Zebrafish imaging reveals TP53 mutation switching oncogene-induced senescence from suppressor to driver in primary tumorigenesis

Most tumours are thought to arise through oncogenic cell generation followed by additional mutations. How a new oncogenic cell primes tumorigenesis by acquiring additional mutations remains unclear. We show that an additional TP53 mutation stimulates primary tumorigenesis by switching oncogene-induced senescence from a tumour suppressor to a driver. Zebrafish imaging reveals that a newly emerged oncogenic cell with the RasG12V mutation becomes senescent and is eliminated from the epithelia, which is prevented by adding a TP53 gain-of-function mutation (TP53R175H) into RasG12V cells. Surviving RasG12V-TP53R175H double-mutant cells senesce and secrete senescence-associated secretory phenotype (SASP)-related inflammatory molecules that convert neighbouring normal cells into SASP factor-secreting senescent cells, generating a heterogeneous tumour-like cell mass. We identify oncogenic cell behaviours that may control the initial human tumorigenesis step. Ras and TP53 mutations and cellular senescence are frequently detected in human tumours; similar switching may occur during the initial step of human tumorigenesis.

T umour cells harbour multiple mutations. Decades of human clinical sample analyses and animal model studies have revealed that mutations in oncogenes and tumoursuppressor genes synergistically promote tumorigenesis 1,2 . The biochemical interactions between these multiple mutations in cancer cell proliferation, survival, and migration have also been widely investigated [3][4][5][6][7] . Based on these studies, it is thought that most cancers originate from a sporadically emerged single oncogenic cell with a mutation in oncogenes or tumoursuppressor genes [8][9][10] and that mutation accumulation facilitates tumorigenesis [11][12][13] . However, the processes by which a single transformed cell primes neoplastic development and the effects of additional mutations on these processes remain unclear. Moreover, although a newly emerged oncogenic cell in healthy tissue is surrounded by normal cells, the interaction between an oncogenic cell and neighbouring normal cells in the early stages of tumorigenesis has not been extensively studied.
Recent studies suggested that cellular senescence, the state of ultimate and irreversible cell cycle arrest, plays both tumoursuppressive and tumour-promoting roles depending on the cellular context. Cellular senescence is triggered by various stresses, including chromosomal instability, telomere shortening, oxidative stress, and DNA damage [14][15][16][17] . Abnormal oncogenic activation, including hyperactivation of Ras signalling, can also induce cellular senescence to exert a tumour-suppressive role by inducing cell cycle arrest. In fact, senescence-associated cell cycle arrest has been detected in human premalignant tumours, and in vivo model studies using Ras signalling-activated or PTEN-deficient mice demonstrated that cellular senescence negatively regulates tumour initiation and progression 14,[18][19][20] . However, senescent cells can also mediate tumorigenic effects by secreting a variety of inflammatory cytokines and growth factors, referred to as the senescence-associated secretory phenotype (SASP) 21 . SASP secretion leads to various effects, including a pathological increase in proliferation of neighbouring premalignant and malignant cells [22][23][24] . However, the roles of SASP in primary tumorigenesis, particularly in newly emerged oncogenic cells in living tissues, are unknown. In addition, how oncogenic cells decide to manipulate cellular senescence for tumour-suppressive or tumorigenic effects is unclear.
In this study, using imaging analyses, we show that a newly emerged Ras G12V oncogenic cell is eliminated from living tissue through cellular senescence. An additional TP53 mutation alters the effects of cellular senescence in Ras G12V cell from tumoursuppressive to SASP-mediated tumorigenic, and SASP secretion from the Ras-TP53 double mutant cell converts its neighbouring normal cells into senescent cells, thereby priming neoplastic development.

Results
Mosaically introduced Ras G12V cells are eliminated. To understand the behaviour of a newly generated oncogenic cell, we used zebrafish larval skin as a model of human epithelial tissue. We previously established a method for oncogenic cell visualisation in zebrafish skin 25,26 , in which fluorescent protein (e.g., mKO2, mCherry, or GFP)-expressing oncogenic cells were artificially introduced into the larval skin of the keratin4 (krt4)-gal4 zebrafish line 27 by injecting UAS promoter-driven expression plasmids (Fig. 1a). Low-dose injection of the UAS-driven plasmids induced a mosaic distribution of fluorescent cells, each of which was surrounded by normal cells in the larval skin (Supplementary Fig. 1a). Ras mutations, observed in~30% of human cancers, are considered to occur in the early stages of tumorigenesis 28,29 . Previous studies demonstrated that forced activation of oncogenic Ras signalling covering a wide area can induce excessive proliferation and dysplasia in the epithelial tissues of model animals, including mice, zebrafish, and flies [30][31][32] . Interestingly, mosaically introduced Ras G12V cells gradually disappeared and therefore did not induce any lesions (Fig. 1b). Ras G12V cells increased their cytoplasmic and nuclear volumes and apically protruded (Fig. 1c, Supplementary Movie 1, 2), but the protruding Ras G12V cells were nonapoptotic ( Supplementary  Fig. 1b). Thus, the newly generated Ras G12V cells surrounded by normal cells were eliminated from the larval skin. This oncogenic cell elimination appears to occur in an immune cell-independent manner. Consistent with this idea, Ras G12V cells were also eliminated from zebrafish larvae that lacked myeloid lineage cells (macrophages, monocytes, and neutrophil) established by injection of morpholino antisense oligo (MO) against spi1b (also known as pu.1) 33 ( Supplementary Fig. 1c, d). In addition, zebrafish larvae have not developed an adaptive immune system, as this system is not functionally mature until approximately 4 weeks post-fertilization 34 . Moreover, we found that the neighbouring cells accumulate F-actin at the contact sites with the protruding Ras G12V cells ( Supplementary Fig. 1e), suggesting that neighbouring epithelial cells are involved in Ras G12V cell elimination. To determine whether similar elimination can occur in other types of oncogenic cells, we mosaically introduced cells overexpressing v-Src, Myc and TP53 hot-spot mutants (TP53 R175H , TP53 R248W or TP53 R273H ), the abnormal activity of which can induce tumorigenesis [35][36][37] . Mosaic v-Src cells were eliminated as observed for Ras G12V cells ( Supplementary Fig. 2a), whereas Myc-overexpressing cells and TP53 mutants-expressing cells remained ( Supplementary Fig. 2b-e), suggesting that this elimination is triggered by a specific type of oncogenic activation.
