The lncRNA MIR31HG regulates the senescence associated secretory phenotype

Senescent cells secrete cytokines, chemokines and growth factors collectively known as the senescence-associated secretory phenotype (SASP) which can reinforce senescence and activate the immune response. However, it can also negatively impact neighbouring tissues facilitating tumor progression. We have previously shown that in proliferating cells the nuclear long non-coding RNA (lncRNA) MIR31HG inhibits p16/CDKN2A expression through interaction with polycomb repressor complexes (PRC1/2) and that during BRAF-induced senescence MIR31HG is overexpressed and translocates to the cytoplasm. Here, we show that MIR31HG regulates the expression of a subset of SASP components during BRAF-induced senescence. The SASP secreted from senescent cells depleted for MIR31HG fails to induce paracrine invasion without affecting the growth inhibitory effect. Mechanistically, MIR31HG interacts with YBX1 facilitating its phosphorylation at serine 102 (p-YBX1 S102 ) by the kinase RSK. p-YBX1 S102 induces IL1A translation which acts as upstream regulator inducing the transcription of the other SASP mRNAs. Our results suggest a dual role for MIR31HG in senescence depending on its cellular localization and points to the lncRNA as a potential therapeutic target in the treatment of senescence-related pathologies.


Introduction
Cellular senescence is an irreversible state of growth arrest that can be driven by several stimuli including telomere shortening due to extensive replication, DNA damage, oxidative stress or oncogene overexpression (Hay ick, 1965, Kuilman, Michaloglou et al., 2010. In addition, it has been recently shown that senescence can play a role in differentiation and tissue regeneration (Munoz-Espin, Canamero et al., 2013, Storer, Mas et al., 2013. Oncogene-induced senescence (OIS) was rstly reported by Serrano et al. when they observed that expressing an oncogenic form of Ras in primary broblast induced senescence (Serrano, Lin et al., 1997). Later work demonstrated that other oncogenes such as mutated BRAF promoted OIS both in vivo and in vitro (Braig, Lee et al., 2005, Michaloglou, Vredeveld et al., 2005. OIS has been considered as a barrier to prevent tumour progression and additional mutations in tumour suppressor genes are required in order to bypass senescence to promote malignancy (Braig et al., 2005, Chen, Trotman et al., 2005, Daniotti, Oggionni et al., 2004. Senescent cells show a characteristic morphology and biochemical features such as halted proliferation, enlarged size, activation of the senescence-associated b-galactosidase, expression of cell cycle inhibitors and the presence senescenceassociated heterochromatin foci (SAHF). Importantly, in several types of senescence, the cells secrete factors such as interleukins, cytokines and metalloproteases, which are part of the senescenceassociated secretory phenotype (SASP) that can impact the cellular environment and homeostasis of the neighbouring tissues (Acosta, Banito et al., 2013, Coppe, Patil et al., 2008, Kuilman, Michaloglou et al., 2008. The downstream effects of the SASP can be bene cial or detrimental depending on the tissue context (Coppe, Desprez et al., 2010, Salama, Sadaie et al., 2014. It has been shown to prevent cancer progression by reinforcing autocrine senescence, inducing paracrine senescence in neighbouring cells (Acosta et al., 2013, Kuilman et al., 2008, and by inducing tissue repair and regeneration (Demaria, Ohtani et al., 2014, Munoz-Espin et al., 2013, Storer et al., 2013. The SASP can also activate the immune system facilitating the clearance of damaged cells (Iannello, Thompson et al., 2013, Xue, Zender et al., 2007. On the other hand, the SASP can promote tumorigenic processes such as angiogenesis and invasion (Coppe et al., 2008, Krtolica, Parrinello et al., 2001. During aging, excessive SASP secretion can induce chronic in ammation that can lead to aged-related pathologies (Baker, Wijshake et al., 2011, Jeyapalan, Ferreira et al., 2007. The composition of the SASP is very variable depending on different aspects such as the senescence stimuli and the cell type (Casella, Munk et al., 2019, Hernandez-Segura, de Jong et al., 2017. De ning the composition of the secretome and identifying new regulators in each biological context is crucial to identify molecular signatures of such a complex phenotype. Previously, the lncRNA TERRA has been shown to be a component of in ammatory exosomes and to modulate transcription of in ammatory cytokines in recipient cells (Wang, Deng et al., 2015) and other lncRNAs have been reported to regulate NF-kB activation and its downstream target genes (Jin, Jia et al., 2016, Ozes, Miller et al., 2016, Rapicavoli, Qu et al., 2013. These ndings raise the hypothesis that lncRNAs could act as key players in the SASP induction during senescence. A few lncRNAs have been directly linked to senescence (Li, van Breugel et al., 2018, Montes & Lund, 2016, Montes, Nielsen et al., 2015. We have previously identi ed the lncRNA MIR31HG to be upregulated in BRAF mediated OIS (Montes et al., 2015). In proliferating cells nuclear MIR31HG represses p16/CDKN2A expression by recruiting Polycomb group complexes. Indeed, we have shown that in melanoma patients harbouring BRAF mutations, MIR31HG negatively correlates with p16/CDKN2A expression (Montes et al., 2015). Upon BRAF induction MIR31HG is upregulated and translocates to the cytoplasm where we now show that it regulates the expression of a part of the SASP repertoire. MIR31HG knockdown in BRAF-induced senescent broblasts reduces expression and secretion of several components of the SASP. Interestingly, conditioned medium (CM) from MIR31HG-depleted senescent cells promotes paracrine senescence but fails to induce cancer cell invasion. Mechanistically, we show that MIR31HG promotes the interaction between YBX1 and the RSK kinase resulting in YBX1 phosphorylation and subsequent induction of IL1A translation. Our results unveil the role of lncRNAs in the regulation of the SASP highlighting the dual role in senescence of MIR31HG by suppressing CDKN2A expression in young cells and facilitating production of a distinct subset of SASP factors in senescent cells.

