Gene expression profiling identifies the zinc-finger protein Charlatan as a regulator of intestinal stem cells in Drosophila

Tissue homeostasis requires a balance between removing dead cells and producing new cells. An imbalance between the two processes can lead to tissue degeneration, inflammation and cancer. Replenishment of lost cells in an adult tissue is usually mediated by resident stem cells. Elucidating the mechanisms that regulate stem cell-mediate homeostasis and regeneration may provide insights into the development of therapeutics for various diseases (Nakada et al., 2011).

The mammalian intestine is essentially a monolayer of epithelial cells with fast cell turnover rate and the tissue integrity is maintained by active stem cells located near the crypt base (Rizk and Barker, 2012). Two overlapping or inter-converting stem cell populations give rise to progenitor cells in the transit-amplifying zone that can replenish cells of various lineages along the crypt-villus axis (Buczacki et al., 2013; Takeda et al., 2011; Tian et al., 2011; Yan et al., 2012).

The adult Drosophila midgut is equivalent to the mammalian stomach and small intestine. The midgut epithelium is largely a monolayer of enterocytes (ECs) and does not have crypt-villus structure. Approximately 1000 intestinal stem cells (ISCs) are distributed evenly along the basal side of the epithelium (Micchelli and Perrimon, 2006; Ohlstein and Spradling, 2006). An ISC divides to generate a renewed ISC and an enteroblast (EB), which ceases division and starts to differentiate. The ISC-EB asymmetry is governed by the Delta-Notch signaling, with high levels of Delta in the renewed ISC activating Notch signaling in the neighboring EB (Bardin et al., 2010; Ohlstein and Spradling, 2007) (see Fig. 1I). Depending on the strength of stimulation on the Notch pathway, the EB may differentiate to become an EC (in wild-type gut 90% of the time) or enteroendocrine cell (EE) (10% of the time) (Micchelli and Perrimon, 2006; Ohlstein and Spradling, 2006,, 2007).

The cells surrounding ISC constitute the niche and secrete growth factors to regulate ISC maintenance and activity. The visceral muscle and mature ECs are sources of Wingless, insulin-like peptides, epidermal growth factor receptor (EGFR) ligands and JAK-STAT pathway ligands called Unpaired (Upd) (Biteau and Jasper, 2011; Buchon et al., 2010; Jiang et al., 2010; Lin et al., 2008; O'Brien et al., 2011; Ragab et al., 2011; Xu et al., 2011). Moreover, the differentiating EBs contribute Upd, Wingless and EGFR ligands to regulate intestinal homeostasis (Cordero et al., 2012; Jiang et al., 2010; Zhou et al., 2013). Recent reports also reveal the secretion of Decapentaplegic/BMP from trachea and ECs to regulate ISC activity (Guo et al., 2013; Li et al., 2013a,b; Tian and Jiang, 2014). Hedgehog signal coming from multiple cell types serves a negative regulatory function in ISC division (Li et al., 2014). Other conserved signaling pathways, including JNK, p38, PVF2 and Hippo, are also required for the regulation of ISCs during homeostasis, tissue damage and aging (Biteau et al., 2008; Bond and Foley, 2012; Jiang et al., 2009; Karpowicz et al., 2010; Park et al., 2009; Ren et al., 2010; Shaw et al., 2010; Staley and Irvine, 2010).

Many ISC-intrinsic factors have been shown to regulate asymmetry and renewal. Osa (as part of the SWI/SNF complex) and the Brahma chromatin remodeling complex regulate Delta expression and ISC proliferation (Jin et al., 2013; Zeng et al., 2013). The histone deubiquitinase Scrawny and growth regulators [including Myc, Target of Rapamycin (TOR) and tuberous sclerosis complex (TSC)] are also required for ISC growth and division (Amcheslavsky et al., 2011; Buszczak et al., 2009; Ren et al., 2013). To search for other intrinsic ISC regulators, we used genetic manipulation to increase the number of ISC-like cells in the adult midgut and performed gene expression profiling. The results revealed many possible proteins that can modulate the function of ISCs or precursor cells (defined as ISCs+EBs). We carried out an in depth analysis of the zinc-finger protein Charlatan (Chn) because its loss of function shows a severe defect in ISC division. Our data further demonstrate that Chn regulation of chromatin is essential for ISC division.

