Cellular senescence limits translational readthrough

ABSTRACT The origin and evolution of cancer cells is considered to be mainly fueled by DNA mutations. Although translation errors could also expand the cellular proteome, their role in cancer biology remains poorly understood. Tumor suppressors called caretakers block cancer initiation and progression by preventing DNA mutations and/or stimulating DNA repair. If translational errors contribute to tumorigenesis, then caretaker genes should prevent such errors in normal cells in response to oncogenic stimuli. Here, we show that the process of cellular senescence induced by oncogenes, tumor suppressors or chemotherapeutic drugs is associated with a reduction in translational readthrough (TR) measured using reporters containing termination codons withing the context of both normal translation termination or programmed TR. Senescence reduced both basal TR and TR stimulated by aminoglycosides. Mechanistically, the reduction of TR during senescence is controlled by the RB tumor suppressor pathway. Cells that escape from cellular senescence either induced by oncogenes or chemotherapy have an increased TR. Also, breast cancer cells that escape from therapy-induced senescence express high levels of AGO1x, a TR isoform of AGO1 linked to breast cancer progression. We propose that senescence and the RB pathway reduce TR limiting proteome diversity and the expression of TR proteins required for cancer cell proliferation.


INTRODUCTION
Cellular senescence is a tumor suppressor mechanism that prevents proliferation in cells bearing oncogenic stimuli (Collado and Serrano, 2010). Senescent cells can efficiently halt tumor progression by remaining out of the cell cycle permanently as benign lesions (Collado and Serrano, 2010;Vernier and Ferbeyre, 2014). Ideally, they also activate their elimination through immune mediated clearance (Kang et al., 2011;Xue et al., 2007). However, if not cleared, some senescent cells can occasionally escape from their dormancy and progress into malignant tumors (Milanovic et al., 2018;Romanov et al., 2001). There is a great heterogeneity in the senescence response depending on the inducer, tissue type and genetic background (Hernandez-Segura et al., 2017). Nevertheless, most senescent cells activate the p53 and retinoblastoma (RB) tumor suppressor pathways that block cell cycle progression (Collado and Serrano, 2010;Vernier et al., 2011;Vernier and Ferbeyre, 2014). In epithelial cells, reactive oxygen species, defective DNA repair and DNA damage were linked to senescence bypass and malignant progression (Gosselin et al., 2009;Nassour et al., 2016). It is commonly accepted that DNA damage and mutations fuel tumor initiation and progression (Harfe and Jinks-Robertson, 2000;Kandoth et al., 2013) by generating both oncogenic drivers and genetic diversity (Bielas et al., 2006). However, little is known about the contribution of translational errors in tumorigenesis.
In normal cells, translational errors in the form of aminoacid misincorporation are estimated to occur at a reduced frequency and unlikely to significantly affect the proteome (Drummond and Wilke, 2009). In microorganisms, mistranslation contributes to adaptive evolution by purging deleterious mutations and inducing stress adaptation (Bratulic et al., 2017;Ribas de Pouplana et al., 2014;True et al., 2004). Moreover, translational recoding is often used by viruses to increase the coding potential of their small genomes (Bidou et al., 2010). For example, retroviruses use frameshifting to control the synthesis of viral replication proteins (Charbonneau et al., 2012). Overall, translational readthrough (TR), originally discovered in viruses and now extended to metazoans, can add a new C-terminal signal to proteins that may change their function and localization (Jungreis et al., 2011;Schueren et al., 2014;Schueren and Thoms, 2016;Singh et al., 2019). The extensions usually contain potential signals to target proteins to the nucleus, peroxisomes or the membrane (Schueren et al., 2014;Schueren and Thoms, 2016;Singh et al., 2019). Taken together, the available evidence suggests that tumor cells could use such recoding mechanisms to generate new protein variants and evolve. Therefore, tumor suppressor pathways could counteract TR as part of their mechanism of action.