MIR31HG knock-down decreases the induction of SASP components during senescence
To assess the role of MIR31HG during OIS we used immortalised human broblasts BJ-hTERT, expressing a constitutively active form of the mouse B-RAF (V600E) that is fused to the oestrogen receptor (BJ ER:BRAF). Addition of 1mM of 4-hydroxitamoxifen (4-OHT) activates BRAF inducing OIS (Pritchard, Samuels et al., 1995). We rst con rmed that MIR31HG was overexpressed in BJ ER:BRAF upon 4-OHT induction ( Figure 1A), as we previously reported for TIG3 ER:BRAF cells (Montes et al., 2015).
In validation of the model, analysis of TGCA data from thyroid and colorectal cancer tumor samples, where BRAF is frequently mutated, demonstrated a higher MIR31HG expression in tumors harbouring BRAF mutations ( Figure 1B). We performed RNA sequencing in control cells and BRAF-induced senescent cells transfected with control siRNA or siRNA targeting MIR31HG. As expected, many genes were differentially expressed comparing control versus senescent cells (Figure 1C and Supplementary Table 1A and 1B). Moreover, the transcriptional pro le of MIR31HG knock-down senescent cells resembled the senescent pro le with several clusters of differentially expressed genes ( Figure 1C and Supplementary  Table 1A and 1B). Among the GO categories signi cantly represented among the genes speci cally downregulated in the MIR31HG knock-down conditions we found cytokine and chemokine-related pathways ( Supplementary Fig. 1A). Furthermore, we observed that senescent MIR31HG-depleted cells failed to upregulate part of the main SASP components previously de ned (Coppe et al., 2010) ( Figure 1D and Supplementary Table 1C and 1D). We validated transcriptomics data by RT-qPCR and con rmed that several interleukins and chemokines such as IL6 and CXCL1 were nearly absent at the RNA level, whereas other factors such as ICAM1 or IL1A were not affected ( Figure 1E). To validate our results in another cellular system, we performed RT-qPCR analysis in TIG3 ER:BRAF cell line. Depletion of MIR31HG during BRAF induced senescence decreases the levels of the SASP RNAs ( Supplementary Fig.1B). In order to analyse the composition of the SASP at the protein level, we knocked down MIR31HG in BJ ER:BRAF cells and induced senescence by 72-hour 4-OHT treatment in serum-free medium followed by mass spectrometry analysis of secreted proteins. Untreated cells transfected with control siRNA were used as control. As expected, control senescent cells, when compared to control proliferating cells, showed differential secretion of factors (Supplementary Fig.1C and Supplementary Table 2A and 2B). The secretome of senescent MIR31HG knock-down cells resembled that of control senescent cells with discrete differences ( Figure 1F and Supplementary Fig. 1C and Supplementary Table 2A and 2B). Among the less secreted proteins upon MIR31HG depletion were in ammatory SASP factors IL6 and CXCL1 ( Figure 1F, and Supplementary Fig.1D and 1E). The reduced secretion of these factors was validated by ELISA and western blotting using two different siRNAs for MIR31HG ( Figure 1G and 1H). Altogether our results demonstrate that MIR31HG knock-down reduces the production of several SASP components during OIS already at the transcriptional level and these changes are re ected in the secretome.
The SASP of senescent MIR31HG knock-down cells induces senescence but not invasion in a nonautonomous manner To determine the paracrine effect of the SASP we induced senescence in cells transfected with control siRNA or siRNAs targeting MIR31HG. After 72h 4-OHT treatment CM was harvested and used for subsequent analysis. The CM from proliferating cells without 4-OHT induction was also harvested at the same time point as control. To assess the potential of the SASP in inducing paracrine senescence, we incubated BJ WT cells with the different CM for 72h. Protein analysis of whole cell extracts by western blot revealed that BJ WT growing in CM from control senescent cells and MIR31HG knock-down senescent were both able to mildly upregulate the senescence marker p53 (Figure 2A) as well as the enzyme β-galactosidase ( Figure 2B) compared to proliferating cells. However, the upregulation of in ammatory cytokines that occurs upon 4-OHT induction is reduced after MIR31HG depletion ( Figure  2C). Acosta et al. reported that TGF-β signalling pathway is responsible for the paracrine senescence effect (Acosta et al., 2013). As expected, RNA levels of TGFβ target genes were upregulated in BJ ER:BRAF upon 4-OHT treatment compared to control proliferating cells (Supplementary Fig. 2A), while MIR31HG knock-down did not affect this increase. Moreover, the protein level of pSMAD2, downstream effector of TGFβ signalling, was upregulated during OIS in control as well as in MIR31HG knock-down senescent cells ( Supplementary Fig. 2B). Similarly, WT BJ cells discretely activated TGFβ signalling when incubated with CM from control senescent cells and also with CM from MIR31HG knock-down senescent cells (Supplementary Fig. 2C and 2D). These results demonstrate that the TGFβ signalling pathway remains active in MIR31HG knock-down senescent cells suggesting that its activation could be responsible for the paracrine senescence induction. To determine the role of the SASP in promoting cancer cell invasion we performed transwell invasion assays using the different CM described above as chemoattractant in the lower chamber of the transwell for MDA-MB-231 breast cancer cells. After 48h incubation a higher number of MDA-MB-231 cells exposed to CM from senescent cells invaded through the matrigel membrane compared to proliferating CM ( Figure 2D). In contrast, the CM of MIR31HG knock-down senescent cells did not promote invasion ( Figure 2D). These results demonstrate that the paracrine senescence effect is retained in the SASP of MIR31HG knock-down senescent cells but not the induction of paracrine invasion. These ndings suggest that the effect of MIR31HG in the SASP production does not occur at a general level but that only a subset of components is inhibited.
MIR31HG knock-down decreases CEBPB protein levels and NF-kB activation during OIS As NF-kB and CEBPB regulate the expression of many SASP genes by stimulating their transcription (Acosta, O'Loghlen et al., 2008, Chien, Scuoppo et al., 2011, Kuilman et al., 2008 we investigated the possibility that MIR31HG knock-down affected the abundance or the activation of these transcription factors. As expected, the protein levels of CEBPB and activated phosphorylated RELA (p-RELA) (component of the NF-kB complex) increased during BRAF activation ( Figure 3A). Interestingly, MIR31HG knock-down in senescent cells reduced CEBPB protein expression and activated p-RELA ( Figure 3A) without affecting their mRNA levels ( Supplementary Fig. 3A). In accordance, binding of CEBPB to the IL6 promoter, a canonical CEBPB target, decreased signi cantly upon MIR31HG knock-down compare to control senescent conditions ( Figure 3B). Consistently, NF-kB translocation to the nucleus was inhibited upon MIR31HG depletion ( Figure 3C). Moreover, the signalling pathway upstream NF-kB activation was affected in MIR31HG knock-down senescent ( Supplementary Fig. 3B). These results suggest that MIR31HG acts upstream of both CEBPB and NF-kB transcriptional activation.
MIR31HG knock-down decreases IL1A protein levels upon OIS induction IL1A has been reported to function as an upstream regulator of the SASP (Orjalo, Bhaumik et al., 2009). In line with this, addition of human recombinant IL1A(hr-IL1A) in our model system induced expression of SASP components at the RNA level ( Supplementary Fig. 3C). Blocking IL1A signalling using siRNA against IL1A reduced the levels of CEBPB expression upon senescence as well as the RNA levels of different components of the SASP (Supplementary Fig. 3D and 3E). Interestingly, while IL1A mRNA was not affected by the levels of MIR31HG ( Figure 1C), IL1A protein levels were strongly reduced in MIR31HG knock-down senescent cells compared to control senescent cells as measured by immuno uorescence staining ( Figure 3D) and by western blotting of the total lysate ( Figure 3E). Addition of human recombinant IL1A (hr-IL1A) rescued the incapacity of MIR31HG knock-down senescent cells to induce SASP RNA transcription ( Figure 3F), IL6 secretion ( Figure 3G) and NF-kB nuclear translocation ( Supplementary Fig. 3F).