Although sorting individual cell types from adult midgut for genomic analysis is possible (Dutta et al., 2014), the number of ISCs present in the midgut is low and the available ISC-specific Delta-Gal4 driver exhibits weak expression (Zeng et al., 2010). Moreover, stem cells may change their properties if dissociated from the niche. Therefore, we artificially increased the pool of ISC-like cells in the midgut by expressing via Gal4/UAS a gain-of-function Raf (Raf^GOF) and a dominant-negative Notch (Notch^DN) (Bardin et al., 2010; Markstein et al., 2014; Micchelli and Perrimon, 2006; Ohlstein and Spradling, 2006,, 2007). Raf^GOF increases the proliferation of ISC, whereas Notch^DN skews the asymmetry to produce precursor cells that are mostly Delta positive (Delta+). The escargot promoter-Gal4 together with the tubulin-Gal80^ts (esg^ts>) provide temporal control, and were used to drive the expression in precursor cells; the UAS-mCD8GFP was included to mark the cells expressing the constructs. In control guts, after a temperature shift to 29°C for 36 h, the GFP+ cells appeared frequently as small clusters with usually one Delta+ cell per cluster, representing the ISC and EB cell nests (Fig. 1A-A″). In guts that overexpressed Raf^GOF, there was a substantial increase in GFP+ precursor cells, indicating high proliferation (Fig. 1B-B″). Overexpression of Notch^DN together with Raf^GOF resulted in not only high numbers of GFP+ cells but also Delta+ staining of these ISC-like cells (Fig. 1C-C‴). By contrast, overexpression of the constitutively active Notch intracellular domain (Notch^IC) together with Raf^GOF caused the formation of larger GFP+ cells that had no Delta staining, indicating a loss of stem cell fate and an increase in differentiation (Fig. 1D-D‴). The number of EEs, as revealed by the Prospero staining, was not affected within the time course of these experiments (Fig. 1E-H). These results demonstrate that we can create fly intestines with highly increased number of Delta+ ISC-like cells or with mostly differentiated cells that have no Delta staining.

We used dissected guts, from the cardia to the midgut/hindgut junction excluding malpighian tubules and crops, from the above-described genetic combinations for gene expression profiling. Total RNA was isolated from control, Raf^GOF, Raf^GOF+Notch^DN and Raf^GOF+Notch^IC fly guts. Then cRNA probes were prepared and hybridized to high-density GeneChip Drosophila Genome 2.0 Arrays (Affymetrix). This type of microarray contains 18,880 probe sets and is able to measure the expression of 18,500 transcripts and variants. Four independent RNA isolation and microarray hybridization were performed.

The expression profile of each genetic background was compared with that of the esg^ts>GFP control. Based on four stringent criteria (see Materials and Methods), a total of 1127 transcripts with significantly increased or decreased expression were identified. To investigate the relationships among the transcripts that showed significant changes, we first carried out a two-dimensional hierarchical clustering analysis (Fig. 2A). This type of clustering algorithm arranges genes or samples according to their similarity in expression profiles across all of the genes or all of the experiments, such that samples or genes with similar expression patterns are clustered together (Eisen et al., 1998; He et al., 2003). The analysis reveals extensive similarity in all three experimental samples, suggesting that many may be potential Raf target genes because Raf^GOF was used in all three experimental strains. The genes that showed opposite expression in the Notch^DN versus Notch^IC samples may represent Notch pathway target genes.

We then derived self-organizing maps (SOM) for our data sets (Fig. 2B). The SOM algorithm clusters genes or transcripts into distinct groups or classes based on the similarity of their relative expression profiles in the different genetic backgrounds (Tamayo et al., 1999). Such analysis yielded 12 expression clusters (c0 to c11). The first noteworthy pattern is low expression in wild-type guts, higher in Raf^GOF alone, highest in Raf^GOF+Notch^DN and lower in Raf^GOF+Notch^IC. Clusters c4, c5 and c8 match this description the best, and they may represent ISC- or precursor-enriched genes. Based on gene ontology analysis, we can further classify most of these genes as involving in cellular metabolism, cell organization, cell cycle, cellular transport and chromosome segregation (Fig. 2C).

A total of 468 genes are included in clusters c4, c5 and c8. Based on our own scientific experience and the preference of higher fold change, we have selected a set of 63 genes that may represent novel or serve as examples of ISC regulators (Table 1). The expression profile of all 1127 genes grouped as clusters is available at http://dx.doi.org/10.6084/m9.figshare.855619.

