The actin cross-linker Filamin/Cheerio mediates tumor malignancy downstream of JNK signaling

Establishment of a malignant cell population within formerly healthy tissue involves reprogramming of intracellular signaling circuits as well as alterations of surrounding tissue, permitting cancer cells to survive, proliferate and disseminate (Friedl and Alexander, 2011; Hanahan and Weinberg, 2011). Cell growth and division, cell shape maintenance and cell motility all rely on actin cytoskeleton remodeling in normal as well as in cancer cells (Nürnberg et al., 2011; Pollard and Cooper, 2009; Richardson, 2011). Actin dynamics depends on a plethora of signaling pathways and actin-binding proteins that mediate actin filament growth, bundling, cross-linking and disassembly. Rho GTPases (Rho, Rac, Cdc42) are well known to direct actin filament growth by regulating actin nucleators, the actin related protein-2/3 (Arp2/3) complex and formins (diaphanous-related formins and formin-like proteins), through WAVE/Scar and WASP proteins (Nürnberg et al., 2011) (Fig. 1A). Rho GTPases, cytokines, growth factors or environmental stress can influence actin dynamics through the Jun N-terminal kinase (JNK) pathway (Brumby et al., 2011; Coso et al., 1995; Marinissen et al., 2004; Minden et al., 1995), which modulates expression of specific actin regulatory genes and phosphorylates cytoskeletal components (Xia and Karin, 2004). Aberrant JNK activity has been associated with various tumor types (Wagner and Nebreda, 2009). However, due to the extreme complexity of mammalian tumors, a direct link between JNK, actin, and actual tumor phenotypes has been missing.

The epithelial tumor model that has been established in the fruit fly Drosophila (Brumby and Richardson, 2003; Pagliarini and Xu, 2003) offers a relevant yet genetically tractable system where to investigate the contribution of actin dynamics to tumor malignancy. In clones of a columnar epithelium (CE) and a squamous peripodial epithelium (PE) of eye/antennal imaginal disc (EAD), overexpression of an activated form of Ras (Ras^V12) induces malignant tumors when combined with disruption of the apico-basal polarity, caused by mutation in one of the tumor suppressors such as Scribble (Scrib), Discs large (Dlg) or Lethal giant larvae (Lgl) (Brumby and Richardson, 2003; Pagliarini and Xu, 2003). These lesions recapitulate hallmarks of mammalian tumors, as the mutant cells overproliferate, resist apoptosis, induce immune response and invade surrounding tissue, ultimately killing the fly larva (Brumby and Richardson, 2003; Cordero et al., 2010; Pagliarini and Xu, 2003; Pastor-Pareja et al., 2008). In this model, activation of JNK together with aberrant Hippo signaling contributes to tumor overgrowth (Doggett et al., 2011; Igaki et al., 2006). At the same time, JNK promotes cell migration and invasion, partly by upregulating expression of matrix metalloprotease 1 (MMP1) (Srivastava et al., 2007; Uhlirova and Bohmann, 2006). The invasive ras^V12scrib^−/− tumor cells display prominent cortical actin filaments (Leong et al., 2009; Uhlirova and Bohmann, 2006). Interestingly, forced activation of Rho GTPases in cells carrying oncogenic Ras causes a phenotype that resembles the invasive behavior of the ras^V12scrib^−/− clones (Brumby et al., 2011), thus linking the perturbed actin cytoskeleton dynamics with tumor malignancy.

Our present study discovers an important function of the actin cross-linking protein Cheerio (Cher), a Drosophila ortholog of mammalian Filamin A (FLNA) (Sokol and Cooley, 1999), in tumor progression. Filamins are large cytoplasmic proteins whose dimers provide cells with mechanical resilience by cross-linking cortical actin filaments into a dynamic three-dimensional structure. In addition to their role as mechanical linkers, filamins interact with transmembrane receptors, adhesion molecules and even transcription factors, serving as a scaffold during signal transduction (Nakamura et al., 2011; Popowicz et al., 2006). The role of FLNA in tumorigenesis remains controversial. Although downregulation of FLNA expression increases the invasiveness of human breast cancer cells (Cunningham et al., 1992; Xu et al., 2010), a positive effect of FLNA on cell migration has also been reported (Castoria et al., 2011; Cunningham et al., 1992). FLNA expression levels in tumors and plasma of patients with breast and primary brain tumors increase with tumor malignancy (Ai et al., 2011; Alper et al., 2009; Bedolla et al., 2009). Finally, a most recent study has demonstrated that FLNA knockout restrains development of K-Ras induced lung tumors in mice (Nallapalli et al., 2012).

Here, we show that Drosophila Filamin/Cher is upregulated in malignant ras^V12scrib¹ tumors in a JNK-dependent manner, and that Cher function is required for tumor growth and invasiveness. In the natural tissue context Cher endows the tumor cells with the lethal capacity to breach the barrier of the PE and invade neighboring organs.