To test whether MIR31HG would be regulating IL1A at a translational level we performed polysome pro ling in control senescent cells and senescent cells upon depletion of MIR31HG. The pro les obtained upon sucrose gradient separation in both conditions showed no major changes in polysome distribution ( Figure 3H). Analysing the distribution of IL1A mRNA through the gradient we observed that the majority of the transcript present in the heavy polysome fractions ( Figure 3I). Remarkably, a signi cant decrease of IL1A mRNA in these fractions was observed in MIR31HG knock-down senescent cells compared to control senescent cells ( Figure 3J and Supplementary Fig. 3G). These results correlate with the decreased IL1A protein we observed upon these conditions. Distribution of ACTB mRNA through the gradient did not show any difference ( Figure 3J and Supplementary Fig 3G). Despite the decrease in mRNA levels of other cytokines such as IL6, the similar distribution of their mRNAs through the gradient indicates an equal translation e ciency suggesting that the effect of MIR31HG on translation is speci c for IL1A ( Figure 3J and Supplementary Fig 3G). MTOR signalling has been recently involved in SASP regulation through IL1A signalling (Laberge, Sun et al., 2015). However, MIR31HG knock-down does not affect mTOR signalling activation ( Supplementary Fig. 3H) indicating an alternative mechanism for MIR31HG regulation of IL1A. These ndings suggest that MIR31HG is implicated in the modulation of the SASP by regulating the levels of IL1A independently of mTOR signalling pathway. HG MIR31HG interacts with YBX1 and YBX1 knock-down phenocopies MIR31HG depletion In order to elucidate the mechanism by which MIR31HG exerts its function in regulating the SASP, we puri ed endogenous MIR31HG with its associated proteins from UV-crosslinked senescent BJ ER:BRAF cells using antisense oligonucleotides containing locked nucleic acid (ASOs) complementary to the MIR31HG sequence coupled to magnetic beads ( Figure 4A). Oligonucleotides against the RNA sequence of luciferase, not expressed in BJ ER:BRAF cells, were used as control. Mass spectrometry analysis of MIR31HG-associated proteins revealed a low number of signi cantly-enriched proteins among which several RNA-binding proteins ( Figure 4B, Supplementary Table 2C and D). MIR31HG has been previously shown to interact with IκBα during osteogenic differentiation regulating NFκB activation (Jin et al., 2016). However, we did not detect IκBα in our pull-down experiments. The majority were heterogeneous nuclear ribonucleoproteins (hnRNPs) or other predominantly nuclear proteins, involved in RNA processing. Recent reports have demonstrated the role of PTBP1 in regulating the SASP composition through the regulation of the alternative splicing of genes implicated in intracellular tra cking (Georgilis, Klotz et al., 2018).
However, since MIR31HG localizes mainly in the cytoplasm during OIS we focused on YBX1 as a protein with de ned cytoplasmic functions ( Figure 4B). YBX1 has been implicated in several cytoplasmic processes, such as translation or mRNA stability among others (Lyabin, Eliseeva et al., 2014). According to previous ndings, YBX1 prevents senescence in epidermal progenitors and knock-down of YBX1 induces senescence features in keratinocytes (Kwon, Todorova et al., 2018). Depletion of YBX1 had a minor impact on cell growth in untreated BJ ER:BRAF (Supplementary 4A and 4B). However, knock-down of YBX1 in BRAF-induced senescent cells mimicked MIR31HG knock-down phenotype. SASP components were decreased at RNA level in YBX1 knock-down senescent cells compared to control senescent cells (Figure 4C and 4D and Supplementary Table 3A and 3B). Likewise, reduction in the secretion of IL6 and CXCL1 upon YBX1 knock-down were comparable to the levels following MIR31HG knock-down ( Figure 1E and 4E). Furthermore, IL1A protein levels were strongly reduced ( Figure 4F) whereas the mRNA level remained unaltered ( Figure 4D).
Formaldehyde crosslinked-RNA immunoprecipitation (RIP) of GFP-tagged YBX1 validated its interaction with MIR31HG ( Figure 4G). As a positive control, we found YBX1 binding its own RNA as previously reported (Skabkina, Lyabin et al., 2005). Interestingly, YBX1 binds IL1A mRNA whereas other cytokines mRNAs are bound to a lesser extent ( Figure 4G). Furthermore, no binding to abundant nuclear RNAs (MALAT1) or the mitochondrial RNAs previously used as negative controls (Matsumoto, Uchiumi et al., 2012) was identi ed.
Altogether, these results suggest that the interaction between MIR31HG and YBX1 might have a role in regulating the SASP during OIS by regulating IL1A translation. To study the biological function of the MIR31HG and YBX1 interaction we analysed the stability and localization of MIR31HG upon YBX1 KD, since stabilization of RNAs is a well-described function of cytoplasmic YBX1 (Evdokimova, Ruzanov et al., 2001, Yang, Wang et al., 2019. We did not observe changes in MIR31HG expression nor localization upon YBX1 knock-down ( Supplementary Fig 4C and 4D). Moreover, protein levels of YBX1 did not change upon depletion of MIR31HG (Supplementary Fig 4E).
YBX1 is phosphorylated at serine 102 by RSK in BRAF-induced senescence Several modi cations which can affect YBX1 function have been reported . Phosphorylation of serine in position 102 has been shown to be involved in the regulation of translation (Evdokimova, Ruzanov et al., 2006). We therefore studied the phosphorylation status of YBX1 and the putative implication of MIR31HG in YBX1 regulation. In order to analyse p-YBX1 in BRAF-induced senescence, we used an antibody that recognizes p-YBX1 at position S102 (p-YBX1 S102 ). Interestingly, we detected increased levels of this modi cation at different time points after BRAF induction ( Figure 5A). Different kinases have been reported to be responsible for S102 phosphorylation such as AKT and RSK (Stratford, Fry et al., 2008, Sutherland, Kucab et al., 2005. We analysed the levels of activated AKT at different time points after 4-OHT induction and we observed that p-Akt decreased over time to nearly undetected levels already after 24h 4-OHT induction (Supplementary Fig 5A). The kinase RSK, however, is activated at early points during senescence induction and remain active at later time points ( Figure 5B) suggesting that YBX1 might be a substrate for RSK during OIS. To investigate the role of RSK in YBX1 phosphorylation in BRAF-OIS, we treated the cells with speci c RSK inhibitors. Treatment of senescent cells with the inhibitor FMK reduced the levels of p-YBX1 S102 ( Figure 5C). Moreover, BI-D780, a more speci c RSK inhibitor, was able to completely abolish phosphorylation of YBX1 in senescent cells ( Figure   5C). Furthermore, overexpression of RSK resulted in an increased level of p-YBX1 S102 (Supplementary Fig  5B). Our results conclude that YBX1 is phosphorylated by RSK kinase in a BRAF-dependent manner.
Phosphorylation of YBX1 at serine 102 induces translation of IL1A It has been previously shown that YBX1 can promote translation of speci c transcripts (El-Naggar, Veinotte et al., 2015, Evdokimova, Tognon et al., 2009. We next wondered whether p-YBX1 S102 could impact IL1A translation. As expected, BRAF-induced senescence resulted in an upregulation of IL1A protein levels compared to control untreated cells ( Figure 5D). Interestingly, treatment with the RSK inhibitor BI-D1780 failed to upregulate IL1A during senescence ( Figure 5D). YBX1 has been recently implicated in cytokine translation by binding the 3'UTR of their mRNAs (Kwon et al., 2018). Using reporter plasmids harbouring the 3'UTR of IL1A after luciferase gene we demonstrated that a WT version of YBX1 is able to increase luciferase translation in response to 4-OHT induction. A mutant that is not able to be phosphorylated at S102 (S102A) failed to induce translation whereas a phosphomimic mutant (S102D) increased translation even in the absence of 4-OHT ( Figure 5E and Supplementary Fig. 5D). These results suggest that p-YBX1 binds IL1A 3'UTR to increase its translation. We assessed the binding capacity of the different phosphorylated versions of YBX1 to IL1A mRNA and to MIR31HG by RIP. All the proteins showed a similar binding to their targets suggesting that phosphorylation at S102 is not affecting YBX1 binding capacity ( Figure 5F).

