Induction, regulation and roles of neural adhesion molecule L1CAM in cellular senescence

Aging involves tissue accumulation of senescent cells (SC) whose elimination through senolytic approaches may evoke organismal rejuvenation. SC also contribute to aging-associated pathologies including cancer, hence it is imperative to better identify and target SC. Here, we aimed to identify new cell-surface proteins differentially expressed on human SC. Besides previously reported proteins enriched on SC, we identified 78 proteins enriched and 73 proteins underrepresented in replicatively senescent BJ fibroblasts, including L1CAM, whose expression is normally restricted to the neural system and kidneys. L1CAM was: 1) induced in premature forms of cellular senescence triggered chemically and by gamma-radiation, but not in Ras-induced senescence; 2) induced upon inhibition of cyclin-dependent kinases by p16INK4a; 3) induced by TGFbeta and suppressed by RAS/MAPK(Erk) signaling (the latter explaining the lack of L1CAM induction in RAS-induced senescence); and 4) induced upon downregulation of growth-associated gene ANT2, growth in low-glucose medium or inhibition of the mevalonate pathway. These data indicate that L1CAM is controlled by a number of cell growth- and metabolism-related pathways during SC development. Functionally, SC with enhanced surface L1CAM showed increased adhesion to extracellular matrix and migrated faster. Our results provide mechanistic insights into senescence of human cells, with implications for future senolytic strategies.


INTRODUCTION
Human population is aging rapidly, raising serious global health and societal concerns, at the same time highlighting the urgent need to better understand the process of aging at the molecular, cellular and organismal levels. Recent research has revealed agingassociated accumulation in diverse tissues of the so-AGING called senescent cells (SC) which, along with their secreted products, appear to causally contribute to aging [1,2]. From a broader perspective, the process of senescence is a cellular stress response leading to persistent activation of cell-cycle checkpoints and longlasting cell-cycle arrest caused predominantly by lasting DNA damage signaling. SC play important pathophysiological roles in a number of processes during ontogenetic development of the mammalian organism [3]; for instance, senescence response takes part in wound healing [4] and tissue regeneration [5]. As a cellular fate commonly triggered by DNA damage response, cellular senescence is considered as a primary tumorigenesis barrier [6][7][8]. DNA damage-induced cellular senescence also accompanies genotoxic cancer therapies [9,10]. Long-lasting persistence of senescent cells in tissues can also induce various pathological changes affecting organismal homeostasis (see, e.g. references [1,2]). These effects are thought to be elicited by secretion from SC of a complex mixture of factors comprising components of the extracellular matrix, its modifiers such as proteases and their regulators, reactive oxygen species [11] and particularly cytokines including morphogens from the TGFβ family, pro-inflammatory species (e.g., IL1, IL6, IL8; [12]), and chemokines (such as MCP-1; [13]). The composition of the SC secretome (denoted as senescence-associated secretory phenotype; SASP) varies according to the cell type and mechanism of senescence induction [14]. Additional factors influencing the secretory response of SC, such as interactions of SC with surrounding cells and specific tissue microenvironment, likely exist at the organismal level. This complexity and variability of SASP is the reason why the effects of SASP on the tissue microenvironment and the pathogenic role of various types of SC remain poorly understood. Nevertheless, it was plausibly demonstrated that senescent cells can affect neighboring cells by induction of oxidative stress and DNA damage resulting in secondary ('bystander') senescence [11,15,16]. This 'domino' effect might contribute to the spread of a mild but chronic inflammatory environment especially in aging tissues [17]. It should also be noted that certain chemotherapeutic drugs are able to induce cellular senescence in normal tissues within the close proximity of tumors, which can also promote local inflammation and self-sustaining genotoxic environment.
Of the various interactions of SC with their milieu, the interaction with the immune system is especially important. SC interact with the immune system in a mutual way -besides being a target of the immune system (senescence immunosurveillance; [18]), SC can modulate the immune system function by either immunoactivation [18] or immunosuppression [19,20].
Intriguingly, observations that genetically engineered removal of SC from the mouse organism induced a 'rejuvenating' effect [2] and the presence of SC affected the rate of mouse aging [1] both indicate agingpromoting harmful effects of SC in vivo. Indeed, there is accumulating evidence that SC contribute to pathogenesis of several aging-associated degenerative diseases including atherosclerosis [21], cartilage degeneration leading to osteoarthritis [22,23], cardiac fibrosis [24], liver fibrosis [25], type 2 diabetes [26], and Alzheimer's and Parkinson's diseases [27,28]. In addition, elimination of senescent cells from the tumor environment is believed to improve the health condition of patients with cancer [29].
Based on the negative effects of SC in tissues, there is a prevailing view that prevention of SC accumulation or support of their elimination would be beneficial for the health span of the organism. The prerequisite for this ambition is our ability to unambiguously detect SC in tissues. However, reliable biomarkers strictly specific to SC are still missing, despite the SC undergo profound changes of their phenotype deviating extensively from their parental cells [30]. The purpose of this study was to identify candidate protein markers differentially expressed on the cell surface of proliferating versus replicatively senescent human fibroblasts using quantitative proteomic analysis. Apart from several previously reported proteins such as PVR, TIMP3, GGT2, and ADAMTS1, we identified the neural adhesion factor L1CAM as being overrepresented on the surface of the replicatively SC. Further analysis of L1CAM whose expression is physiologically restricted to neural and kidney tissues revealed that except for oncogenic Ras-induced senescence, other prosenescence stimuli can induce L1CAM. Also intrigued by the known association of L1CAM with cancer and poor prognosis of cancer patients [31][32][33][34][35][36][37], we then pinpointed several cellular signaling and metabolic pathways involved in regulation to L1CAM expression, as well as functional impact of elevated L1CAM expression, using several human models of SC. Both the overall spectrum of the cell-surface proteins identified in our unbiased proteomic screen, and the insights into L1CAM regulation and potential roles in senescence are presented in this dataset.

