Senescent tumor cells lead the collective invasion in thyroid cancer

Cellular senescence has been perceived as a barrier against carcinogenesis. However, the senescence-associated secretory phenotype (SASP) of senescent cells can promote tumorigenesis. Here, we show senescent tumour cells are frequently present in the front region of collective invasion of papillary thyroid carcinoma (PTC), as well as lymphatic channels and metastatic foci of lymph nodes. In in vitro invasion analysis, senescent tumour cells exhibit high invasion ability as compared with non-senescent tumour cells through SASP expression. Collective invasion in PTC is led by senescent tumour cells characterized by generation of a C-X-C-motif ligand (CXCL)12 chemokine gradient in the front region. Furthermore, senescent cells increase the survival of cancer cells via CXCL12/CXCR4 signalling. An orthotopic xenograft in vivo model also shows higher lymphatic vessels involvement in the group co-transplanted with senescent cells and cancer cells. These findings suggest that senescent cells are actively involved in the collective invasion and metastasis of PTC.

I nvasion and metastasis are hallmarks of cancer 1,2 . Invasion is a critical step in the progression to metastasis. For invasion, tumour cells modify not only their shape, but also their attachment to other cells and to the extracellular matrix (ECM). This alteration is known as the 'epithelial-mesenchymal transition' (EMT) and is characterized by loss of cell to cell adhesion molecules (E-cadherin) and upregulated expression of adhesion molecules associated with cell migration (N-cadherin) 3,4 . Through the EMT, tumour cells can detach from the main mass, and the separated tumour cells can invade into the ECM, as well as blood or lymphatic vessels as individual single cell. Therefore, the EMT is supposed to be involved in most steps of tumour progression, from invasion to metastasis, by conferring the abilities to invade, resist apoptosis and disseminate to tumour cells 1 . However, the underlying mechanism of invasion and metastasis varies depending on the type of cancer. Although certain types of high-grade and mesenchymal tumours infiltrate by single-cell migration with EMT characteristics, most low-grade tumours retain cell-to-cell adhesions and invade as cohesive multicellular strands. This type of invasion is known as 'collective invasion.' In carcinomas, originating from breast, colon, prostate and the thyroid gland, cancer cells invade cohesively with features of collective invasion 5 .
In collective invasion, most cancers are composed of varying degrees of heterogeneous subpopulations with distinct biologic properties involving proliferative ability, genetic alterations, signal pathways, drug or immune response, angiogenic potential, cell metabolism, motility, secretome and senescence, as well as different abilities for invasion and metastasis; certain cancer cells invade in the front of collective invasion as leaders whereas others are located in the rear and follow [6][7][8] . Among these biological properties, cellular senescence has been suggested as a barrier against carcinogenesis, because senescence induced by oncogenic activation (oncogene-induced senescence; OIS) is commonly observed in premalignant tumours, but rare in their malignant counterparts 9 . However, recent evidence indicates that cellular senescence can promote carcinogenesis by producing various growth factors, cytokines and proteases, collectively referred to as the senescent-associated secretory phenotype (SASP) 10 . Although senescent cells are rarely observed in cancers, the existence of isolated senescent cells in cancers has also been reported [11][12][13][14][15] . In our previous study involving papillary thyroid carcinoma (PTC), we demonstrated the presence of senescent cells in PTC 16 . Furthermore, our preliminary investigation frequently detected senescence associated-b-galactosidase (SA-b-Gal) positive senescent tumour cells in the invasive borders of PTC, lymphatic channels and metastatic foci of lymph nodes displaying features of collective invasion. These observations led us to hypothesize that senescent cells could participate in PTC invasion and metastasis. To explore this hypothesis, we analysed BRAFV600E-expressing PTC tissues from patients and employed an in vitro senescent thyrocyte model using BRAFV600E oncogenic activation, which is known as the most common oncogenic driver in PTC 17 , and applied this in vitro model and an orthotopic xenograft nude mouse model to characterize senescent cells and determine their involvement in collective invasion of PTC.