TR results from the recognition of stop codons (UAA, UAG and UGA) by near cognate tRNAs instead of the release factors (RFs) that can also recognize the termination codons to end translation (Tate et al., 2018). The efficiency of translation termination depends on the identity of the stop codon (Firth et al., 2011;Grentzmann et al., 1998;Loughran et al., 2014) and is modulated by metabolic stress in bacteria (Zhang et al., 2020) or oxidative stress in yeast (Gerashchenko et al., 2012). However, little is known about the processes controlling TR in mammalian cells.

Identification of readthrough candidates in mammalian cells using bioinformatic analysis
In order to look for endogenous readthrough candidates, we looked in silico for peptides encoded after termination codons detected in proteomics experiments. We first built a database of peptide sequences located between the first arginine/lysine after the canonical stop codon and the next stop codon, always in the same reading frame using sequences deposited in RefSeq hg38.2. We discarded sequences shorter than five aminoacids. Candidate readthrough peptides were then matched to the data set of peptides identified by trypsin digestion of human proteins and mass spectrometry published in (Rosenberger et al., 2014). We found 278 candidates for endogenous readthrough above the false discovery rate (FDR) threshold (Table S3). Following a readthrough propensity predictor algorithm (Schueren et al., 2014), four candidates showing the highest readthrough propensity values were chosen. This algorithm assigns regression coefficients to the stop codon and all possible nucleotides in the stop codon context based on experimental data. The stop codon context comprises the nucleotide sequences from −6 to +9 positions surrounding the stop codon. The new readthrough candidates are: Vasodilator-stimulated phosphoprotein (VASP, NM_003370.3), Aspartate beta hydroxylase (ASPH, NM_004318.3), Hepsin (HPN, NM_182983.2) and Fibrillarin (FBL, NM_001436.3). Oligonucleotides containing 30 up-stream and 30 down-stream nucleotides flanking the stop codon, as well as the stop codon of those readthrough candidates were chemically synthesized (Biocorp, Canada) and subcloned in XhoI/ApaI restriction sites between Rluc and Fluc genes of our pMSCV-Rluc-Fluc reporter construct. The sequences of all new readthrough candidates' regions are provided in Table S2.

Dual-luciferase assays
IMR-90 transduced cells were washed with 1X PBS, scratched and lysed in 1X Passive lysis buffer supplied in the Dual Luciferase Stop & Glo ® Reporter Assay System (Promega). A 20 μl cell lysate sample was used for luminescence measurements with Lumin Triathler-Hidex. After adding 50 μl of the Fluc reagent (Substrate+Buffer) to the samples, luminescence was measured for 10 s. Addition of 50 μl of the Rluc reagent-Fluc quenching (Stop & Glo ® ) preceded another 10 s measurement. The efficiencies of readthrough (% Readthrough) were determined by averaging the relative Fluc activities (Fluc/Rluc ratio) from luciferase reporters UGA or AQP4, divided by the relative luciferase activities of their non-readthrough controls.
We also compared the readthrough fold-changes, relative Fluc activities (Fluc/Rluc) of each condition were normalized to the relative Fluc activities from non-senescent cells: Normalized Fluc/Rluc ¼ Fluc=Rluc ðx conditionÞ Fluc=Rluc(non À senescent cells) : To measure readthrough after therapy-induced senescence, PC3 and MDA.MB.231 cells were retrovirally transduced with luciferase reporter AQP4 or UGA. Transduced cells were treated 24 h using 100 nM of camptothecin or vehicle (DMSO), to induce senescence. Following the treatment, cells were washed with 1X PBS and fresh medium was added.
Luciferase activities were measured at day 7 post treatment. All assays were performed in technical triplicates. Normalized relative Fluc fold changes and Fluc/Rluc ratios are provided in Table S4 as well as all biological replicates.
Cell proliferation assay and senescence-associated β-galactosidase assay Relative density of cells was assessed from estimations of cell number according to crystal violet retention assay (Vernier et al., 2011). The senescence-associated β-galactosidase (SA-β-Gal) activity was measured at day 7, 12 or 35 post-infection as previously described (Vernier et al., 2011). All assays were performed in technical triplicates.