Ras G12V cells undergo cellular senescence. Cytoplasmic and nuclear swelling is a common phenotype in senescent cells 38,39 . In addition, activation of oncogenic Ras signalling can induce cellular senescence 38 . Therefore, we tested whether cellular senescence occurred in swollen Ras G12V cells in the zebrafish epithelia. The senescence phenotype is often characterised by several unique features: (i) activation of a chronic DNA damage response, such as phosphorylation of the histone H2AX (γH2AX), (ii) senescence-associated β-galactosidase (SA-β-gal) accumulation, (iii) epigenetic alternation such as H3K9 trimethylation (H3K9me3), and (iv) irreversible cell-cycle arrest and its associated expression of various cyclin-dependent kinase inhibitors, including CDKN2A (p16 INK4a ), CDKN2B (p15 INK4b ), and CDKN1A (p21 CIP ) 17,39 . In the zebrafish genome, CDKN2A and CDKN2B are integrated into only one gene, cdkn2a/b. Accumulation of γH2AX in the nucleus (Fig. 2a), SA-β-gal accumulation (Fig. 2b), H3K9me3 upregulation ( Supplementary  Fig. 3a), cell-cycle arrest (reduction of EdU incorporation) (Fig. 2c), and high cdkn2a/b expression (Fig. 2d) were detected in mosaically introduced Ras G12V cells, suggesting that the Ras G12V cells were senescent. Interestingly, high cdkn2a/b expression and nuclear size enlargement were not detected in zebrafish larval skin ubiquitously expressing Ras G12V ( Supplementary Fig. 4a, b), indicating that Ras activation is not sufficient to induce cellular senescence in the skin. These results also suggest that neighbouring normal cells are required to induce senescence in mosaic Ras G12V cells.
Senescence drives Ras G12V cell elimination. To examine the involvement of cellular senescence in the Ras G12V cell elimination, we inhibited both CDKN2A/B-and TP53-mediated senescence-inducing pathways 17 by injecting MOs against cdkn2a/b and tp53. Injection of both MOs inhibited the swelling and elimination of mosaically introduced Ras G12V cells (Fig. 2e,  suggesting that senescence-induced cell swelling generates a driving force for oncogenic cell elimination. We also examined the status of cell adhesion, as mosaically introduced Ras G12V cells were apically delaminated. As expected, the membrane-localizing activity of E-cadherin, a major component of cellular adhesion 40 , was decreased in the Ras G12V cells, whereas the surviving Ras G12V cells in larvae injected with cdkn2a/b and tp53 MOs retained membrane E-cadherin ( Supplementary Fig. 5a), suggesting that Ras G12V -induced cellular senescence reduces adhesion between Ras G12V cells and neighbouring normal cells. Consistent with the above findings showing that ubiquitous expression of Ras G12V cells was insufficient to induce cellular senescence in the skin, a reduction of membrane E-cadherin levels was not detected in the skin ubiquitously expressing Ras G12V ( Supplementary Fig. 5b), suggesting that neighbouring normal epithelial cells are required to induce senescence-mediated E-cadherin reduction and mosaic Ras G12V cell elimination. These results indicate that cellular senescence suppresses tumorigenesis through immune cellindependent oncogenic cell elimination.
Ras G12V -TP53 R175H double mutant cells form a tumour-like cell mass. Although mutation accumulation is thought to facilitate tumorigenesis [11][12][13] , the effects of additional mutations on the behaviour of a newly generated oncogenic cell remain unclear. Recently, we reported that a newly generated Wnt signallinghyperactivated oncogenic cell is eliminated through apoptosis in zebrafish embryos, whereas loss of the tumour-suppressor gene smad4, which is associated with Wnt signalling hyperactivation in human cancers, blocked this elimination 25 . Therefore, we tested whether Ras G12V cell behaviour is influenced by additional mutations, including gain or loss of TP53, gain of Myc, loss of Smad4, and loss of adenomatous polyposis coli, events which are known to interact genetically and/or biochemically with Fig. 1 Newly generated Ras G12V cell surrounded by normal cells is eliminated from zebrafish epithelia. a Schematic illustration of experimental introduction of fluorescent oncogenic cells in zebrafish larval skin in a mosaic manner through the Gal4/UAS system. b Mosaic Ras G12V cell disappears from larval skin. Representative images show the larvae with mosaically introduced cells expressing mKO2 with Ras G12V (mKO2 + Ras + cells) (red). Skin cells were visualised with krt4p:gal4; UAS:EGFP (green). Scale bar: 200 μm. Bottom box plots of relative mKO2 + cell number of 2 days postfertilisation (dpf)/1 dpf ratio show first and third quartile, median is represented by a line, whiskers indicate the minimum and maximum, and outliers are shown as dots outside of the box. Each dot represents one larva. Unpaired two-tailed t-test was used. Note that oncogenic cells were introduced not only to the head region but also to the trunk and tail regions, and their disappearance occurred in all areas. c Cytoplasmic and nuclear swelling occurs in mosaic Ras G12V cells. Representative confocal images show mosaic mCherry + (control) or mCherry + Ras G12V (Ras + ) cells (magenta) and nucleus (upper; blue, lower; grey) in the larvae. Arrows indicates nuclei of mCherry + cells. Scale bar: 20 μm. Box plots of estimated nuclear or cytoplasmic size (μm 3 ) show first and third quartile, median