Results
Senescent tumour cells are identified in thyroid cancer. We examined senescent cells in various tumour types, including thyroid, breast, colon and stomach cancers by SA-b-Gal staining ( Supplementary Fig. 1), a standard biomarker of senescence, and found that senescent cells were frequently identified in thyroid cancer, and mostly in BRAFV600E-expressing PTC. To characterize senescent cells in PTC, we performed SA-b-Gal staining using snap-frozen tissue sections from 70 cases of BRAFV600E-expressing PTC. Interestingly, however, senescent cells were not evenly distributed in the tumours, but more frequently observed at the invasive borders of PTC ( Fig. 1a and Supplementary Fig. 2). Since p16 INK4A is a classic marker of OIS and have proved useful marker of senescence in vivo, we performed immunostaining for p16 INK4A using serial snap-frozen sections of the corresponding PTC and compared the expression of p16 INK4A with that of SA-b-Gal. We observed that the intensity and distribution of p16 INK4A -immunopositive cells were correlated with those of SA-b-Gal positive cells (Fig. 1b). Real-time polymerase chain reaction (PCR) analysis also indicated an increase in p16 INK4A mRNA expression in PTC ( Supplementary Fig. 3). Since cellular senescence is characterized by a stable cell cycle arrest, we also performed Ki67 (MIB-1) immunostaining to evaluate the proliferative activity in p16 INK4A -immunopositive cells and confirmed the absence of Ki67 (MIB-1) expression in these cells (Fig. 1b). These data together revealed that senescent cells in thyroid cancer were frequently present at the invasive borders. To determine whether p16 INK4A immunopositive senescent cells were thyroid tumour cells or stromal cells, we assessed the expression of BRAFV600E using VE1, a BRAFV600E-specific monoclonal antibody and thyroid transcription factor-1 (TTF-1) in formalin-fixed paraffin embedded (FFPE) tissues of BRAFV600E-expressing PTC. We found that p16 INK4A expressing senescent cells were also immunopositive for TTF-1 and BRAFV600E, indicating that senescent cells were not stromal cells, but neoplastic epithelial cells (Fig. 1c).
Senescent cells exist at the collective invasion border. In PTC invasive regions, neoplastic epithelial cells were organized into multicellular units and exhibited well preserved the E-cadherin expression, indicating collective invasion (Fig. 2a). Furthermore, collective invasion was maintained through lymph node metastasis from the primary cancer lesion. The expression of E-cadherin was preserved not only in primary PTC, but also in the metastatic foci of lymph nodes, and E-cadherin was present at tumour emboli in the D2-40 positive lymphovascular channels (Fig. 2a). Interestingly, senescent tumour cells were frequently located at the front region of E-cadherin conserved collective invasion (Fig. 2b). Since the EMT is known to broadly regulate cancer invasion and metastasis, we next assessed whether the EMT occurred in senescent cells by investigating the expression of EMT markers at the invasive border of PTC where senescent cells were detected. Twist1 and Zeb1 immunohistochemistry was performed using FFPE tissues of PTC, but no increase in the expression of Twist1 and Zeb1 at the invasive border was observed as compared with the corresponding cancer centre. In addition, E-cadherin expression was retained not only at the cancer centre, but also at the invasive border, whereas N-cadherin was hardly detected (Fig. 2c). Taken together, these data revealed that BRAFV600E-expressing senescent tumour cells were present at the invasive borders of PTC with features of collective invasion.
Senescent tumour cells have invasive ability via SASP. Since senescent tumour cells are present at the invasive borders of PTC, we next determined whether senescent cells have the ability to invade the matrix. Thus, we isolated tumour cells from BRAFV600E-expressing PTC. The isolated primary tumour cells were composed of both senescent and non-senescent tumour cells (senescent cells comprised 22% of the total). However, the invasion assay showed that 62% of invaded cells were SA-b-Gal positive, whereas only 9.5% of remaining cells (non-invasive cells) in the upper portion of the membrane were SA-b-Gal positive (Fig. 3a). Although the possibility that senescent cells enable non-senescent cells to migrate/invade and the non-senescent cells undergo senescence after migration/invasion is present, these data suggested that senescent cells exhibited a higher invasive ability than non-senescent tumour cells. Senescent cells acquire many changes in gene expression, mostly as alterations in mRNA expression, resulting in increased expression of numerous proteins, a phenomenon referred to as SASP. Within the SASP, we focused on the expression of proteases and chemokines, since senescent tumour cells face with ECM at the invasive border. We first evaluated SASP expression in the entire cancer area of PTC tissues which were composed of senescent and non-senescent tumour cells and compared the results with those of corresponding non-neoplastic thyroid tissues by real-time PCR. Although wide variations in SASP expression existed across PTC cases, the expression of interleukin (IL)-6, IL-8 and matrix metalloproteinases (MMPs) was markedly upregulated in cancer tissue as compared with levels