Epithelial Tumors Originate in Tumor Hotspots, a Tissue-Intrinsic Microenvironment

Malignant tumors are caused by uncontrolled proliferation of transformed mutant cells that have lost the ability to maintain tissue integrity. Although a number of causative genetic backgrounds for tumor development have been discovered, the initial steps mutant cells take to escape tissue integrity and trigger tumorigenesis remain elusive. Here, we show through analysis of conserved neoplastic tumor-suppressor genes (nTSGs) in Drosophila wing imaginal disc epithelia that tumor initiation depends on tissue-intrinsic local cytoarchitectures, causing tumors to consistently originate in a specific region of the tissue. In this “tumor hotspot” where cells constitute a network of robust structures on their basal side, nTSG-deficient cells delaminate from the apical side of the epithelium and begin tumorigenic overgrowth by exploiting endogenous Janus kinase/signal transducer and activator of transcription (JAK/STAT) signaling activity. Conversely, in other regions, the “tumor coldspot” nTSG-deficient cells are extruded toward the basal side and undergo apoptosis. When the direction of delamination is reversed through suppression of RhoGEF2, an activator of the Rho family small GTPases, and JAK/STAT is activated ectopically in these coldspot nTSG-deficient cells, tumorigenesis is induced. These data indicate that two independent processes, apical delamination and JAK/STAT activation, are concurrently required for the initiation of nTSG-deficient-induced tumorigenesis. Given the conservation of the epithelial cytoarchitecture, tumorigenesis may be generally initiated from tumor hotspots by a similar mechanism.


Introduction
In epithelial tissues, cells communicate with their neighbors and receive information from the surrounding environment through signaling networks, adhesion molecules, and junctional molecules in order to form complex organs and maintain their integrity and morphology. This robust, self-organizing system, however, is progressively disrupted during tumor development. In tumorigenesis, transformed mutant cells evolve into a malignant neoplasm through a multistep process whereby the transformed cells acquire traits that enable them to become tumorigenic and ultimately malignant [1]. Although many genes have been identified as involved in different steps of cancer cell progression, little is known about the beginning of tumorigenesis, in which only small subsets of mutant protumor cells deviate from the robustly organized microenvironment to evolve into aggressive tumors.
Studies of a group of Drosophila tumor-suppressor genes, lethal giant larvae (lgl), discs large (dlg), and scribble (scrib), highlighted the critical relationship between loss of epithelial organization and tumor development. These genes play key roles in regulation of apical-basal cell polarity and cell proliferation in epithelial tissues [2]. In developing imaginal discs, epithelial tissues that are homozygous mutant for any of these three genes cause cells within the normally monolayered epithelia to lose structure and polarity, fail to differentiate, and overproliferate, thus becoming multilayered amorphous masses that fuse with adjacent tissues [2]. Similarly, loss or alteration in expression of these genes in mammals has been shown to be involved in development of malignant tumors [3,4]. The neoplastic phenotypes exhibited by the mutant tissues have led to the classification of these three genes as conserved neoplastic tumor-suppressor genes (nTSGs) [2,3]. However, when nTSG mutant fly cells are generated in otherwise wild-type epithelia using the FLP/FRT-mediated mitotic recombination technique, they undergo cell-competition-dependent apoptosis and are eliminated from the epithelial tissue [5][6][7][8]. This cell-competition-dependent elimination of nTSG-depleted cells has recently been confirmed in mammalian cells [9], suggesting cell competition is an evolutionarily conserved tissue-homeostatic mechanism that ensures elimination of pro-tumor cells and epithelial integration.
Here, we show that nTSG-deficient pro-tumor cells evade competitive pressure to take a first step toward evolving into aggressive tumors when they are located in a specific region within an epithelium and how this tissue-intrinsic local microenvironment has a decisive role for the life-or-death fate of pro-tumor cells.