is represented by a line, whiskers indicate the minimum and maximum, and outliers are shown as dots outside of the box. Each dot represents one cell. Unpaired two-tailed t-test was used. Source data are provided as a Source Data file. oncogenic Ras mutation in tumorigenesis [4][5][6]41,42 . Our analyses revealed that addition of TP53 R175H , TP53 R248W , or TP53 R273H , all of which are considered as TP53 gain-of-function hot-spot mutations 37 , had the most significant effects on the Ras G12V cells. We mosaically introduced oncogenic cells co-expressing Ras G12V with each TP53 mutant into zebrafish larval skin ( Supplementary   Fig. 6a). Although cells with Ras G12V alone were actively eliminated at 2 days postfertilisation (dpf), addition of TP53 R175H , TP53 R248W , or TP53 R273H mutation to Ras G12V cells significantly increased their viability (Fig. 3a, Supplementary Fig. 6b). Consistent with these results, additional TP53 R175H mutation blocked Ras G12V cell-induced F-actin accumulation in neighbouring cells   Fig. 5d). In contrast, mosaically introduced cells with either Ras G12V or TP53 R175H or with both Ras G12V and TP53 wild-type did not efficiently form the cell mass (Fig. 3d, e). Further, knockdown of endogenous tp53 using tp53 MO injection did not enhance Ras G12V cell-induced cell mass formation (Fig. 3d). Coexpression with TP53 DD , a TP53 dominant-negative mutant 43 that can inhibit the activity of TP53 wild-type ( Supplementary  Fig. 7a, b), in Ras G12V cells did not reinforce cell mass formation (Fig. 3d). Thus, gain-of-function, but not loss-of-function, of TP53 cooperates with the Ras G12V mutation to promote the evolution of a few oncogenic cells into a tumour-like cell mass. Notably, the introduction of cells co-expressing Ras G12V with a transcriptionally-inactive TP53 R175H-QSQS mutant 44,45 , did not induce cell mass formation (Fig. 3d), indicating that TP53 R175H requires its transcriptional activity to facilitate double-mutantderived cell mass formation.

Double-mutant cells form tumour-like cell masses through SASP.
Although active cell proliferation should occur during cell mass formation, almost all Ras G12V -TP53 R175H double-mutant cells were EdU-negative non-proliferative cells (Fig. 2c) and EdU-positive double-mutant cells were not detected in the cell masses (Supplementary Fig. 8a). Interestingly, EdU-positive or phospho-histone H3 (pH3)-positive proliferating cells lacking Ras G12V -TP53 R175H mutations were detected in the double-mutant cell-surrounding area (Fig. 5a, Supplementary Fig. 8b). Surviving double-mutant cells in zebrafish skin express SASP factors, suggesting that SASP factors secreted from these cells stimulate the proliferation of the neighbouring cells and consequent cell mass formation. To test this, we knocked down the mitogenic SASP factor il1b 50-52 using MO blocking il1b mRNA translation ( Supplementary Fig. 9a). As expected, il1b knockdown blocked the double-mutant cell-induced increase in pH3-positive proliferating cells in the surrounding area ( Fig. 5a) and cell mass formation (Fig. 5b), suggesting that the double-mutant cell-derived Il1b stimulates cell proliferation in neighbouring normal cells, leading to cell mass formation. We also found that treatment with the ROS scavenger N-acetyl-L-cysteine (NAC) reduced ROS production in the double-mutant cells (Fig. 4a) and cell mass formation (Fig. 5c), suggesting that, in addition to Il1b, other SASP factor ROS are involved in cell mass formation. To further evaluate the contribution of Il1b and ROS, we artificially increased the levels of Il1b and ROS in mosaic Ras G12V cells. Interestingly, co-expression of exogenous Il1b in Ras G12V cells reduced F-actin accumulation in the neighbouring cells ( Supplementary  Fig. 9b) and induced cell mass formation in a small number of larvae (Fig. 5d), indicating that addition of exogenous Il1b partially mimicked the effects of TP53 R175H mutation addition. Moreover, exposure to hydrogen peroxide (H 2 O 2 ), an ROS that can be generated in senescent cells 17 , further facilitated cell mass formation induced by cells co-expressing Ras G12V cells and exogenous Il1b (Fig. 5d). However, H 2 O 2 treatment did not affect Ras G12V cellinduced cell mass formation in the absence of exogenous Il1b (Fig. 5d). These results suggest that the double mutation-induced Fig. 2 Newly generated oncogenic cells undergo cellular senescence. a-d Representative confocal images show cells with GFP (a, d) or mCherry (b, c) alone (control), or with Ras G12V , TP53 R175H , or Ras G12V and TP53 R175H (GFP + , mCherry + , Ras + , TP53 RH+ , or Ras + TP53 RH+ cells) (green in a, d, magenta in b, c). Scale bar: 20 μm. In a, arrows indicate γH2AX (magenta)-accumulated GFP + oncogenic cells. Violin plots of γH2AX intensities in GFP + cells show 75th, 50th (median), and 25th percentiles. In b, SA-β-gal was visualised by SPiDER-βGal (green). Arrowheads indicate SA-β-gal-expressing mCherry + oncogenic cells, and arrows indicate cells SA-β-gal-expressing neighbouring cells. Violin plots of SA-β-gal intensities in mCherry + cells show 75th, 50th (median), and 25th percentiles. In c, EdU + cells (green) were observed in mCherry + or TP53 RH+ cells, but not Ras + or Ras + TP53 RH+ cells (magenta). Yellow broken lines: nuclei of mCherry + cells. Bar plots show EdU + mCherry + cells (mean ± SEM). Each dot