To explore the role the actin filament network plays in tumor malignancy, we searched for actin cytoskeleton-regulating genes that would show altered expression in the eye/antennal imaginal discs (EAD) carrying invasive ras^V12scrib¹ clonal tumors (Brumby and Richardson, 2003; Pagliarini and Xu, 2003). Functions of candidate proteins in actin dynamics are depicted in Fig. 1A. The previously established enrichment of matrix metalloprotease 1 (mmp1) (Uhlirova and Bohmann, 2006) and downregulation of the protease inhibitor Reck mRNAs (Srivastava et al., 2007) in ras^V12scrib¹ tumors served for reference (Fig. 1B). We found increased levels of diaphanous (dia) and cher transcripts in ras^V12scrib¹ tumors compared with EAD bearing control clones; expression of the other candidates did not change appreciably (Fig. 1B). A role for diaphanous-like formins in cancer progression is supported by previous studies in mammalian models (Nürnberg et al., 2011), and effects of Drosophila Dia on cell motility (Liu et al., 2010), and regulation of imaginal disc growth (Sansores-Garcia et al., 2011) have been established. We therefore concentrated on the less explored function of the Filamin A (FLNA) ortholog Cher. Using quantitative RT-PCR (qRT-PCR), we confirmed that cher mRNA was significantly upregulated in aggressive ras^V12scrib¹ tumors (Fig. 1C). Importantly, cher levels correlated with tumor invasiveness as overgrown but noninvasive tumors, in which JNK activity was inhibited by expression of a dominant-negative form of the Jun kinase Basket (Bsk^DN), displayed control levels of cher mRNA (Fig. 1C).

The EAD consist of a basal CE and an apical PE that face each other with their apical sides (Pallavi and Shashidhara, 2005) and are surrounded by the basal membrane. In addition, circulating blood cells, hemocytes, populate the outer surface of the EAD (Holz et al., 2003). Hemocyte numbers have been shown to increase dramatically in the presence of ras^V12scrib¹ clonal tumors, contributing to their invasiveness (Cordero et al., 2010; Pastor-Pareja et al., 2008). To determine which cell type accumulated Cher, we stained EAD bearing control, ras^V12scrib¹ or ras^V12scrib¹bsk^DN clones with Cher, H2 and Fasciclin III (FasIII) antibodies. Whereas H2 marks hemocytes (Kurucz et al., 2003), FasIII decorates the lateral membranes of both PE and CE epithelia, thus revealing overall EAD morphology.

Control Cher staining was restricted to the apical part of the EAD, marking the PE monolayer (Fig. 2A). Its distribution throughout the EAD was even, with a slight enrichment in the antennal part of the disc. In contrast, Cher expression increased in numerous ras^V12scrib¹ cells located within the PE (Fig. 2B). Disturbed pattern of Cher and FasIII staining indicated that already in early tumors, observed 5 days after egg laying (AEL), the PE morphology became compromised with cells piling atop each other (Fig. 2B″). As clonal tumors continued to grow (7 days AEL) and generate multilayered cell masses, the Cher expression domain expanded basally and the integrity of both the PE and CE deteriorated (supplementary material Fig. S1A). Cher showed a similar pattern of enrichment in invasive ras^V12dlg¹ tumors (supplementary material Fig. S1B).

As expected, few hemocytes attached to control EAD or those carrying noninvasive ras^V12scrib¹bsk^DN tumors (Fig. 2A,C), whereas invasive ras^V12scrib¹ or ras^V12dlg¹ tumors attracted great numbers of hemocytes (Fig. 2B, supplementary material Fig. S1B). However, the hemocytes expressed Cher regardless of the tumor genotype (Fig. 2). Therefore, both the PE and the associated hemocytes contributed to normal Cher expression within the EAD and to its increase upon induction of ras^V12scrib¹ or ras^V12dlg¹ tumor clones. However, although Cher from the hemocytes accumulated due to higher hemocyte population, elevated cher expression in the clonal tumors required JNK activity.

Inhibition of JNK signaling, achieved either by expressing Bsk^DN or removing the transcription factor Fos (encoded by kayak), reduced the Cher signal within ras^V12scrib¹bsk^DN (Fig. 2C) and ras^V12scrib¹kay³ (supplementary material Fig. S1C) clones, respectively. The reduction was to levels seen in the surrounding wild-type tissue, suggesting that basal cher expression in EAD did not depend on JNK. To test whether cher might be activated by JNK signaling directly, we examined four putative AP-1 sites that occur within 1.4 kb upstream of the first untranslated cher exon (Fig. 1D). AP-1 motifs are recognized by Activating Protein 1, a dimer of bZIP proteins such as Jun and Fos, acting downstream of JNK (Kockel et al., 2001). Wild-type and mutated versions of the 1.4-kb region were tested for transcriptional activity in Drosophila S2 cells that possess a functional AP-1 complex, as evidenced by activation of a TPA-dependent TRE::RFP reporter (Fig. 1E) (Chatterjee and Bohmann, 2012). Consistent with the presence of endogenous Cher in S2 cells (Fig. 1F), we observed high activity of the cher^wt::mRFP reporter (Fig. 1G). When all four AP-1 sites were mutated (cher^mut::mRFP) the expression was lost (Fig. 1H), indicating that the AP-1 motifs were essential.

