show that although SGs can assemble in response to oxidative stress (AS) during all the steps leading to senescence, their number significantly increases only when cells have become fully senescent (Figs. 1 and 2 and supplemental Figs. 3-5 and 7). One of the main consequences of this increase is the rapid decrease in the expression levels of p21 protein (Figs. 3, C and D), a key player in the activation of the senescence phenotype (2,54). Although the AS-induced recruitment of p21 mRNA to SGs (Fig. 6) correlates with an increase in its half-life (Figs. 4, B-D), this translocation coincides with the rapid inhibition of p21 translation only at late senescence (Figs. 3, C and D and 4). We show that this translation inhibition may be explained by the co-localization of p21 mRNA and CUGBP1 protein to SGs (Figs. 6 and supplemental Fig. 7). The increase in the number of SGs in response to AS also coincides with a decrease in the ability of senescent cells to recover the normal expression levels of p21 protein upon stress removal (Fig. 7). Collectively, our data argue that the formation of SGs and the reduction in the translation levels of p21 protein represent two main events through which senescent cells respond to stress conditions.
Although their ability to divide or to differentiate is impaired, senescent cells are still able to adapt to different growth conditions. However, their adaptive capacity to environmental stresses decreases (55, 56). SGs are believed to act as a storage area for many poly(A) ϩ mRNAs in response to various stresses. The translation of these mRNA is temporarily blocked, and as soon as the stress is removed they will be either redirected to the translation machinery or eliminated by rapid decay (20 -22, 57). Our data show that this is also the case for senescent cells, where SGs can be induced in response to different stresses (Figs. 1 and 2 and supplemental Figs. [3][4][5]. Additionally we observed that the number of SGs increases considerably with senescence (Figs. 1 and 2 and supplemental Fig. 3-5). One of the main events required for AS-induced SG assembly is the phosphorylation of eIF2␣ (23,48). Here we observed that although the general level of eIF2␣ decreases during senescence (Fig. 3A), the phosphorylation of eIF2␣ increases in response to AS (Fig. 3B). These observations indicate that the same cellular mechanisms (the phosphorylation of eIF2␣) required for SG assembly in response to oxidative stress in proliferative cells remain active in senescent cells.
Although, it is well established that senescent cells recover from stresses more slowly than proliferative cells (58 -60), the molecular mechanism behind this delay is still elusive. Our data indicate that the increase in the number of SGs could in part explain this delay. Indeed, we observed that the disassembly of SGs occurs at 2 h after removal of AS for proliferative and early senescent cells (supplemental Fig. 5, panels 9 -12 and 25-28). However, in late senescent cells this disassembly occurs only 4 h after the stress is removed (supplemental Fig. 5, panels [45][46][47][48]. Interestingly, our data also indicate that under these conditions, the expression levels of p21 protein (61) cannot fully recover after stress removal (Fig. 7). It has been shown that reducing the levels of p16 protein in senescent cells allowed the resumption of their growth upon p53 inactivation (62). There- FIGURE 7. After the removal of AS treatment, the expression levels of p21 protein recover to lower maximal levels in cells at late senescence when compared with their counterparts at an earlier senescent state. A and B, early (day 4) and late (day 10) senescent IDH4 cells were incubated with AS (0.5 mM) for 30 min. Cells were subsequently washed twice with PBS, replenished with fresh media, and incubated for various periods of time at 37°C. Western blots, performed with total cell extracts prepared from these cells, were analyzed using antibodies specific to p21 and G3BP (used as the loading control). B, graph illustrating the quantified p21 protein levels in A normalized against G3BP loading control for each time points during recovery. Data are represented as means of three independent experiments Ϯ S.E. (error bars). O/N, overnight.
fore, it is possible that a similar decrease in p21 levels, such as the one mediated by AS treatment, could lead to the same outcome. Exploring this possibility could help better define the role of p21 in the senescence-based cell transformation that is triggered by a variety of acute or chronic stresses.
In this study we show that SGs may play an important role in modulating the expression of p21 in senescent cells that have been exposed to oxidative stress conditions. As expected (12), we observed that the expression of p21 protein increases during early senescence to a maximum level in both stressed and unstressed cells (Figs. 3, C and D). However, we were surprised to observe that in response to AS treatment, the expression level of p21 protein was considerably reduced only in late senescent cells (Figs. 3, C and D). Because AS did not affect the steady state levels of p21 mRNA and increased its half-life independently of the senescence state of the cell (Fig. 4), we concluded that the observed reduction in p21 protein level is due to an effect on its translation. Our data confirm this assumption and showed that SGs, by co-recruiting p21 mRNA and its translation activator CUGBP1 (Figs. 6 and supplemental Fig. 7), are part of the regulatory mechanisms that interfere with the translation of p21 mRNA in response to oxidative stress. This conclusion was further supported by the fact that in response to AS, only fully senescent cells reduce the levels of p21 mRNA that are normally associated with heavy polysomes (Fig. 5). It is not clear, however, why the translation of p21 is affected only in cells that are fully senescent and not during the early steps of senescence. Addressing this question will help better understand why senescent cells react differently to a variety of stresses.
Much work is needed to define the impact of repeated exposures to stress on the ability of senescent cells to adapt and respond to environmental changes. Such studies could address whether prolonged or repeated exposure to stress may explain pro-malignant function acquired by senescent cells in some old individuals (2,4,63). It is, thus, critical to identify the molecular pathways as well as the key players that mediate how senescent cells respond to stress to advance the design of anti-cancer drugs.