MIR31HG knock-down in senescent cells reduces cytoplasmic p-YBX1 inhibiting its interaction with RSK
We next examined whether MIR31HG might have an impact on YBX1 phosphorylation. We performed cellular fractionation to analyse the localization of phosphorylated YBX1 in senescent cells and in senescent cells where MIR31HG was depleted. Interestingly, we observed that upon MIR31HG knockdown p-YBX1 is reduced in the cytoplasm ( Figure 6A and Supplementary Fig. 6A) suggesting a role for MIR31HG in the phosphorylation process of YBX1. Several cytoplasmic lncRNAs have been shown to act as scaffolds for bringing molecules into close proximity. In order to analyse whether MIR31HG was binding both YBX1 and its kinase RSK, we performed native RIP using GFP-tagged version of RSK. We could not detect RSK binding to MIR31HG (Supplementary Fig. 6B and 6C). To further analyse the role of MIR31HG in the interaction of YBX1 with its kinase RSK, we performed proximity ligation assays (PLA) using antibodies against total YBX1 and total RSK. Importantly, the interaction was higher in senescence compared to control ( Figure 6B). Interestingly, MIR31HG knock-down reduced the level of interaction con rming our hypothesis that MIR31HG mediates YBX1 interaction with its kinase RSK during OIS ( Figure 6B).