To test whether activation of JNK signaling was sufficient to stimulate cher transcription, we overexpressed the JNKK Hemipterous (Hep^wt) in third-instar larvae using the conditional TARGET system, which enables heat-inducible ubiquitous expression of UAS-driven transgenes (McGuire et al., 2003). Quantitative RT-PCR showed that expression of the known JNK target gene puckered (Martín-Blanco et al., 1998) and of cher was induced in response to the temporary JNK activation in vivo (supplementary material Fig. S2). Together, the results show that the JNK pathway is necessary and sufficient for activating cher transcription above its basal level.

The strong JNK-dependent upregulation of Cher expression in ras^V12scrib¹ tumors raised the question whether Cher might be required for JNK-mediated tumor growth and invasiveness (Igaki et al., 2006; Uhlirova and Bohmann, 2006; Leong et al., 2009; Brumby et al., 2011). To this end, we utilized a UAS-cher^IR (RNAi) transgenic fly line and a previously described protein-null allele cher¹ (Robinson et al., 1997; Sokol and Cooley, 1999), and tested their ability to modify tumor phenotypes. Removing Cher using either method in mitotic clones induced within the developing EAD had no impact on clone number or morphology, and resulted in adults with regular eyes (supplementary material Fig. S3A–C). Cher is therefore dispensable for normal eye development.

However, loss of cher markedly interfered with the growth of ras^V12scrib¹ tumors. On day 7 AEL, the size of ras^V12scrib¹ clones increased dramatically at the expense of surrounding wild-type tissue in both the eye and antenna parts of the EAD (Fig. 3A). The tumor mass became strongly reduced in Cher-deficient ras^V12scrib¹cher¹ clones, mainly within the antenna part (Fig. 3B; supplementary material Fig. S4). FACS sorting of cells from EAD dissected on days 6 and 7 AEL confirmed that the apparent decline in number of GFP-positive cells in ras^V12scrib¹cher¹ EAD clones was indeed significant (Fig. 3C).

To discern whether the absence of cher function impaired tumor cell viability or proliferation, we quantified cells positively staining with apoptotic (activated caspase 3) and mitotic (phospho-histone 3, pH 3) markers in clonal (GFP⁺) and surrounding (GFP⁻) EAD tissue upon FACS sorting (Fig. 3D). Although caspase-3-positive cells occurred in EAD harboring both ras^V12scrib¹ and ras^V12scrib¹cher¹ clones, particularly at the clonal boundaries (supplementary material Fig. S4A,B), the loss of Cher caused no change in the incidence of apoptosis in either the clonal or the surrounding EAD cell populations (Fig. 3D). Therefore, Cher deficiency did not compromise tumor cell viability. Normally developing eye discs show two rows of mitotic cells along the morphogenetic furrow, and asynchronously dividing cells anterior to the furrow (supplementary material Fig. S5A). The presence of overgrowing ras^V12scrib¹ clones doubled the incidence of pH 3-positive cells predominantly within the clonal (GFP⁺) tissue (Fig. 3D; supplementary material Fig. S5B), and this increase required Cher (Fig. 3D). Interestingly, EAD carrying ras^V12scrib¹cher¹ clones also displayed extra cell divisions which, however, mainly occurred within the non-clonal surrounding tissue (Fig. 3D; supplementary material Fig. S5C). Based on these data we conclude that Cher is cell-autonomously required for proliferation of ras^V12scrib¹ tumors but not for their survival.

Although the absence of cher function limits tumor cell proliferation, it promotes cell differentiation within the EAD clonal tumors. Scarce Elav-positive cells confirmed that most ras^V12scrib¹ cells failed to differentiate into neurons (Fig. 3A) (Brumby and Richardson, 2003; Uhlirova et al., 2005). Despite sizable ras^V12scrib¹cher¹ clones located posterior to the morphogenetic furrow, the number of Elav-positive differentiating photoreceptors increased and their organization improved relative to ras^V12scrib¹ clones (Fig. 3A,B). Consequently, well-ordered arrays of ommatidia appeared in ras^V12scrib¹cher¹ EAD compared with the indiscernible pattern in ras^V12scrib¹ discs (supplementary material Fig. S5B,C). Further, removal of Cher from ras^V12scrib¹ clones improved expression of the roughoid and sevenless genes that promote photoreceptor specification (Fig. 3E). Expression of both genes was similarly restored in noninvasive ras^V12scrib¹bsk^DN clones (Fig. 3E) that displayed reduced cher activity (Fig. 1C; Fig. 2C).

A closer examination of EAD uncovered differences in morphology and distribution of ras^V12scrib¹ versus ras^V12scrib¹cher¹ clones. The former were found within both the PE and CE (Fig. 2B″; Fig. 4C; supplementary material Fig. S1A). Many apically localized ras^V12scrib¹ cells were elongated and besides strong Cher expression they also showed upregulation of MMP1 and massively enriched cortical actin fibers (Fig. 4A; supplementary material Fig. S6A–C). In contrast, cross-sections of ras^V12scrib¹cher¹ EAD revealed that large tumors only resided within the basal compartment of the CE (Fig. 4D). Smaller ras^V12scrib¹cher¹ clones were confined between the PE and CE (Fig. 4D); their round, cyst-like shape was highlighted by cortical actin (Fig. 4B). Finally, only few Cher-deficient clones found within the PE were often surrounded by non-clonal cells extending long, Cher-rich protrusions between and over the clonal cells (Fig. 4E,F). Consequently, the PE appeared largely intact (Fig. 4D). The infrequent ras^V12scrib¹cher¹ apical clones and those enclosed within the epithelia still expressed MMP1 (supplementary material Fig. S6D–F). mmp1 mRNA was only slightly lowered in EAD bearing ras^V12scrib¹cher¹ relative to ras^V12scrib¹ tumors (supplementary material Fig. S6G), indicating that Cher was dispensable for JNK-mediated MMP1 expression.