observed in corresponding non-neoplastic thyroid tissues (Fig. 3b, Supplementary Table 1). Some PTC cases exhibiting high SASP expression shows more frequent lymph metastases, but there were no differences in histologic grade between PTC cases. To investigate SASP expression in senescent cells more precisely, we compared the genes expression profiles between the cancer centre and invasive borders ( Â 10 microscopically dissected from the cancer margin by a pathologist; n ¼ 3), where senescent cells were predominantly present, by RNA sequencing and detected an increase in SASP expression in the cancer invasive area (Fig. 3c, right panel). We validated SASP expression using real-time PCR analysis (Fig. 3c, left panel). Furthermore, we performed immunohistochemical staining of MMPs in ten cases of PTC, and found that immunoexpression of MMPs was higher than levels observed at the cancer centre, which agreed with mRNA expression (Fig. 3d). Taken together, these data indicated that senescent tumour cells in BRAFV600E-expressing PTC exhibited invasive ability and expressed SASP, including MMPs. Since we could not separate the senescent tumour cells from the primary cancer, we employed an in vitro senescent model to mimic the senescence phenotype occurring in BRAFV600Eexpressing PTC. We generated a BRAFV600E-mutant lentivirus and induced OIS in isolated primary human thyrocytes, and further confirmed that the senescence phenotype was diminished by shBRAF lentivirus infection (Fig. 3e). The gene expression profiling in BRAFV600E-induced senescent thyrocytes showed similar result as those of senescent tumour cells in PTC (Fig. 3f,g and Supplementary Fig. 4). In addition, we found that BRAFV600E-induced senescent thyrocytes exhibited a higher invasive ability ( Supplementary Fig. 5). Therefore, we considered that our in vitro senescent model might be useful for determining of the roles of senescent cells in BRAFV600E-expressing PTC.

Senescent cells in lymphovascular invasion and metastasis.
During collective cell migration and invasion, a coordinated movement between leader and follower cells is essential, and the guidance chemical or mechanical signals are required for directional migration 18 . First, we attempted to evaluate the mechanical guidance signals by analysing polarization of the actin cytoskeleton in PTC, but failed to identify this polarization ( Supplementary Fig. 6). Since senescent cells produce abundant SASP, including chemokines, we hypothesized that the chemical guidance signals are involved in the collective cell migration and invasion of PTC (Fig. 4a). As shown in Fig. 4b, the cluster of senescent tumour cells was located in the front region of collective invasion, whereas non-senescent tumour cells were located in the rear area of the senescent cells, appearing like that senescent tumour cells led the invasion, followed by non-senescent cells. We then investigated whether senescent cells were present during metastasis: senescent cells were actively involved in the cancer invasion and metastasis. Therefore, we performed SA-b-Gal staining, p16 INK4A and VE1 immunohistochemistry in 10 cases of PTC with metastatic lymph nodes and found that p16 INK4A immunopositive senescent cells existed in the metastatic foci of lymph nodes in all PTC cases (Fig. 4c, Supplementary Table 2). We then identified SA-b-Gal-positive senescent cells in tumour emboli of peritumoral lymphovascular spaces in primary PTCs. p16 INK4A and VE1 immunohistochemistry indicated that BRAFV600E-expressing tumour emboli in lymphatic channels included p16 INK4A -immunopositive senescent cells (Fig. 4d). These data demonstrated that BRAFV600Eexpressing senescent cells were not temporarily present during the initial process of collective invasion, but existed during the collective lymphovascular invasion and metastasis.
CXCL12/CXCR4 signalling in the collective invasion. We then determined whether senescent tumour cells in PTC could lead the collective invasion. To address this question, we performed an organoid three-dimensional (3D) culture assay using primary tumour tissues from BRAFV600E-expressing PTC as previously described 6 . After the emergence of collectively migrating cells from the tumour organoids, we analysed the tumour organoid by SA-b-Gal staining. SA-b-Gal positive senescent cells were identified at the front of invasive foci (Fig. 5a), with 77% of leading cells showing SA-b-Gal positivity in organoid cultures.
To verify that the senescent cells induced chemogradient collective invasion (Fig. 4a), we analysed chemokines expression at the invasion border of PTC in vivo and BRAFV600E-induced senescence in vitro model by RNA sequencing. RNA-sequencing analysis revealed upregulation of several chemokines, among which we focused on the expression of CXCL family, since C-X-C-motif ligand (CXCL)/CXCR signalling is an essential mechanism of collective invasion and lymph node metastasis 19 . We found that most of CXCLs/CXCRs and chemokine ligands (CCLs)/chemokine receptors (CCRs) showed no expression or very lower FPKM (Fragments Per Kilobase of transcript per Million fragments mapped) values in PTC (Supplementary Table 3). We further analysed RNA expression ratios between the cancer centre and the invasive border and found increased expression of several chemokines in the invasive area (Fig. 5b).