represents one larva. A two-tailed one-way ANOVA test with Sidak correction (a, b, c) was used. In d, fluorescent in situ hybridization visualised cdkn2a/b mRNA (magenta). Arrows or arrowheads indicates cells expressing cdkn2a/b weakly or strongly, respectively. Bottom graphs show percentages of GFP + cells with weak or strong cdkn2a/b expression. Fisher's exact test with Bonferroni correction was used. e Representative confocal images show cells expressing mKO2 + Ras G12V (Ras + ) (magenta) and nucleus (upper: blue, lower: grey). Skin cells visualised with krt4p:gal4; UAS:EGFP (green). Scale bar: 20 μm. Box plots of estimated nuclear or cytoplasmic size (μm 3 ) show first and third quartile, median is represented by a line, whiskers indicate the minimum and maximum. Each dot represents one cell. Unpaired two-tailed t-test was used.f Representative images show larvae with GFP + Ras G12V (Ras + ) cells (green) with control MO or cdkn2a/ b + tp53 MOs. Scale bar: 200 μm. Box plots of relative Ras + cell number between 2/1 dpf ratio show first and third quartile, median is represented by a line, whiskers indicate the minimum and maximum, and outliers are shown as dots outside of the box. Each dot represents one larva. Unpaired two-tailed t-test was used. Source data are provided as a Source Data file.  (2 dpf / 1 dpf) (%) n of larvae: (24) (31) 2.6e-6 Fig. 3 TP53 R175H mutation attenuates mosaic Ras G12V cell elimination and stimulates heterogeneous cell mass formation. a Mosaic Ras G12V TP53 R175H cells remain in larval skin. Representative images show larvae with mosaically introduced GFP + Ras G12V (GFP + Ras + ) or GFP + Ras G12V TP53 R175H (GFP + Ras + TP53 RH+ ) cells (green). Scale bar: 200 μm. Box plots of relative GFP + cell number of 2 dpf/1 dpf ratio show first and third quartile, median is represented by a line, whiskers indicate the minimum and maximum. Each dot represents one larva. Unpaired two-tailed t-test was used. b-e TP53 gain of function, but not loss of function, stimulates heterogeneous cell mass formation. In b, representative images show cell mass in the trunk region of larvae with mosaically introduced mKO2 + Ras G12V TP53 R175H (Ras + TP53 RH+ ) cells (red) at 1.5 dpf. Skin surface layer cells were visualised with krt4p:gal4; UAS:EGFP (green). In c, representative image of haematoxylin and eosin staining of cell mass in the trunk region of larvae with mosaically introduced Ras + TP53 + cells. Scale bar: 50 μm. Graph in d and e shows the percentage of 1.5 dpf embryos with cell mass (Means ± SEM). Each dot represents an independent experimental result. In d, A two-tailed one-way ANOVA test with Sidak correction was used. In e, Unpaired two-tailed t-test was used. Note that cell mass formation was induced not only in the head region but also in the trunk and tail regions. Source data are provided as a Source Data file.
high-level production of Il1b and ROS in oncogenic cells enables cell mass formation in cooperation with neighbouring cells.
Propagation of senescence promotes tumour-like cell mass formation. Notably, the introduction of the double mutant cells upregulated ROS levels not only in themselves but also in neighbouring cells, and introduction of cells with either Ras G12V or TP53 R175H mutation did not induce such ROS diffusion (Fig. 4a, yellow arrows in top panels and bottom right graph). ROS were also detected in cells with and without double mutations inside the growing cell mass (Fig. 5e) breaks induction 53 , we hypothesised that diffused ROS induce senescence in neighbouring cells. Consistently, the growing cell mass contained γH2AX-, or cdkn2a/b-positive cells lacking Ras G12V -TP53 R175H mutations (Fig. 5f, g, arrowheads). In addition, mosaically introduced Ras G12V -TP53 R175H double-mutant cells, but not cells with either Ras G12V or TP53 R175H , induced SA-β-gal activity and γH2AX in neighbouring cells (Fig. 2b, arrowheads, Fig. 5h). All γH2AX-positive neighbouring cells were pH3-negative non-proliferative cells (Supplementary Fig. 8b), and NAC treatment reduced the double-mutant-induced γH2AX in neighbours (Fig. 5h)  To further analyse neighbouring cell senescence, we generated cellular senescence reporter zebrafish Tg(cdkn2a/b-hs:Achilles) ( Supplementary Fig. 10a), which express fast-maturing YFP variant protein, Achilles 54 , in cells expressing endogenous cdkn2a/b (Supplementary Fig. 10b). As expected, the cdkn2a/bhs:Achilles reporter was upregulated in mosaic Ras G12V singlemutant or Ras G12V -TP53 R175H double-mutant cells, and almost all of these cells were EdU-negative arrested ( Supplementary  Fig. 10c), suggesting that the cdkn2a/b-hs:Achilles reporter can be used to detect senescent cells. Time-lapse imaging of live Tg(cdkn2a/b-hs:Achilles) larvae captured the beginning of cdkn2a/b reporter expression in Ras G12V -TP53 R175H cells (Supplementary Fig. 10d, Supplementary Movie 4) and subsequent expression of the cdkn2a/b reporter in the neighbours (Supplementary Fig. 10e, Supplementary Movie 5). These observations are consistent with our model that the senescent double-mutant cells propagate senescence to the neighbours. We also confirmed that both cdkn2a/b reporter-positive senescent cells and EdUpositive proliferating cells were present in the Ras G12V -TP53 R175H cell neighbours, but these two populations were mutually exclusive ( Supplementary Fig. 10f).