L1CAM is enriched on the cell surface of replicatively senescent fibroblasts
In our search to identify proteins differently expressed on the plasma membrane of senescent cells, we performed quantitative proteomic analysis of cell surface proteins of proliferating (PD 28) and replicatively AGING senescent (PD 84) BJ fibroblasts (see Supplementary Figure 1A, B for activity of senescence-associated βgalactosidase and markers of persistent DNA damage response, respectively). For enrichment of cell surface proteins we utilized biotin labeling of cell surface proteins (see Materials and Methods for details). Out of 650 quantified proteins identified by mass spectrometry, 151 proteins showed differential expression (Supplementary Table 1). Within this group we focused on the candidate proteins whose expression on the surface of senescent fibroblasts was increased more than four fold over the level in proliferating counterpart cells. In addition to several proteins already reported as overexpressed in senescent cells, such as metallopro-teinase inhibitor 3 (TIMP3) [38], γ-glutamyltranspeptidase 2 (GGT2) [39] and ADAMTS1 [40] , our  analysis revealed some previously unrecognized  proteins such as THY1, BST1, FAT1 (for full list of  differentially changed proteins, see Supplementary  Table 1) and neural cell adhesion molecule L1, L1CAM ( Figure 1A). Thirty six most upregulated proteins were selected and analyzed for mRNA levels, of which more than dozen was significantly elevated (see Supplementary Figure 1C). The changes of total protein level of five selected proteins (fibronectin, L1CAM, collagen IV, integrin α2 and PVR) in replicative senescent BJ cells were verified by immunoblotting (see Supplementary Figure 1D). L1CAM attracted our attention as it is expressed physiologically in neural and renal tissues and pathologically in several types of human tumors (see Discussion for further details). To validate the mass spectrometry data, we compared the L1CAM mRNA, total and surface protein expression in young/proliferating versus senescent BJ cells. As shown by immunoblotting of biotin/streptavidin immunoprecipitates, the total level of L1CAM was increased in replicatively senescent BJ cells compared to the proliferating cells (compare lanes 'L' for proliferating and replicatively senescent cells in Figure 1B). The increased cell surface protein level in replicatively senescent cells was verified by comparing the amount of eluted biotinylated L1CAM from streptavidin matrix (see lines 'E' in Figure  1B). This was further confirmed by flow cytometry analysis using live cell staining with L1CAM antibody, where the surface protein level of L1CAM was markedly elevated in replicatively senescent BJ cells ( Figure 1C). Note, similar cell surface enhancement of L1CAM in replicative senescent BJ cells was observed also for fibronectin and PVR (Supplementary Figure  1E). The L1CAM protein elevation was concordant with the changes of the L1CAM mRNA levels ( Figure  1D and Supplementary Figure 1C, D). Using indirect immunofluorescence, we observed marked heterogeneity of L1CAM cell surface levels among individual cells in both populations of proliferating and senescent BJ cells ( Figure 1E; for validation of antibody specificity, see Supplementary Figure 1F, G; note that the L1CAM antibody staining of formaldehyde-fixed cells was nonspecific).
Altogether, our data show that L1CAM is overrepresented on the cell surface of replicatively senescent BJ fibroblasts; and L1CAM's enhanced surface protein level corresponds to the increased expression of L1CAM mRNA.

L1CAM expression is cell type-and senescence stimulus-dependent
During serial cultivation of cells in vitro there is a likelihood of selection of clones with enhanced replicative potential [41]. To examine whether increased L1CAM expression in replicatively senescent AGING BJ cells was due to clonal selection of cells bearing higher L1CAM expression, we followed the expression of L1CAM in a scenario of prematurely induced senescence in BJ cells triggered by ionizing radiation (IR) [42], 5-bromo-2'-deoxyuridine (BrdU) [43], and interferon-γ (IFNγ) [16,44], or overexpression of oncogenic H-Ras(V12) [46]. With the exception of H-Rasinduced senescence, the cell surface expression of L1CAM was increased in BJ fibroblasts upon exposure to all other stimuli (Figure 2A, B; see Supplementary Figure 1A for SA-β-gal staining), indicating that cell surface expression of L1CAM is not the result of a clonal selection during serial passaging. The lack of L1CAM induction in H-Ras oncogene-induced senescence suggested the dependence of L1CAM expression on the type of senescence-inducing stimulus. Moreover, we observed that the L1CAM transcript level remained unchanged after BrdU treatment despite enhanced L1CAM cell surface expression ( Figure 2C), indicating that both de novo synthesis and/or enhanced (re)localization of L1CAM to the cell surface can take part in a mechanism of its enhanced cell surface expression. The heterogeneity of L1CAM expression in the population of SC was apparent among prematurely senescent cells as well.
To determine whether the heterogeneous expression of L1CAM in senescent cells stems from clonal heterogeneity present already in proliferating BJ cells, we sorted proliferating BJ cells according to their surface L1CAM level by FACS to populations with low (L1CAM low ) and high (L1CAM high ) expression (Supplementary Figure 2A) and followed the L1CAM levels for several population doublings. Notably, the differences in L1CAM levels between the sorted subpopulations balanced out after approximately ten population doublings (Supplementary Figure 2B) indicating that epigenetic rather than genetic factors likely determine the L1CAM heterogeneity. No differences in proliferation of L1CAM 'high' versus 'low' cells were observed (Supplementary Figure 2C), consistent with the notion that L1CAM expression is not linked with proliferation advantage of any subpopulation. In addition, there were no significant differences in the occurrence of DNA damage foci (detected as 53BP1 and serine 139 phosphorylated histone H2A.X) between L1CAM high and L1CAM low cells (see Supplementary Figure 2 D and E). We also did not observe any marked morphological differences among senescent L1CAM high and L1CAM low populations of BJ cells, except that L1CAM high cells were slightly larger in size (mean area ± SEM: 327.6 ± 5.317 μm 2 for L1CAM low and 344.9 ± 4.113 μm 2 for L1CAM high ; two-tailed paired test, p = 0.0123; Supplementary Figure 2F). Furthermore, downregulation of L1CAM in replicatively senescent cells did not result in escape of cells from senescence (Supplementary Figure 2G), indicating that L1CAM is not involved in senescent cell cycle arrest.
Next we tested a panel of other human cell types, including normal and cancerous cells, for L1CAM expression during the development of premature senescence. Figure 3 summarizes L1CAM mRNA, total and cell surface protein levels in a variety of cell types brought to senescence by IR or BrdU (Supplementary Figure 3A). In accord with a previous study [47], pancreatic carcinoma PANC-1 cells featured high L1CAM cell surface expression even under unperturbed conditions (see Supplementary Figure 3B for the relative transcript level of L1CAM in all cell types analyzed). IR but not BrdU induced both L1CAM transcript level and cell surface expression in PANC-1 and similarly in osteosarcoma U2OS cells ( Figure 3A, B). In contrast, BrdU induced L1CAM expression in prostate cancer PC3 and melanoma A375 cells more potently than IR. Normal diploid fibroblasts MRC-5 and HSF-1, and immortalized retinal pigment epithelial RPE-1 cells responded by elevation of L1CAM mRNA and total protein levels in dependence on the stimulus but without concomitant protein expression on the cell surface, indicating again that cell surface exposure of L1CAM is regulated. Note that the expression of L1CAM is the lowest in RPE-1 among the tested cell lines (Supplementary Figure 3B); therefore, it is possible that the observed elevation of the L1CAM level after senescence induction was not sufficient for detection of the surface protein level. The transcript levels of L1CAM were only slightly induced in prostate cancer DU145 cells exposed to either IR or BrdU; and the L1CAM protein remained undetectable ( Figure 3C). In general, our data showed that higher levels of L1CAM transcriptional induction appear to be necessary for the increase of the total protein level; however, even an increased total protein level is not always expressed as the enhanced presence of protein on the cell surface.
To investigate whether the L1CAM is also induced during senescence in different species, we triggered premature senescence in three mouse cell lines TRAMPC2, TC1 and B16F10 by exposure to docetaxel (7.5 μM, 4 days) as reported previously [48,49]. In all three cell lines, L1CAM transcript levels were elevated after docetaxel treatment (see Supplementary Figure 4A for L1CAM level and Supplementary Figure 4B for senescence-associated beta-galactosidase staining) indicating that L1CAM induction during senescence is not restricted to human cells.
To conclude, the senescence-associated expression of L1CAM on the cell surface is feature shared in normal and cancerous cells of human and mouse origin, but AGING dependent on cell type and senescence-inducing stimulus. Our findings also indicate that L1CAM cell surface expression is a complex process regulated both at the level of transcription and posttranscriptionally.

