Characterization of novel markers of senescence and their prognostic potential in cancer

Cellular senescence is a terminal differentiation state that has been proposed to have a role in both tumour suppression and ageing. This view is supported by the fact that accumulation of senescent cells can be observed in response to oncogenic stress as well as a result of normal organismal ageing. Thus, identifying senescent cells in in vivo and in vitro has an important diagnostic and therapeutic potential. The molecular pathways involved in triggering and/or maintaining the senescent phenotype are not fully understood. As a consequence, the markers currently utilized to detect senescent cells are limited and lack specificity. In order to address this issue, we screened for plasma membrane-associated proteins that are preferentially expressed in senescent cells. We identified 107 proteins that could be potential markers of senescence and validated 10 of them (DEP1, NTAL, EBP50, STX4, VAMP3, ARMX3, B2MG, LANCL1, VPS26A and PLD3). We demonstrated that a combination of these proteins can be used to specifically recognize senescent cells in culture and in tissue samples and we developed a straightforward fluorescence-activated cell sorting-based detection approach using two of them (DEP1 and B2MG). Of note, we found that expression of several of these markers correlated with increased survival in different tumours, especially in breast cancer. Thus, our results could facilitate the study of senescence, define potential new effectors and modulators of this cellular mechanism and provide potential diagnostic and prognostic tools to be used clinically.

Apoptosis and senescence are the two main processes that prevent the emergence of transformed cells. 1 Senescence is usually defined as a permanent cell cycle arrest in which cells remain metabolically active and adopt characteristic phenotypic changes. 2 Senescent cells often appear multinucleated, large and extended, and exhibit spindle and vacuolization features. 3 The onset of this phenotype is believed to be either the result of telomere shortening after a number of cell divisions (replicative senescence) or a response to a diverse range of stress stimuli (stress-induced premature senescence). 3,4 Expression of oncogenes, such as Ras, cyclin E, E2F3 and Raf, can also trigger senescence, underscoring its tumoursuppressing properties. [5][6][7] Moreover, presence of senescent cells in vivo is often observed in the pre-malignant stages of a tumour; they gradually disappear, suggesting that the senescent barrier needs to be overcome in order to progress into full malignancy. [8][9][10] Cell senescence has also been associated with age-dependent organismal changes in rodents and primates, [11][12][13] and accumulation of senescent cells has been shown to contribute to the functional impairment of different organs. 14 This has led to the hypothesis that senescence is an antagonistically pleiotropic process, with beneficial effects in the early decades of life as a tumour suppressor but detrimental to fitness and survival in later stages, because of its contribution to age-related pathologies. 15 Despite the considerable knowledge accumulated in the 50 years since Leonard Hayflick first described the phenomenon, 16 the molecular pathways involved in senescence have not been fully characterized. 17 One of the wellknown features of both replicative and stress-induced senescence is the participation of the p53-p21 and/or p16-RB axis in the phenotype. Although in vivo suppression of p53 and/or its upstream regulator ARF is enough to prevent senescence in some models, 18 other cell types rely primarily on p16 for its induction. 19 The p53 target gene, p21, has often been considered critical for establishing senescence, whereas p16 may be more involved in the maintenance of the phenotype, 20 an effect also achieved by an increase in intracellular reactive oxygen species. 21,22 Cellular senescence is associated with the secretion of growth factors, chemokines and cytokines, collectively known as the senescence-associated secretory phenotype (SASP). SASP has an effect on cell proliferation and angiogenesis, as well as a possible role in promoting aging and tumourigenesis. 23,24 It can also promote migration of leukocytes and tumour cells, which in turn may induce tumour metastasis. 25 Increased expression of intracellular and/or secreted proteins, such as p21, p16, macroH2A, IL-6, phosphorylated p38 MAPK, PPP1A, Smurf2 or PGM, [26][27][28][29][30] has been used as a surrogate marker of senescence, although it does not provide a specific measurement. 18 Senescent cells display different modifications in the organization of chromatin that can help identify them as well. In normal cells, DNA staining reveals completely uniform colour outlines, whereas senescent cells usually show dot-like patterns, known as senescenceassociated heterochromatic foci (SAHF), which appear because of intensive remodelling in the chromatin and a lower susceptibility for digestion by nucleases. 