Discussion
The SASP is a hallmark of senescent cells and responsible for mediating the patho-physiological effects in the surrounding tissues. For a long time, efforts in the eld have been focused on trying to induce senescence in cancer cells in order to prevent cancer progression (Braig et al., 2005, Ventura, Kirsch et al., 2007, Wu, van Riggelen et al., 2007. In contrast hereto, more recent reports have shown that targeting senescent cells strongly improve age-related diseases (Baker et al., 2011, Farr, Xu et al., 2017, Xu, Palmer et al., 2015. Although recent work has extended our knowledge of the signalling network upstream the transcriptional induction of the SASP components (Freund, Patil et al., 2011, Herranz, Gallage et al., 2015, Hoare, Ito et al., 2016, Kang, Xu et al., 2015, Tasdemir, Banito et al., 2016, the complexity and diversity of the SASP suggest that additional regulators might be involved. It is crucial to acquire further knowledge of the detailed mechanism to be able to design therapies for cancer, in ammation and aging treatment. Here, we contribute to understanding the SASP regulation by describing the molecular mechanism by which the lncRNA MIR31HG regulates a subset of the SASP components during BRAF-induced senescence by modulating IL1A translation. It is known that most of the SASP components are regulated at the transcriptional level, although a small fraction may additionally be regulated by post-transcriptional mechanisms (Herranz et al., 2015, Laberge et al., 2015. In this study, we focus on the senescence process initiated following the expression of a mutated BRAF. This is of importance as somatic mutations in BRAF occur in approximately 7% of human cancer (Davies, Bignell et al., 2002). We observed that MIR31HG expression is higher in thyroid and colorectal cancer tumor samples harbouring BRAF mutations, as compared to BRAF wild type, suggesting the relevance of this lncRNA in cancer.
We nd that the secretome from BJ ER:BRAF senescent cells and BJ ER:BRAF MIR31HG knock-down senescent cells show differences in the levels of key SASP components. Whereas most of the genes deregulated at transcription level are also altered at the protein level, a small proportion is regulated only at RNA or protein level. This, together with the comparable polysome distribution pro les of these cell lines, suggest no changes in global translation upon MIR31HG depletion. Interestingly, IL1A, an upstream regulator of the SASP, is reduced at the protein level following MIR31HG knock-down whereas the mRNA level is not signi cantly altered. No IL1A was detected in the conditioned media by mass spectrometry or by ELISA (data not shown), likely due to the fact that IL1A remains attached to the membrane during OIS as previously described (Orjalo et al., 2009). In our cellular system addition of hr-IL1A induces the transcription of several cytokines and instigates the SASP. Intriguingly, it also rescues the decreased transcription of SASP components caused by MIR31HG depletion during BRAF-induced senescence. These results strongly support the idea of MIR31HG regulating IL1A as an upstream regulator of the SASP transcriptional program. Work from the Campisi lab has demonstrated that the translation of IL1A during RAS-induced senescence is regulated by mTOR (Laberge et al., 2015). Although the mTOR pathway is increased in our OIS model, it is not altered in MIR31HG knock-down conditions. Instead, we found the lncRNA MIR31HG to affect IL1A translation through YBX1. Endogenous MIR31HG pull-down and subsequent validations by RIP show that YBX1 is a MIR31HG binding partner. Importantly, YBX1 depletion mimics the MIR31HG knock-down phenotype in BRAF-induced senescence. This includes the reduced translation of IL1A and therefore the downregulation of downstream SASP components at RNA level, such as CXCL1, IL8 or IL6 among others. Interestingly, recent work has shown that YBX1 prevents cytokine translation in cycling keratinocytes by binding their 3'UTR. In these studies, YBX1 knock-down results in increased translation of several cytokines and therefore induction of senescence (Kwon et al., 2018). In proliferating BJ cells, knock-down of YBX1 does induce a slight decrease of cell growth consistent with this report. However, knock-down of YBX1 in cells subjected to BRAF-induced senescence resulted in reduced RNA levels of most of the cytokines previously annotated as SASP components (Coppe et al., 2010). Interestingly, in BRAF-induced senescence we observe that YBX1 binds a high fraction of the IL1A mRNA while binding other cytokines to a lesser extent. YBX1 binds AU-rich motifs present in the 3'UTR of cytokine mRNAs (Beiter, Hoene et al., 2015, Kwon et al., 2018, which is consistent with our IL1A 3' UTR luciferase reporter assays. However, the stronger binding of YBX1 to IL1A mRNA compared to other cytokines show some degree of speci city towards IL1A mRNA. The mechanism underlying this speci city remains unknown. In a recent study, Dominguez and colleagues determined the importance of contextual features in RNA recognition by RBPs (Dominguez, Freese et al., 2018). Perhaps RNA secondary structures, the anking nucleotide composition or the proximity to other RNA-binding proteins might be in uencing the binding of YBX1 to the IL1A mRNA.
We observe an increase in p-YBX1 S102 in BRAF-induced senescence and, importantly, p-YBX1 S102 is reduced in the cytoplasm upon MIR31HG knock-down. Our results indicate that this post-transcriptional modi cation of YBX1 is involved in promoting IL1A translation during BRAF-induced senescence through its 3'UTR, which could explain the previously reported opposite roles of YBX1 in cytokine translation (Kwon et al., 2018). The role of p-YBX1 in translation remains to be fully understood. It is known that YBX1 binds the mRNA cap structure by displacing the eukaryotic translation factor 4E (eIF4A) and eIF4G and hence promotes translational repression (Evdokimova et al., 2001, Nekrasov, Ivshina et al., 2003. p-YBX1 S102 on the other hand binds the mRNA cap structure less tightly allowing translation of repressed mRNAs (Evdokimova et al., 2006). However, not much is known about its role in promoting translation. Our luciferase reporter assay indicates that during senescence, p-YBX1 102 may have a role in promoting IL1A translation through the 3'UTR since the phosphomutant YBX1 S102A fails to induce luciferase induction. Mutation of S102 did not alter YBX1 binding capability to IL1A or other mRNAs, consistently with previous reports demonstrating that phosphorylation at S102 does not affect the general RNA binding capacity of YBX1 (Evdokimova et al., 2001) . Phosphorylation of translation factors are determining events occurring during translation regulation (Proud, 2019) and perhaps p-YBX1 is involved in differentially recruiting other translation factors with a role in IL1A mRNA translational regulation. Further phospho-proteomic studies would help to address this question. We cannot exclude that phosphorylation at other residues occur during senescence. In fact, S165 and S176 has been shown to activate NF-kB signalling (Martin, Hua et al., 2017, Prabhu, Mundade et al., 2015. It would be interesting to investigate whether these modi cations have any impact in senescence or in translation regulation. LncRNAs can modulate post-transcriptional modi cations of proteins (Li, Liu et al., 2017, Lo Piccolo, Mochizuki et al., 2019, Wang, Xue et al., 2014, Yoon, Abdelmohsen et al., 2013. Our results suggest that during BRAF-induce senescence RSK appears to be the main kinase responsible for YBX1 phosphorylation. PLA experiments show an increase in RSK-YBX1 interaction during OIS correlating with an increased level of p-YBX1. Despite at a lower level, the presence of interaction in proliferating conditions suggests that the interaction may occur without BRAF induction. The signalling cascade activated during OIS might be necessary to phosphorylate YBX1 at least at S102. We cannot exclude that RSK phosphorylates other residues in proliferating cells. Nevertheless, our results indicate that MIR3HG is important for the YBX-RSK interaction during senescence, when the lncRNA is expressed and predominantly located in the cytoplasm (Montes et al., 2015). We do not detect MIR31HG directly binding RSK, likely excluding a role for MIR31HG as scaffold bridging together YBX1 and its kinase. Instead, MIR31HG interaction with YBX1 might infer an optimal conformation to facilitate its phosphorylation. Another possibility is that the interaction between YBX1 and MIR31HG brings YBX1 to another protein or protein complex required for the phosphorylation. Structural studies or more exhaustive phosphoproteomic analysis could improve our understanding on how MIR31HG promotes YBX1 phosphorylation under OIS.
The SASP has been reported to either limit or promote tumor progression. Dissociation of the good and bad sides of the SASPs is di cult since many of its components can have both roles depending on the cellular context (Salama et al., 2014). Besides, many of the senescence effectors are activated by the same pathways that activate the SASP. Therefore, the identi cation of factors that can affect the SASP without altering the tumour-suppressive effects associated with senescence is a promising strategy for senescence-related therapies. Several reports have described factors that can uncouple the senescence growth arrest from the SASP (Georgilis et al., 2018, Herranz et al., 2015, Laberge et al., 2015, Tasdemir et al., 2016. Here, we identify a lncRNA that, when located in the cytoplasm during OIS, is responsible for the production of a distinct subset of SASP components. Its depletion during OIS leads to a decrease of interleukins, chemokines and other factors preventing invasion in vitro without reverting growth arrest. Interestingly, we have previously described a mechanism for the nuclear MIR31HG in proliferating cells in repressing p16 INK4A expression by recruiting polycomb group proteins (Montes et al., 2015). Altogether, our results suggest that the lncRNA MIR31HG has a dual role in inducing and buffering senescence depending on the cellular localization, which highlights the complexity of the regulation of senescence and the SASP. Interestingly, MIR31HG has been reported to be upregulated in different types of cancer (Eide, Eilertsen et al., 2019, Shih, Chiang et al., 2017, Yang, Liu et al., 2016. The fact that depletion of MIR31HG induces p16 expression without activation a SASP response (Montes et al., 2015) and decreases pro-invasive factors during SASP induction makes it an interesting target for therapeutic purposes. Inhibition of MIR31HG as well as other anti-SASP therapies could be potentially used to ameliorate the detrimental effects of senescence cells.