L1CAM expression is a downstream event linked to inhibition of cyclin-dependent kinases by p16ink4a
Next, to decipher the mechanisms mediating L1CAM expression during induction of senescence, we first investigated to which phase of the cell cycle checkpoint cascade the expression of L1CAM is linked. To uncouple the inhibition of cell cycle from upstream events of DNA damage response we utilized ectopic expression of tetracycline-regulatable inhibitor of cyclin-dependent kinases p16 ink4a (p16) in human mesothelioma H28 cells in comparison to ectopic expression of p21 waf1 (p21) known to induce DNA damage response [50]. Doxycycline induction of both p16 and p21 in two single cell-derived clones (p16) and one single-derived p21 clone led to cell cycle arrest (Supplementary Figure 5A; for senescence-associated β-galactosidase staining, see Supplementary Figure 5B), as observed previously [51]. In contrast to p21, no signs of enhanced DNA damage response detected as 53BP1/γH2AX foci were observed after p16 induction when compared to non-induced cells (Supplementary Figure 5C). A reproducible increase of L1CAM mRNA ( Figure 4A) and protein ( Figure 4B) was observed after induction of both p16 ( Figure 4C) and p21 (not shown), however, without detectable increase of the L1CAM cell surface protein level. These findings indicated that the induction of L1CAM expression is a downstream event following induction of cdk protein inhibitor(s).

Cell type-dependent and mutual interaction of L1CAM and Erk signaling pathways
It has been reported that activation of the L1CAM signaling pathway by antibody-mediated L1CAM crosslinking or manipulation with its level affect the activity of Erk1/2 [52][53][54]. Indeed, we observed that the basal activity of Erk1/2 was higher in L1CAM low human melanoma A375 and BJ cells when compared to L1CAM high cells ( Figure 5A). Additionally, knockdown of L1CAM in BJ cells increased Erk1/2 phosphorylation ( Figure 5B). This data obtained in unperturbed cell culture conditions underscored the role of L1CAM in the regulation of activity of Erk1/2.
The other way around, inhibition of Erk1/2 activation by selumetinib, a chemical inhibitor of MEK, in A375 ( Figure 5C) and HeLa (data not shown) cells resulted in marked elevation of L1CAM mRNA, total and cell surface protein expression ( Figure 5D, E), which indicated that the MAPK pathway may operate in a negative feedback loop to suppress L1CAM, thus balancing its own activity. Further, combined knockdown of Erk1 and/or Erk2 by RNA interference resulted in elevation of L1CAM mRNA ( Figure 5F) and surface protein expression ( Figure 5G), further supporting the interplay between L1CAM and Erk pathways. Nevertheless, the exposure of BJ cells to selumetinib did not show any effect on L1CAM mRNA and protein levels (not shown), indicating cell typespecific regulation of the L1CAM expression.
These findings fit well with the absence of L1CAM expression during H-RAS oncogene-induced senescence (OIS). To exclude the possibility that our strain of BJ fibroblasts with tetracycline-inducible RAS oncogene lost the competence to induce L1CAM (due to cloning of cells with aberrant L1CAM regulation), we exposed Ras-induced senescent BJ cells presorted for low level of L1CAM to MEK inhibitor selumetinib. As shown in Figure 5J, the inhibition of MEK resulted in an increased L1CAM protein level, indicating unperturbed sensitivity of L1CAM to Erk1/2 inhibition in OIS cells, supporting the suppressive role of the Ras/MAPK pathway on L1CAM expression. AGING Overall, our results indicate a crosstalk between L1CAM and Ras/MAPK pathways, where the elevated level of L1CAM is associated with lower Erk1/2 activity and the Ras/MAPK signaling exerts a suppressive effect on L1CAM gene expression.