Although several chemokines, such as CCL4, CCL19, CCL21, CXCL2 or CXCL12, were upregulated at the invasive border than cancer centre, expressions of their receptors were very low in PTC, except for CXCL12 receptor (Supplementary Tables  3 and 4). We further investigated the expression of several members of the CXCLs/CXCRs family in 13 Fig. 5c, right panel), indicating that CXCL12/CXCR4 signalling was significantly upregulated at the invasive border and resulting in gradients of expression between the cancer centre and invasive border. We performed p16 INK4A and CXCL12/CXCR4 immunohistochemistry on tissue from 13 cases of BRAFV600E-expressing PTC to evaluate the relationship between senescent cells and CXCL12/CXCR4 expression. The expression of CXCL12, the unique CXCR4 ligand, significantly increased in the collective invasion area where p16 INK4A -immunopositive senescent cells were present, whereas CXCR4 was diffusely overexpressed in all cases of cancer tissues (Fig. 5d). These data revealed that senescent cells might induce a CXCL12mediated chemoattractant gradient between leader (senescent tumour cells) and follower (non-senescent tumour cells) cells. An in vitro BRAFV600E-induced senescence model showed marked induction of CXCL12 mRNA expression and secretion of CXCL12 protein in medium (Fig. 5e).

Senescent cells lead the collective invasion.
To determine whether senescent cells were capable of attracting cancer cells through the CXCL12/CXCR4 signalling, we performed an in vitro cell migration assay using two kinds of thyroid cancer cell lines: SNU790-CXCR4 and HTH83 ( Supplementary Fig. 7). BRAFV600E-induced senescent thyrocytes enhanced the migratory ability of CXCR4-expressing SNU790 and HTH83 cells (Fig. 6a, Supplementary Fig. 8), and results of a transwell assay were consistent with those of the cell migration assay (Fig. 6b, Supplementary Figs 9 and 10). To rule out the possibility that other cytokines were involved in cancer cell migration, we created two kinds of CXCL12 knockdown senescent cells by shRNA or application of the AMD3100 which is well known CXCR4 antagonist ( Supplementary Fig. 11). Our results indicated marked inhibition of cancer cells attraction by inhibition of CXCR4 signalling (Fig. 6a,b, Supplementary Figs 8 and 10). These data demonstrated that BRAFV600E-induced senescent thyrocytes can attract thyroid cancer cells through CXCL12/CXCR4 signalling.
We next determined whether CXCL12 expressing senescent cells could serve as leaders of collective invasion. To address this question, we first performed a 3D-matrix invasion assay (Fig. 6c, upper panel). When SNU790-CXCR4 thyroid cancer cells were co-cultured with normal thyrocytes, cancer cells invaded slowly with broad margins. However, in the presence of BRAFV600E-induced senescent thyrocytes, we observed protruding multicellular invasive strands from the tumour margin, indicating collective invasion (Fig. 6c, lower panel). We then performed live image analysis of the BRAFV600E-induced senescent thyrocytes/thyroid cancer cell co-culture system. When normal primary thyrocytes (green fluorescent protein (GFP)) and thyroid cancer cells (mCherry) were co-cultured, only a small number of thyroid cancer cells (mCherry) invaded the matrigel (Fig. 6d, upper panel), however, in co-cultured BRAFV600Einduced senescent thyrocytes (GFP) and cancer cells (mCherry), the BRAFV600E-induced senescent thyrocytes invaded the matrigel as leader cells, with formation of a canal (bright field), and cancer cells (mCherry) followed senescent cells along the canal (Fig. 6d, middle panel, high magnification field, Supplementary Fig. 12). When CXCL12 in senescent cells was knocked down by shRNA, the invasion of cancer cells was markedly decreased (Fig. 6d, lower panel, Supplementary Fig. 12). These data demonstrated that BRAFV600E-induced senescent thyrocytes lead collective invasion through CXCL12/CXCR4 signalling.