RT-qPCR
Senescent and non-senescent IMR-90 s were collected at 12-or 35-days post-infection, with either pBabe-H-RasV12 or pBabe control vector in TRIzol reagent (Invitrogen) and total RNA was extracted following manufacturer's instructions. Two μg of total RNA were then used for the RT-qPCR reactions as described in (Igelmann et al., 2021). Primers for qPCR were designed using the Universal Probe Library Assay Design Center (https://www.roche-applied-science.com/sis/rtpcr/upl/index/jsp) and are provided in Table S5. Analysis for indicated genes were done using the ΔΔCT method. HMBS and TBP were used as reference genes (Vernier et al., 2011). All assays were performed in technical triplicates.

Statistics and reproducibility
Statistical analysis (one-way ANOVA with post-hoc Tukey's honestly significant difference (HSD) test) was done using the one-way ANOVA test calculator at: http://astatsa.com/OneWay_Anova_with_TukeyHSD/. Twotailed Student's t-tests were performed using GraphPad Prism version 6.0c software. A P<0.05 was considered statistically significant. Each experiment was repeated at least three times, except for those in Fig

Translation termination is improved in oncogene-induced senescence
To investigate the effect of cellular senescence on TR efficiency, we first used a model of oncogene-induced senescence (OIS) in human primary cells. Primary fibroblasts were retrovirally transduced with H-RasV12 oncogene (senescence inducer) or the control empty vector. They were also transduced with luciferase reporter plasmids containing either: (1) a TGA within an artificial intercistronic region (UGA readthrough); (2) the stop codon context from AQP4, a well-known mammalian readthrough candidate (Loughran et al., 2014); or (3) the non-readthrough controls (TGA mutated to CGA) between Rluc and Fluc fusion protein gene (Fig. 1A). Immunoblotting shows a band at 35 kDa corresponding to Renilla luciferase, as well as a weak band at 100 kDa indicating the fusion protein Renilla-Firefly luciferase produced after a ribosomal readthrough in cells transduced with the AQP4 readthrough reporter (Fig. 1B, right). For UGA reporter, the fusion protein is not detectable. On the other hand, in cells transduced with the nonreadthrough control only the fusion protein can be detected (Fig. 1B, left). At early stages of H-RasV12 oncogene expression, cells present a cancer-like behavior characterized by a hyperproliferation phase. However, around 6 days after introduction of oncogenic ras, cells stop proliferation and enter a well-characterized and stable cell cycle arrest (Serrano et al., 1997). We confirmed this senescent cell cycle arrest at day 12 post infection, using an assay for the senescence-associated β-galactosidase activity that allowed visualizing characteristically large and flat cells that stained positive for this biomarker (Fig. S1A). RT-qPCRs were carried out to evaluate the decrease of Ki67 expression, indicating the proliferation arrest, and the activation of p53 targets (CDKN1A/ p21, GADD45A) and RB pathway (CDKN2A/p16INK4a, MCM6) in senescent cells (Fig. S1B). Global translation was not affected in senescent cells since their polysome profiles were similar to control growing cells (Fig. 1C). This is consistent with previous reports and with the fact that senescent cells are actively secreting multiple proinflammatory mediators (Young et al., 2009). In addition, the nonreadthrough (CGA) controls presented similar relative Fluc activities both in senescent and proliferating cells (Fig. S1C). Cap-dependent Rluc expression increased with time (Fig. 1D). Readthrough-dependent translation of the firefly luciferase reporter increased early after introduction of oncogenic ras in normal human fibroblasts when cells proliferate rapidly but it decreased when the cells underwent OIS as measured at days 12 or 20 after H-Rasintroduction (Fig. 1E). The relative efficiency of TR was calculated by dividing Fluc by Rluc values, then normalizing to the Fluc/Rluc ratio obtained in the same cells using the non-readthrough (CGA) control. The results clearly show that the senescent program reduced TR efficiency at UGA codons without any particular context and in the sequence flanking the AQP4 translation termination signal (Fig. 1F). Similar results can be seen after plotting the Fluc/Rluc ratio of senescent Ras-expressing cells relative to cells with empty vectors (Fig. 1G). In contrast, control cells did not change TR efficiency during the same time in culture (Fig. S1D). To discard the possibility that TR efficiency might be decreased due to a cell-cycle arrest independent from the senescent phenotype, proliferating cells were starved for a week to induce quiescence. Starved normal cells compared to cells supplied with fetal bovine serum (FBS) showed no differences in the efficiency of TR. Besides, starved and FBSsupplied senescent cells showed similar TR values (Fig. S1E). These observations lead us to conclude that the decrease of TR efficiency is senescence-specific. Immunoblots showed the gradual reduction of phosphorylated RB, the absence of the E2F target MCM6 and the reduction of the mitotic marker H3( pS10) in H-Ras cells, confirming the establishment of senescence. P53 accumulates in senescent cells but decreases after 20 days post-infection in H-Ras cells (Fig. 1H).
