Dynamic MAPK signaling activity underlies a transition from growth arrest to proliferation in Drosophila scribble mutant tumors

Proteins essential for maintaining epithelial structures, such as cell polarity complexes, are involved in growth control (Bilder, 2004; Boggiano and Fehon, 2012; Sun and Irvine, 2016). For example, the basolateral Scribble complex, composed of Scribble (Scrib), Discs large (Dlg) and Lethal giant larvae [L(2)gl], was discovered as a group of ‘neoplastic tumor suppressor genes’ (nTSGs) in Drosophila (Bilder et al., 2000; Gateff, 1978; Woods and Bryant, 1991). Drosophila larvae homozygous mutants for any of the nTSGs grow into giant larvae with tumorous imaginal discs and optic lobes. These mutant tumors fail to differentiate and grow into masses that survive serial transplantations, induce cachexia and eventually kill the hosts (Figueroa-Clarevega and Bilder, 2015; Gateff, 1978). Studies of Drosophila nTSGs over decades have provided valuable insights into the mechanisms of growth control and tumorigenesis (Bilder, 2004; Gonzalez, 2013; Pastor-Pareja and Xu, 2013; Richardson and Portela, 2018; Sonoshita and Cagan, 2017). For example, analyses of nTSG mutant clonal growth have revealed the cooperative actions of multiple conserved signaling pathways during tumor development, cell competition-mediated tumor suppression mechanisms and tumor microenvironment influences (Brumby and Richardson, 2003; Chen et al., 2012; Cordero et al., 2010; Igaki et al., 2006; Katheder et al., 2017; Pagliarini and Xu, 2003; Vaughen and Igaki, 2016; Yamamoto et al., 2017). In particular, JNK and ERK signaling pathways, subgroups of the MAPK pathway (frequently deregulated in human cancers), have long been known as crucial for switching the growth outcome of scrib mutant cells when they are generated as mosaic loss-of-function clones surrounded by wild-type cells (Brumby and Richardson, 2003; Cordero et al., 2010; Igaki et al., 2006; Pagliarini and Xu, 2003; Uhlirova and Bohmann, 2006).

Interestingly, although Drosophila nTSG mutant tumors have successfully modeled many aspects of human epithelial cancers, it remains unclear whether and how the fast-growing fly nTSG mutant tumors change over time. The majority of mammalian tumors undergo evolution fueled by intratumor heterogeneity (McGranahan and Swanton, 2017). Many sources can contribute to intratumor heterogeneity. For example, individual tumor cells can acquire different types of mutations and grow into distinct subclones within the same tumor (de Bruin et al., 2014; Gerlinger et al., 2012). Moreover, most tumor cells co-exist with various immune and stromal cells, and different niches further contribute to intratumor heterogeneity (Jimenez-Sanchez et al., 2017). Whether the fly scrib mutant tumors exhibit some degree of intratumor heterogeneity is unclear.

Through quantitative analysis of Drosophila scrib mutant tumor growth, we found that over time scrib mutant tumors display different growth rates and cell cycle profiles. Moreover, quantitative changes in MAPK signaling activity underlie the change in growth rate. In particular, high JNK signaling activity is a primary cause of growth arrest in early scrib mutant tumors. We further found that JNK signaling activation is highly heterogeneous in scrib mutant tumors. JNK signaling activity displays a radial polarity, with JNK^high cells located at the tumor periphery, which underlies a decrease in JNK signaling activity over time. Together, these data reveal that fly scrib mutant tumors are highly dynamic on the temporal axis.

To explore temporal changes in scrib mutant tumors, we monitored the growth of wing imaginal discs derived from eggs collected from a scrib¹ allele over 3 h (Bilder et al., 2000) (Fig. 1A,B). We quantitatively measured the volumes of the scrib mutant tumors, which are known to grow over an extended larval period (Bunker et al., 2015; Vaccari and Bilder, 2005). On average, at 4 and 5 days after egg laying (AEL), the volumes of the scrib mutant tumors were around 15-30% of those of control imaginal discs raised under identical conditions (Fig. 1A,B). We calculated the growth rate (defined as the average volume change per day) in scrib mutant and wild-type imaginal discs based on the volume measurements in Fig. 1B and found that the growth rate of scrib mutant tumors increased over time and, starting from day 7, became comparable with that of the control group at 4 days AEL (Fig. 1C). Note that the scrib mutant tumors monitored in our experiments eventually grew to sizes consistent with previous reports (Bilder et al., 2000; Gateff, 1978).