CXCL12/CXCR4 signalling is involved in anoikis resistance. We further analysed CXCL12/CXCR4 signalling in cancer progression. In lymphovascular spaces, cancer cells are detached from the ECM, and a loss of cell-matrix interactions induces apoptosis, referred to as anoikis. Therefore, the acquisition of anoikis resistance is a critical step required for metastasis, and anoikis-resistant tumour cells are associated with high incidence of metastatic lesions and increased cell survival in the blood 20 . EMT is among the intrinsic mechanisms used to escape anoikis, however, in this study, we observed well-preserved E-cadherin expression in tumour emboli, implying that EMT is not the principal mechanism in PTC (Fig. 7a). Interestingly, p16 INK4A -positive senescent cells in CXCR4-expressing tumour cells expressed CXCL12 (Fig. 7a,b,  Supplementary Fig. 13). CXCR4 and its ligand, CXCL12, can promote metastasis by preventing anoikis in cancer cells 21 . To determine whether senescent cells could increase the anoikis resistance of thyroid cancer cells, we performed an anoikis resistance assay using HEMA-coated plates (Fig. 7c, upper panel).      When we cultured thyroid cancer cells alone (SNU790-CXCR4 or HTH83), anoikis was rapidly induced in cancer cells in suspension. However, when we co-cultured thyroid cancer cells with senescent thyrocytes, the survival of cancer cells increased after a prolonged period of forced detachment. Furthermore, CXCL12 knockdown by application of two different shCXCL12 or the CXCR4 antagonist AMD3100 decreased the survival of cancer cells (Fig. 7c, lower panel, Supplementary Fig. 14a). We further analysed anoikis resistance by assessing caspase and apoptosis-related proteins expression, with results confirming that BRAFV600E-induced senescent thyrocytes increased the anoikis resistance of thyroid cancer cells (Fig. 7d,e and Supplementary Fig. 14b,c). These data revealed that senescent cells in tumour emboli of PTC can increase anoikis resistance through CXCL12/CXCR4 signalling.

Senescent cells enhance collective lymphovascular invasion.
To verify the invasion and metastatic ability of senescent cells in vivo, thyroid cancer cells together with BRAFV600E-induced senescent or normal thyrocytes were orthotopically transplanted into the thyroid gland of female nude mice (Fig. 8a). Since the PTC cell line (SNU790, SNU790-CXCR4) did not form a tumour mass in nude mice ( Supplementary Fig. 15), we used the anaplastic thyroid cancer cell line (HTH83; BRAF wild type and HRASQ61R þ / À ) 22 , which exhibited higher CXCR4 expression ( Supplementary Fig. 7). We observed that 58% (14/24) of nude mice developed thyroid cancer (6/10, normal thyrocytes/cancer transplanted group; 8/14, BRAFV600E-induced senescent thyrocytes/cancer transplanted group; Fig. 8b). There was no significant difference in terms of tumour development (60% versus 57%) and tumour size (7.5±2.81 mm versus 6.7±3.64 mm; P ¼ 0.414, Mann-Whitney U-test) between groups (Fig. 8c). We also examined neck lymph nodes and lymphovascular invasion in the peritumoral region of transplanted mice. Metastatic lymph nodes were detected in both groups (cancer positive lymph node/total identified lymph node; control, 1/14 (7%); senescence, 3/26 (12%), but no statistical  significance was observed between the groups. Interestingly, lymphovascular invasion by tumour cells was more frequent in BRAFV600E-induced senescent thyrocytes/cancer transplanted mice than in normal thyrocytes/cancer transplanted mice, as we identified 22 foci of lymphovascular invasion in BRAFV600Einduced senescent thyrocytes plus cancer transplanted mice, but only 6 foci of lymphovascular invasion in normal thyrocytes plus cancer transplanted mice (P ¼ 0.043, Mann-Whitney U-test) (Fig. 8d). In addition, BRAFV600E-expressing senescent thyrocytes were observed in a multicellular unit of tumour emboli in lymphovascular channels (Fig. 8e). Collectively, these data suggested that BRAFV600E-expressing senescent thyrocytes might be associated with increased lymphovascular invasion and increased anoikis resistance in thyroid cancer.

Discussion
Cellular senescence can be induced by loss of telomeres after extensive proliferation, as well as exposure to a variety of stresses, such as oxidative stress, DNA-damaging agents or oncogene activation 9,23-27 . Senescence, especially OIS, is considered as a barrier to tumorigenesis, given that OIS is commonly observed in premalignant tumours, such as melanocytic nevi, lung adenomas or pancreatic intraductal neoplasia, but is rare in their malignant counterparts 9,28,29 . However, previous studies indicated that a significant proportion of senescent cells are present in various types of cancers, such as B cell lymphoma, lung, breast, colorectal and thyroid cancers [11][12][13][14][15][16]28,[30][31][32] , indicating that senescent cells do not completely disappear. We also found that senescent cells were present in BRAFV600Eexpressing PTC. Cellular senescence in cancer cells can be induced by cell-autonomous mechanisms, such as overexpression of oncogenes or loss of tumour-suppressor genes [33][34][35] . In addition, certain types of tumour-infiltrating immune cells can induce senescence in a non-cell-autonomous manner [36][37][38] .