Delta protein is a well-characterized ISC-specific marker in the Drosophila midgut, whereas esg is well known for its expression in the precursors. From our analysis, both esg and Delta showed their expression profiles belonging to the clusters c4, c5 and c8. Furthermore, ISCs are the only dividing cells in the adult Drosophila midgut and many of the genes in the c4, c5 and c8 clusters are cell cycle-associated genes, including Polo, Cdc2, Pendulin, Cyclin B and string (http://dx.doi.org/10.6084/m9.figshare.855619 and Fig. 3A). By using a protein trap line (FlyTrap CC01326), we confirmed that the Polo-GFP fusion protein was detectable by GFP antibody staining in Delta+ ISCs, as well as in the neighboring EBs (Fig. 3B). Cdc2 is another essential component for mitotic entry, and the Cdc2 protein was detectable using antibody staining in Delta+ ISCs and in adjacent EBs (Fig. 3C). These results further suggest that the transcripts in the c4, c5 and c8 clusters are associated with precursor cells in the midgut.

We used the esg-Gal4 driver to perform RNAi knockdown of 10 of the transcripts identified from clusters c4, c5 and c8. Four lines that targeted RhoGEF4, kugelkern, ballchen and roughest did not show detectable phenotype, whereas six RNAi lines showed mild to severe phenotype (Fig. 3D-I). The knockdown of the transcript for the zinc-finger protein Charlatan (Chn) resulted in the most severe loss of the esg>GFP signal that marked the precursors (Fig. 3E). Pendulin as an Armadillo repeat-containing protein has similarity with Importin and may play a role in nuclear import during proliferation (Küssel and Frasch, 1995); the knockdown of pendulin led to an observable reduction of the esg>GFP signal (Fig. 3F). Scute has been shown to be an upstream activator of Chn in the developing peripheral neurons (Escudero et al., 2005; Reeves and Posakony, 2005), and knockdown of scute caused a mild defect in the midgut precursors (Fig. 3G). The high-mobility group proteins (HMG D and Z) function to regulate the compaction of chromatin (Ragab et al., 2006); the knockdown of HMG D alone caused only a mild disorganization, although we did not attempt the double RNAi for D and Z (Fig. 3H). Another chromatin protein Histone 2A variant (H2Av) serves as a signal for the recruitment of repair proteins at the site of DNA damage (Madigan et al., 2002; Park et al., 2012). The knockdown of H2Av showed a severe disruption of midgut organization (Fig. 3I). Further analysis of H2Av RNAi-induced phenotype showed that the Delta+ ISCs (arrows in Fig. 3J,K) did not divide as efficiently so the cell nests were smaller, whereas the EBs had bigger cell size, indicating abnormal differentiation (open arrows in Fig. 3K). Therefore, a sample of the genes identified from the profiling have biological functions in ISCs or precursors.

We performed an in-depth analysis of chn because the RNAi of this transcript exhibited a severe defect in ISCs/EBs. Chn regulates peripheral neuron development by forming a positive regulatory loop with the achaete-scute complex genes (Escudero et al., 2005; Kania et al., 1995; Reeves and Posakony, 2005). The predicted Chn protein has 1108 amino acids and contains five C2H2-type zinc fingers. Chn has no extensive homology with any mammalian proteins, but the sequence of the Chn zinc fingers has 35% identity and 55% similarity to those of the human zinc-finger protein 462. Meanwhile, the functional property of Chn as a transcriptional repressor in photoreceptor neurons resembles that of another human zinc-finger protein, NRSF/REST (Escudero et al., 2005; Tsuda et al., 2006).

A strong loss-of-function allele chn⁹ (Tsuda et al., 2006) was used to perform a clonal analysis by the mosaic analysis of repressible cell marker (MARCM) technique. chn⁹ mutant ISCs marked by GFP expression remained as single cells, whereas wild-type clones had increased cell numbers over the same period of 8 days (Fig. 4A). Moreover, chn⁹ mutant cells remained small and the Delta staining was not detectable, within 2 days of clone induction (Fig. 4B,C). We also analyzed in further detail the adult midgut precursors after chn RNAi using the esg^ts> driver. The number of mitotic cells in the midgut, as revealed by staining for phosphorylated histone 3 (pH3) was reduced after 2 days. A concomitant gradual decrease of GFP+ cells that represent precursors was observed (Fig. 4D,E). This mitotic defect was still dominant when two different damaging agents were used to stimulate ISC division (Fig. 4F). These defects were corroborated with the lack of EdU incorporation, which reveals both ISC replication and EB endoreplication, and the loss of staining of the mitotic marker Cdc2 (Fig. 4G-J). The inclusion of a transgenic chn cDNA rescued the RNAi-induced Delta staining and cell proliferation phenotype (supplementary material Fig. S1A-D). Together, these results demonstrate that Chn has an intrinsic function in ISCs to maintain their ability to divide.