To test whether the growth rate change we observed was specific to the scrib¹ allele, we also quantified the growth rate of tumors depleted of Scrib or Dlg protein at the posterior region of wing discs using engrailed-Gal4-mediated RNAi. At 4 and 5 days AEL, the volume of scrib RNAi and dlg RNAi tumors was much smaller than the size of the posterior region in control wing discs (Fig. S1). However, the anterior regions of the scrib RNAi, dlg RNAi and control imaginal discs had comparable volumes (Fig. S1), indicating that the growth arrest phenotype observed in early scrib tumors is likely to be inherent and independent of an overall delay in larval development. Interestingly, from 4 to 6 days AEL we detected an increase in phosphorylated histone H3-positive (PH3⁺) cell number in the posterior scrib RNAi and dlg RNAi tumors (Fig. S1), indicative of changing growth rates during tumor progression. The fly larvae harboring scrib RNAi or dlg RNAi imaginal discs turned into pupae by day 7 AEL, preventing further measurement of growth rates. Notably, the cell competition process mediated by Minute mutations and differential Myc levels does not cross compartment boundaries (Johnston, 2009; Morata and Ripoll, 1975; Simpson, 1979; Simpson and Morata, 1981). However, tumorigenic dCsk-deficient cells generated in the patched expression domain are known to undergo apoptosis at the anterior–posterior boundary (Vidal et al., 2006). It is therefore likely that the growth of posterior scrib RNAi and dlg RNAi tumors is largely independent of the influence of cell competition, except for cells at the anterior–posterior boundary.

It is possible to attribute the growth arrest phenotype of early scrib mutant tumors to increased apoptosis, defects in cellular growth or defective cell proliferation. However, we found that prevention of apoptosis by overexpressing p35 did not rescue the growth arrest of posterior scrib RNAi cells (Fig. S2). Moreover, although we were able to detect apoptotic cells in the posterior scrib RNAi region, a similar number of apoptotic cells could also be detected in the anterior control region (data not shown). Therefore, the growth arrest phenotype of early scrib mutant tumors is unlikely to be caused by increased apoptosis. The cell volume of individual scrib mutant cells was larger than that of the control wing disc cells (Fig. S3). Therefore, the growth arrest of early scrib mutant tumors is also unlikely to be caused by defects in the growth of individual cells.

We then performed flow cytometry analysis with wild-type imaginal discs and scrib mutant tumors expressing fluorescence ubiquitin cell cycle indicator (FUCCI) sensors to examine possible cell proliferation defects (Zielke et al., 2014). In wild-type discs at day 4 AEL, about 60% of cells were in the G0/G1 phase, 25% in the S phase and 15% in the G2/M phase (Fig. 1D,G), in agreement with previous reports (Milan et al., 1996). However, in scrib mutant tumors at 4 days AEL about 25% of cells were in the G0/G1 phase, 25% in the S phase and 50% in the G2/M phase (Fig. 1E,G). The results suggest that a population of cells in early scrib mutant tumors is arrested at the G2/M stage.

Drosophila larval brain neuroblasts are an excellent model for analyzing mitosis because of their accessibility for live imaging (Cabernard and Doe, 2013). We observed that scrib mutant neuroblasts frequently displayed a significantly prolonged entry into mitosis (Fig. S4), consistent with the G2/M arrest defects observed in early scrib mutant wing disc cells. Notably, by day 8 AEL, the cell cycle distribution of scrib mutant cells was comparable with that of day 4 wild-type imaginal discs (Fig. 1F,G). Therefore, the growth arrest phenotype in early scrib mutant tumors is caused by cell cycle defects, which can resolve over time.