The reduced number of apical clonal tumors and improved PE integrity suggested that removal of Cher might curb tumor spreading, as fewer MMP1-producing cells were in the position to attack the basal membrane. Indeed, we found that in contrast to highly invasive ras^V12scrib¹ clonal tumors, fewer ras^V12scrib¹cher¹ clones invaded the brain and the ventral nerve cord (Fig. 5A,B). Unbiased estimation of tumor grade (see Materials and Methods) showed that invasive behavior of ras^V12scrib¹cher¹ tumors was significantly reduced (P≤0.001) compared with cher⁺ clones. Suppression of tumor invasiveness was also achieved by Cher depletion in ras^V12scrib¹ clones using RNAi (P≤0.001) (Fig. 5C).

Strikingly, elimination of Cher from ras^V12scrib¹ clones improved animal viability. Larvae bearing ras^V12scrib¹ tumors all die in the third instar and never pupate (Pagliarini and Xu, 2003; Uhlirova and Bohmann, 2006; Brumby and Richardson, 2003). In contrast, on day 8–9 AEL, 30% of ras^V12scrib¹cher¹ larvae began to wander and pupate (Fig. 5D), albeit 2–3 days later than control animals. Thirty percent of ras^V12scrib¹cher¹ pupae then emerged as adults. Interestingly, these flies had abnormally enlarged eyes with folded surface, accommodating surplus ommatidia (Fig. 5E–G). Consistent with impaired proliferation of clonal cells in EAD of third-instar larvae, the adult eyes contained only few GFP⁺ cells (Fig. 5F′), suggesting that the superfluous adult tissue most likely originated from extra mitotic activity that was observed outside the mutant clones (Fig. 3D; supplementary material Fig. S5C).

Processes affected by Cher deficiency in the tumor cells, including changes in cell shape, proliferation and motility, likely involve actomyosin contractions. It has been well established that epithelial tumors exhibit increased cell-generated forces through actomyosin contractility that enhance their growth, survival and invasiveness (Butcher et al., 2009). It is therefore plausible that the increased Cher levels provide ras^V12scrib¹ tumors with the advantage of extra-resilient actomyosin network. Cher has been previously co-localized with nonmuscle myosin II (MyoII) heavy chain, encoded by the zipper (zip) locus, to the cleavage furrow of dividing Drosophila cells (Field and Alberts, 1995). Our immunostaining of EAD bearing control and ras^V12scrib¹ clones revealed overlapping patterns of Cher and Zip proteins (Fig. 6A,B). However, ras^V12scrib¹ tumors showed enrichment of both Cher and Zip relative to the surrounding non-clonal tissue (Fig. 6B).

These results prompted us to test whether Cher and Zip form a complex. Indeed, we were able to co-precipitate Cher with Zip in lysates from EAD bearing control clones or ras^V12scrib¹ tumors, the latter showing apparently higher amounts of the individual proteins and of their complex (Fig. 6E). Besides enrichment for Cher and Zip, ras^V12scrib¹ clones also accumulated phosphorylated myosin II regulatory light chain, a product of the spaghetti squash gene (sqh) (Karess et al., 1991) that marks sites of active actomyosin contraction (Fig. 6C). Strikingly, no such enhancement of p-Sqh relative to the non-clonal tissue was found in ras^V12scrib¹cher¹ clones (Fig. 6D), although these clones still contained increased levels of the myosin II heavy chain Zip (supplementary material Fig. S7). Therefore, loss of Cher appears to affect MyoII activity.

To address the importance of the Cher-Zip interaction for malignancy of ras^V12scrib¹ cells, we removed Zip from the tumor clones using an RNAi UAS-zip^IR transgene. Depletion of Zip coincided with decreased p-Sqh levels whereas Cher and cortical actin remained enriched within ras^V12scrib¹zip^IR clones (Fig. 6F,G). Importantly, reduced occurrence of Zip lessened tumor invasiveness and improved pupation rate to an extent similar to the effect of Cher deficiency (Fig. 5C,D). Taken together, these data indicate that enrichment of the mutually interacting Cher and Zip proteins and enhanced MyoII contractile activity promotes growth and invasiveness of ras^V12scrib¹ tumor clones.