nTSG-Knockdown Cells Undergo Site-Specific Tumor Growth in Wing Imaginal Discs
Through close examination of nTSG-deficient cells in Drosophila wing imaginal discs, we found RNA interference (RNAi)-induced silencing of a single nTSG caused tumorigenic overgrowth in specific areas of the disc (Fig 1A and 1B). At the same time, many of the nTSGknockdown cells adjacent to wild-type normal cells underwent cell-competition-induced apoptosis ( Fig 1C, S1A and S1B Fig), as has been previously shown [6][7][8]. When lgl was knocked down by ptc-Gal4-driven lgl-RNAi along the anterior-posterior (AP) boundary of developing wing imaginal discs, dysplastic overgrowth was induced in the dorsal hinge region where the epithelial sheet is folded (Fig 1B, 1C and 1D). In contrast, the ptc-Gal4-induced misexpression of a constitutively active form of Yorkie (YkiM123), a transcriptional co-activator of the Hippo tumor-suppressor pathway, did not cause dysplastic overgrowth and instead showed hyperplastic overgrowth throughout the entire region of misexpression (S1C Fig). This clear phenotypic difference between lgl knockdown and Yki over-activation can be explained by their functions in two different tumor suppressor pathways; nTSGs are involved in epithelial organization through maintenance of apical-basal polarity [3] and mitotic spindle orientation [10][11][12], while the Hippo hyperplastic tumor suppressor pathway is involved in regulation of cell proliferation and apoptosis [13]. However, the difference between the phenotypes induced by lgl knockdown in the dorsal hinge and other regions in the same epithelial tissue was unexpected.
We therefore hypothesized that specific regions in the imaginal epithelial tissue are intrinsically susceptible to tumorigenic stimuli. To test the hypothesis of site-specific tumorigenesis, we generated random clones of nTSG-knockdown cells using the heat-shock-activated flip-out Gal4 system [14]. The dysplasia induced by lgl or scrib knockdown became clear in the hinge region 4 d after heat-shock-induced RNAi expression, and dysplastic tumor growths in the region were clearly observed 7 d after RNAi induction (Fig 1E-1H). These tumors showed strongly up-regulated expression of both phosphorylated c-Jun N-terminal kinase (pJNK) and matrix metalloproteinase-1 (MMP1) (S2A and S2B Fig), indicating an invasive cellular behavior [15]. In the hinge area, the occurrence ratio of nTSG-knockdown-induced dysplastic cell mass (each cell mass was counted as a single event of tumorigenesis) was the highest in the dorsal hinge (Fig 1I). In the wing pouch region, however, these nTSG-knockdown cells did not show dysplastic tumor growth (0%, n = 85). Apoptosis was detected in some nTSG-knockdown cells in both the wing pouch and hinge regions of the mosaic wing disc 3 d after RNAi induction (Fig 1J and 1K). These apoptotic cells, detected by cleaved Dcp-1 staining, were mostly at the boundary between the nTSG-knockdown cells and wild-type cells (81.7%, n = 278 apoptotic lgl-knockdown cells), suggesting cell competition is involved in eliminating nTSG-knockdown cells, as described previously [6][7][8]. Despite cell-competition-induced apoptosis, surviving nTSG-knockdown cells showed dysplastic tumor growth in the hinge region but not the pouch area. These results suggest the tumorigenic potential of pro-tumor cells with apicalbasal polarity defects is dependent on their local environment in the epithelial tissue. We therefore refer to the wing pouch region where pro-tumor cells do not show dysplastic tumor growth as a "tumor coldspot," whereas the hinge region where pro-tumor cells induce tumorigenesis is referred to as a "tumor hotspot."