We then tested whether increasing the expression level of Chn might increase ISC division. A UAS-chn transgene (Tsuda et al., 2006) was crossed with the esg^ts>GFP and adult flies were shifted to 29°C to allow Gal4-driven expression of Chn. There was an approximately 1.5-fold increase of pH3+ cells in the midguts at every time point examined (Fig. 5A). The number of GFP+ cells also increased, whereas the number of Delta+ cells remained similar (Fig. 5B,C, arrows). These results suggest that increased expression of Chn promotes stem cell division but does not change the stem cell number by altering the asymmetric division.

Tissue homeostasis requires the coordination of stem cell division and daughter cell differentiation. Although overexpression of Chn increased the number of GFP+ daughter cells, these cells had relatively small nuclei and were tightly associated with each other (Fig. 5C,C′). This suggests that they behave as precursor-like cells and may be arrested at an early stage of differentiation. The marker Pdm1 is expressed in late EBs and mature ECs (Fig. 5D,F, arrows) (Zhou et al., 2013). The GFP+ cells in Chn overexpression guts did not express Pdm1 (Fig. 5E, encircled). This arrest of differentiation could continue for 12 days (Fig. 5G, encircled). If we downshifted to the permissive 23°C from day 6 to 12 to allow the Gal80^ts to function again to inhibit Gal4-driven Chn expression, then a substantial number of GFP+ cells became Pdm1+ (Fig. 5H-J for quantification). By using the EE marker Prospero, we also found that the number of Prospero+ cells was lower after Chn overexpression but increased again after temperature downshift (Fig. 5K). These results demonstrate that high levels of Chn can block differentiation, but the process can resume when Chn overexpression is halted.

Some stem cell maintenance factors function by antagonizing differentiation (Fuchs and Chen, 2013; Stergiopoulos and Politis, 2013). Even though we show in Fig. 5 that overexpressed Chn can block differentiation, if the endogenous Chn acts by such a mechanism, the loss of Chn should lead to differentiation at the expense of stem cell number. We carried out staining with antibodies for various cell fate markers. Similar to that in the chn⁹ mutant clones, the expression of Delta rapidly disappeared after esg^ts>chn RNAi (Fig. 6A-C). The GFP+ cells after chn RNAi remained small and did not show the Su(H)LacZ staining, suggesting that they are not differentiating EBs (Fig. 6D,E). Moreover, the overall number of Pdm1+ cells or Prospero+ cells in chn RNAi guts did not increase at any time point examined and decreased slightly over time (Fig. 6F-K). These results suggest that, in the absence of Chn, the ISCs merely lose their stem cell properties but do not automatically enter differentiation. The data shown in Figs 4-6 together suggest a model in which the normal function of Chn is to promote ISC and precursor cell properties, while differentiation is governed by a separate set of factors that are not directly regulated but can be antagonized by Chn.

Previous reports show that Chn represses Delta gene expression in the eye imaginal discs and binds to promoter sequences of Delta and other genes in vitro (Tsuda et al., 2006; Yamasaki et al., 2011). However, the direct repression of Delta by Chn does not correlate with our midgut phenotypic analysis, because Chn is required positively within the ISCs for Delta expression. Moreover, Chn is essential for ISC division, whereas Delta is required for establishing ISC/EB asymmetry but not for ISC division (Ohlstein and Spradling, 2007). Therefore, we surmised that Chn might regulate other ISC properties that indirectly influence Delta expression.