To investigate the role of cellular senescence in PTC, we utilized a BRAFV600E-induced senescence in vitro model, a cell-autonomous senescence model, for several reasons. First, BRAFV600E mutation is the most common driving oncogene mutation in PTC tumorigenesis. Second, all senescent cells in BRAFV600E-expressing PTCs expressed BRAFV600E protein.
Last, senescent cells could be identified in the areas of rare or no infiltration of immune cells. Senescent cells present a steady status, typically arrested in the G1 phase of the cell cycle, but metabolically and transcriptionally active, secreting a group of factors collectively termed the SASP 10 . The composition and amounts of the SASP may vary according to the duration of senescence and mechanisms inducing senescence 10,39 . In the present study, certain types of SASP, such as MMPs and chemokines, identified in our in vitro model also overlapped significantly with those of in vivo senescent tumour cells. These data suggested that SASP expressed at the invasive borders of PTC could at least in part originate from BRAFV600E-expressing senescent cells. Among the SASP, we focused on chemokines and observed that various chemokines were upregulated in in vitro senescent cells. We deeply considered several kinds of possible chemokines signalling in cancer invasion and metastasis using RNA-sequencing analysis. Although several chemokines were upregulated at the invasive border of PTC, the expression of their receptors were very low, except for the CXCL12 receptor. These chemokine expression patterns were also identified in in vitro model (RNA-sequencing data was deposited in Gene Expression Omnibus: GSE81144 and GSE83560). Finally, we found that the expression of CXCL12/CXCR4 was significantly upregulated in both senescent cells in vitro and in PTC tissues, therefore, we focused on CXCL12/CXCR4 signalling. CXCR4 activation stimulates the directed migration of cancer cells and increases their invasiveness 21 . Migration and invasion of CXCR4-expressing tumour cells depend upon a gradient of the chemokine, CXCL12, the only known ligand for CXCR4 (ref. 40). Interestingly, CXCL12 expression was not homogeneous across the tumour areas of PTC, but prominently increased in areas of collective invasion in PTC. Moreover, CXCL12-expressing cells were clearly correlated with senescent cells in areas of collective invasion. These data suggested that senescent cells in PTC can generate a chemotactic gradient at invasive borders through increased CXCL12 expression. Generation of a front-rear CXCL12 gradient can guide collective migration through the activation of CXCR4 signalling, and beyond guidance, chemotactic signalling can reinforce cell-cell adhesion and cell clustering 18 . Here, we demonstrated that CXCR4-expressing PTC cells were able to follow CXCL12-expressing senescent thyrocytes, and that CXCL12-expressing senescent thyrocytes effectively attracted CXCR4-expressing cancer cells in the migration assay. In addition, 3D invasion assay showed that some senescent thyrocytes formed the route in collagen containing matrigel by expressing MMPs and subsequently led cancer cells. Interestingly, when we administered the CXCR4 antagonist (AMD3100) to the organoid culture, we observed marked decreases in senescent cells invasion ability (Supplementary Fig. 16a). We also found that MMPs expression was markedly downregulated in AMD3100 treated cells ( Supplementary Fig. 16b), which was consistent with the results of previous reports 41,42 . Collectively, these data suggested that senescent cells work as a 'storming party' in the collective invasion of cancer through SASP, including MMPs and CXCL12/CXCR4 signalling. PTC is a well-differentiated malignancy, mostly presenting an indolent clinical behaviour and a good prognosis. However, PTCs are frequently associated with lymph nodes metastasis. Regional lymph node metastases are identified in 30 to 50% of PTC cases at the time of initial diagnosis 43,44 . In our series of BRAFV600E-expressing PTC, lymph node metastasis was detected in 48 of 79 cases (61%) (Supplementary Table 5). Although BRAFV600E mutation is one of the prognostic markers related to lymph node metastasis in PTC 45,46 , mechanisms related to frequent lymph node metastasis in BRAFV600E-expressing PTC has not yet been clearly explained. Here, we found neoplastic senescent cells in metastatic foci of regional lymph nodes. This finding raised a question of whether senescent cells in metastatic foci were from the primary tumour together with cancer cells in the manner of collective invasion, or cancer cells metastasized from the primary lesion and then became senescent cells after arrival at the metastatic site in lymph nodes. To answer this question, we examined lymphatic channels, the middle part of metastasis, in PTC tissues and micrometastatic foci in the lymph nodes. SA-b-Gal and p16 INK4A immunopositive neoplastic senescent cells were detected in tumour emboli, as well as micrometastatic foci (Supplementary Fig. 17). Although we cannot completely exclude the possibility of the latter hypothesis, these data supported that senescent tumour cells might move from the primary tumour mass to regional lymph nodes via lymphovascular circulation together with non-senescent tumour cells. Senescent cells in tumour emboli of lymphovascular channels and in metastatic lymph nodes expressed CXCL12. These findings provided another important point that senescent cells could be involved in or contributed to the survival of cancer cells in the lymphovascular system. Some studies suggested that CXCL12 might be involved in tumour cell growth and survival in several types of cancers 47,48 . CXCL12 increases anoikis in colon cancer 49 , whereas CXCR4 activation by CXCL12 in breast cancer cells decreases their sensitivity to anoikis 21 . Here, senescent cells enhanced anoikis resistance in thyroid cancer cell lines via CXCL12/CXCR4 signalling. Taken together, these data strongly indicated that cancer cells could invade into lymphovascular systems together with senescent cells, and that senescent cells might be involved in cancer cells survival, invasion, as well as metastasis.