Endogenous JAK/STAT Activity Is Required to Induce Tumorigenesis of nTSG-Knockdown Cells
Tumorigenesis is normally associated with activation of growth-promoting signaling pathways [16], so we suspected the tumor hotspots in the wing disc might be related to local activation of these pathways. The Janus kinase/signal transducer and activator of transcription (JAK/STAT) pathway has been shown previously to be endogenously active in the hinge region of developing wing imaginal discs [17], and its dysregulation has been implicated in diverse types of human cancers [18] and in several tumor models in Drosophila [19,20]. Its secreted cytokine- like ligand Unpaired (Upd) [21], a Drosophila homolog of mammalian Interleukin-6 (IL-6), is endogenously expressed in dorsal, anterior lateral, and ventral posterior compartments of the hinge area [22]. The expression pattern of 10xSTAT92E-GFP, a JAK/STAT activity reporter, in wild-type tissues confirmed the endogenous activity of this pathway in these same tumor hotspots, particularly in the dorsal hinge region where its endogenous activity appears to be the highest (Fig 2A). The dorsal hinge has three epithelial folds: proximal, medial, and distal folds. The endogenous activity displayed by the JAK/STAT pathway is high in the medial fold, weak in the proximal fold, and barely detectable in the distal fold (Fig 2A and 2B) [23]. If tumorigenesis depends on JAK/STAT activity in the tumor hotspot, we predicted it would occur most obviously in the medial fold, where JAK/STAT activity is the highest. Indeed, the dysplastic tumor growth induced by lgl-or scrib-knockdown clones was mostly observed in the medial fold (Fig 2C and 2D). This site-specific tumorigenesis was further examined by suppressing nTSGs using different Gal4 drivers (act-Gal4, sd-Gal4), each of which has a broad expression pattern, including both the wing pouch and hinge regions in the wing disc ( Fig 2E, S3A and S3B Fig). We found that the dysplastic tumor growth induced by these different drivers occurred most frequently in the medial fold of the dorsal hinge (71.19%, n = 59 tumors in dorsal hinge), with strong up-regulation of 10xSTAT-GFP in the tumorous tissue ( Fig 2E). Conversely, no dysplastic tumor growth was detected in the wing pouch area where JAK/STAT signaling is inactive (0%, n = 59), and the nTSG-knockdown cells in this tumor coldspot did not show up-regulation of 10xSTAT-GFP ( Fig 2E and S3C Fig). These findings suggest a strong correlation between endogenous JAK/STAT activation and tumorigenesis induced by nTSG knockdown in the wing imaginal disc.
To determine whether JAK/STAT activation is required for tumor growth, we suppressed STAT92E, the Drosophila homolog of mammalian STAT3 and STAT5 [24,25], in lgl-RNAiexpressing mosaic discs by co-expression of STAT92E-RNAi. The mosaic wing disc bearing clones expressing STAT92E-RNAi alone did not show any obvious phenotype ( Fig 3A). Depletion of STAT92E in lgl-knockdown cells, however, blocked the dysplastic tumor growth ( Fig  3C), as has been shown in scrib and dlg mutant cells [26], indicating STAT activation is necessary for nTSG-knockdown-induced tumorigenesis. Next, we asked whether the activation of the JAK/STAT pathway is sufficient for the dysplastic tumor growth of nTSG-knockdown cells. To address this question, we ectopically activated JAK/STAT signaling in lgl-RNAiexpressing mosaic discs by co-expression of a constitutively active form of STAT92E (STA-T92E ΔNΔC ) [27]. First, we confirmed that ectopic expression of STAT92E ΔNΔC without lgl-or scrib-knockdown showed no dysplastic tumor growth in wing discs ( Fig 3B). However, overactivation of STAT in lgl-or scrib-knockdown cells in tumor hotspots, including the distal fold of the dorsal hinge, dramatically enhanced the tumor size (Fig 3D-3F) when compared with nTSG-knockdown-induced tumors without STAT over-activation from the same time point post-clone induction (Figs 1F and 3F). Three days after heat-shock induction, clones with coexpression of lgl-(or scrib-) RNAi and STAT92E ΔNΔC showed clear dysplasia in the hinge regions ( Fig 3D and 3E). This dysplasia developed into large tumor masses a day later ( Fig 3F). In contrast, nTSG-knockdown cells without STAT over-activation only started to display dysplasia 4 d after clone induction ( Fig 1F). To determine whether ectopic activation of STAT can transform a tumor coldspot into a hotspot, STAT92E ΔNΔC was co-expressed with lgl-(or scrib-) RNAi in the pouch region. Still, tumors were not found in this coldspot, where apoptotic cell death was prevalent ( Fig 3G). These results indicate that JAK/STAT activity is required yet not sufficient to induce tumorigenesis of nTSG-knockdown cells when they reside in tumor coldspots. Therefore, we hypothesized that additional factor(s) are responsible for the difference in tumorigenesis between coldspots and hotspots.