ROS and il1b and il6 transcripts were detected in both Ras G12V -TP53 R175H double-mutant cells and cells lacking Ras G12V -TP53 R175H mutations inside the cell mass (Fig. 5e, i, Supplementary Fig. 11a), indicating that SASP factors are generated not only in the double-mutant cells but also in the senesced neighbours. To test this possibility, we prevented senescence in neighbouring normal cells by treatment with NAC to inhibit ROS, a senescence inducer secreted from doublemutant cells, or by MO-mediated knockdown of tp53, which is a mediator of ROS-induced cellular senescence 55 . NAC treatment reduced the expression of SASP factors, il1b and il11b, in larvae mosaically introduced with Ras G12V -TP53 R175H double-mutant cells (Supplementary Fig. 3e). In TP53-knockdown zebrafish skin, mosaic introduction of the double-mutant cells did not efficiently induce ROS production and il1b expression in the surrounding area ( Supplementary Fig. 11b, c) and cell mass formation (Fig. 5j), suggesting that neighbouring cell senescence and its associated production of SASP factors are required for cell mass formation. These findings also indicate that ROS generated in senescent double-mutant cells induce senescence and ROS production in the neighbours, suggesting that a senesced neighbouring cell also induces senescence in its neighbours, which may propagate senescence. Consistent with this idea, mosaic introduction of the double-mutant cells induced strong SA-β-gal activity in adjacent cells (~20 μm radius from the double-mutant cells) and moderate SA-β-gal activity in second adjacent cells (20-40 μm radius from the double-mutant cells) (Fig. 5k, Supplementary Fig. 11d). In the double-mutant cell-induced cell mass, the senescence markers γH2AX and cdkn2a/b were broadly expressed in not only the area immediately neighbouring the double-mutant cells but also 2-3 cells away from these cells (Fig. 5f, g). Thus, in addition to Il1bmediated neighbouring cell proliferation, ROS-mediated senescence propagation may also contribute to cell mass formation.
Mosaic Ras G12V cells form a cell mass in damaged skin. Next, we tested the behaviour of oncogenic cells on unhealthy epithelia with abnormal TP53 activity or senesced epithelia. Mosaically introduced of Ras G12V mutation in larval skins ubiquitously expressing TP53 R175H efficiently generated tumour-like cell masses (Fig. 6a), in which γH2AX, ROS, and il1b were detected in cells with and without Ras G12V mutation ( Supplementary  Fig. 12a-c); this cell mass formation was supressed by il1b MO or NAC treatment (Fig. 6b, c), suggesting that a newly generated Ras G12V mutation can prime tumorigenesis by inducing Il1bmediated proliferation and ROS-mediated senescence in its neighbours in the epithelia with abnormal TP53 activity. To produce senesced epithelia, we treated zebrafish larvae with doxorubicin, which induces DNA damage and consequent TP53 activation and cellular senescence 56,57 . Doxorubicin treatment partially induced γH2AX-positive senescent cells in larval skin ( Supplementary Fig. 12d), whereas it did not induce apoptotic cells ( Supplementary Fig. 12e). Compared to mosaically introduced Ras G12V cells in normal skin, those in doxorubicin-treated skin relatively strongly expressed cdkn2a/b, il1b, 1l11b, and ROS ( Fig. 6d-g, Supplementary Fig. 12f) and efficiently induced tumour-like cell masses (Fig. 6h). Mosaic Ras G12V cell-induced cell mass formation in doxorubicin-treated skin was blocked by MO-mediated tp53 knockdown (Fig. 6i). These results indicate that the mechanism of Ras G12V cell-induced cell mass formation in senesced epithelia is similar to that of double-mutant cellinduced cell mass formation in the healthy epithelia, and that senesced epithelia lose the potential to eliminate oncogenic cell and become prone to tumorigenesis initiation compared to healthy epithelia.

Discussion
We demonstrated the potential mechanisms of tumour initiation and roles of cellular senescence in this process. In healthy zebrafish epithelial tissue, a newly emerged oncogenic Ras G12V cell becomes senescent, which stimulates cell swelling and loss of intercellular adhesion with neighbouring cells and consequent Ras G12V cell elimination (Fig. 7, left). However, the addition of a TP53 gain-of-function mutation (TP53 R175H ) prevents Ras G12V cell elimination by restoring adhesion with neighbouring cells and reinforces Ras G12V cell senescence and consequent production of Fig. 5 SASP-mediated senescence propagation is involved in Ras G12V TP53 R175H cell-driven cell mass formation. a Representative confocal images show phospho-histoneH3 (pH3) (magenta) and nucleus (blue) in the larvae with GFP + (control) or Ras + TP53 + cells (green) with control or il1b MO. Scale bar: 20 μm. Graph shows the percentage of GFP + cells with pH3 + neighbours (mean ± SEM). b, c Larvae with Ras + TP53 RH+ cells with control or il1b MO (b) or untreated or treated with NAC (c). Graph shows the percentage of 1.5 dpf embryos with cell mass (mean ± SEM). d Representative images show cells coexpressing Ras G12V and zebrafish Il1b (Ras + IL-1β + cells) (green) with or without H 2 O 2 treatment. Scale bar: 50 μm. Graph shows the percentage of 1.5 dpf embryos with cell mass (mean ± SEM). Each dot in graph (a-d) represents an independent experimental result. e-g, i Representative images show ROS (e; green), γH2AX (f; magenta), cdkn2a/b mRNA (g; magenta), or il1b mRNA (i; magenta) and nucleus (f; grey) in cell mass on the larvae with Ras + TP53 RH+ cells (e; magenta, f, g, i; green). Dot lines surround cell mass. Scale bar: 50 μm. h Representative confocal images show γH2AX (magenta) and nucleus (blue) in larvae with Ras + TP53 RH+ cells (green) untreated or treated with NAC. Scale bar: 20 μm. Violin plots of γH2AX intensities in neighbours (<20 μm radius from GFP + , Ras + , TP53 RH+ , or Ras + TP53 RH+ cells) show 75th, 50th (median), and 25th percentiles. j Schematic illustration (right) shows experimental inhibition of secondary senescence in neighbouring cells. Larvae with Ras + TP53 RH+ cells were injected with control or tp53 MO. Graph show percentage of 1.5 dpf embryos with cell mass (means ± SEM). k Violin plots of SA-β-gal intensities in neighbouring cells around mCherry + cells show 75th, 50th (median), and 25th percentiles. adj, mid, and dist indicates SA-β-gal activity-positive adjacent cells (<20 μm radius from the double-mutant cells), second adjacent cells (20-40 μm radius), and distant cells (more than 40-μm radius). A two-tailed one-way ANOVA test with Tukey's correction (a) or Sidak correction (d, h, k) or unpaired two-tailed t-test (b, c, j) was used. Source data are provided as a Source Data file.  (Fig. 7, middle). Additional TP53 mutation-mediated survival of mutant cells would assist in the continuous high-level secretion of IL-1β and ROS to the neighbours. Neighbouring cells undergo proliferation or senescence in response to IL-1β or ROS, respectively. Secondary senesced cells also secrete IL-1β and ROS, which can propagate cellular senescence and stimulate aberrant cell proliferation nearby, resulting in formation of a heterogeneous tumour-like cell mass. Similar heterogeneous cell mass formation depending on IL-1β and ROS occurs in abnormal zebrafish skin with TP53 R175H mutation or DNA damage (Fig. 7, right).