The aminoglycoside gentamicin disrupts prokaryotic protein synthesis by binding to 16S ribosomal RNA, but also affects eukaryotic ribosome proofreading, inducing a conformational change in the ribosome-mRNA complex stimulating the efficiency of TR (Manuvakhova et al., 2000). We sought to investigate whether OIS can decrease TR stimulated by gentamicin. We added 900 μg/ml of gentamicin sulfate to OIS cells 24 h before luciferase assays. The non-readthrough controls were unaffected by gentamicin stimulation (Fig. S2A). Gentamicin increased TR efficiency and this effect was reduced in senescent cells both at UGA stop codons and the natural readthrough sequence of AQP4 ( Fig. 2A; Fig. S2B,C). Of note, for the UGA reporter the TR efficiency was very small (0.1%) but increased more than 50-fold for the AQP4 reporter (6.9%). In both cases, the levels of TR were reduced by about a factor of 2 in senescent cells and we obtained similar results after gentamicin treatment (Fig. S2B,C). Strikingly, TR levels were the same in senescent and proliferating cells with the reporter containing the programmed readthrough region from Moloney Murine Leukemia Virus (MMuLV) ( Fig. 2A). This region contains a UAG stop codon followed by a pseudo-knot structure (Houck-Loomis et al., 2011) and it allows up to 10% of ribosomes to suppress translation termination in order to synthesize the Gag-Pol polyprotein in cells infected with the virus (Yoshinaka et al., 1985). Also, the aminoglycoside gentamicin did not affect MMuLV TR efficiency.
We also treated cells 72 h before cells lysis with 600 μg/ml of gentamicin and 25 μM of CDX5-1, a novel small molecule that potentiates TR efficiency when combined with aminoglycosides although it has by itself no effect (Baradaran-Heravi et al., 2016). We observed that while single gentamicin treatment increased UGA TR efficiency up to 6.6-fold in control cells, the combination increased it up to 16.5-fold. Surprisingly, the TR efficiency was still limited in senescent cells after the combination, which improved the efficiency only 1.7-fold over gentamicin treatment alone (Fig. S2D). Together, these results indicate that senescence engages a significant barrier for TR.
To investigate more broadly whether senescence affects natural readthrough signals, we made luciferase reporters using the stop codon context of several mRNAs, which we predicted in silico as candidates for readthrough (see Materials and Methods). We tested both their basal and gentamicin-stimulated TR. We found that the TR of VASP and HPN were significantly reduced during OIS while the TR of FBL showed a tendency to be reduced (Fig. S2E). ASPH TR was undetectable neither in proliferating nor in senescent fibroblasts. Nevertheless, after gentamicin stimulation, we could observe the TR decrease in senescent cells (Fig. S2F).