31,32 SAHF development is not necessary for the establishment of senescence and its presence depends on cell type and the triggering stimuli. 33 Apart from these factors, the most distinctive measurable feature of senescent cells is the presence of a specific β-galactosidase enzymatic activity at pH 6.0, different from the normally observed at pH 4.0 within lysosomes. 34 This has been named senescence-associated β-galactosidase (SA-β-Gal) and it is thought to be a consequence of the enlargement in the structures of lysosome in senescent cells, without having a known role in the establishment or maintenance of the phenotype. 35 Although it is currently the standard for detecting senescent cells, several conditions, such as high cell confluence or treatment with hydrogen peroxide, can also stimulate SA-β-Gal activity, leading to many false positives. 36 In summary, none of the currently available markers are sufficient on their own for conclusively identifying senescent cells in vivo or in vitro, which underscores the need for better characterization tools. 30 Here, we describe and validate a list of novel senescentspecific proteins associated with the plasma membrane, uncovered through a proteomic screening, which define a profile that can easily be interrogated in a specific and quantitative manner using different techniques. We propose to use them as potential selective markers of senescence and we also anticipate that they may have a role as effectors and/or modulators, which would uncover novel pathways involved in the process. Moreover, we explored their prognostic potential and found a correlation between their expression and patient survival in different types of cancer, consistent of the role of senescence as an important tumour-suppressor mechanism.

Results
Proteomic analysis of the expression of proteins associated with the plasma membrane in senescent cells. In order to characterize the profile of proteins selectively associated with the cell membrane after the induction of senescence, we used a bladder cancer cell line, EJ, with a tetracycline (tet)-regulatable p21 or p16 expression system ( Figure 1a). These cells, named EJp21 and EJp16, respectively, 22,37 irreversibly senesce after prolonged expression of the induced protein ( Figure 1b and Supplementary Figure 1A). We isolated the membrane fraction from lysates of these cells (Figure 1c) and performed a mass spectrometry screen comparing the senescent cells with their non-induced counterparts. As shown in Figure 1d, 107 proteins were exclusively present in membranes of senescent EJp21 and 132 in EJp16. Seventeen were present in both senescent cells but in none of the controls. Among these proteins, DCR2, NOTCH3 and ICAM1 were detected, all of which had been previously reported to be increased in senescence. 10,38,39 This confirmed the suitability of the proteomics protocols used for our screen. We then selected 10 proteins from the analysis for further validation: DEP1, NTAL, EBP50, STX4, VAMP3, ARMCX3, B2MG, LANCL1, VPS26A and PLD3. They were chosen because none of them had been previously been associated with senescence and they were all predicted to be present on the plasma membrane using available algorithms (http://www.enzim.hu/ hmmtop/html/submit.html).
Validation of potential membrane markers of senescent cells. We next confirmed that the selected proteins were indeed expressed preferentially in the membranes of senescent cells. To this end, we used lysates with the cell membrane fraction from EJp16 and EJp21 induced to senesce. As shown in Figure 2a, basal levels of DEP1, NTAL, EBP50, STX4 VAMP3 and ARMCX3 were low in membrane lysates of uninduced EJp16 cells. After 5 days of p16 expression, when cells are known to be irreversibly senescent, 22 expression of these proteins was significantly increased, except for VAMP3, which only show minor induction (Figure 2a and Supplementary Figure 1B). DEP1 and NTAL were notably expressed in EJp21 in basal conditions, and were slightly upregulated after 5 days of p21 induction. EBP50, STX4 and ARMCX3 displayed low basal levels of expression followed by a substantial increase after EJp21 entered senescence. VAMP3 only showed a small increase in induced EJp21 cells. As shown in Figure 2b (and Supplementary Figure 1B), B2MG, VPS26A and LANCL1 and PLD3 were not induced significantly in any senescent model. Finally, DCR2 was shown to be induced in both p16-and p21-dependent senescence, as expected, 10,40 although its increase was much higher in EJp16. The results were similar using whole-cell lysates and none of the markers tested were present in the parental EJ cell line (Supplementary Figure 2A). All these results together confirmed that six of the potential markers (DEP1, NTAL, EBP50, STX4, VAMP3 and ARMCX3) were upregulated in senescent cells, although at different levels, and three more (B2MG, LANCL1 and VPS26A) were not significantly induced, according to western blots. There were also p21and p16-specific patterns of expression.