Recent studies in Drosophila show that increased mechanical tension promotes tissue overgrowth via inhibition of Hippo (Hpo) signaling (Fernández et al., 2011; Sansores-Garcia et al., 2011). Suppression of Hpo allows the transcriptional co-activator Yorkie (Yki) to turn on genes such as expanded (ex) and dally (Baena-Lopez et al., 2008; Hamaratoglu et al., 2006). Enhanced Yki activity has been previously detected in ras^V12scrib¹ clones (Doggett et al., 2011). Our data indicated that Cher may reinforce mechanical tension within the tumor tissue by interacting with the actomyosin network. We therefore tested whether expression of Yki targets in ras^V12scrib¹ clones might require Cher. Indeed, we found that the activity of an expanded reporter (ex::lacZ) (Boedigheimer and Laughon, 1993) as well as the levels of ex and dally mRNAs were clearly lower in EAD carrying Cher-deficient clones relative to those with ras^V12scrib¹ clones (Fig. 7). Immunostaining with an anti-β-Gal antibody showed that upregulated ex::lacZ expression colocalized with ras^V12scrib¹ clones and that cher deficiency resulted in cell-autonomous reduction of the reporter activity (Fig. 7F,G). Consistent with the previous report (Doggett et al., 2011), inhibition of Hpo signaling in ras^V12scrib¹ mosaic EAD appeared independent of JNK (Fig. 7C,E). Thus, Cher is necessary in ras^V12scrib¹ tumors to suppress the output from the Hpo pathway. These data suggest a possible link between Cher-dependent mechanical tension and attenuation of Hpo signaling in the tumor context.

Although actin cytoskeleton has been primarily linked with cell migration of normal as well as cancer cells, roles of the actin network in numerous other processes have emerged. The involvement of the actin cross-linking protein FLNA in cancer development has been suggested but poorly defined, as FLNA expression can correlate both positively and negatively with malignant tumor progression (Alper et al., 2009; Xu et al., 2010). Latest data from a mouse model support a tumor-promoting role of FLNA in K-Ras induced lung cancer (Nallapalli et al., 2012). In this study we demonstrate the contribution of the Drosophila FLNA ortholog, Cher, to aggressive growth and invasiveness of malignant clonal tumors.

We propose a model for Cher function in tumor malignancy (Fig. 8). In the presence of Cher, ras^V12scrib¹ clones form tumors that rapidly proliferate within both the PE and CE. JNK-dependent upregulation of MMP1 activity in apical (PE) clones and increased mechanical force generated by the quickly expanding tumor mass cooperatively destroy the basement membrane barrier, allowing tumor cells to depart from the disc and invade surrounding organs. In addition, destruction of the basement membrane and secretion of JAK/STAT activating cytokines induce proliferation of circulating hemocytes (Pastor-Pareja et al., 2008), which further enhance tumor growth and spreading (Cordero et al., 2010). Enrichment in the cortical actin filaments and the Cher complex with MyoII provide the tumor cells with resilience sufficient to withstand increased mechanical stress and with extra contractility to promote their proliferation and invasion. Conversely, removal of Cher from ras^V12scrib¹ clones slows their proliferation down, particularly within the PE, so that fewer MMP1-producing cells directly face the basement membrane and may attack its integrity. Consequently the reduced tumor mass remains contained within the largely intact PE, which therefore presents a physical barrier to the tumors.

Interestingly, tumors but not normal eye/antennal imaginal epithelia require Cher function. We show that in the EAD of third-instar larvae, Cher protein localizes primarily to the PE, a monolayer of polarized cells that lie between the CE and the basal membrane. Consistent with the absence of an externally visible eye phenotype in cher mutant flies (Li et al., 1999; Sokol and Cooley, 1999), loss of Cher from EAD clones had no obvious effect (supplementary material Fig. S3). Cher function, however, becomes needed in the context of rapidly growing and invasive tumors. We found that similar to MMP1 (Srivastava et al., 2007; Uhlirova and Bohmann, 2006), Cher was upregulated in EAD carrying ras^V12scrib¹ clones in JNK and Fos dependent manner. The enhanced Cher expression originated from tumor cells within the PE, yet with tumors growing into undifferentiated multilayered masses, its expression expanded also basally. In this regard Cher expression pattern highly correlated with that of MMP1 (Uhlirova and Bohmann, 2006; supplementary material Fig. S6B,C).

Our genetic analysis clearly demonstrated the cell-autonomous requirement of Cher in the ras^V12scrib¹ tumors as clonal-specific cher removal suppressed aggressive tumor growth by reducing tumor cell proliferation, and it also suppressed tumor invasiveness while improving cell differentiation. Interestingly, in accordance with the pattern of Cher protein expression, ras^V12scrib¹ tumors originating in the PE seemed to be much more sensitive to Cher deficiency than those in the CE. We based this conclusion on the fact that we consistently found only few rather small clones within the PE of EAD carrying ras^V12scrib¹cher¹ tumors while numerous smaller clones were enclosed between the two epithelial layers. Our observation of filopodia-like protrusions emanating from the non-clonal cells over the ras^V12scrib¹cher¹ cells suggests that even clones originated in the PE may be forced out from the PE basally towards the CE. Sizable clones were only found at the base of the CE. Although these ras^V12scrib¹cher¹ clones grew in time, their proliferation rate was slower relative to aggressive ras^V12scrib¹ clones, thus permitting some animals to complete development. Even when the ras^V12scrib¹cher¹ clones reached a size of ras^V12scrib¹ tumors (in larvae that failed to pupate), their invasive potential was reduced.