L1CAM expression is linked to metabolic changes
Given that L1CAM transcription is regulated by hypoxia in tumor cells [55], we asked whether metabolic changes associated with senescence (see e.g. references [56][57][58]) contribute to regulation of L1CAM gene expression. We noted that the concentration of glucose in culture medium affected the expression of L1CAM in U2OS and HeLa cells. Both cell types cultured in higher concentration of glucose (4.5 g/L) expressed lower levels of L1CAM compared to cells cultured in 1 g/L glucose, indicating that the expression of L1CAM might be linked to glycolytic energy metabolism of tumor cells ( Figure 6A, B). Furthermore, the level of L1CAM transcript and protein increased after inhibition of the mevalonate pathway (chosen here for being one of the pathways that utilize acetyl-CoA, derived from glucose metabolism) in BJ fibroblasts by cerivastatin ( Figure  6C). As reported previously, various forms of cellular senescence are accompanied by repression of solute carrier family 25 member 5 (SCL25A5) coding for ADP/ATP translocase 2 (ANT2; [59,60]), which has been implicated in glycolytic metabolism of tumor cells [61][62][63]. RNA interference-mediated downregulation of ANT2 but not ANT3 (Supplementary Figure 6A) in HeLa and BJ cells led to marked elevation of L1CAM both at mRNA and protein levels ( Figure 6E). The intrapopulation heterogeneity of L1CAM expression AGING was diminished after ANT2 knockdown (data not shown). Interestingly, the inhibition of MEK by selumetinib in A375 cells resulted in suppression of the ANT2 mRNA level associated with the increase of L1CAM expression (see Figure 5C and Figure 6F), suggesting a reciprocal link between ANT2 and L1CAM expression. In support of this notion, there was also an inverse relationship between the mRNA levels of both genes in A375 cells exposed to IFNγ and TGFβ ( Figure 6G), the cytokines reported to induce senescence [16,60]. Note that TGFβ alone strongly induced cell surface expression of L1CAM in A375 cells ( Figure 6H), which was associated with stronger cytostatic effect compared to IFNγ (Supplementary Figure 6B).
Altogether our data indicate that the induction of L1CAM during senescence might be linked to accompanying metabolic changes, specifically to suppression of the SCL25A5/ANT2 gene expression.

L1CAM increases cell migration and adhesion of both proliferating and senescent BJ cells
Several studies reported the role of L1CAM in enhanced migration of tumor cells (see, e.g., references [31,64]). Indeed, in a wound healing assay, young BJ cells with downregulated L1CAM using lentivirally transduced short hairpin RNA recovered the disrupted area significantly slower than control cells ( Figure 7A), indicating that migration of normal cells can also be affected by L1CAM surface expression (see reference [54]). Next we asked whether enhanced expression of L1CAM is linked with the migratory properties of senescent cells. Utilizing heterogeneity of L1CAM expression in the cell population of BJ senescent cells, we tested migration of BJ cells sorted by FACS according to L1CAM cell surface expression. Replicatively senescent BJ cells sorted for higher surface L1CAM levels healed the scratch more efficiently than their counterparts with lower cell surface expression of AGING L1CAM ( Figure 7B). To further corroborate these findings, we employed a 3D migration assay to follow cell motility induced by fetal bovine serum as attractant.
Using time-lapse cell tracking, several parameters such as velocity, Euclidean distance (length of migration that the cell overcomes directly from the start point to the endpoint), and accumulated distance (cell's trajectory) were evaluated. Notably, senescent BJ cells with high L1CAM cell surface expression migrated faster and covered longer Euclidean and accumulated distances AGING ( Figure 7C). To exclude the possibility of an impact of growth conditions on the tested subpopulations, a 3D random migration assay of non-sorted but L1CAM antibody-labeled senescent BJ fibroblasts was also performed. The results revealed a similar trend as in the case of L1CAM-sorted BJ cells. L1CAM high BJ cells migrated faster and travelled further distances ( Figure  7D) compared to L1CAM low BJ cells. The migration of proliferating BJ fibroblasts in the wound healing assay was faster compared to replicatively senescent cells, where young BJ covered the disrupted area approximately three times faster (by 16 hours) compared to senescent cells (48 hours; see references [65,66]). The proliferation of young cells contributed to faster wound recovering as well; however, the difference in the velocity was significant ( Supplementary Video 1 and 2).
Next we asked whether the cell surface level of L1CAM correlates with the cell adhesive properties. As shown in Figure 7E, replicatively senescent L1CAM high BJ cells showed 10, 17 and 20% higher adhesion to the components of extracellular matrix laminin, fibronectin, and collagen 1, respectively, compared to senescent L1CAM low BJ cells. To confirm the contribution of L1CAM to cell adhesive properties, L1CAM was downregulated in senescent BJ fibroblasts by RNA interference (Figure 7F; for the level of L1CAM downregulation, see Supplementary Figure 6C). Indeed, control cells adhered more efficiently compared to the cells with a decreased L1CAM level. Taken together, the enhanced level of L1CAM correlates with increased migratory properties not only of proliferating but also senescent BJ cells. The expression of L1CAM affects the adhesion of BJ cells to the components of extracellular matrix.