Translation termination is improved in PML-induced senescence
The tumor suppressor PML controls p53 and RB which are central for the establishment of OIS response (Vernier et al., 2011). Also, PML plays a role in antiviral responses (El Asmi et al., 2014;Regad et al., 2001) and, since many viruses use readthrough, we wanted to investigate whether PML modulates this process. Normal human fibroblasts IMR-90 cells were retrovirally transduced with a vector that allows expression of a conditionally inactive PML (pBabe-PML-IV-Estrogen Receptor) as well as with UGA, MMuLV and AQP4 luciferase reporters, in order to analyze TR efficiency in this model. Transduced cells were stimulated with 100 nM of the estrogen antagonist 4-hydroxy-tamoxifen (4-OHT), to induce PML-IV nuclear translocation and senescence (Acevedo et al., 2016). We found that induction of PML reduced basal or gentamicininduced TR at UGA stop codons (Fig. 2B). Intriguingly, the effect of PML on gentamicin-induced readthrough was much more important than the effect of OIS, suggesting that PML could be one important regulator of TR efficiency. TR was also reduced after PML induction with 4-OHT, with or without gentamicin, in AQP4 and MMuLV transduced cells (Fig. 2C,D). The decrease of TR efficiency in MMuLV reporter further suggest an antiviral role of PML-IV.

Circumventing senescence decreases the fidelity of translation termination
Although senescence in response to oncogenes includes a very stable cell cycle arrest, some cells escape from the process and progress towards malignant transformation. Cells that circumvent OIS display a gene expression profile typical of malignant cells and chromosomal aberrations (Deschenes-Simard et al., 2016). We sought to determine whether these cells also lose their tight control over TR. We obtained cell populations that bypassed OIS from longterm cultures (35 days) of IMR-90 cells expressing oncogenic ras. Cells that escaped senescence revert their morphology to that of normal growing cells (not shown) and expressed lower levels of oncogenic ras (Fig. 3A), indicating a mechanism that allow them to prevent pro-senescence ras/ERK signaling. This is consistent with our previous observation that decreasing ERK signaling accelerates senescence bypass in H-Ras-expressing cells (Deschenes-Simard et al., 2013. Cells that bypassed senescence also had lower levels of the CDK inhibitor p16INK4a and higher levels of the proliferation marker KI67 (Fig. 3A). We next measured TR in these cells using the UGA stop codon or the natural signals for AQP4 (Loughran et al., 2014), VASP and HPN. In all cases, the cells that bypassed senescence displayed an increased TR even higher than control non-senescent cells (Fig. 3B).

E7 Oncoprotein increases TR
Since low translation fidelity and increased TR are associated to escape from senescence and malignant transformation (Marcel et al., 2013), we next investigated which tumor suppressor pathways activated in senescent cells could modulate TR. To investigate whether p53 and/or RB pathways are implicated in translation termination efficiency, we used E6 (inhibits p53) and E7 (inhibits RB) oncoproteins from HPV-16 virus. Transduction of E6 did not affect the TR levels in senescent cells. However, RAS-senescent cells infected with E7 oncoprotein or with both E6 and E7, showed a significant increase in TR at UGA stop codons and the natural readthrough signals from AQP4 and VASP (Fig. 4A-C). These results indicate that E7 targets, likely the pocket proteins family including RB, p107 and p130, mediate the readthrough decrease in senescent cells. We also knocked down RB mRNAs' expression with shRNAs in H-Ras cells, but we failed to see the increase of TR levels (data not shown) that we observed in E7 expressing cells, suggesting that RB in coordination with p107 and p130 must influence the reduction in TR.