Early and late scrib mutant tumors display different growth rates and cell cycle profiles, indicative of potential evolving capacity. We therefore performed bulk RNA sequencing (RNA-seq) for wild-type imaginal discs and scrib mutant tumors collected at different time points to gain a global view of transcriptomic changes associated with scrib tumor progression (Fig. 2). The biological repeats were well clustered per sample for each time point and the scrib mutant tumors formed a transition trajectory along time on the principal component analysis (PCA) plot (Fig. 2A). Both PCA and hierarchical clustering analyses suggested that the scrib mutant tumors could be grouped into three stages with distinctive transcriptomic signatures: an early stage at 4-6 days AEL, an intermediate stage at 8 days and a late stage at 10-14 days (Fig. 2A,B).

We performed linear regression analysis for each gene in the scrib tumor time series and ranked the genes by the slope of the regression line (Fig. S5). For the 10% of genes ranked at the top and bottom, we performed pathway enrichment analysis with the KEGG pathway database (Kyoto Encyclopedia of Genes and Genomes, https://www.genome.jp/kegg/) (Fig. 2C,D) as well as the Drosophila GLAD database (Gene List Annotation for Drosophila, https://www.flyrnai.org/tools/glad/web/) (Hu et al., 2015) (Fig. S6). For example, the expression level of genes related to oxidative phosphorylation decreased over time and the expression level of glycolytic genes increased over time in the scrib mutant tumors, consistent with a glucose metabolism drift widely observed in tumors called the Warburg effect (Eichenlaub et al., 2018; Wang et al., 2016). Of the several conserved signaling pathways that show significant changes in scrib mutant tumors over time, the pathway enrichment analysis with both KEGG and GLAD databases pointed us to the MAPK signaling pathway (Fig. 2C; Fig. S6A). Notably, two of the three Drosophila MAPK pathway subgroups, the JNK and ERK pathways, have long been known as crucial for regulating the growth outcome of scrib mutant mosaic clones (Brumby and Richardson, 2003; Cordero et al., 2010; Igaki et al., 2006; Pagliarini and Xu, 2003; Uhlirova and Bohmann, 2006). Therefore, although the transcriptomic analysis suggested that scrib mutant tumors have extensive changes besides the MAPK pathway, including several metabolic pathways and other signaling pathways such as the Notch and JAK/STAT pathways, we first focused on examining whether JNK and ERK signaling activities change over time and whether these changes have functional implications regarding scrib tumor growth.

We first examined JNK signaling activity because the expression levels of two widely used JNK signaling target genes, Mmp1 and puckered (puc) (Igaki et al., 2006; Uhlirova and Bohmann, 2006), as measured in the bulk RNA-seq experiments (Fig. 2), decreased significantly in scrib mutant tumors over time (Fig. 3A). Although JNK signaling activity is elevated in scrib tumors of all stages compared with wild-type wing discs (Fig. 3A,B), in agreement with previous reports (Bunker et al., 2015; Igaki et al., 2006), we found that the protein levels of Mmp1 and TRE-DsRED, a JNK signaling reporter (Chatterjee and Bohmann, 2012), decreased significantly in scrib tumors at 8 days AEL compared with scrib tumors at 5 days AEL (Fig. 3B). These data suggest that JNK signaling activity decreases as the scrib tumors grow over time.

We also found that the growth arrest phenotype in early scrib or dlg tumors was effectively rescued through overexpression of a dominant-negative form of Tak1 (Tak1^DN) or Basket (Bsk^DN), which block JNK signaling activity (Fig. 3C-H; Fig. S7). Notably, overexpression of Ras^V12, NICD or Yki^S168A (an active form of Yki), which are known to promote scrib clonal growth in a mosaic setting (Brumby and Richardson, 2003; Chen et al., 2012; Pagliarini and Xu, 2003), did not rescue the growth arrest phenotype in early scrib tumors (Fig. S8). Note that we used a C855a-Gal4 driver that expresses early in whole scrib or dlg tumors and does not cause lethality when Ras^V12 or NICD are co-expressed (Fig. S9). Together, the data suggest that high JNK activity is a potent inhibitor of tumor growth in early scrib tumors. This result is consistent with a recent study showing that injury-induced JNK activation in wing imaginal discs also leads to cell cycle arrest at G2 (Cosolo et al., 2019).