We further validated these results using fractionation by sucrose gradient of whole-cell lysates of senescent EJp16. Figure 3 shows that DEP1, NTAL, EBP50, STX4 and ARMCX3 colocalize in the same fraction as cell membrane markers Na/K ATPase and Calnexin. B2MG shows low levels of expression, consistent with Figure 2. This supports the localization of these proteins in the plasma membrane. We also used immunofluorescence microscopy to study expression of these proteins ( Figure 4). DEP1, NTAL, EBP50 and STX4 showed induction in senescent EJp16, similar to the positive control, DCR2. VAMP3 and ARMCX3 also showed upregulation, but at lower levels. In EJp21, all markers were significantly increased, except STX4, which only showed a moderate elevation, and EBP50. The expression of these proteins in IMR90 human fibroblasts was also measured, comparing early passage cells with those induced to senesce after serial passaging (see SA-β-Gal staining in Supplementary Figure 1C) or in normal diploid fibroblasts after ras-mediated oncogene-induced senescence. All the proteins tested showed low basal levels in dividing fibroblasts and increased expression in senescent ones ( Figure 4 and Supplementary Figure 2B), confirming that they could also be used as markers of replicative senescence in normal cells.
Defining a protocol for rapid detection of senescence cells by fluorescence-activated cell sorting (FACS) analysis. Using the information from the validation experiments described above, we chose two of the novel membrane proteins (DEP1 and B2MG) to define a simple and specific protocol using flow cytometry that would allow for the rapid detection of senescent cells in culture. DEP1 and B2MG were selected because they had large predicted extracellular epitopes recognized by commercially available fluorescent-tagged antibodies. As a positive control, we used NOTCH3, which fulfils the same requirements and it is already known to be induced in senescent cells. 38 Non-permeabilized cells were exposed to a mix of three fluorescently tagged antibodies and the fluorescence intensity of the sample was measured by a cytometer (see Materials and methods section for protocol details). The total time needed to measure the presence of senescent cells in cell cultures was under 2 h. As shown in Figure 5, there was a consistent two to threefold increase in the mean fluorescence intensity of all markers in EJp16 induced to senesce. We confirmed this result using another model of p21-induced senescence HT1080p21-9 (refs 41,42) (see SA-β-Gal staining in Supplementary Figure 1C), which showed a~3-fold increase of all three markers. Selective expression of these and other markers in HT1080p21-9 was also confirmed by western blot (Supplementary Figures 2C and D). Moreover, normal human diploid fibroblasts that entered replicative senescence after serial passaging also showed upregulation of the markers, although at lower levels ( Figure 5), consistent with a lower percentage of SA-β-Gal-positive cells (see Supplementary Figure 1C). Of note, a control staining with a fluorescently tagged actin antibody did not show any increase in expression after the induction of senescence in any of  Figure 3A). These results together confirm that validated membrane markers of senescence from our proteomic screen can be successfully used to determine the presence of senescent cells in samples and could provide a faster and more selective detection tool than those currently available.