To further implicate the RB pathway in the control of TR in senescent cells, we used an in-frame deletion mutant of E7 whose interactions with RB, P107 and p130 are disrupted (E7 Δ21-24) (Helt and Galloway, 2001). As expected, H-Ras cells transduced with wild-type E7 showed a twofold increase in TR compared to senescent cells but the mutant E7 Δ21-24 failed to do so (Fig. 4D). Besides, the proliferating cells transduced with E7 presented a tendency to increase TR, while the TR was not affected in proliferating cells transduced with the mutant E7 Δ21-24 (Fig. S3A). In addition, RT-qPCRs were carried out to evaluate the expression of E2Fs target genes in cells transduced with E7 or the E7 Δ21-24 mutant. As expected E7 but not E7 Δ21-24 increased E2F target genes expression (Fig. S3B). Next, we used a retroviral vector for CDK4 overexpression to block pocket proteins activity in senescent cells (Sherr et al., 2016) and, as expected, we observed an increase in TR (Fig. 4E). Relative cell growth and RT-qPCRs show that CDK4 overexpression does not affect proliferation arrest markers at day 12 post infection in senescent IMR-90 cells (Fig. S3C,D). Taken together, these results indicate that the observed reduction in TR is not a simple consequence of the growth arrest of senescent cells and strongly suggest that the activation of the RB tumor suppressor pathway increases the fidelity of translation termination during OIS.

RB pathway activation reduces TR
Having demonstrated that RB pathway inhibition increased TR we next investigated whether activation of the RB pathway would inhibit TR. First, we used the CDK inhibitor palbociclib (Law et al., 2015). Treatment of proliferating fibroblasts with palbociclib reduced TR of the AQP4 reporter and gentamicin-stimulated TR at the UGA stop codon (Fig. 5A,B). Interestingly, palbociclib also reduces TR in cells that spontaneously escaped from senescence (Fig. 5C). As expected, palbociclib also inhibits cell proliferation and strongly activates the RB pathway in proliferating and H-Ras-escaped IMR-90 cells ( Fig. 5D; Fig. S4) restoring their senescent phenotype (Fig. 5E,F).
We recently demonstrated that extra-ribosomal functions of RPS14/uS11 and RPL22/eL22 were linked to the regulation of the cell cycle and senescence. We found that these two proteins interact with the CDK4-Cyclin D1 complex, inhibiting its activity and consequently activating the RB pathway (Bury et al., 2021;Del Toro et al., 2019;Lessard et al., 2019;Lessard et al., 2018). These results suggest that RPS14/uS11 and RPL22/eL22 could, as palbociclib, affect TR. We transduced IMR-90 cells with Myc tagged RPL22/eL22 or RPS14/uS11 as well as with the AQP4 readthrough reporter and measured TR efficiency at day 7, 12 and 14 after transduction. The expression of RPL22/eL22 or RPS14/ uS11 was confirmed by immunoblots against the Myc tag (Fig. 5G). We also measured biomarkers of senescence such as RB, the E2F target MCM6 and the mitosis marker phospho-H3 (Fig. 5G). Cells bearing RPS14/uS11 showed a 15% decrease of TR at days 7 and 12 after transduction, and 30% decrease of TR 14 days after transduction compared to control cells (Fig. 5H). In contrast, RPL22/eL22-transduced cells showed a peak of TR reduction at day 7 post infection, and a progressive recovery 12and 14-days post-infection (Fig. 5I). We already reported that RPS14/uS11 is more efficient than RPL22/eL22 for inhibition of CDK4 and induction of senescence (Del Toro et al., 2019), explaining this difference. These results suggest that RPS14/uS11 haploinsufficiency in the 5q-syndrome (Ebert et al., 2008) and frequent RPL22/eL22 hemizygous gene deletions found in cancer (Ajore et al., 2017) could play a role in proteome diversity and tumorigenesis. Taken together, our results suggest that the RB pathway is implicated in translation termination accuracy (Fig. 5J).