Tumor Hotspots Have Structurally Robust Cellular Organizations in the Basal Side
One key difference between coldspots and hotspots is the morphology of columnar epithelial cells composing the pseudostratified monolayer, such that cells in the wing pouch (coldspot) have a long and straight shape along their apical-basal axis, whereas cells in the valley-folded hinge regions (hotspot) are shorter ( Fig 4A). Therefore, we examined subcellular localization patterns of various proteins that function in cytoarchitectural construction of epithelial cells, which might serve to characterize the differences between morphologies. Measurement of proteins involved in the actin cytoskeleton (F-actin), components of the cell cortex (α-Spectrin and FasIII), apical complexes (aPKC), focal adhesion complexes (βPS-Integrin), adherens junctions (E-cadherin and Armadillo), and septate junctions (Dlg and Lgl) showed no significant differences in subcellular localization between the coldspot and hotspot (Fig 4A, 4B, and 4G). Interestingly, however, we found that intrinsic subcellular distribution of cortical microtubules (MTs), a cytoskeletal component crucial for the maintenance and modification of epithelial cell structures [28], showed distinct pattern differences between the coldspot and hotspot ( Fig  4A and 4C-4F). MTs were localized primarily at the apical side of epithelial cells in the coldspot (Fig 4E). By contrast, the hotspot area lacked this apical enrichment of MTs. Instead, MTs appeared enriched at the basal side ( Fig 4F).
Furthermore, in transmission electron micrographs (TEM), we found other striking differences in cytoarchitecture between the hotspot and coldspot. The basement membrane, a sheetlike form of extracellular matrix (ECM) underlying epithelial tissues, is composed of approximately ten thin laminae in the imaginal epithelia. These ECM laminae were loosely organized in the coldspot, whereas they were tightly aligned in the hotspot (Fig 5A, 5B, 5D and 5E). Another remarkable difference was also identified in the basal side of the epithelial layer; in the valley-folded hotspot regions, cellular membranes displayed a complicated set of bends at the basal side, whereas in the coldspot they appeared straight along the apical-basal axis (Fig 5A  and 5D). In addition, the TEM images revealed many circular-shaped double membrane structures at the basal side of the hotspot, but very few in the coldspot (Fig 5A and 5D). These circular membrane structures, which appeared similar to cross-sections of filopodial protrusions, varied in size (50-500 nm in radius) and had microtubules passing inside ( Fig 5E). Indeed, when single cells were marked by randomly expressed membrane-GFP, laterally-elongated filopodial protrusions (5-10 μm) were consistently detected at the basal side of cells in the hotspot (Fig 5F). The web of protrusions appeared intricately intertwined among neighboring cells. In contrast, only shorter filopodial protrusions (1-5μm) were found at the basal side of coldspot cells (Fig 5C). Taken together, these observations suggest that the basal side of the hotspot contains robustly organized membrane and ECM structures, which may prevent delamination of pro-tumor cells from its surface.