DISCUSSION
Despite that cellular senescence is a phenomenon well characterized and recognizable in vitro and the detection of SC in tissues is now less problematic due to the recently published robust method for their detection [67], there is a lack of strictly specific markers that would unambiguously allow targeting of SC for therapeutic purposes [68]. In search of cell-surface determinants overrepresented on senescent cells, we identified adhesion molecule L1CAM as being enhanced on replicatively senescent BJ fibroblasts. Further analysis of L1CAM expression in different senescence-inducing scenarios revealed that L1CAM expression depends on cell type and senescenceinducing stimulus, it shows marked intra-population heterogeneity and multiple levels of control, including regulation of the cell surface exposure. Our data also show that the L1CAM signaling pathway is interlinked with Erk signaling in a reciprocal manner. Stimuli changing cellular metabolism such as glucose levels, inhibition of the mevalonate pathway, and suppression of ANT2 (SCL325A5) all affected the level of L1CAM expression, suggesting that L1CAM gene expression might be controlled by the metabolic changes accompanying the development of cellular senescence. The L1CAM level correlated with increased migratory properties of both proliferating and senescent BJ cells and with enhanced adhesion of cells to the components of extracellular matrix.
Our proteomic comparative analysis of cell surface proteins of proliferating and senescent BJ cells based on surface protein biotinylation revealed more than 70 proteins significantly enhanced on the plasma membrane of senescent cells. In previous studies using different methodological approaches, several of these proteins, such as tissue inhibitor of metalloproteinase 3 (TIMP3) [38], γ-glutamyltranspeptidase 2 (GGT2) [39] and ADAMTS1 [40] have already been reported as being overrepresented in senescent cells, thereby validating the technique employed in this study.
The L1CAM/L1 cell adhesive molecule is a transmembrane glycoprotein originally discovered as a cell adhesion molecule in the nervous system [69], where it plays a crucial role during the brain development, being specifically involved in the neurite outgrowth [70] and fasciculation [71], adhesion of neurons and astrocytes [72], and cell migration [73]. L1CAM attracted our attention due to its aberrant expression in several types of human tumors, where it acts as a promalignant factor and progression marker. For instance, L1CAM expression increases with progression of breast cancer [31]. Further, L1CAM promotes tumor progression and metastasis in melanoma [32] and ovarian [33], gastric [34], lung [35], and pancreatic cancers [36,37], where its presence is associated with a poor prognosis. Besides promoting cancer cell migration (see, e.g., [36]), including perineural invasion of cancer cells (for a review, see ref. [74]), L1CAM confers anti-apoptotic protection, chemoresistance [75] and radioresistance [76], stimulates cell survival [77] and acts as a pro-angiogenic factor [78].
Analysis of L1CAM induction in various cell types and senescence conditions revealed that the increase of the L1CAM total protein level during senescence was mostly associated with the increase of L1CAM transcript, indicating senescence-induced de novo transcription. Nevertheless, as in the case of HSF and RPE-1, the elevation of the total L1CAM protein must not necessarily be accompanied by the increase of its mRNA level indicative of senescence-mediated AGING posttranscriptional regulation of L1CAM. Moreover, the increase of the L1CAM total protein level was not always associated with its enhanced cell surface presence, indicating that the exposure of L1CAM on the cell surface of senescent cells is a complex process involving regulation at the transcriptional, posttranscriptional, and posttranslational levels.
Regulation of the L1CAM gene expression is controlled by diverse stimuli such as hypoxia or TGFβ signaling. It has been shown that inhibition of HIF-1 leads to inhibition of L1CAM and subsequent blockage of tumor growth and metastasis in lung cancer [55].
The cell cycle block responsible for premature senescence provoked by BrdU or IR is mediated by DNA damage response-induced expression of p21 and/or p16. In our attempt to understand the mechanism of L1CAM induction in premature senescence we found out that artificial expression of p16 itself is capable of elevating L1CAM transcript and protein levels, indicating that upstream DNA response signaling is not a prerequisite for senescence-mediated transcriptional induction of L1CAM. The notion that induction of L1CAM transcription was entirely absent in oncogenic Ras-induced senescence (Figure 2) led us to investigate the role of Ras/MAPK signaling in the control of L1CAM gene expression. Several reports point to the role of L1CAM signaling in augmentation of Erk activity ( [52][53][54]). Based on observed intrapopulation variability where the L1CAM levels inversely correlated with Erk1/2 'basal' activity which increased after L1CAM downregulation, our results showed rather opposite relationship indicating cell type differences. In contrast, chemical inhibition of MAP kinase MEK or downregulation of Erk1 or Erk2 resulted in elevation of the L1CAM transcript and protein levels. This suggests the suppressive role of the Ras/MAPK pathway on the L1CAM expression and the existence of a regulatory circuit controlling the mutual balance of L1CAM and the Ras/MAPK signaling, where the expression of L1CAM is controlled by the Erk pathway in a negative feedback loop. It is therefore likely that the level of L1CAM expression in a particular cell type depends on the equilibrium of interlinked activities of Erk and L1CAM pathways. Furthermore, we found that L1CAM expression was also upregulated in A375 cells exposed to TGFβ in concordance with previous reports [79,80]. Multiple interactions between TGFβ/SMAD and the Ras/MAPK pathways have been described (reviewed, e.g., in [81]). For instance, the Ras/MAPK pathway can suppress the activity of SMAD3 by Erk-dependent phosphorylation of SMAD3 [82]. It is therefore likely that chemical inhibition of MEK relieves the Ras/MAPK-mediated inhibition of the TGFβ pathway, resulting in increased expression of L1CAM; however, this warrants further investigation. The reciprocal expression of SCL25A5/ANT2 and L1CAM genes could reflect their reverse transcriptional regulation by the TGFβ pathway. The ANT2 gene belongs to early immediate genes induced by mitogens such as plateletderived growth factor and epidermal growth factor [83], indicating the role of the MAPK pathway in ANT2 gene expression. Expression of ANT2 is strictly growthdependent, and ANT2 is suppressed during growth cessation mediated by TGFβ-activated NFI/SMAD transcription factor complexes [59,84]. In contrast, L1CAM expression is induced by TGFβ ( [79,80] and this study) and suppressed by Ras/MAPK signaling ( Figure 5). However, it seems that L1CAM expression might be ANT2-dependent, as RNA interferencemediated knockdown of ANT2 resulted in L1CAM induction. As L1CAM is induced by metabolic signals, we propose that changes in the energy metabolism in response to the cell cycle arrest induce L1CAM. Further study is needed to reveal the mechanism of L1CAM induction in SC.
The cause of heterogeneous expression of L1CAM in proliferating and senescent cell populations in vitro is unclear. We observed that cells sorted according to the L1CAM surface level sustained its level of expression for several consecutive population doublings prior the establishment/return to the original heterogeneity, indicative of the role of epigenetic rather than genetic factors (clonal evolution) in the L1CAM intrapopulation heterogeneity. Based on our results we propose that the individual cell-specific sensitivity to Ras/MAPK and TGFβ signaling sets the level of the L1CAM surface expression.
In accord with previous study showing that manipulation of the level of L1CAM affects the migration properties of breast cancer cells [31], we observed that downregulation of L1CAM attenuates migration of proliferating BJ cells, supporting the role of L1CAM in cell locomotion. Intriguingly, also SC with a higher level of L1CAM surface expression migrated faster, which was accompanied by increased adhesion to extracellular matrix proteins laminin-1, collagen-1, and fibronectin. Senescent cells migrated significantly slower compared to proliferating cells, which may reflect overall changes in the adhesion apparatus and cytoskeleton of senescent cells, and likely an aberrant cell polarity and response to chemotactic stimuli. We attempted to decipher the mechanism of enhanced locomotory properties of cells with elevated L1CAM. To this end, we analyzed the number and shape of focal adhesions, the RhoA activity, integrin α2β1 and integrin αV levels, and the cell polarization to a wound; however, we did not find any significant AGING differences among L1CAM high and low cell subpopulations (see Supplementary Figure 7A, B, C, D, E, and F). Our results are also broadly consistent with the L1CAM participation in axon guidance and neuronal migration, in a pathway culminating in MEK and Erk activation [85]. Furthermore, a very recent study showed that ionizing radiation reduces ADAM10 expression in brain microvascular endothelial cells and this indirectly leads to enhanced L1CAM protein level [86], suggesting yet another level of stress-responding L1CAM regulation.
Senescent cells represent premalignant cellular forms [6], and the expression of the L1CAM oncoprotein in senescent cells may provide them with specific features that can be inherited by tumor cells bypassing the senescence state [87]. There is accumulating evidence supported by experimental animal models that cancer cells can disseminate early during tumor development to form metastases later, after a shorter or longer period of dormancy (see, e.g. [88,89], for a review, see [90]), sometimes even without observable formation of a primary tumor. Although it has not yet been proved experimentally that senescent cells can spread to distant tissues from the site of their origin, such possibility cannot be ruled out. Provided cellular senescence may be bypassed, as supported by several experimental findings (see, e.g., refs. [91,92]), the migratory and invasive properties of senescent cells might become important. Based on the correlation between the L1CAM level and tumor invasiveness it cannot be excluded that migration of senescent cells in the tissues might also be affected by L1CAM expression. Given the emerging roles of SC in promoting both tumorigenesis and organismal aging, our present results may inspire future senolytic approaches through exploiting the cell-surface expression of L1CAM in cellular senescence.