Establishing the clinical relevance of the validated markers. We next expanded our in vitro results to tissue obtained from mouse models and human biopsies. Figure 6a shows that lung adenomas in V600E BRAF mutant mice, which have been shown to consist mostly of senescent cells, 5 are positive for DEP1, STX4 and B2MG expression, whereas they are only weakly positive for NTAL. Of note, the level of expression of these markers was comparable to that of p16, a commonly used senescent marker. Non-adenoma cells were negative for all markers (data not shown). Moreover, human naevi, which are rich in senescent melanocytes, 9 also showed positive staining for the same markers, especially DEP1 and STX4 (Figure 6b and Supplementary Figure 3B). STX4 also reacted with other cell types, thus showing a higher background than DEP1. This indicates that proteins in  All these data together suggest that our panel of markers could be used clinically to detect the presence of senescent cells in tissues and thus provide diagnostic and/or prognostic information for diseases such as cancer. To confirm this hypothesis, we used PPISURV, a novel online tool that correlates gene expression with survival rates in cancer patients using publicly available data. 43 As shown in Supplementary Table 1, high expression of our validated markers correlated with increased survival in glioma, liposarcoma, chronic lymphocytic leukaemia, colon, breast and lung cancers, among other gene expression omnibus (GEO) data sets. This is consistent with senescence being an important tumour-suppressor mechanism in vivo. 2 Of note, negative correlations were also observed, suggesting that the prognostic potential of some targets may be tumour specific. Indeed, breast cancer showed the strongest correlation with the expression of our markers, as all 10 were associated with increased patient survival in different data sets (Figure 7). Interestingly, two data sets of breast cancer showed a better prognostic associated with the combined increased expression of four to six of the markers together (Supplementary Table 2). This indicates that the panel of senescent markers that we describe here could be used as a prognostic tool in cancer and underscores the clinical relevance of our findings.

Discussion
Senescence is a well-defined cellular mechanism with a critical role in processes such as ageing 44 and cancer. 45 Despite having been studied for decades, the mechanisms involved in senescence are not fully understood. One of the features of senescent cells that had not been previously characterized was the profile of expression of proteins on their surface. Such proteins have the potential to be especially relevant for three reasons. First, they could contribute to define the interactions of these cells with the microenvironment and help explain how the mechanisms of senescent cell clearance work. This is important in the context of the tumour-suppressor functions of senescence, as well as its involvement in the symptoms associated with ageing. 46,47 Second, specific cell membrane proteins with extracellular epitopes would be useful to rapidly detect senescent cells. Given the fact that the current protocols for these analyses are not ideal, identifying extracellular epitopes of the senescent proteome could greatly improve this field of study. Finally, uncovering novel upregulated proteins could enhance our understanding of the processes that determine the senescent phenotype.
Using a proteomics approach, we identified an average of 935 proteins associated with the plasma membrane of either control or senescent EJp21 and EJp16 cells, with 107 being specific of the senescent cells. From this screen, we then selected for validation 10 proteins that were preferentially expressed in both senescent cells (and not in either of the  Figure 6 Expression of putative senescent markers in mouse and human tissues. Immunohistochemical staining of mouse (a) and human skin samples (b) with DEP1, NTAl, STX4 and B2MG antibodies. p16 is used as a known marker of senescence. Magnification: × 10 (mouse) and × 20 (human) controls) or highly expressed in one of them. Some, like the DEP1 phosphatase, has already been associated with tumour-suppressor mechanisms. 48 Others, such as STX4, VAMP3, VPS26A and PLD3, may all have a role in vesicle trafficking in the cell, [49][50][51][52][53][54] perhaps contributing this way to some aspects of the SASP. We are currently performing further experiments to determine whether any of these proteins actively participates in senescence or their expression is an epiphenomenon.
We studied the expression of these targets in different models, mainly two inducible EJ cell lines that undergo senescence through activation of only one of the main pathways involved in the process, those mediated by p16 or p21. The proteins were upregulated in at least one of the models, with some clearly induced in both. Moreover, the results were also validated in normal human fibroblasts, thus confirming the relevance of the data in both replicative and stress-induced pathways of senescence. Our data suggest that these 10 proteins have the potential to be used as markers of senescence, perhaps together with those previously described (p21, p16, p15, DCR2, NOTCH3, etc.). It is likely that their expression profile would differ between tissues and Figure 7 Correlation between senescent markers expression and survival in breast cancer. Kaplan-Meier survival curves of patients with breast cancer, segregated according to high (red) or low (green) expression of the genes from our panel of senescent markers, obtained from public databases through a bioinformatics analysis using PPISURV (www.bioprofiling.de). Each graph represents a different GEO data set A new profile of membrane proteins expressed in senescent cells M Althubiti et al depending of the triggering stimuli. For instance, EBP50 and STX4 are better induced in the p21 model, whereas DEP1, NTAL and ARMCX3 seem more specific for p16-induced senescence. Additional studies will be required to determine which combination of markers particularly defines senescent cells in each situation. This would greatly increase the specificity of any protocols to identify these cells in vitro and in vivo.