nTSG-Knockdown Cells Delaminate Apically and Undergo Tumor Growth in Hotspots
The aforementioned morphological observations led us to examine the early-phase behavior of pro-tumor cells in different regions of the wing disc through three-dimensional confocal imaging. In the wing pouch, the lgl-RNAi-or scrib-RNAi-expressing cells underwent apoptosis ( Fig  1J and 1K) and were extruded from the basal side of the epithelial layer 2 d after clone induction (Fig 6A-6C). Apoptotic nTSG-knockdown cells were similarly observed in the basal side of the folded hinge hotspot region, yet remained within the epithelial layer (Fig 6A, 6B and 6D). These results suggest that apoptotic pro-tumor cells, found in both coldspot and hotspot regions, tend to be extruded toward the basal side of epithelial sheets, likely by myosin-IIdependent pulling forces generated by apoptotic cells, as has been previously shown [29]. Consistent with this observation, sporadic wing-disc cells misexpressing a pro-apoptotic gene, Reaper, were also extruded toward the basal side in both coldspots and hotspots (S4 Fig). Interestingly, we found some surviving nTSG-knockdown cells in the hotspot delaminated from the apical side of the epithelial layer (Fig 6F). In contrast, no apical delamination of the Fig 4. Subcellular localization of microtubules shows distinct pattern differences between the coldspot and hotspot. (A) Vertical section along the AP boundary of a wing disc labeled with a basement membrane marker Collagen IV/Vkg-GFP (blue). The disc was also stained for F-nTSG-knockdown cells was detected in the coldspot (Fig 6E and 6H), consistent with the previous report that mitotic spindle misorientation caused by nTSG knockdown induces live-cell delamination from the basal side in the wing pouch. These basally delaminated cells then undergo apoptosis [10]. In the hotspot, by contrast, apically-delaminated cells appeared to have proliferated 5 d after clone induction, forming an amorphous mass of cells at the apical side (Fig 6G-6I). Indeed, the dysplastic tumor growth induced by nTSG knockdown always occurred at the apical side of the epithelial layer (100%, n = 127).

Both Apical Delamination and Endogenous JAK/STAT Activity Are Required for nTSG Depletion-Induced Tumorigenesis
As has been described above, the tumorigenic overgrowth was detected most frequently in the medial fold where JAK/STAT activity was high (Fig 2D), suggesting apically delaminating protumor cells are exposed to endogenous JAK/STAT signaling activity that is specifically activated in the medial fold for their survival and growth. Similar to reports in mammalian epithelial cells [30,31], Drosophila Upd is secreted from the apical surface of epithelial cells to transduce the signal to the neighboring cells, where it binds the receptor Domeless, which is also localized on the apical membrane [32]. Therefore, apical delamination in the valley-folded hotspot where the JAK/STAT ligand abundantly accumulates favors the survival and proliferation of the pro-tumor cells. If this hypothesis is true, we predicted that nTSG-knockdown cells delaminated apically would not undergo tumorigenesis in the coldspot where JAK/STAT signaling is not active. The guanine nucleotide exchange factor RhoGEF2, an activator of the Rho family small GTPases [33,34] has been implicated in apical constriction and the cell-shape change of epithelial cells during gastrulation [35,36], actomyosin contraction during blastoderm cellularization [37,38], and segmental groove formation during embryogenesis [39] in Drosophila. Furthermore, knockdown of p115 RhoGEF, a mammalian homolog of RhoGEF2, has been shown to disrupt the apically directed extrusion of apoptotic cells in cultured epithelial monolayers from human bronchia [40]. Therefore, we examined whether a defect in Rho-GEF2 activity could disturb the delamination direction of nTSG-knockdown cells in the wing imaginal disc. Using the en-Gal4 driver to knockdown RhoGEF2 in the posterior compartment of a wing disc (Fig 7A-7C), we found that both apical enrichment of F-actin and basal enrichment of βPS-integrin were decreased in the pouch cells (Fig 7D and 7E). Unexpectedly, Rho-GEF2 knockdown also reversed the subcellular localization of cortical MTs in the coldspot: MTs became enriched at the basal side, opposite to their normal apical localization in the same region ( Fig 7F). Moreover, the thickness of basement membranes visualized with a functional GFP-tagged Collagen IV (Vkg-GFP) was strongly reduced by RhoGEF2 knockdown in the coldspot (Fig 7G and 7H) where ECM laminae were loosely organized (Fig 5A and 5B). These observations suggest that RhoGEF2 knockdown induces hotspot-like cytoarchitectures in the coldspot area.