Previous studies revealed two tumour-suppressive mechanisms of cellular senescence. The first is induction of cyclin-dependent kinase inhibitor-mediated irreversible cell cycle arrest in oncogenic cells [14][15][16][17] . The second is stimulation of immune cellmediated senescent cell elimination 58,59 . In this study, we found that immune cell-independent oncogenic cell elimination is a third antitumour mechanism of cellular senescence. Interestingly, similar elimination was triggered by activation of Src in the zebrafish epithelia. Cells with hyperactive Src may be eliminated through cellular senescence because these signalling pathways are also involved in senescence induction 60 . In contrast, zebrafish . Graph shows the percentage of 1.5 dpf embryos with cell mass (mean ± SEM). Each dot in graph (a-c) represents an independent experimental result. d-g Doxorubicin promotes senescence and SASP factor production in mosaic Ras + cells. In d, representative images show fluorescence in situ hybridization for cdkn2a/b mRNA (magenta) in Ras + cells (green) with or without doxorubicin treatment, and nucleus (blue). Arrow and arrowhead indicate cdkn2a/b + Ras + cells weakly or strongly, respectively. Graph shows percentage of cdkn2a/b + Ras + cells. In e-f, representative images show fluorescence in situ hybridization for il1b (e) or il11b mRNA (f) (magenta) in Ras + cells (green) with or without doxorubicin treatment. Arrows indicate il1b + Ras + or il11b + Ras G12V cells. Graph shows percentage of il1b + Ras + or il11b + Ras G12V cells (mean ± SEM). Each dot represents one larva. In g, representative images show ROS-probe, CellROX (magenta), -stained larvae with Ras + cells (green) with or without doxorubicin treatment. Box plots of CellROX relative intensities in Ras + cells show first and third quartile, median is represented by a line, whiskers indicate the minimum and maximum. Each dot represents one cell. h Representative images show cell mass in the head region of doxorubicin-treated larvae with GFP + Ras + (Ras + ) cells (green). Right graph shows percentage of embryos with cell mass (mean ± SEM). i Doxorubicin-treated larvae with Ras + cells with control or tp53 MO. Graph shows percentage of 1.5 dpf embryos with cell mass (mean ± SEM). Each dot in graph (h, i) represents an independent experimental result. Note that cell mass formation was induced not only in the head but also in the trunk and tail in TP53 R175H -expressing and doxorubicin-treated skin. Scale bar: 50 (a, h) or 10 (d-g) μm. Unpaired two-tailed t-test (a-c, e-i) or Fisher's exact test with Benjamini-Hochberg correction (d) was used. Source data are provided as a Source Data file.
epithelia could not eliminate oncogenic cells with TP53 hot-spot mutations (TP53 R175H , TP53 R248W and TP53 R273H ), high Myc levels, and abnormally high Wnt/β-catenin activity 25 , suggesting that this elimination system functions against a specific type of oncogenic activation.
Our study also showed that a newly generated Ras G12V cell stimulates senescence neighbouring normal cells, indicating that communication with surrounding cells is involved in Ras G12V cell elimination. Notably, mammalian epithelia can also eliminate sporadically emerging oncogenic cells by communicating with neighbouring normal cells; in Madin-Darby canine kidney cell culture, mosaically introduced oncogenic cells with abnormal Ras and Src activities were apically extruded through communication with neighbouring normal cells 61,62 . This neighbouring cellmediated oncogenic cell elimination system is known as cell competition or epithelial defence against cancer (EDAC) 63 . A similar phenomenon occurs in cultured cells in the mouse small intestine, pancreas, and lung 64,65 . The mechanisms of EDAC are relatively well-studied. For example, recent studies showed that neighbouring cell-mediated upregulation of pyruvate dehydrogenase kinase in Ras G12V cells and the calcium wave, which is propagated from Ras G12V cells to surrounding cells via the mechanosensitive calcium channel and IP3 receptor, promote Ras G12V cell elimination 26,64 . Interestingly, we found that treatment with 2-aminoethoxydiphenylborane (inhibitor of mechanosensitive calcium channel and IP3 receptor) or dichloroacetate (specific inhibitor of pyruvate dehydrogenase kinase family) significantly suppressed cellular senescence in Ras G12V cells and their elimination ( Supplementary Fig. 13a, b), suggesting that these EDAC mediators function upstream of Ras G12V cell senescence and that senescence-driven Ras G12V cell elimination in zebrafish is mediated by mechanisms similar to that of EDAC in mammals. In addition, our findings suggest that cellular senescence-mediated elimination functions in mammalian epithelial tissues.