Preparation of tet-one p16 and p21 constructs
cDNA sequences for p16 and p21 were synthesized in Genescript. The cDNAs were subcloned to the Lenti-X™ Tet-One™ Inducible Expression System, which was obtained from Clontech (631847).

Preparation of recombinant lentiviruses, transduction of cells and clonal expansion
Recombinant lentiviruses were obtained from calciumphosphate transfected HEK 293T cells using packaging plasmids psPAX2 (Addgene, 12260) and pMD2.G (Addgene, 12259) together with either tet-one empty or tet-one p16 and p21 constructs. The medium containing lentiviral particles was harvested 36 to 48 h posttransfection, and the viral particles were precipitated using PEG-it (System Biosciences). Target cells H28 were transduced with viruses at multiplicity of infection MOI 5-10 and selected for puromycin resistance (2 μg/ml; Invivogen). For clonal expansion of each transduced cell type, single cells were sorted by BD FACSaria and the clones were then selected according to the expression profile of p16 or p21 protein.

Live cell immunofluorescence
Cells grown on glass coverslips were incubated with primary antibodies at room temperature for 15 minutes in PBS supplemented with Ca 2+ and Mg 2+ (PBS + ). After washing with PBS + , incubation with secondary antibody was performed at room temperature for 15 minutes. After washing with PBS + , cells were fixed with 4% formaldehyde at room temperature for 15 minutes. Coverslips were mounted in Mowiol containing DAPI to counterstain nuclei and viewed by fluorescence microscope Leica DMRXA.

Indirect immunofluorescence
Cells grown on glass coverslips were fixed with 4% formaldehyde and permeabilized with 0.1% Triton X-100 in two consecutive steps, each at room temperature for 15 minutes, and blocked with 10% FBS at room temperature for 30 minutes. After washing with PBS, cells were incubated with diluted primary antibodies at room temperature for 1 hour and then extensively washed with PBS/0.1% Tween 20. The incubation with secondary antibodies was performed at room temperature for 1 hour. Coverslips were mounted in Mowiol containing DAPI to counterstain nuclei and viewed by fluorescence microscope Leica DMRXA.

Cell sorting
Cells were washed two times with PBS and detached from the cultivation plates by incubation with accutase (Accumax, Merck Millipore, USA) at 37°C and 5% CO 2 for 2 to 3 minutes. To stop the protease, cultivation medium was added to the detached cells. Cell suspension was centrifuged at 700 x g for 5 minutes. The pellet was once washed with cultivation medium to get rid of the accutase, and the suspension was again centrifuged. Cell pellets were then stained with L1CAM antibody diluted 1 : 100 in cultivation medium on ice for 20 minutes, centrifuged, washed twice with culture medium and incubated with secondary antibody (diluted 1 : 500 in cultivation medium) on ice for 15 minutes. Cells were then washed twice with culture medium and after the last wash, they were resuspended in cultivation medium without serum and subjected to cell sorting.