To determine whether RhoGEF2 knockdown has an effect on the delamination direction of nTSG-deficient cells in the disc, we co-knocked down RhoGEF2 in scrib-RNAi-or lgl-RNAiexpressing mosaic discs. Consequently, loss of RhoGEF2 resulted in apical delamination of both scrib-and lgl-knockdown cells in the coldspot (Fig 8A and 8C). Because nTSG-

actin (red) and α-tubulin (green). (B-C) Signal intensities of actin (B) and α-tubulin (C) staining in (A) plotted using Interactive 3-D Surface Plot. (D) Line drawings trace the apical (red) and basal (blue) sides of the epithelial layer. (E-F) Magnified images of the wing pouch region (E) and the folded dorsal hinge region (F) indicated in (B).
Left panels: merged images of F-actin (red) and α-tubulin (green) staining. Right panels: subcellular localization of α-tubulin (green). (G) Vertical sections along the AP boundary of wild-type wing discs were stained for the indicated proteins (white). Nuclei were labeled with DAPI (blue). Scale bars represent 10 μm.
doi:10.1371/journal.pbio.1002537.g004 knockdown alone resulted in only basal delamination in this same region (Figs 6H and 8A), we conclude that depletion of RhoGEF2 can disturb the delamination direction of nTSG-deficient cells. However, these apically delaminated scrib-knockdown cells did not show tumorigenic overgrowth in the lumen of coldspots 4 d after clone induction (Fig 8C). Since endogenous JAK/STAT signaling is required for tumorigenesis in the hotspot (Figs 2 and 3C), we asked whether ectopic activation of JAK/STAT signaling can induce tumorigenic overgrowth of nTSG-deficient cells delaminated apically at the coldspot. To this end, we misexpressed STA-T92E ΔNΔC together with scrib-RNAi and RhoGEF2-RNAi and found that scrib-knockdown cells with RhoGEF2 suppression and STAT ectopic-activation showed apical delamination and tumor growth in the lumen of coldspots 4 d after clone induction (Fig 8B and 8D). These tumors showed strongly up-regulated expression of MMP1 (Fig 8E), which was not detected in scrib-and RhoGEF2-co-knockdown cells located in the coldspot (Fig 8D). These results indicate that both apical delamination and JAK/STAT activity are required for nTSG depletioninduced tumorigenesis in this Drosophila epithelial tissue. These data suggest that the effects of each modification on nTSG-knockdown protumor cells are discrete and that collaboration of these two independent processes (apical delamination and JAK/STAT activation) is required for the initiation of nTSG-depletioninduced tumorigenesis. Considering that ectopic activation of STAT alone in basally extruded nTSG-knockdown cells did not induce tumorigenesis in the coldspot, apical delamination provides the pro-tumor cells with a crucial survival advantage. Apoptosis of pro-tumor cells, such as nTSG mutant cells, is known to be induced by Eiger, the Drosophila homolog of mammalian tumor necrosis factor (TNF)-α, which is produced by circulating hemocytes recruited to tumor tissues [41]. Because hemocytes directly associate with cells along the basal side of the epithelial layer [42], apical delamination could have prevented the pro-tumor cells from receiving the death signal.