Tumour formation in thyroid and metastatic foci was not observed when we performed orthotopic transplantation using SNU790 PTC cells. This might be due to the relatively indolent and low tumorigenic nature of SNU790 cells. Compared with SNU790 cells, HTH83 cells were more capable of developing tumour formation in the thyroid gland. However, the growth rate was high, resulting in tumour mass compression of the trachea and esophagus in mice 3 weeks after transplantation. Subsequently, some of the transplanted mice died of respiratory failure. Although the 3-week-period was not long enough to observe metastasis into regional lymph nodes, we could not maintain the mice for more than 3 weeks after orthotopic transplantation. This limitation may explain why we could identify only a small number of metastatic regional lymph nodes. Nonetheless, cancer invasion into lymphatic vessels was observed more frequently in the senescent thyrocytes co-transplanted mice than control mice. These data suggested that senescent cells were capable of promoting lymphovascular invasion and are involved in metastasis across the types of thyroid cancer.
Cancer invasion and metastasis are complex processes. To complete these processes, cancer cells must separate from the primary cancer, invade through the ECM, lymphatic and hematogenous systems, survive in the circulation, undergo extravasation into distant tissues or organs, and colonize them 50,51 . In recent years, cancer invasion has emerged as a hot spot in cancer research, and it has been suggested that tumour cells can acquire the ability to invade, resist apoptosis and disseminate through the EMT 1 . However, in the present study, few signs of EMT were detected in PTC. Our observations through morphological and immunohistochemical analyses, as well as 3D organoid assays, provided strong evidence of collective invasion in PTC. First, cancer cells at the invasion focus were not individual single cells, but maintained gland and/or papillary structures. Second, expression of the epithelial marker, E-cadherin, remained in invasive epithelial cells, lymphatic tumour emboli and the lymph node metastatic region. Third, EMT markers, such as Twist1 and Zeb1, were not upregulated in invasive areas. Interestingly, instead of EMT, senescent cells were located at the front region of collective invasion in PTC. Senescent cells secrete various SASP, which can disrupt normal tissue structure and function and promote malignant phenotypes in nearby cells, as well as modify the extracellular microenvironment and remodel ECM components. In addition, senescent cells attracted cancer cells and increase cancer cell survival. Therefore, senescent cells in PTC are fully qualified as leader cells in collective cancer invasion.
In conclusion, our results demonstrated that senescent cells are actively involved in the progression of cancer and provided a new concept for the collective cancer invasion model. Senescent cells promoted collective cancer invasion and metastasis via expression of SASP. Our present observations predict a new strategy for cancer treatment, including interventions targeting cellular senescence and SASP, to improve patient outcome by inhibiting metastasis.

Methods
PTC samples from patients. This study was approved by the Institutional Review Board of Ajou University Hospital (AJIRB-BMR-OBS-15-133). We obtained informed consents from all the patients who were included in the present study. PTC samples were obtained from patients at Ajou University Hospital after surgical resection. Fresh tumour and normal tissues were separately sampled in the representative areas by an experienced pathologist and snap frozen in liquid nitrogen immediately after resection, according to the tissue specimen regulation of Ajou University Hospital. Patients who had a past history of chemotherapy or radiation therapy before the surgery were excluded from the study.