Marked restoration of neuronal differentiation in EAD carrying ras^V12scrib¹cher¹ clones suggests a non-autonomous effect for Cher on cells within the CE. Several studies postulated a functional involvement of PE cells in disc patterning apart from their function in providing mechanical support during disc eversion (Cho et al., 2000; Gibson and Schubiger, 2000; Pallavi and Shashidhara, 2003). We speculate that disturbed integrity of PE epithelia in EAD containing ras^V12scrib¹ tumors might interfere with transmission of inductive signals from PE to the disc proper, thus preventing photoreceptor differentiation. The largely intact PE in EAD bearing ras^V12scrib¹cher¹ clones may still provide sufficient signal required for proper patterning. Finally, however, we cannot exclude an alternative that Cher participates in dedifferentiation of ras^V12scrib¹ cells.

Deficiency of Cher alters the shape of tumor cells. Although ras^V12scrib¹cher¹ cells enriched F-actin to an extent similar to invasive ras^V12scrib¹ cells, their shape was markedly different. Regardless of their position within the EAD, most of the ras^V12scrib¹cher¹ clonal tumors assumed round morphology, contrasting with the elongated ras^V12scrib¹ cells, and separated themselves from the surrounding wild-type tissue, in some cases resulting in their extrusion from the host epithelia. Work in mammalian tissue cultures and genetic studies in Drosophila provided evidence that even modest differences in adhesive properties guide cell sorting and segregation (Dahmann et al., 2011; Steinberg and Takeichi, 1994; Wei et al., 2005). Interestingly, endothelial cells isolated from Flna^−/− mice showed unaltered F-actin levels and motility but displayed abnormal adherens junctions and reduced levels of VE-cadherin (Feng et al., 2006). Thus, it is plausible that cell shape changes and clonal extrusion from epithelial sheets occurring in the absence of Cher result from altered adhesive properties.

However, more recent hypotheses and experiments implicate differential mechanical tension along the cell borders as the driving force of cell sorting (Landsberg et al., 2009; Monier et al., 2010). Cellular tension is generated locally by myosin motors pulling against the actin filaments, and can act over a distance due to assembly of actin filaments into three-dimensional networks, mediated by cross-linkers such as Filamins (Gorlin et al., 1990). It has been shown that FLNA can potentiate actomyosin ATPase activity in vitro (Sosiński et al., 1984) and that it binds regulators of myosin-mediated contractility such as the small Rho GTPases (Rho, Rac, Cdc42) or the ROCK kinase (Nakamura et al., 2011). We have shown here that Cher binds Zip, the Drosophila ortholog of the nonmuscle myosin II heavy chain, and that the Cher-Zip complex is enriched in ras^V12scrib¹ clonal tumors. Simultaneous upregulation of p-Sqh indicates enhanced myosin ATPase and motor activity, and hence increased actomyosin contractility of tumor cells. Through its cross-linking action on the actomyosin cytoskeleton, Cher might therefore enable the tumors to maintain cell shape, proliferate and migrate even in an environment of increased mechanical stress, generated by the expanding tumor mass within the host epithelium. Thus, as a non-exclusive alternative to distorted cell adhesion discussed above, we propose that Cher deficiency might compromise tumor cell shape stability, proliferation and spreading by reducing cell surface tension. In this regard, it is noteworthy that human melanoma cells lacking FLNA are deformed and display blebbing (Cunningham, 1995; Cunningham et al., 1992).

We have identified cher as a target of JNK signaling, which is required for overgrowth and invasiveness of ras^V12scrib¹ clonal cells. It was therefore not surprising that loss of Cher affected invasive ras^V12scrib¹ tumors in a way similar to inhibited JNK activity, namely by limiting tumor growth, suppressing tumor invasiveness, and restoring cell differentiation (Igaki et al., 2006; Leong et al., 2009; Uhlirova and Bohmann, 2006). However, the impact of removed cher function on reduced malignity of ras^V12scrib¹ tumors exceeded the effect of compromised JNK activity. This was reflected by survival to adulthood of animals bearing ras^V12scrib¹cher¹ clones but not ras^V12scrib¹bsk^DN clones. One explanation for the stronger effect of cher mutation is that whereas a basal uninduced level of Cher expression persisted in the presence of the dominant-negative form of Bsk, the cher¹ allele caused a complete loss of function. Alternatively, Cher could promote the growth of ras^V12scrib¹ tumors through other mechanisms.

The loss of scrib has been recently shown to contribute to the overgrowth of the ras^V12scrib¹ clonal tissue independently of JNK, by inhibiting the tumor-suppressor Hpo pathway (Doggett et al., 2011). A link between actin cytoskeleton dynamics and Hpo signaling has been supported by studies in Drosophila and mammalian cell culture systems (Fernández et al., 2011; Sansores-Garcia et al., 2011; Wada et al., 2011; Zhao et al., 2012). Furthermore, manipulating filamentous actin in developing imaginal discs results in Yki-dependent tissue overgrowth (Fernández et al., 2011; Sansores-Garcia et al., 2011). Here we have demonstrated that the actin cross-linker Cher promotes tumor growth, and based on Cher-dependent deregulation of two Yki targets, expanded and dally, we speculate that Cher does so at least in part by modulating the Hpo/Yki pathway. Clearly, the exact mechanism how Filamin/Cher interacts with this pathway awaits elucidation. It is conceivable that some components of Hpo/Yki signaling sense mechanical cues effected by Cher on the actomyosin network. Our results hint that it might be the tension generated by myosin motor activity driving tumor proliferation of ras^V12scrib¹ tumors. This would agree with a recent discovery that activity of mammalian Yki homologues YAP and TAZ requires stress fibers and Rho (Dupont et al., 2011). Research in this direction can elucidate how growth-promoting signaling pathways hijack the cytoskeletal network during tumorigenesis.