Therapy-induced senescence and TR
To investigate whether a reduction in readthrough also applies to the senescent response induced by chemotherapeutic drugs in cancer cells (therapy-induced senescence: TIS) we treated PC3 prostate cancer cells with camptothecin for 24 h and confirmed that most of the cell population was senescent using the SA-β-Gal assay and western blots for phospho-RB, MCM6 and pH3(S10) (Fig. 6A,B). As found for OIS, TIS also reduced TR on both the UGA and the AQP4 reporters (Fig. 6C) and similar results were found in MDA.MB.231 breast cancer cells (Fig. 6D,E). TIS is not a stable phenotype (Guillon et al., 2019;Milanovic et al., 2018;Saleh et al., 2020). After 20 days in culture, a population of growing cells emerged from the TIS population and continued to grow without signs of senescence. These cells that bypass TIS are known to be more malignant (Milanovic et al., 2018;Saleh et al., 2020) and they have an increased TR (Fig. 6F) along with higher phosphorylation of RB (Fig. 6G).
Senescence suppresses the expression of AGO1x, an AGO1 isoform generated by TR AGO1x is an isoform of AGO1 that is expressed by TR of the AGO1 mRNA (Eswarappa et al., 2014;Ghosh et al., 2020;Singh et al., 2019). AGO1x acts as an inhibitor of the miRNA pathway (Singh et al., 2019) and in breast cancer cells it is required to maintain cell proliferation through inhibition of interferon gene expression (Ghosh et al., 2020). Senescent cells upregulate interferon gene expression (Fridman and Tainsky, 2008;Frisch and MacFawn, 2020;Moiseeva et al., 2006) suggesting that a reduced TR of AGO1 mRNA may lead to low levels of AGO1x and high levels of interferon genes in these cells. To test this idea, we triggered therapy induce senescence (TIS) in MDA.MB.231 breast cancer cells with Camptothecin (Fig. 7A,B). As expected, the expression of AGO1x was reduced in these TIS cells (Fig. 7C) but was dramatically induced after cells escaped from senescence ( Fig. 7C,D). Senescence and the escape from senescence was characterized using KI67 (Fig. 7E,F) and immunoblots for RNR (a repression target of p53), p21 (an activation target of p53), MCM6 (a repression target of RB), phospho-H3 (a marker of cell proliferation) and γH2AX (a marker of DNA damage) (Fig. 7G). These results demonstrate that senescence can lead to repression of cancer driver proteins like AGO1x by reducing TR. DISCUSSION We show here that cellular senescence induced by oncogenic ras, the tumor suppressor PML, CDK4 inhibitors or chemotherapy reduces TR. Conversely, we also report that cells that bypass senescence have an increase in translational TR (Fig. 7H). The effect of senescence was demonstrated in a variety of reporters containing stop codons in a common context, within regions that stimulate endogenous TR and on an endogenous RT target, AGO1x. We propose a novel caretaker tumor suppressor activity for cellular senescence controlled mainly by the RB pathway in preventing TR and the generation of C-terminally extended proteins (Schueren et al., 2014;Stiebler et al., 2014). The potential for TR to confer advantages to cells in stress was previously reported in yeast cells expressing the prion PSI that enhances both TR and resistance to several stresses (True et al., 2004).
Senescence also reduced the stimulatory effect of aminoglycosides and the enhancer effect of CDX5-1 on TR (Manuvakhova et al., 2000). It has been reported that CDX5-1 potentiates G418-induced TR up to 180-fold compared to the aminoglycoside G418 alone (Baradaran-Heravi et al., 2016). Green and Goff reported that aminoglycosides can increase TR of the MMuLV gag-pol junction in HEK-293 cells (Green and Goff, 2015). However, using normal fibroblasts, we found that the same drugs did not have such effect. HEK-293 cells have an inactivation of the RB and PML tumor suppressor pathways due to expression of adenovirus E1A oncoprotein. In our normal cells, disruption of these pathways increases TR. It is thus plausible that in the context of an altered control in HEK293 cells, the MMuLV gag-pol junction TR is stimulated by aminoglycosides. In addition, Green and Goff used several aminoglycosides in their studies, gentamicin being the less active and its effects were reported as dose-independent and not consistent. Since gentamicin does increase TR in other stop codon contexts, we suggest that the pseudoknot structure of the viral signal may interfere with the binding or the action of gentamicin on the ribosome.