We identified membrane E-cadherin localization as a key factor in mutant cell elimination. Mosaically introduced Ras G12V cells blocked membrane localization of E-cadherin in a cellnonautonomous manner, which drives apical extrusion of Ras G12V cells from the zebrafish larval skin. Additional TP53 mutation recovers membrane E-cadherin and allows retention of mutant cells. However, the detailed mechanisms of how Ras G12V and TP53 mutations control E-cadherin localization were unclear. Notably, Ras G12V cells induced F-actin accumulation in its surrounding cells, whereas Ras G12V -TP53 R175H double-mutant cells did not, indicating that the levels of F-actin accumulation in neighbouring cells are inversely correlated with E-cadherin levels in mutant cells. One possibility is that F-actin accumulation in neighbours is involved in regulating E-cadherin in the mutant cells. It will be interesting to study the neighbouring cell's machinery controlling E-cadherin in mutant cell and its elimination.
We also showed that inside the growing cell mass, some double-mutant cells and neighbouring cells had reduced E-cadherin levels, whereas the remaining cells did not. It remains unclear how this E-cadherin heterogeneity is induced. Previous reports showed that a SASP factor, including ROS, can disrupt the membrane-localization of cadherin 66 . ROS, which are continuously produced inside the cell mass, may be involved in reducing E-cadherin levels in this context.
Addition of a TP53 gain-of-function mutation prevented Ras G12V cell elimination by restoring intercellular adhesion with neighbours and blocking F-actin accumulation in neighbours to reinforce Ras G12V cell senescence and SASP factor production, resulting in tumour-like cell mass formation in zebrafish epithelia. Ras and TP53 mutations are common genetic mutations in various cancers, and human pancreatic ductal adenocarcinoma and colorectal cancer cells acquire both mutations during tumour progression 67,68 . Interestingly, recent studies revealed that human TP53 hot-spot point mutations, including TP53 R175H , TP53 R248W , and TP53 R273H , are gain-of-function 69 and enhance oncogenic potential 70,71 . In fact, mice with both Ras and TP53 gain-of-function mutations efficiently form more malignant tumours compared to in mice with either Ras or TP53 mutation 3,7,72,73 . Notably, a previous study reported that an additional TP53 R175H mutation prevents the disappearance of senescent cells with oncogenic Ras mutations and stimulates tumorigenesis in the mouse pancreas 74 . Although this previous study did not examine the mechanisms of senescent cell disappearance and involvement of SASP in tumorigenesis, the mechanisms may be similar to those we observed in the zebrafish model and in mammals.
We observed a new mode of Ras-TP53-mediated primary tumorigenesis. TP53 mutation stimulates senescent Ras G12V cells not only to survive but also to induce proliferation and secondary senescence in the surrounding area. In mouse liver and human fibroblast cultures, senesced oncogenic cells also induced secondary senescence in neighbouring cells 75 . In addition, in human patient specimens, signs of senescence, including oxidative damage accumulation and TP53 upregulation, were detected in normal tissues adjacent to the tumours [76][77][78][79][80][81] , suggesting that secondary senescence occurs in human tumorigenesis. However, the role of secondary senescence in tumorigenesis has not been examined. Here, we demonstrated that secondary senescence in the neighbouring cells of oncogenic cells is required for tumour initiation in a zebrafish model. Thus, our work provides insight into the interactions between oncogenic cells and their surrounding cells during the early stages of cancer development.
Human tumours generally exhibit heterogeneity and include a diverse collection of cells harbouring distinct molecular signatures 82 . Although cancer heterogeneity is the largest barrier to cancer treatment, mainly because of its resistance to therapy 83 , the mechanisms that generate heterogeneity in cancer have not been completely explained. The currently acceptable view is that differentiation and/or additional mutations in cancer cells generate tumour heterogeneity 82 . In our zebrafish model, mutant cells converted neighbouring cells to abnormal proliferative or senescent cells through the secretion of SASP factors, thereby forming a heterogeneous cell mass. Based on these findings, SASP-mediated propagation of senescence and abnormal proliferation in the neighbourhood may be a new model of heterogeneous tumour generation. Thus, by using zebrafish imaging analyses, we revealed potential mechanisms of primary tumorigenesis.

Methods
Zebrafish maintenance. Zebrafish wild-type strain (AB), Tg(krt4p:gal4; UAS:EGFP) 27 , Tg(krt4p:gal4; UAS:mKO2), Tg(UAS:GAP43mKO2-T2A-H-Ras G12V ), Tg(UAS:GAP43EGFP-T2A-TP53 R175H ), Tg(cdkn2a/b-hs:Achilles), Tg(UAS:zE-cadherin-GFP), and Tg(il1b:EGFP) 84 were raised and maintained under standard conditions. All experimental animal care was performed in accordance with institutional and national guidelines and regulations. The study protocol was approved by the Institutional Animal Care and Use Committee of the respective universities (Gunma University Permit# 17-051; Osaka University, RIMD Permit# R02-04). One-cell stage embryos were used for cell injection to generate transgenic fish lines or mosaic larvae, with the latter processed for up to 2 dpf. To avoid excess tumour formation, zebrafish larvae were sacrificed within 24 h after a day after the initial cell mass formation, in our study.