Isolation and culture of normal thyrocytes. Normal thyrocytes were isolated from human thyroid tissues. Normal thyroid tissue was obtained after surgical resection. After confirmation with in situ smear cytology, normal thyroid tissue was washed five times with phosphate-buffered saline (PBS) and cut into small pieces, followed by treatment with 0.2% collagenase I (Worthington Biochemical, Lakewood, NJ) in HEPES buffer for 6 h at 32°C. Cells were resuspended in complete F12K media (Gibco BRL, Bethesda, MD; thyroid stimulating hormone (TSH), 10 mU ml À 1 ; insulin, 0.01 mU ml À 1 ; hydrocortisone, 10 nM; transferrin, 0.005 mg ml À 1 ; somatostatin, 10 ng ml À 1 ; glycyl-L-histidyl-lysineacetate, 10 ng ml À 1 ) and placed in 10 cm culture dishes. Next, we confirmed thyrocytes with thyroid specific markers including thyroglobulin and thyroid transcription factor-1. We further confirmed absence of BRAFV600E mutation by VE1 immunostaining.
Thyroid cancer cell lines. PTC cell line (SNU790) was purchased from Korean Cell Line Bank (KCLB, Seoul, Korea). Anaplastic thyroid carcinoma cell line (HTH83: established by Dr Nils-Erik Heldin, University of Uppsala, Sweden) was kindly provided by Dr Yoon Woo Koh (Department of Otorhinolaryngology, Yonsei University, Korea, cells authentication was performed by short tandem repeat (STR) analysis) and maintained in complete Roswell Park Memorial Institute medium (RPMI) media with 10% fetal bovine serum (FBS; Gibco BRL). Mycoplasma contamination of the cell lines used for this study was examined with Mycoplasma-PCR Detection Kit according to manufacture's instructions (#25235, iNtRON Biotechnology Inc., Seongnam, Korea).
Orthotopic xenograft of cancer cells to nude mice. All animal experiments were approved by the institutional animal research ethics committee at Ajou University Medical Center (Approval number: 2015-0008). An orthotopic xenograft of the thyroid cancer model was established by inoculating HTH83 cells into the lobe of thyroid gland 52 . Female BALB/c-nude mice (9-weeks-old) were anesthetized by intraperitoneal injection of ketamine (0.01 ml g À 1 of body weight). HTH83 cells (9 Â 10 5 ) and thyrocytes (normal or senescent; 1 Â 10 5 ) were resuspended with 10 ml of PBS and then injected into thyroid gland with 30-Gauge needle. After 3 weeks, the mice were sacrificed and analysed tumour size, lymph node metastasis and lymphovascular invasion.
Immunohistochemistry and immunocytochemistry. Immunohistochemical staining was performed with primary antibodies on 4 mm-thick representative tissue sections of formalin fixed paraffin-embedded tissue section in the Benchmark XT automated immunohistochemistry stainer (Ventana Medical Systems Inc., Tucson, AZ, USA). The primary antibodies used were as follows: anti-BRAFV600E (VE1), predilution (#790-4855,Ventana Medical Systems Inc); p16 INK4A , predilution (#705-4713, Ventana Medical Systems Inc); Anti-Human NATURE COMMUNICATIONS | DOI: 10.1038/ncomms15208 ARTICLE Statistical analysis. Statistical analysis on SA-b-Gal staining in different areas of PTC frozen samples was performed by w 2 test. Analyses on H score of immunohistochemical staining between cancer centre and invasive borders were performed using Wilcoxon signed rank test. Analyses of pathologic data from in vivo mice models were performed by Mann-Whitney U-test. Other data are presented as mean±s.d. of independent determinations, and were analysed using Student's t-test, and a P valueo0.05 was considered as significant. IBM SPSS ver. 22 (IBM Corp., Armonk, NY, USA) was used for all statistical analysis.
Microscope image acquisition. For image acquisition of histology and immunohistochemical staining, we used ScanScope CS system (Aperio Technologies, Inc., Vista, CA, USA). The cell images were acquired using an Olympus microscope mounted with an Olympus DP70 digital camera and DP-Manager software (Olympus Microscope Corp., Tokyo, Japan) at room temperature. Immunofluorescence images were collected on a Zeiss LSM 510 microscope and analysed with Zeiss Axio Imager software (Carl Zeiss, Jena, Germany) at room temperature.
Data availability. The Raw RNA-sequencing data have been deposited in the Gene Expression Omnibus database under the accession code GSE81144 and GSE83560. The RNA-sequencing data referenced during the study are available in a public repository from the https://www.ncbi.nlm.nih.gov/geo/. The authors declare that all the other data supporting the findings of this study are available within the article and its Supplementary Information files and from the corresponding author upon reasonable request.