Cell cycle measurement
Cells were washed with PBS, trypsinized, and subsequently collected into fresh medium. After centrifugation (500 x g at 4°C for 3 min), the cell pellet was resuspended in PBS. For fixation, a suspension drop was let drop into a centrifuge tube with -20°C 100% ethanol while vortexing and kept in -20°C for at least 2 hours. Fixed cells were collected by centrifugation, the cell pellet washed with PBS, resuspended in PBS containing RNase A (final concentration 0.2 mg/ml; Thermo Fisher Scientific, Waltham, MA, USA) and incubated at RT for 30 min. Prior to FACS measurement, propidium iodide (final concentration 12.5 μg/ml)/NP40 (final concentration 0.1%) solution was added to the samples.

Surface protein purification and mass spectrometry
BJ cells were labeled with 13 C 6 arginine. The surface proteins were isolated using the biotinylation method (according to manufacturer's protocol, Thermo Fisher Scientific). Briefly, cells growing on a 10 cm 2 dish were washed with ice-cold DPBS and incubated with 0.25 mg/ml sulfoNHS-S-S-biotin (Thermo Fisher Scientific) solution in DPBS at 4°C for 15 min. Cells were washed three times and the remaining unreacted NHS ester was quenched with 100 mM glycine in DPBS. After 10 min at 4°C, the unreacted NHS ester was washed away with DPBS. Cells were then lysed on AGING the dish by adding 500 μl of lysis buffer (0.5% SDS, 500 mM NaCl, 50 mM Tris, pH 7.4). The lysate was denatured at 95°C for 5 min and sonicated. Total protein in light and heavy labeled samples was measured by the BCA method and samples were mixed in a 1 : 1 protein concentration ratio. One hundred μl of pre-equilibrated streptavidin-sepharose (GE Healthcare, USA) was mixed with the lysate and agitated at room temperature for 1 hour. After extensive washing four times with 1 ml of lysis buffer followed by washing four times with 1 ml of 1 M NaCl, 50 mM Tris (pH 7.4) and 2 × 1 ml of distilled water, the bound proteins were eluted three times with 200 µl 0.5% SDS, 50 mM Tris (pH 6.8), 10% glycerol, 50 mM TCEP at 95°C for 5 minutes. The pooled eluate was concentrated on a 10 kDa cut-off Microcon ultrafiltration column (Merck Millipore, USA) and loaded on the SDS-PAGE gel. After electrophoresis, each gel lane was cut into six slices, the protein reduced and alkylated by iodoacetamide, and digested with trypsin.
Digested peptides were desalted and peptide mixtures were measured using LC-MS consisting of a Dionex UltiMate 3000 RSLCnano system (Thermo Fisher Scientific) coupled via an EASY-spray ion source (Thermo Fisher Scientific) to an Orbitrap Elite mass spectrometer (Thermo Fisher Scientific). Purified peptides were separated on 15 cm EASY-Spray column (75 µm ID, PepMap C18, 2 µm particles, 100 Å pore size; Thermo Fisher Scientific). For each LC-MS/MS analysis, about 1 μg peptides were used for 165 min runs. First 5 min, peptides were loaded onto 2 cm trap column (Acclaim PepMap 100, 100 µm ID, C18, 5 µm particles, 100 Å pore size; Thermo Fisher Scientific) in loading buffer (98.9%/ 1%/0.1%, v/v/v, water/ acetonitrile/ formic acid) at a flow rate of 6 µl/min. Thereafter has been switched valve and peptides were loaded in buffer A (0.1% v/v formic acid in water) and eluted from EASY-Spray column with a linear 120 min gradient of 2% -35% of buffer B (0.1% v/v formic acid in acetonitrile), followed by a 5 min 90% B wash at a flow rate 300 nl/min. EASY-Spray column temperature was kept at 35°C. Mass spectrometry data were acquired with a Top12 data-dependent MS/MS scan method. Target values for the full scan MS spectra were 1 x 10 6 charges in the 300-1700 m/z range, with a maximum injection time of 35 ms and resolution of 120,000 at m/z 400. A 2 m/z isolation window and a fixed first mass of 110 m/z was used for MS/MS scans. Fragmentation of precursor ions was performed by CID dissociation with normalized collision energy of 35. MS/MS scans were performed in an ion trap with ion target value of 5 x 10 4 and maximum injection time of 100 ms. Dynamic exclusion was set to 70 s to avoid repeated sequencing of identical peptides.
MS raw files were analysed by MaxQuant software (version 1.4.1.2), and peptide list were searched against the human Uniprot FASTA database and common contaminants database by the Andromeda search engine with cysteine carbamidomethylation as a fixed modification and N-terminal acetylation, methionine oxidations and thioacyl-lysine as variable modification. The false discovery rate was set to 0.01 for both proteins and peptides with a minimum length of six amino acids and was determined by searching a reverse database. Trypsin was set as protease, and a maximum of two missed cleavages were allowed in the database search. Peptide identification was performed with an allowed initial precursor mass deviation up to 7 ppm and an allowed fragment mass deviation of 0.5 Da. Matching between runs was performed. Quantification of SILAC pairs was performed by MaxQuant with standard setting using a minimum ratio count of 2.
Bioinformatics analyses were performed with the Perseus software.

Quantitative real time reverse transcription polymerase chain reaction (qRT-PCR)
Total RNA was isolated using RNeasy Mini Kit (Qiagen, MD, USA) according to the manufacturer's protocol. First strand cDNA was synthesized from 200 ng of total RNA with random hexamer primers using TaqMan Reverse Transcription Reagents (Applied Biosystems). qRT-PCR was performed in ABI Prism 7300 (Applied Biosystems) using SYBR Green I Master Mix (Applied Biosystems).
The relative quantity of cDNA was estimated by the ΔΔCt method [96]; data were normalized to GAPDH. Samples were measured in triplicates.

SDS-PAGE and immunoblotting
Cells were washed two times with PBS and then harvested into Laemmli SDS sample lysis buffer, sonicated, and centrifuged at 1600 x g for 10 min. Concentration of proteins was estimated by the BCA method (Pierce Biotechnology Inc., Rockford, USA).