We next examined why JNK signaling activity decreases as the scrib tumors grow. From single-cell transcriptomic data generated from scrib mutant tumors, we found that expression of the JNK signaling reporter gene Mmp1 is highly heterogeneous (Fig. 3I). This data led us to re-examine the Mmp1 expression pattern; we found that Mmp1 showed strongest signals at the periphery of scrib mutant tumors (Movies 1, 2). To exclude the possibility that the Mmp1 staining pattern was an artifact from poor antibody penetrance, we examined the expression pattern of a JNK signaling reporter TRE-DsRED without using antibody staining. Similarly, we found that TRE-dsRED displayed a radial polarity, with the strongest fluorescence signals at the periphery of scrib mutant tumors (Fig. 3J-M). Moreover, single-cell transcriptomic analysis demonstrated that Mmp1 expression shows high correlation with a number of cytoskeletal genes, such as cindr (Table S1; Fig. S10). A protein-trap line of cindr also revealed a radial polarity in scrib mutant tumors (Movies 3, 4). Therefore, JNK signaling activation in homozygous scrib mutant tumors exhibited a radial polarity, with JNK^high cells located at the surface of the scrib tumors. Because scrib mutant tumors grow in a compact manner, the surface-to-volume ratio drops as the tumor volumes increases, leading to a decrease in the percentage of JNK^high G2/M-arrested cells over time.

The bulk scrib tumor time series transcriptomic data demonstrated that components of the ERK signaling pathway also showed significant changes (Fig. 4A). Moreover, when we ordered genes by the change in the percentage of cells expressing the specific gene in single-cell transcriptomic data, we noticed that kekkon1 (kek1), a well-established ERK signaling target (Ghiglione et al., 2003, 1999), was ranked at the top. Few kek1⁺ cells existed in early scrib tumors and the majority of cells were kek1⁺ in late scrib tumors (Fig. 4B; Fig. S11). Using a kek1-lacZ line (Musacchio and Perrimon, 1996), we confirmed that late scrib tumors indeed harbored many more kek1⁺ cells than early scrib tumors (Fig. 4C-E). Similarly, the expression of sprouty (sty) (Casci et al., 1999) and argos (aos) (Golembo et al., 1996), two other genes induced by ERK activity, also increased over time (Fig. 4A; Fig. S12). Moreover, the expression of vein (vn), an EGFR ligand promoting patterning and proliferation in wing imaginal discs (Schnepp et al., 1996), also increased over time (Fig. 4A; Fig. S12). Together, these data suggest that ERK activity increases over time as scrib mutant tumors grow.

ERK signaling activity is required for both growth and patterning in wing imaginal discs (Diaz-Benjumea and Hafen, 1994; Prober and Edgar, 2000; Zecca and Struhl, 2002a,b). Here, we focused on examining whether increasing ERK activity over time influences scrib tumor growth. Interestingly, although overexpression of Ras^V12 did not rescue early scrib mutant tumors from growth arrest (Fig. S8), it significantly increased the size of scrib mutant tumors at later stages (Fig. 4F-H). Conversely, blocking ERK signaling activity through expression of a dominant-negative form of EGFR (Buff et al., 1998) (Fig. 4I-N) and ERK and Ras RNAi constructs (Fig. S13) led to a reduction in scrib and dlg mutant tumor sizes at later stages. These data suggest that an increase in ERK signaling activity promotes scrib tumor growth at later stages.

Interestingly, ERK and JNK pathways have extensive cross-talks in scrib mutant tumors, as indicated by the high correlation of a number of ERK pathway components with Mmp1 expression at the single-cell level (Table S1; Fig. S14). Moreover, when we blocked JNK signaling activity in early scrib mutant tumors, we observed an increase in kek1⁺ cell number (Movies 5, 6; Fig. S15) in addition to an increase in tumor size (Fig. 3C-H), indicating that high JNK signaling activity possibly represses ERK activity in early scrib mutants. However, the molecular mechanism mediating JNK repression of ERK activity remains unclear.

Here we have demonstrated that dynamic changes in JNK and ERK activities underlie a transition from a growth arrest state to a proliferation state over time in Drosophila scrib mutant tumors.