Treatments and siRNA transfections
Senescence was induced by treatment with 1µM 4-hydroxytamoxifen (4-OHT, Sigma) 24h after cell seeding for at least 48h (unless otherwise indicated). siRNA oligonucleotides were transfected at a nal concentration of 50nM by reverse transfection using RNAiMAX (Invitrogen) according to the manufacturer's instructions for 48h or 72h. In case of senescence induction, 1µM 4-OHT was added 24h after transfection to fresh media for 48h for RNA analysis or 72h for protein analysis. Primer sequences are listed in Supplementary Table 4. BI-D1780 (Axon 1528) and FMK (Gift from Morten Frodin) Rsk inhibitors were used at 10uM for the time indicated in each experiment.

MIR31HG expression analysis in tumor samples
Gene expression (RNAseq) and somatic mutation data for tumors in The Cancer Genome Atlas (TCGA Pan-Cancer) were downloaded from UCSC Xena platform (Goldman, Craft et al., 2019). Thyroid carcinoma (THCA) and colo-rectal adenocarcinoma (COAD & READ) were used for analysis because these tumors have high frequency of BRAF mutations (>10%) and also express MIR31HG. Expression values for all the samples were converted to log2(fpkm+1) unit. Wilcoxon test was performed to compare the expression differences in the BRAF mutant and BRAF wild type tumors, using R.

YBX1-GFP and RSK-GFP overexpressing cell lines
To generate lentiviral constructs, we cloned the coding sequence of YBX1 ampli ed from cDNA, into the inducible vector pLVX-TetOne Puro Vector (Clontech) that already contained GFP-fusion protein. We use In-Fusion HD Cloning designing primers according to the manufacturer's indications.
After 48h supernatants containing the viral particles were lter through a 0.45μm lter. Con uent BJ ER:BRAF cells were then transduced with 1/3 of the viral supernatant and 8 μg/μl polybrene. 24h posttransduction the cells were selected with 1 μg/ul puromycin. Cells were maintained in tetracycline-free tested serum (Clontech). For the expression of YBX1, 100ng/ml doxycycline was used for 72h unless otherwise indicated.
To generate the phosphomutant (S102A) and phosphomimic (S102D) versions of the protein, directed mutagenesis was performed using QuikChange II XL Site-Directed Mutagenesis Kit (Agilent) directly from the pLVX-YBX1-GFP construct. Primers with the respective mutations were designed according to the manufacturer's indications. The pLVX-RSK-GFP cell line was generated as described for YBX1, amplifying the coding sequence from cDNA with speci c primers. Primer sequences are listed in Supplementary   Table 4.

RNA extraction and qRT-PCR analysis
Total RNA was isolated using Trizol reagent (Invitrogen), treated with TURBO DNase (Ambion, Lifetechnologies) and reverse transcribed using TaqMan Reverse Transcription kit (Applied Biosystems) with random hexamer primers. Quantitative real-time PCRs were performed using Syber Green PCR Fast PCR Master Mix 2x (Applied Biosystems). The housekeeping genes HPRT1 and RPLP0 were used for normalization of qRT-PCR data, unless otherwise stated. Primer sequences are listed in Supplementary   Table 4.

Western blot analysis
Cells were seeded and reverse transfected in 6-well plates (NUNC). In case of senescence, cells were treated the following day with 1 µM 4-OHT. After 72h cells were harvested, washed once with PBS and the pellets lysed in RIPA buffer (150 nM NaCl, 0.1% sodium deoxycholate, 0.1% sodium dodecyl sulfate, 50 mm Tris-HCl (pH 8), 1 mM EDTA) containing protease inhibitors (Complete Mini Protease Inhibitor Cocktail; Roche Applied Science). Proteins were separated by electrophoresis in 4-12% NuPAGE Bis-Tris gels (Invitrogen) and transferred to nitrocellulose membranes (Amersham). Antibodies used are listed in Supplementary Table 5.