AP-1 binding sites in the Drosophila cher gene were predicted using the Transcription Element Search System (TESS) (Schug, 2008) (http://www.cbil.upenn.edu/cgi-bin/tess). For construction of cher^wt::RFP, TPA DNA response elements in TRE::RFP vector (Chatterjee and Bohmann, 2012) were replaced with a 1.4-kb sequence upstream of the first cher exon, amplified using SphI-cher-prom for and XhoI-cher-prom rev primers (supplementary material Table S1). cher^mut::RFP was prepared by mutating all four AP-1 binding sites within the cher^wt::RFP plasmid using the QuikChange Lightning Multi Site-Directed Mutagenesis Kit (Agilent Technologies) with primers cher-prom-ap1-A, B, C and D for (supplementary material Table S1). To raise anti-Cher antibody, a Drosophila cher cDNA fragment encoding amino acids 189–482 was cloned into the pET28b plasmid. Rats were immunized (Eurogentec) with bacterially expressed protein, which was purified using the hexahistidine tag.

Generation of mosaics in eye/antennal imaginal discs using the Mosaic analysis with a repressible cell marker method (MARCM) (Lee and Luo, 2001) was carried out as described (Uhlirova et al., 2005) using the following mutant and transgenic fly strains: (1) ey-FLP1; Act >y⁺ >Gal4, UAS-GFP; FRT82B, Tub-Gal80, (2) w;; FRT82B, (3) w; UAS-ras^V12; FRT82B scrib¹/TM6B, (4) w; UAS-ras^V12; FRT82B scrib¹ UAS-bsk^DN/TM6B, (5) w; UAS-ras^V12; FRT82B scrib¹kay³ (6) w; UAS-cher^IR (VDRC, Tr. ID-107451), (7) w; UAS-cher^IR; FRT82B/TM6B, (8) w; UAS-ras^V12 UAS-cher^IR/CyO; FRT82B scrib1/TM6B att., (9) w;; FRT82B cher¹/TM6B, (10) w; UAS-ras^V12; FRT82B scrib¹ cher¹/TM6B, (11) w; UAS-hep^act, UAS-ras^V12; FRT82B, (12) w; UAS-zip^IR (VDRC, Tr. ID-7819), (13) w; UAS-ras^V12, UAS-zip^IR; FRT82B scrib¹, (14) ex⁶⁹⁷ (ex::lacZ). All crosses were carried out at 25°C. cher¹ (Robinson et al., 1997) and kay³ mutant lines were a gift from Lynn Cooley (New Haven, CT) and Constantin Yanicostas (Institut Jacques Monod, Paris, France), respectively. For TARGET inducible expression, T80-Gal4, UAS-EGFP/CyO; tub-Gal80^ts/TM6B att. females were crossed to control (w¹¹¹⁸) and w; UAS-Hep^wt males, respectively. Progeny was kept at 18°C until early third-instar. Larvae were heat-shocked at 37°C for 60 min. Total RNA was isolated after additional 6 h of incubation at 29°C.

More than 100 third-instar larval brains (7 days AEL) from each genotype were dissected, fixed and mounted. The degree of malignancy was determined by analyzing blind samples under a fluorescent microscope. The genotypes were disclosed to the observer only after the count to ensure unbiased results. Arbitrary degrees of invasive tumors were: (0) noninvasive, (1) clonal cells overgrow one of the optic lobes and invade the VNC, (2) both optic lobes are overgrown by mutant tissue and cells enter the VNC from both sides, and (3) optic lobes and most of the VNC are invaded by clonal tissue. Statistical significance was determined using a chi-square test (Prism). Pupae were counted on days 8 and 9 AEL. Percentage was calculated from 100 animals per experiment with at least three replicates. Statistical significance was determined using one-way ANOVA followed by the Unequal H HSD test.

Dissected imaginal discs and brains were fixed in 4% paraformaldehyde in PBST (PBS and 0.1% Triton X-100) for 25 min, washed and blocked in PBST containing 0.5% BSA. The following primary antibodies were used for overnight staining at 4°C: rat anti-Cher (1:1000), rat anti-N-Fil (1:1000) (Sokol and Cooley, 1999), mouse anti-MMP1 (1:300, DSHB 14A3D2), mouse anti-H2 (1:500) (Kurucz et al., 2003), chicken anti-Zipper (1:500, a gift from Eric Wieschaus), rabbit anti-pH 3 (1:100, Cell Signaling Technology), rabbit anti-activated caspase 3 (1:500, Cell Signaling Technology), rabbit anti-phospho MRLC (Ser19) (1:100, Cell Signaling Technology), and rat anti-Elav (1:200; 7E8A10) and mouse anti-Fasciclin III (1:300, 7G10), both from Developmental Studies Hybridoma Bank (Iowa). After washing, samples were incubated with a corresponding secondary antibody coupled to Rhodamine (TRITC), Cy3 or Cy5 (Jackson ImmunoResearch) for 2 h. Samples were counterstained with Alexa 456-phalloidin (Invitrogen) and DAPI or Hoechst (Invitrogen) to visualize actin filaments and nuclei, respectively. The ex::lacZ activity was detected in mosaic EAD using a standard X-Gal (5-bromo-4-chloro-3-indolyl-β-D-galactopyranoside) staining procedure.