We found that JNK signaling activation exhibited a radial polarity, with JNK^high cells located at the surface of homozygous scrib mutant tumors. We do not yet know the underlying reason for the heterogeneous JNK activation pattern. One possible reason is that the active JNK ligand Eiger might be mostly provided from external sources. This might involve hemocytes (Cordero et al., 2010) or the fat body, as recently shown for another nTSG mutant alg3 (de Vreede et al., 2018). Another possible input is potential mechanical stresses that the tumor surface cells experience because JNK signaling is activated at leading-edge cells, which assemble supracellular actin cables and experience high actomyosin-dependent tensile force during dorsal closure (Martin-Blanco et al., 1998).

JNK activation induced by injury in the wing imaginal discs also leads to cell cycle arrest at G2, which can be restored by overexpression of Cdc25/String (Cosolo et al., 2019). We found that expression of Cdc25/string(stg) and Cks30A anti-correlates with Mmp1 in the single-cell data. Moreover, we could not rescue growth arrest in early scrib mutant tumors by overexpression of String alone (data not shown), indicating that the cell cycle defects in early scrib mutant tumors are probably mediated by multiple factors downstream of high JNK signaling activity.

Clonal scrib mutant cells are eliminated through cell competition when they are surrounded by wild-type neighbors (Brumby and Richardson, 2003; Vaughen and Igaki, 2016; Yamamoto et al., 2017). It is noteworthy that clonal scrib mutant cells are likely to be in a different state from growth-arrested homozygous scrib mutant cells. Studies have shown that overexpression of Ras^V12, NICD and p35 can effectively block the clonal scrib mutant cells from undergoing apoptosis induced by cell competition. In our study, we found that overexpression of Ras^V12, NICD and p35 had little effect in relieving the early scrib mutant tumors from growth arrest. It is likely that interactions between clonal scrib mutant cells and wild-type cells during cell competition induce an additional layer of complexity into determination of clonal scrib mutant cell state, as the effects of JNK activation are known to be highly context dependent (Pinal et al., 2019).

In human solid tumors, the tumor margin and core were also shown to experience different immune environments (Kather et al., 2018). Our study highlights that tissue topological factors (peripheral versus center) can be an inherent source of diversity in cell populations in growing tumors and that dynamic signaling rewiring during the processes of early tumorigenesis does not necessarily require the generation of de novo mutations or new cell clones.

The fly strains used in this study were scrib¹ FRT82B/TM6B (Bilder et al., 2000), dlg^GH19 FRT19A/FM7c (a kind gift from the Schupbach laboratory, Princeton University; a C-to-T mutation led to a stop codon at Q42), UAS-scrib RNAi on the second chromosome (Bloomington/BL38199), UAS-scrib RNAi on the third chromosome (BL35748), UAS-dlg RNAi on the third chromosome (BL35772), UAS-Ras85D RNAi (BL34619), UAS-ERK RNAi(BL34855), y[1] v[1]; P{y[+t7.7]=CaryP}attP2 (BL36303, the third chromosome TRiP line background strain), y[1] v[1]; P{y[+t7.7]=CaryP}attP40 (BL36304, the second chromosome TRiP line background strain), engrailed-Gal4 UAS-GFP (BL25752), engrailed-Gal4 (BL30564), UAS-p35 (Hay et al., 1994) (BL6298), Fly-FUCCI (BL55098), worGal4, UAS-cherry::Jupiter, Sqh::GFP (Cabernard and Doe, 2013), c855a-Gal4 (Hrdlicka et al., 2002) (BL6990), UAS-Ras^V12 (Karim and Rubin, 1998), UAS-Yki^S168A (Oh and Irvine, 2009) (BL28818), UAS-NICD (Rebay et al., 1993), UAS-Bsk^DN (Igaki et al., 2006) (a kind gift from Jose C. Pastor-Pareja School of Life Sciences, Tsinghua University, China), UAS-Tak1^DN (BL58811), TRE-dsRED (Chatterjee and Bohmann, 2012), kek1-lacZ (Ghiglione et al., 2003) and UAS-EGFR^DN (Buff et al., 1998) (BL5364), cindr^CA06686 (BL50802).