DEP1, NTAL, ARMCX3, LANCL1, B2MG, PLD3 and VPS26A have at least one predicted extracellular domain. This suggests that they could be detected with specific antibodies without the need to permeabilize cells. Using this information, we selected two of them, DEP1 and B2MG, to develop a proof of principle staining protocol that could help determine the amount of senescent cells present in a sample. The goal was to achieve higher specificity and shorter experimental times than the current standard, the SA-β-Gal staining assay, which has many false positives and it is not proportional to the intensity of the arrest. We believe that our results show that a fast detection method based on specific antibodies against extracellular epitopes is feasible. As mentioned above, further optimization will be needed to decide the best targets and conditions for different cell types and tissues. Increasing the simultaneous number of markers detected could also augment the specificity of the protocol, if needed. Moreover, markers specific to either the p16 or p21 pathways could help determine which of the two is preferentially activated in response to each senescence-inducing stimulus.
As senescence stops the progression of cancer in vivo 2 and it is known to be increased in response to many therapies, 45 the presence of senescent cells in tumours could be considered an indication of a controlled or less aggressive/ advanced disease. Thus, we reasoned that our proteins could have a utility as prognostic tools in different types of cancer. We demonstrated this using a bioinformatics approach. We assessed a clinical application of the validated markers uncovering a positive correlation between their expression and increased survival in several malignancies. This shows that the characterization of novel markers of senescence has not only an experimental relevance in the lab but also a potential impact at the bedside. Indeed, our results suggest that the detection of senescent cells in cancer samples using one or more of our markers could be used to predict survival in breast cancer, and perhaps also in other types of tumours.
In summary, our results provide new information regarding the mechanisms involved in senescence, and we showed that this can be used experimentally to rapidly detect senescent cells, with important clinical implications. The exact role of these markers in the senescent pathways will be investigated in the future, thus contributing to our better understanding of this intricate cellular process. Such information could be important to define new therapeutic interventions that could increase the positive impact of senescence on human health and/or diminish its negative effects.

Materials and Methods
Cell culture. EJp21 were maintained in DMEM supplemented with 10% foetal bovin serum (FBS, Gibco, Paisley, UK), penicillin-streptomycin (50 unit/ml), hygromycin (100 μg/ml) and genticin (750 μg/ml). EJp16 cells were maintained in DMEM supplemented with 10% FBS, penicillin-streptomycin (50 unit/ml), hygromycin (100 μg/ml) and puromycin (2 μg/ml). In order to inhibit p21 or p16 expression, tet was added to the medium every 3 days to a final concentration of 1 μg/ml. To induce p21 and p16 expression, cells were washed three times and seeded directly in culture medium in the absence of tet. 37 IMR90 (human fibroblasts derived from lungs of a 16-week female foetus) and normal human diploid fibroblasts (Cellworks, San Jose, CA, USA) were maintained in DMEM supplemented with 10% FBS, and penicillinstreptomycin (50 unit/ml) until they reached replicative senescence. HT1080p21 were maintained in in DMEM supplemented with 10% FBS and penicillinstreptomycin (50 unit/ml). To induce p21 expression, 100 μM isopropyl β-D-1thiogalactopyranoside was added to the medium. To induce ras expression, cells were infected with a retroviral construct containing ras (gift of Stuart A Aaronson, Mount Sinai School of Medicine, New York, NY, USA).
Plasma membrane protein extraction. This protocol was performed according to the Abcam Plasma Membrane Protein Extraction Kit (ab65400; Abcam, Cambridge, UK).