Our data describe how two different causal factors are closely involved in initiation of nTSG defect-induced tumorigenesis in tumor hotspots: apical delamination of surviving pro-tumor cells and high levels of endogenous JAK/STAT activity. The different sensitivity to tumorigenic growth between the distal pouch and proximal hinge regions in Drosophila wing imaginal discs has been reported in some previous studies [8,43,44]. However, the tumors induced by lgl mutant clones in those studies were created under specific experimental conditions in which the competitive pressure was removed or attenuated by conferring oncogenic activity on lgl mutant clones via overexpression of oncogenic Ras V12 or Yki, or lowering competitiveness of the surrounding cells via the Minute technique. In contrast, the site-specific tumorigenesis phenotype reported here was induced by knockdown of an nTSG alone, revealing an important role played by the endogenous local microenvironment on determining where tumors can grow. The tumor hotspots display a network of robust basal structures, including a web of intertwining filopodia and tightly laminated basement membranes, which could force protumor cells to delaminate from the apical side and enter the lumen formed by the disc epithelium and peripodial membrane. In the disc lumen where JAK/STAT activity is high, such as the medial fold, tumorigenesis takes place in this niche-like environment. The cells in the medial fold of the dorsal hinge region also show intrinsic resistance to ionizing radiation (IR)-and drug-induced apoptosis and help tissue regeneration, further suggesting the intrinsic tumor susceptibility of this subpopulation [45]. A recent in vitro study using organotypic human mammary acini has shown apical translocation of single cells from the epithelial layer allows the oncogenic mutant cells to undergo luminal outgrowth [46]. In addition, chronic inflammation is linked to all stages of tumor development. It is well known that activation of STAT family members (particularly STAT3), which are closely linked to inflammatory processes in multiple tissues, is implicated in tumorigenesis, promoting proliferative and survival signaling [47]. In pathological histology, it is also known that tumors frequently arise in transitional zones where two different types of epithelial tissue meet, resulting in the appearance of a distinct abrupt transition, which can be found in numerous locations within various tissues [48]. For example, tumors in the anal canal were found to arise primarily in a transitional zone between stratified squamous epithelium of anal skin and mucosal epithelium of the large intestine [49]. Intriguingly, even in wild-type mice, the epithelium of the transitional zone in the anus intrinsically shows many features reminiscent of hyperproliferative epidermis including aberrant expression of differentiation markers, enhanced Ras-MAPK signaling, and locally increased inflammation [49,50]. Future studies will determine if the transitional zones in mammalian tissues have the tumor-hotspot-specific cytoarchitectures that we found in the Drosophila imaginal discs. Besides the transitional zones, however, it is also likely that tumor-hotspot microenvironments are easily formed by small alterations in epithelial cytoarchitectures. Given the widely conserved cytoarchitectures of metazoan epithelia, disruption of tissue organization initiated from intrinsic tumor hotspots could explain a general mechanism of tumorigenesis.

Immunohistochemistry and Image Analysis
Immunofluorescent stainings of imaginal discs were performed according to standard procedures for confocal microscopy as described previously [52]. The following antibodies were

Transmission Electron Microscopy
The sample preparations were performed according to standard procedures for TEM as described previously [54]. Dissected imaginal discs were fixed in a solution of 2.5% glutaraldehyde and 2% paraformaldehyde buffered with 0.1 M sodium cacodylate to pH7.4. They were post-fixed with 1% osmium tetroxide in the same buffer, stained en bloc with 0.5% uranyl acetate in distilled water, dehydrated in ethanol, and embedded in Epon. Ultra-thin sections (70 nm) were stained with 2% uranyl acetate solution and Reynold's lead citrate solution. Images were obtained with a VELETA CCD Camera (Olympus Soft Imaging Solutions) mounted on a JEM 1010 transmission electron microscope (JEOL).

Quantification of Tumorigenesis Occurrence Ratio
Tumorigenesis was induced by random expression of nTSG-RNAi using the heat-shock activated flip-out Gal4-UAS system, and the occurrence ratio of tumorigenesis was counted 4 to 7 d after clone induction. Each occurrence of tumorigenesis includes a dysplastic cell mass deviating from the disc-proper epithelial layer with a diameter larger than four cells. The percentages of tumor growths were quantified from more than three independent experiments for each genotype. The numerical data used in Fig 8A and 8B are included in S1 Data.

Statistical Analysis
Two-tailed unpaired t tests assuming equal variances were performed for all statistical analyses. p-Value < 0.001 was considered statistically significant for all analyses.
Supporting Information S1 Data. Excel spreadsheet containing, in separate sheets, the underlying numerical data for Fig 8A and 8B.
Funding acquisition: YT ES WMD.