To search for other possible mechanisms, we stained the midgut with various markers. The JAK-STAT is a major growth regulatory pathway in the midgut and the 10XSTAT-GFP reporter in individual precursor cells did not show an obvious decrease after chn RNAi (supplementary material Fig. S1E-J). Meanwhile, the acetylated histone 3 (AcH3) staining that was normally higher in precursor cells became undetectable after chn RNAi (Fig. 7A,B). HP-1 is associated not only with heterochromatin and gene silencing but also euchromatin to positively regulate transcription (Agarwal et al., 2007; Kwon and Workman, 2011). In the midgut, HP-1 was detected in all cell types, mainly as one strong focus per cell (Fig. 7C), as previously described (Lee et al., 2011). In esg^ts>chn RNAi midguts, the HP-1 staining in the precursor cells became weaker and was present as more foci (Fig. 7D). Higher magnification confocal images revealed that this abnormal HP-1 staining correlated closely with the DNA staining (supplementary material Fig. S2A,B). In comparison, the HP-1 staining remained similar to that in control samples when the Myo1A-Gal4 driver was used to induce chn RNAi in ECs (Fig. 7E,F). Moreover, we stained for H3K4me3, another chromatin marker associated with transcriptional initiation. This marker showed only a slight decrease after esg^ts>chn RNAi (supplementary material Fig. S2C-D). These results suggest that Chn is required specifically in precursor cells, but not in mature ECs, to maintain a normal chromatin structure, especially that marked by histone acetylation.

We assessed the functional importance of the reduced histone acetylation observed in the Chn-depleted precursor cells. Various transgenic RNAi constructs targeting the transcripts of different histone deacetylases (HDACs) were crossed together with the chn RNAi construct (supplementary material Fig. S2E). There was a significant rescue of the mitotic cell count, particularly when HDAC2 was also knocked down (supplementary material Fig. S2E and Fig. 7G). The antibody staining for overall histone acetylation was visibly improved after the double knockdown of HDAC2 and chn (Fig. 7H-J). The HP-1 and DNA staining was also visibly improved after the double knockdown (supplementary material Fig. S2F-H). Because reducing the expression of an HDAC is sufficient to rescue most of the chn RNAi-induced midgut phenotypes, it suggests that Chn-mediated regulation of histone acetylation is an event upstream of other chromatin modification and is essential for ISC division.

We have used microarray analysis of genetically transformed adult Drosophila midguts to identify and characterize potential ISC/precursor cell regulators. A total of 1127 genes showed significant changes in at least one of the modified genetic backgrounds when compared with control midguts (http://dx.doi.org/10.6084/m9.figshare.855619). Analysis of the data in the various backgrounds has identified a group of genes (in the clusters c4, c5 and c8 in Fig. 2) that should have enriched expression in ISCs or precursors. Examples of these genes are Delta and escargot, two genes already well known to exhibit this expression pattern. Because ISCs are the only mitotic cells in the midgut, it is logical to expect that mitotic regulators should have enriched expression in ISCs. A substantial number of genes in clusters c4, c5 and c8 are indeed involved in cell cycle or mitotic regulation. Moreover, the Raf^GOF used probably pushed the cell cycle to become much faster and further increased the presence of these mitosis-associated transcripts in the ISC-like cells. Overall, the analyses presented in this report suggest that our approach has identified potentially important ISC or precursor regulators.

In depth analysis of one of the genes on the list, Chn, demonstrates its function to modulate the chromatin that is essential for ISC division and for EB precursor state (Fig. 7K). Previous reports suggest that Chn as a zinc-finger protein binds to the enhancers of achaete-scute complex genes, as well as other genes, including Delta (Escudero et al., 2005; Kania et al., 1995; Reeves and Posakony, 2005; Tsuda et al., 2006; Yamasaki et al., 2011). Further results suggest that Chn can function during development as either an activator or repressor of transcription in a context-dependent manner (Yamasaki et al., 2011). However, the binding sequences of Chn are highly divergent. Therefore, Chn may also have another means by which to regulate gene expression. We show that the prominent histone H3 acetylation in midgut precursor cells is dependent on Chn. Furthermore, the rescue of this acetylated H3 pattern by knocking down HDAC2 correlates with the rescue of ISC division.

A recent study reported that the epigenetic stability of Drosophila follicle stem cells increases during differentiation (Skora and Spradling, 2010). In the same study, the overall epigenetic status of differentiated cells was found to be more consistent than that of stem cells. Therefore, the differential effect of chromatin modulation in precursor cells but not in ECs after chn knockdown may be due to differences in the epigenetic flexibility of ISCs/precursor cells versus differentiated cells. It is possible that a chromatin structure in precursor cells allows Chn to bind to and maintain the chromatin in a precursor cell-specific state. When Chn is absent, the chromatin structure is altered to an extent that some important stem cell properties such as stem cell division are compromised and Delta expression is lost.