SDS-PAGE separation, extraction and analysis of proteins from gel lanes by data-independent LC/MSE mass spectrometry. Senescent and growing EJp21 and EJp16 plasma membrane samples were separated by 10% SDS-PAGE. After staining with the Coomassie blue, the gel was cut to obtain separate sample lanes. Gel lanes were cut sequentially into slices of approximately 1.5 mm and transferred to a 96-well low binding PCR plate. Each slice was destained, digested with trypsin and peptides extracted for Mass Spectrometry analysis as previously described. 55 Nanoscale LC was used to separate the complex peptide mixtures using a Waters nanoACQUITY UPLC (Waters, Manchester UK). Chromatography was performed using a 50 min reversed-phase gradient (formic acid (0.1%)/acetonitrile) and a 75 μm × 25 cm C18 column (Waters, BE130) operated at 300 nl/min. Mass spectrometry analysis was performed using a SYNAPT G2S (Waters) operated in a data-independent (MSE) manner. The selected analysis mode enabled precursor and fragment ions from the tryptic digest to be analysed simultaneously. The data acquired were processed and searched using ProteinLynx Global Server (Waters) and visualized and reanalyzed using Scaffold (Proteome Software, Portland, OR, USA).
SA-β-Gal staining. Cells were washed three times with PBS and fixed with 4% formaldehyde for 5 min at room temperature, then stained as previously described. 34 Immunoblot analysis. In all, 1 μg/ml Protease Inhibitor Cocktail Set III (Calbiochem, Billerica, MA, USA) was added to cell lysates. Protein concentrations were determined using Bradford protein assay (Fermentas, Thermo Scientific, Waltham, MA, USA). Twenty microgram of total protein per sample was subjected to 10% or 6% SDS-PAGE and transferred to Immobilon-P membranes (Millipore, Billerica, MA, USA). An ECL detection system (Thermo Scientific) was used to visualize the results. Alternatively, an Odyssey CLx Infrared Imaging System (Li-COR, Lincoln, NE, USA) was used. See Supplementary Table 1 for antibodies used.
Immunofluorescence. Cells were split into six-well plates containing sterile coverslips. After 24 h, media was aspirated from the plates and cells were washed three times with PBS. Cells were fixed using 1 ml of 4% paraformaldehyde for 30 min with gentle shaking. After fixing, cells were washed three times with PBS and permeabilized with 1 ml 0.1% Triton X-100 for 10 min. Cells were then washed three times with PBS and blocked with 1% BSA for 30 min. Coverslips were incubated with 100 μl 1 : 100 primary antibody overnight at 4°C. The following day, coverslips were washed three times with PBS and incubated with 100 μl secondary anti-rabbit or antimouse antibodies (Alexa Fluor 488 and 594, Invitrogen, Paisley, UK) for 45 min in the dark. After incubation, coverslips were washed three times with PBS and stained with 4′,6-diamidino-2-phenylindole, dihydrochloride (DAPI, Invitrogen) for 10 min. Slides were labelled and the coverslips were mounted and sealed with transparent nail varnish. Slides were analysed using a Nokia TE300 semi-automatic microscope (Nokia, Keilaniemi, Finland). See Supplementary Table 1 for antibodies used.
Immunohistochemistry. Lung adenoma (from a conditional V600E BRAFknock-in mouse model) 5 and human naevi (from clinical samples obtained by GSS) were fixed, paraffin-embedded, sectioned and stained with haematoxylin and eosin A new profile of membrane proteins expressed in senescent cells M Althubiti et al following standard protocols. Tissue immunostaining was performed as previously described. 56 See Supplementary Table 1 for antibodies used.
FACS analysis of senescent-associated cell surface proteins.
Plates were washed with cold PBS and cells were collected by gently scraping them in 0.5 ml cold PBS, and then kept on ice. The use of trypsin was avoided to prevent internalization of extracellular proteins. Cells were centrifuged (200 g for 5 min at 4°C) and the supernatant discarded. Cells were then resuspended in 200 μl of blocking buffer (0.5% BSA in PBS) and incubated 15 min on ice, then transferred to 96 rounded bottom multi-well plate. These were centrifuged (500 g for 5 min at 4°C) and the supernatant was discarded. Cells were resuspended with a mix of the required antibodies (see Supplementary Table 1 for antibodies used), appropriately diluted, and incubated at 4°C in the dark for 30-45 min. Cells were next washed twice with blocking buffer (150 μl per well) and centrifuged for 500 g for 5 min at 4°C. The cell pellet was then resuspended in 300-500 μl of blocking buffer and fluorescence was read by a flow cytometer.