Conditioned media
To generate conditioned media the cells transfected and treated as indicated were growing in serum-free growth media for 24h. The media was ltered through a 0.45uM lter, centrifuged at 500g for 5min and placed on the corresponding recipient cells for the time indicated in each experiment.

Enzyme-linked immunosorbent assay (ELISA)
To measure the secretion of IL6 and CXCL1 the CM from different conditions were analysed by ELISA using commercially available kits (Thermo Fisher Scienti c).

Secretome analysis
The CM for proteomics analysis was harvested from 6 cm plates as indicated above. CM was concentrated using Centricon (Milipore) 3kDa lter, precipitated by TCA and washed in acetone. Protein pellets were resuspended in 6M GndHCl in 10mM Tris/HCl pH 8.0 with 2mM DTT and incubated at 56°C for 30min. In-solution digestion was performed after sample dilution with 50mM TEAB using 250ng of LysC (Wako) (at 2M GndHCl) and followed by 500ng of trypsin (Promega) (at 0.6M GndHCl). Reduced, alkylated and acidi ed peptides were desalted on 100ul C18 stage tips (Thermo) and subjected to LC-MS/MS analysis.
Tryptic peptides were identi ed by LC-MS using an EASY-nLC 1000 (ThermoFisher Scienti c) coupled to a Q Exactive HF (ThermoFisher Scienti c) equipped with a nanoelectrospray ion source. Peptides were separated on an in-house packed column of ReproSil-Pur C18-AQ, 3µm resin (Dr Maisch, GmbH) using a 90-minute gradient of solvent A (0.5% acetic acid) and solvent B (80% acetonitrile in 0.5% acetic acid) and a ow of 250 nL/min. The mass spectrometer was operated in positive ion mode with a top 12 datadependent acquisition, a resolution of 60,000 (at 400 m/z), a scan range of 300 -1700 m/z and an AGC target of 3e6 for the MS survey. MS/MS was performed at a scan range of 200-2000 m/z using a resolution of 30,000 (at 400 m/z), an AGC target of 1e5, an intensity threshold of 1.0e5 and an isolation window of 1.2 m/z. Further parameters included an exclusion time of 45 sec and a maximum injection time for survey and MS/MS of 15 ms and 45 ms respectively.
The raw les obtained from LC-MS were processed using the MaxQuant software 72 version 1.5.3.30.
Peak lists were searched against the human UniProt database using the Andromeda search engine incorporated in MaxQuant with a tolerance level of 7 ppm for MS and 20 ppm for MS/MS (Cox & Mann, 2008). Trypsin was chosen as digestion enzyme with max 2 missed cleavages allowed. Variable modi cations were set to methionine oxidation, protein N-terminal acetylation, deamidation of asparagine and glutamine. Carbamidomethylation of cysteine was set as xed modi cation and other parameters were kept as default.
Statistical analysis was conducted in R environment (https://www.r-project.org) using DEP (v. 1.8.0) bioconductor package for proteomics analysis, and visualizations were made using the ggplot2 R package (Wickham, 2016). Imputation of missing data was performed using MinProb setting based on minimal intensity values observed for each sample.

Crystal violet staining
Cells were seeded and reverse transfected in 6-well or 12-well plates (NUNC). 24, 48 and 72h after transfection or treatment cells were washed twice in PBS and xed with 10% formalin for 10 min and stained with 0.1% crystal violet solution for 30 min. Excess crystal violet stain was removed by several washes with water. The plates were allowed to dry and crystal violet was extracted by the addition of 10% acetic acid. The amount of crystal violet staining was quanti ed by measurement of the absorbance at 570 nm.
Senescence associated β-galactosidase staining Cells were seeded and transfected in 12-well plates (NUNC). At 72h post-transfection they were xed and stained using the β-galactosidase staining kit (Cell signaling) according to the manufacturer's protocol.
Invasion assay 50 ul of 0.5ug/ml Matrigel LDEV-Free (Corning)was added to the transwell. 400ul of the desired CM was placed at the bottom of the transwell with 8.0 μm pore size (Croning). 50.000 MDA/MB-231 cells were resuspended in 200ul DMEM without FBS and place on top of the transwell. After 48h cells across the matrigel membrane were washed in PBS and xed in cold 70% ethanol for 15 mins and stained using crystal violet staining as indicated above.

Cell fractionation
Cells were grown in 15 cm dishes (Nunc). Nuclear/cytoplasmic fractionation was performed using Nuclei EZ Lysis Buffer (Sigma) following the manufacturer's protocol.

Immuno uorescence
Cells were seeded on multichamber slides (Nunc) and transfected or treated. 72h cells were washed with PBS 1x and xed in 4% paraformaldehyde (PFA) (Sigma) for 15 min. The cells were permeabilized by incubation with 0.1% Triton X-100 in PBS (Sigma) for 10min and blocked for 30 min with 5% normal goat serum in PBS before incubation with the primary antibodies in 2% normal goat serum overnight. After intensive washing with PBS 1x, the secondary antibody was incubated in 2% goat serum for 1h at room temperature in the dark. The coverslips were washed, drained and mounted on glass slides using Prolong Gold Antifade Reagent with DAPI (ThermoFisher Scienti c). Images were taken using a Zeiss uorescence microscope.