Confocal stacks were acquired at room temperature with Olympus FV1000 confocal microscope equipped with 20× UPlan S-Apo (NA 0.85), 40× UPlan FL (NA 1.30) and 60× UPlanApo (NA 1.35) objectives. Maximum projections were generated using Fluoview 2.1c Software (Olympus) and Image J (Abramoff et al., 2004), and transversal sections using Imaris 7.0.0 (Bitplane). Final image processing including panel assembly, brightness and contrast adjustment were done in Photoshop CS4 (Adobe Systems, Inc.). Z-stacks of adult eyes were taken using motorized Leica M165 FC fluorescent stereomicroscope equipped with DFC490 CCD camera and GFP2 (Ex. 480/40 nm) filter set. Images were processed using the Multifocus module of LAS 3.7.0 software (Leica).

S2 Schneider cells were cultured at 25°C in Shields and Sang M3 insect medium (Sigma-Aldrich) containing 8% fetal bovine serum and antibiotics (Pen/Strep; Gibco). Prior to transfection, cells were starved in serum-free medium for 30 min and subsequently transfected using Fugene HD (Roche Applied Science) according to manufacturer's instructions. Immunostaining was performed as described (Jindra et al., 2004).

Total RNA was isolated from EADs dissected from third-instar larvae (6 or 7 days AEL) using TRIzol (Invitrogen), and 2 µg of DNase-treated RNA were transcribed using Superscript III reverse transcriptase with oligo (dT) primers (Invitrogen). Semi-quantitative PCR was carried out using ExTaq Polymerase (Takara Bio Inc.) in 26 standard temperature cycles. Quantitative RT-PCR was performed with Power SyBR Green PCR master mix (Applied Biosystems) using the 7900HT Fast Real-Time PCR System (Applied Biosystems). All qPCR primers (supplementary material Table S1) were designed to anneal at 62°C. All data were normalized to rp49 transcript levels, and fold changes in gene expression were calculated using the Relative standard curve method (Larionov et al., 2005). At least 4–6 biological replicates were analyzed per experiment. Statistical significance was determined using the unpaired two-tailed Student's t-test with unequal variance.

Dissected EAD were dissociated into cell suspension by incubating in 10×Trypsin/EDTA/PBS for 45 min and by subsequent passing through insulin syringe with a 0.33-mm diameter needle (Terumo). Cell counting based on GFP fluorescence was performed using a FACSCalibur Flow Cytometer (BD Biosciences). For analysis of mitosis and apoptosis EAD cells were fixed and permeabilized using the IntraSure kit (BD Biosciences). Upon staining with the following antibodies: rabbit anti-cleaved caspase-3 (PE Conjugate), rabbit anti-pH 3 (Alexa-647 Conjugate, both from Cell Signaling Technology) and anti-GFP (AL-488-conjugated, Invitrogen), for 30 min at RT, cells were incubated in BD Stabilizing Fixative (BD Biosciences) and analyzed on a FACS Aria III flow cytometer (BD Biosciences). Data (>10,000 events/sample, four biological replicates) were analyzed using FACS Diva software (BD Biosciences).

Dissected EAD (80 per genotype) were lysed in 50 mM TRIS (pH 7.8), 150 mM NaCl, 1 mM EDTA (pH 8.0), 1% Triton-X100, 0.01% Igepal, protease inhibitors (Roche). After 10 minutes of incubation on ice, lysates were centrifuged at 12,000 g for 10 min at 4°C. Protein concentration of supernatant was determined using the BioRad protein assay, and supernatant (500 µg of total protein) was incubated overnight with a rat anti-Cheerio antibody (1:200) at 4°C. The antibody complexes were captured using 30 µl of equilibrated Dynabeads Protein G (Invitrogen) for 4 h at 4°C. After five washes in lysis buffer, proteins were recovered in two consecutive elution steps, each with 50 µl of 0.1 M glycine-HCl (pH 3.0) for 5 min. After neutralization with 10 µl of 0.5 M Tris-HCl (pH 7.8) and 1.5 M NaCl, proteins were subjected to SDS-PAGE and detected by immunoblotting with the appropriate antibodies, followed by enhanced chemiluminescence using ImageQuant LAS4000 reader (GE Healthcare).

We are grateful to Lynn Cooley, Dirk Bohmann, Eric Wieschaus, Constantin Yanicostas, István Andó, the Bloomington Center (Bloomington, USA), the VDRC (Vienna, Austria) and the DSHB (Iowa, USA) for fly stocks, plasmids and antibodies. We thank Christoph Göttlinger, Gunter Rappl and Hinrich Abken for help with FACS analysis, Tobias Lamkemeyer for help with protein purification, Peter Frommolt for help with qRT-PCR analyses, Astrid Schauss for advice with image quantifications, Sabrina Jung for technical help, and Marek Jindra for discussion and critical reading of the manuscript. The authors declare no conflict of interest.