About 50 embryos collected within 3 h were put into an individual vial of fly food to avoid crowding and the larvae raised in an incubator at 25°C for appropriate lengths of time before dissection. Imaginal discs were fixed and stained according to standard protocols. The primary antibodies used were mouse anti-phospho-Histone3 (1:1000; Cell Signaling Technology), mouse anti-Mmp 1(1:30; a mixture of 5H7B11, 3A6B4 and 3B8D12, DSHB) and mouse anti-beta-galactosidase (1: 25, 40-1a, DHSB). The secondary antibodies conjugated with various Alexa Fluor dyes (Thermo Fisher Scientific) were used at 1:500. Phalloidin conjugated with Alexa Fluor dyes (1:1000, Thermo Fisher Scientific) and Hoechst 33342 (1:10,000, Thermo Fisher Scientific) were used to stain F-actin and DNA, respectively. All images were acquired on a Leica TCS SP8 confocal microscope.

About 30 larvae were dissected in PBS. Cell lysates were homogenized in 1× RIPA (Millipore) with protease inhibitors (Roche). The primary antibodies used were mouse anti-MMP1 (1:100; a mixture of 5H7B11, 3A6B4 and 3B8D12, DSHB), rabbit anti-DsRed (1:1000; Takara) mouse anti-beta-galactosidase (1:250; 40-1a, DHSB) and mouse anti-alpha-tubulin (1:5000; 12G10, DSHB).

Images were taking as z-stacks with a step size of 1 μm. Tissue volume was measured with Measure Stack plugin in Fiji using the actin-staining channel; the wand tool was used to select regions of interest, where manual adjustment of tolerance value and recognition of edges were made for each slice. PH3⁺ cell number was calculated with the Cell Counter plugin in Fiji.

Wing imaginal discs were dissociated in 0.25% trypsin-EDTA solution at 37°C for 10 min and stained with Hoechst at 37°C for 45 min. Cells were sorted with BDFACSAria IIIu and the data analyzed with FlowJo.

Female virgins of hsFLP; worGal4, Sqh::GFP,UASCherry::jupiter; FRT82B/TM6B were crossed with males of scribFRT82B/TM6B. Progeny were heat-shocked at 38°C (in the water bath) for 1 h and subsequently raised at 25°C until imaging. Larvae (5 days AEL) were then dissected and imaged according to standard protocols (Cabernard and Doe, 2013). For wild-type controls, larvae expressing wt; worGal4,Sqh::GFP,UASCherry::jupiter; Dr/TM6B were dissected and imaged with the same laser setting as that of the scrib mutant neuroblasts.

Total RNA was extracted from control and scrib¹ wing imaginal discs with the RNeasy Mini Kit (Qiagen). Construction of cDNA libraries and 150 bp paired-end sequencing on Illumina HiSeq platform were performed by Novogene. Cleaned raw reads obtained from Novogene were mapped to the reference genome (BDGP6) using STAR (2.5.4a) (Dobin et al., 2013). Default parameters were used except for the max intron size, which was set at 100,000. Counts were generated by featureCounts available in Subread package (1.32.2) with the default setting (Liao et al., 2013). PCA analysis was performed in DESeq2 (1.22.2) after low expression genes were filtered (baseMean>100) (Love et al., 2014). Count normalization to counts per million (cpm) was performed using edgeR (3.24.3) before hierarchical clustering (hclust function in R) (Robinson et al., 2010). Linear regression was performed using the lm function in R (3.5.2). Pathway enrichment analysis was performed using the default setting of clusterProfiler (3.10.1) (Yu et al., 2012). The GEO accession number for these data is GSE130243.

We thank Dr Chris Doe and Dr Jose C. Pastor-Pareja, Bloomington Stock Center and Developmental Biology Hybridoma Bank for providing fly stocks and reagents. We thank Dr Ting Xie for providing the 10× single-cell RNA-seq protocol. We thank Dr Zilong Wen and Dr Mingjie Zhang for sharing their confocal microscope. We thank Dr Gang Wang and Dr Zhenguo Wu for their help with FACS experiments. We thank Dr Chris Doe and Dr Trudi Schupbach for helpful comments on the manuscript.