Detection of proliferating cells using EdU. To detect proliferating cells, zebrafish larvae were incubated with 500 μM EdU/10% DMSO in Egg water for 1 h, and larvae were washed three times in egg water. Larvae were fixed with 4% PFA in PBS for 1 h at RT. The larvae were rinsed in double-distilled water, placed in acetone at 20°C for 7 min, rinsed in double-distilled water, and permeabilized for 1 h in PBS/ 1%DMSO/1%Triton. For EdU detection, larvae were processed according to the Click-iT EdU Cell Proliferation Assay Kit (Invitrogen, Carlsbad, CA, USA; Alexa Fluor 488: C10637 and Alexa Fluor 647: C10640). After loading, larvae were washed three times in 0.1% Tween 20 in PBS, and then immunostaining was performed.
Whole-mount immunostaining. At 26-32 hpf, Tg(krt4p:gal4) larvae with mosaically introduced fluorescent cells were manually dechorionated and then fixed with 4% PFA overnight at 4°C. As for H3K9me3, antigen retrieval was performed 96 . The larvae were washed four times with 0.5% Triton X-100 (PBST) and incubated with blocking reagent (10% foetal bovine serum, 4% Block Ace (Megmilk Snow Brand, Tokyo, Japan), and 1% DMSO in 0.1% PBST) for 1 h (γH2AX-stained larvae were incubated with blocking reagent without Block Ace). The larvae were incubated with primary antibodies overnight at 4°C, washed, and incubated with AlexaFluor-conjugated secondary antibodies and with Hoechst33342 (Invitrogen) overnight at 4°C. Immunostained larvae were visualised with a M205FA fluorescent stereomicroscope (Leica), LSM700, or FV1000 (Olympus, Tokyo, Japan) confocal laser-scanning microscope. Images were prepared and analysed using ImageJ software (NIH, Bethesda, MD, USA). A list of primary and secondary antibodies is shown in Supplementary Table 1.
Whole-mount in situ hybridisation. Whole-mount in situ hybridisation was performed according to a standard protocol. Fluorescence in situ hybridisation was performed according to a previously described protocol 97 . Digoxigenin-or FITClabelled RNA antisense probes were prepared from plasmids containing GFP, Achilles, mKO2, lcp1, npsn, il1b, il6, il11b, cxcl8a, and cdkn2a/b. Images were taken using a M205A stereomicroscope (Leica) and FV3000 confocal laser scanning microscope.
Haematoxylin and eosin staining. The fixed larvae were embedded in paraffin and serially sectioned. Sections of 4-μm thickness were stained with haematoxylin and eosin according to a standard procedure.
Estimation of cytoplasm and nuclear size. The cytoplasm/nuclear areas in each cell were measured using ImageJ software. To estimate the size of the cytoplasm/ nuclear area, the sum of the measured areas was multiplied by the thickness between the slices.
Quantitative PCR (qPCR). UAS promoter-driven plasmids (Ras expression plasmid doses were as follows: 25 pg in Supplementary Fig. 3c, e, 15 pg in Supplementary Fig. 7d, 12.5 pg in Supplementary Fig. 12 f) were injected into one-cellstage embryos. Total RNA from 25 larvae was purified using TRIzol reagents (Invitrogen), and cDNAs were synthesised using ReverTra Ace qPCR RT Master Mix (Toyobo, Osaka, Japan). qPCR was performed on an Mx3000P QPCR system (Agilent Technologies, Santa Clara, CA, USA) with THUNDERBIRD SYBR qPCR Mix (Toyobo); the specific primers are described in Supplementary Table 1; actb1 was used as a loading control. qPCR cycling conditions were as follows: 95°C for 1 min, [95°C for 10 s, 60°C for 30 s] (45 cycles), followed by dissociation curve analysis. The primer list is shown in Supplementary Table 1.
Generation of Tg(cdkn2a/b-hs:Achilles) zebrafish. To generate a donor plasmid containing a heat-shock promoter and cDNA encoding zebafish-codon-optimized and nuclear-localization-signal-tagged Achilles (Tbait-hs-zAchilles-NLS), zAchilles-NLS cDNA was synthesised (FASMAC, Atsugi, Japan) and replaced with GFP sequence in Tbait-hs-GFP plasmid (a gift from Dr. S. Higashijima) 98 . Tbait-hs-zAchilles-NLS was co-injected with a single-guide RNA (sgRNA) targeted for digestion of cdkn2a/b, an sgRNA targeted for donor plasmid digestion, and Cas9 mRNA into one-cell stage wild-type embryos. A transgenic fish was outcrossed with a wild-type fish to produce the founder line. sgRNA sequences are described in Supplementary Table 1.
Luciferase assays. HEK293 cells were grown in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum and a 100 U/ml penicillin-streptomycin mixed solution. Cells were seeded into 35 mm-diameter plates and transfected with the expression plasmids and TP53 reporter plasmids using polyethylenimine MW 25000 (Polysciences, Warrington, PA, USA). After 24 h, firefly and Renilla luciferase activities were determined with the Promega Dual Luciferase Assay System (Promega, Madison, WI, USA). The pRL-EF vector, which expresses Renilla luciferase under the control of the EF-1α promoter, was used to normalize the transfection efficiency of the luciferase reporters. The values shown are the averages of experiments repeated three times, with each transfection performed in duplicate for each experiment.
For time-lapse confocal live imaging, larvae were manually dechorionated by forceps and mounted in 1% low melting agarose with egg water onto glass-bottom dishes. Live imaging was performed using an FV3000 confocal laser scanning microscope (Olympus, Tokyo, Japan) equipped with a 20 × 1.0 NA objective. Two laser lines, 488 and 561 nm were used. The recording interval was 20 min. At each time point, 40-50 confocal slices through the z-axis were acquired. Collected image data were processed using ImageJ.