AGING
Before separation by SDS-PAGE (9% and 14% gels were used), 100 mM DTT and 0.01% bromophenol blue were added to the cell lysates. The same protein amount (40 μg) was loaded into each well. After electrophoresis, proteins were electrotransferred from the gel onto a nitrocellulose membrane using the wet transfer method and detected by specific antibodies combined with horseradish peroxidase-conjugated secondary antibodies (donkey anti-mouse, Bio-Rad, Hercules, CA, USA). Peroxidase activity was detected by ECL (Pierce Biotechnology Inc.). GAPDH or pan-actin were used as a marker of equal loading, as indicated.

Wound healing assay
A confluent monolayer was disrupted with 10 μl micropipette tip. To remove floating cells, the culture medium was removed and replaced by fresh medium. Recovery of the disrupted area was monitored with time-lapse Leica Microscope DMI6000 for 32 to 48 hours. Cell migration was quantified by comparing the corrupted area with the recovered area according to formula: % of recovering = [(corrupted area -recovered area) / corrupted area] × 100. The size of the area was measured using ImageJ Software; velocity was analyzed with tracking plug-in.

3D migration assay
BJ cells sorted according to the L1CAM surface level were serum-starved for 24 hours and then seeded into collagen-containing Matrigel inserted in the column (μ-Slide Chemotaxis 3D, Ibidi, Munich, Germany) according to the manufacturer's protocol. Migration directed towards serum-containing medium was monitored with time-lapse Leica Microscope DMI6000 for 8 hours. Cell tracking was performed using ImageJ Software. For estimation of velocity and distance of the tracked cells, Chemotaxis and Migration Tool Software (Ibidi) was used. To evaluate random migration, the same procedure was used except that BJ cells were stained with antibody against L1CAM prior to seeding into Matrigel (to discriminate cells according to the L1CAM level based on the intensity of L1CAM surface staining (no fluorescence signal referred to cells with a low L1CAM level; in contrast, high L1CAM cells were those that showed strong fluorescence signal).

Adhesion assay
A 96-well cell culture plate was coated with laminin-1 (10 μg/ml in Hank's balanced salt solution), fibronectin (20 μg/ml in H2O), or rat tail collagen 1 (2 μg/ml in H2O) at 37°C at 5% CO2 for 1 hour. After washing with 0.1% BSA dissolved in DMEM (WB), the wells were blocked with blocking buffer (0.5% BSA in DMEM) at 37°C at 5% CO2 for 1 hour. After subsequent washing with WB, the plate was chilled on ice for 5 minutes, cell suspensions were seeded into the wells and left to adhere in a CO2 incubator at 37°C for 30 minutes. Next, the plate was shaken at 200 rpm for 10 to 15 seconds and washed three times with WB. Cells were then fixed with 4% formaldehyde at room temperature for 15 minutes, washed with WB and stained with crystal violet (5 mg/ml in 2% ethanol) for 10 minutes. The plate was subsequently washed several times with H2O to get rid of the remains of crystal violet. After drying up the plates, 2% SDS (in H2O) was added and incubated at room temperature for 30 minutes. Optical density at 550 nm was estimated by a spectrophotometer (Multiskan EX, Thermo Scientific, USA).

RNA interference
siRNAs targeting L1CAM (sense: GCAAGAGACAUA UCCACAAtt, antisense: UUGUGGAUAUGUCUCUU GCtg) were introduced into the cells using Lipo-fectamine™ RNAiMAX (Invitrogen, Carlsbad, CA, USA). Nonsense siRNA sequences (siNT; Ambion, CA, USA) were used as a negative control. shRNAs targeting L1CAM were introduced into the cells using a lentiviral vector. The shRNA vector targeting L1CAM was obtained by insertion of double-stranded oligo: 5'-CGGCAGCAAGAGACATATCCACAACTCGAG TTGTGGATATGTCTCTTGCTGTTTTT-3' into the AgeI and EcoRI sites in the pLKO.1 vector. A nontarget shRNA plasmid was obtained from Sigma (cat. No. SHC016). Lentiviral particles were produced by cotransfection with pMD2.G and psPAX2 (a kind gift from Didier Trono) in HEK293T cells as described previously [93]. Stably transduced cells were selected after 7 days by growing in media containing 2.5 µg/ml of puromycin.

RhoA activity measurement
RhoA activity was measured using G-LISA Activation Assay (Cytoskeleton, Inc). The procedure was performed according to manufacturer's protocol.

Determination of cell nuclei polarization
Confluent cell monolayer was scratched by pipette tip. After 4 hours, cells were stained (live) with the L1CAM antibody, then fixed, permeabilized and stained with antibody against a marker of MTOC, γ-tubulin. For determination of cell polarization, the position of MTOC was measured relatively to perpendicular lines (90°) positioned toward the scratch line [97]. The nuclei with MTOC present within the perpendicular lines were considered as polarized.

AUTHOR CONTRIBUTIONS
BM performed the research, analyzed the data and wrote the paper. RD, TI, LK, PB, PD, DH, MH, ZN, LA, PH, PV and OS performed the research and analyzed the data. JB wrote the paper. ZH designed the study, analyzed and interpreted the data, and wrote the paper.
AGING Supplementary Figure 6. (A) mRNA of ANT3, ANT2, and L1CAM after downregulation of ANT3 using RNA interference in HeLa and BJ cells (upper chart) and total L1CAM protein in Hela and BJ cells after downregulation of ANT3 using RNA interference (lower panel). GAPDH was used as a loading control. (B) Phase contrast microscopic images of A375 cells after exposure to 500 U/ml IFNγ, 10 ng/ml TGFβ, or their combination for 4 days. (C) Effectiveness of L1CAM downregulation by L1CAM siRNA in replicatively senescent BJ fibroblasts detected by Western blotting (siNT, control; L1CAM). GAPDH was used as a loading control. Scale bar, 100 μm.