Our results indicate that translation termination tends to be inaccurate in cancer-like cells where the RB pathway is dysfunctional but is particularly efficient in senescent cells. Early work by Stanners and colleagues showed that SV40-mediated transformation increased mistranslation in mammalian cells  (Pollard et al., 1982). SV40 encodes for the large T antigen that binds and inactivates the RB family of tumor suppressors (DeCaprio et al., 1988) like the E7 oncoprotein we used in this study. Recently, Ruggero et al. found that loss of SNORA24, a small nucleolar RNA that mediates pseudouridylation of rRNA, leads to bypass of RAS-induced senescence in the liver promoting hepatocellular carcinoma. Intriguingly, cells lacking SNORA24 had an increase in TR and mistranslations (McMahon et al., 2019). Also Diaz and

RESEARCH ARTICLE
Biology Open (2021) 10, bio058688. doi:10.1242/bio.058688 colleagues found that translational fidelity is decreased in the most aggressive breast cancer cell lines (Belin et al., 2009). They also reported that the p53 tumor suppressor controls translational fidelity and in particular the translation of oncogenic proteins from IRES (Marcel et al., 2013). In our experimental system, the RB pathway was more important to control TR levels in OIS. Consistent with this notion, we show that TR was reduced after TIS in p53 mutant PC-3 and MDA.MB.231 cells. However, we did notice that inactivation of p53 in cells where RB is also inactivated further increased TR, suggesting also an important role for p53 (Fig. 4A,C). Taken together, the evidence indicates that preventing translational errors and TR is a tumor suppressor mechanism. The RB tumor suppressor pathway may repress genes that modulate TR. TR occurs when near cognate tRNAs outcompete the translation termination factors eRF1/eRF3 (Beissel et al., 2019;Cassan and Rousset, 2001). RB represses tRNA expression (Gjidoda and Henry, 2013) and this could prevent TR by near cognate suppressor tRNAs. Also, the translation initiation factor eIF3 interacts with post-initiation ribosomes and controls TR (Beznosková et al., 2013(Beznosková et al., , 2015. RB could repress the expression of eIF3 subunits or sequester them in PML bodies (Vernier et al., 2011). Consistent with the latter mechanism the eIF3 subunits EIF3K and EIF3E (Int-6) localize to PML bodies (Morris- Desbois et al., 1999;Salsman et al., 2013). Viral proteins such as HTLV-I Tax can delocalize EIF3E from PML bodies (Desbois et al., 1996), an event that could facilitate TR.
Of special interest for a plausible pharmacological modulation of readthrough levels, we found that the CDK4 inhibitor palbociclib was able to reduce TR and re-induce a stable cell cycle arrest in cells that bypassed senescence. If the phenotypic plasticity conferred by C-terminal extended proteins plays a causal role in the origin of human cancers or their resistance to chemotherapy, palbociclib could be used to halt the progression of these lesions. Fig. 7. Senescence affects endogenous TR target AGO1x. (A) Images of senescence-associated β galactosidase assay and its quantification in MDA.MB.231 treated with 40 nM camptothecin (therapy-induced senescence (TIS)) or Vehicle (control) for 96 h. Data shows mean of three biological replicates and error bars show SD Students-t-test was performed and P value is indicated. (B) Relative mRNA levels of indicated mRNA in cells as in A. Data shows mean of three biological replicates and error bars show SD Students-t-test was performed and P value is indicated. (C) Immunofluorescence of AGO1x in MDA.MB.231 cells that were treated with 40 nM camptothecin (TIS) or Vehicle (control) or cells that were treated with camptothecin and left to recover for 14 days to allow bypass of senescence (Bypass of TIS). Scale bar=50 µm. A representative image of three biological replicates is shown. D. Quantification of results in (C). A minimum of 50 cells were scored for nuclear staining intensity using Image J. Statistical analysis was performed by ANOVA with Tukey post-test. A.U., arbitrary units.