Chromatin Immunoprecipitation
Cells were xed with 1% formaldehyde for 10 min. Crosslinking was arrested by adding glycine (0.125 M) for 5 min at room temperature. The cells were subsequently harvested in SDS lysis buffer (0.5% SDS, 100mM NaCl, 50mM Tris-Cl pH8.1, 5mM EDTA pH 8.0, protease inhibitor mixture Complete [Roche], and 1 mM phenylmethylsulfonyl uoride [PMSF]. Nuclei were pelleted and resuspended in IP buffer (2volumes SDS lysis buffer:1 volume Triton-X buffer [100mM Tris-Cl, pH 8.6, 100mM NaCl, 5mM EDTA pH 8.0, 5% Triton X-100]). The lysates were sonicated using BIORUPTOR sonicator for 12 cycles of 30sec and centrifuged at maximum speed. The sheared chromatin was diluted to 1ml with IP buffer and precleared with salmon sperm DNA/recombinant protein A-agarose (ThermoFisher Scienti c) for 2 h. 1% of the sample was used as the input control, and the remaining precleared chromatin was incubated overnight with 10 μg of antibody by incubation with salmon sperm DNA/protein-A agarose (50% slurry) and centrifugation. The bead pellets were washed in low or high salt conditions (0.1% SDS, 1% Triton X-100, 2 mM EDTA pH 8.0, 20 mM Tris-HCl pH 8.0, and 150 mM [low]/500 mM [high] NaCl). The beads were then washed once with LiCl buffer (0.25 M LiCl, 1% NP-40, 1% Na-deoxycholate, 1 mM EDTA, and 10 mM Tris-HCl pH 8.0) followed by two washes with Tris-EDTA buffer. Elution buffer (0.1% SDS, 0.1 M NaHCO 3 ) was added to the samples and the crosslinking was reverted by incubation at 68°C overnight. Samples were incubated 1h at 37°C with RNAse A (Sigma) and 45 min at 50°C with proteinase K (Ambion). The DNA was puri ed using Minelute PCR Puri cation Kit (Qiagen) and then ampli ed by qPCR. Primer sequences are listed in Supplementary Table 4. RNA sequencing and bioinformatic analysis RNA integrity was con rmed on Agilent 2100 Bioanalyzer using Agilent RNA 6000 Nano kit (Agilent Technologies). RNA-seq libraries were prepared from 2 µg total RNA with CATS mRNA-seq Kit (with polyA selection) v2 x24 (Diagenode) according to the manufacturer's protocol. Concentrations of the libraries were measured using the Qubit uorometer (Invitrogen) and fragment size was assessed on Agilent 2100 Bioanalyzer using Agilent High Sensitivity DNA kit (Agilent Technologies). The libraries were sequenced on Illumina NextSeq500 with 75 bp single-end. Raw reads were trimmed using Cutadapt (Martin, 2011) to remove adapters and minimum read length after trimming was set to 18. Trimmed reads were mapped to hg38 using STAR aligner (Dobin, Davis et al., 2013) (version 2.5.1a). Uniquely mapped reads were counted towards genes using featureCounts (Liao, Smyth et al., 2014) (version 1.5.1). Differential expression analysis was performed with DESeq2 (Love, Huber et al., 2014) using FDR < 0.01. Heatmaps were generated using pheatmap package in R with default settings and relative expression as RPM calculated by Z-score scaling. GO-term analysis was carried out using PANTHER (Thomas, Campbell et al., 2003) (http://pantherdb.org/about.jsp, version 14.1) using Fischer's exact test with FDR multiple test correction.
Only genes differentially expressed between Senescence and Senescence MIR31HG-KD and |log 2 fold-change| > 0.75 were used for GO-term analysis.

MIR31HG oligonucleotide-based pull down
Cells were grown in 150 and 500 cm2 dishes in 25 or 90ml media respectively up to ~90% con uency.
Supernatants were transferred to 2mL tubes and pre-heated to 65°C with 1200rpm shaking, then oligonucleotide-coated beads were added to lysates for 4h incubation at 65°C with shaking (250ul beads/0.5g initial cell pellet distributed in 2ml tubes). The beads were washed four times with RP buffer at room temperature and then washed with 1mL 50mM TEAB buffer to remove detergents. For mass spectrometry sample preparation the beads were resuspend in 100ul 50mM TEAB buffer (Sigma) including 2mM DTT. Trypsin was added (500ng) for O/N incubation at 37°C, then followed by treatments with DTT (10mM) and IAA (55mM) and acidi cation with TFA. 100ul StageTips (Thermo Fisher Scien c) were used for desalting and puri ed peptides were subjected to LC-MS analysis (as described in the above section).
Native RNA immunoprecipitation Cells were seeded in 15cm plates and treated with 1µM 4-OHT the day after for 48 hours. The cells were washed twice with PBS and centrifuged at 500g for 5 minutes. The pellet was resuspended in lysis buffer (50mM TRIS/HCl, pH 7.4, 100mM NaCl, 0.5% Triton X-100, 5mM EDTA, 0.25% Na-deoxycholate, Protease Inhibitor (Roche), RNase Inhibitor (NEB)). After 15 minutes centrifugation at 16000g the supernatant was collected in a new tube. 10% of the extract was saved as RNA input. The rest of the cell extract was incubated with 10ul of GFP-Trap ® Magnetic Agarose beads (Chromotek) rotating at 4°C for 2 hours. The beads were then collected using a magnetic rack and washed three times with lysis buffer. RNA was extracted adding 600ul of TRIzol reagent (Invitrogen, ThermoFisher) to the beads using manufacturers protocol.

Statistical analysis
Student's t-test was used for statistical analysis and performed using GraphPad Prism. Signi cance was determine at p < 0,05. Data are represented as mean mean ± s.d. Figure legends show the number of independent biological replicates (n).

Data availability
Raw data for the RNA-seq are deposited to GEO (GSE144752). Raw data for proteomics experiments are deposited to PRIDE ProteomeXchange (PXD017475). The data and reagents that support the ndings of this study are available from the corresponding author upon reasonable request. Source data is available for Figs  Vinculin. Molecular weight marker is shown in kDa (n>4). (E) BJ ER:BRAF cells expressing a doxycyclineinduced empty, wild type, mutant S102A or mutant S102D versions of YBX1 were transfected with a siRNA against YBX1. 24h later cells were transfected with luciferase reporter constructs pGL3-promoter (control) or pGL3-promoter-3'UTR (3'UTR) containing the 3'UTR of IL1A mRNA and treated with ethanol (control) or with 1 M 4-OHT for 48h. The graph shows the the luciferase values normalized to renilla and relative to the empty cell line in the absence of 4-OHT set as 1 (n=4). (F) RIP analysis using a GFP-tagged version of WT YBX1, S102A mutant or S102D mutant in formaldehyde crosslinked cells induced with doxycycline and treated with 1 M 4-OHT for 72h. The graph shows YBX1 binding to MIR31HG, IL1A and MALAT as negative control. The results are shown as percentage of input relative to an empty GFP cell line treated in the same conditions (n=4). All statistical signi cances were calculated using two-tailed Student t-tests, ***P< 0.001; **P < 0.01; *P < 0.05; ns, non-signi cant. All error bars represent mean ± s.d.