A proposed scenario for stem cell maintenance is that stem cells are kept in a state that prohibits them from differentiation (Fuchs and Chen, 2013; Stergiopoulos and Politis, 2013). Therefore, the loss of ‘stemness’ would directly lead to differentiation and therefore loss of stem cells. However, our analysis of Chn argues that stemness and initiation of differentiation can be a multi-step process. Although loss of Chn causes loss of stem cell properties required for ISC division, these chn⁻ ISCs do not enter differentiation automatically. Meanwhile, overexpression of Chn does not convert precursor cells into stem cells but can halt differentiation, and temperature-shift experiments demonstrate that this is a reversible process. This suggests that there is a separate set of differentiation-promoting factors that are antagonized by the overexpressed Chn. The expression of these differentiation promoting factors are under separate controls, perhaps established during the asymmetric division to form the EBs. Overall, our analyses demonstrate that Chn in the adult midgut is a novel regulator required for maintaining a precursor cell chromatin structure that is compatible with ISC division.

RNAi lines of chn, scute, pendulin, HMG D, H2AvD, HMG Z, RhoGEF4, ballchen, roughest and kugelkern were from Bloomington Stock Center (TRiP #26779, 26206, 27692, 31344, 28966, 26219, 31178, 31350, 28672 and 28750, respectively). UAS-chn and chn⁹ mutant lines were as described previously (Tsuda et al., 2006). Other stocks have been previously described (Amcheslavsky et al., 2011,, 2009). Viability tests and feeding experiments were as previously described (Amcheslavsky et al., 2011,, 2009). For lineage analysis, GFP-marked intestinal stem cell clones from MARCM were generated as previously described (Amcheslavsky et al., 2009; Lee and Luo, 2001). Fly stocks were crossed to generate the following genotype: hsFLP,UASGFP,tubGal4;FRT42Dchn⁹/FRT42D,tubGal80;+/+. The final cross and offsprings were maintained at 18°C. To induce MARCM clones, flies were heat shocked in a 37°C water bath for 45 min. The flies were then kept at 29°C for an additional 1-8 days, as indicated in the figure legends, before dissection.

For gut dissection, female flies were used routinely because of their bigger size, but male flies were also used to check the phenotypes. The entire gastrointestinal tract was pulled from the posterior end directly into fixation medium containing 1×PBS and 4% formaldehyde (Mallinckrodt Chemicals). Guts were fixed in this medium for 3 h, except for Delta staining where the fixation was for 0.5 h. Subsequent rinses, washes and incubations with primary and secondary antibodies were carried out in a solution containing 1×PBS, 0.5% BSA and 0.1% Triton X-100. For RNA isolation, the malpighian tubules, crops, esophagus and rectum were removed, leaving only the midgut. Approximately 15 midguts were put into a 1.5 ml centrifuge tube on dry ice and frozen immediately for subsequent RNA isolation.

For EdU labeling experiments, wild-type and chn RNAi flies were fed on 100 μM EdU (Invitrogen) in a 5% sucrose solution for 24 h. Guts were dissected and EdU incorporation was detected using Alexa Fluor 555 heat-shock assay (Click iT, Invitrogen) following the manufacturer's instruction. The immunostaining procedure has been described previously (Amcheslavsky et al., 2011,, 2009). The following antisera were used: anti-Delta (monoclonal, 1:100 dilution, DSHB, Iowa), anti-Prospero (monoclonal, 1:100, DSHB), anti-Armadillo (monoclonal, 1:100, DSHB), anti-HP1 (monoclonal, 1:50, DSHB), anti β-galactosidase (rabbit polyclonal, 1:50,000, 55976; Cappel, MP Biomedicals), anti-Cdc2 (rabbit polyclonal, 1:500, sc-53; Santa Cruz Biotechnology), anti-acetylated-Histone 3 (rabbit polyclonal, 1:500, 06-599; Millipore), anti H3K4me3 (rabbit monoclonal, 1:2000, C42D8; Cell Signaling), anti-GFP (rabbit polyclonal, 1:1000, A11122; Molecular Probes) and anti-Pdm1 (1:500, from Dr Xiaohang Yang, Zhejiang University, Hang Zhou, China). Secondary antibodies were used in 1:2000 dilutions as follows: goat anti-mouse IgG conjugated to Alexa 488, Alexa 568 or Alexa 555 (A11017, A11019, A21422; Molecular Probes), and goat anti-rabbit IgG conjugated to Alexa 488 or Alexa 555 (A11070, A21428; Molecular Probes). DAPI in Vectashield (Vector Labs) was used at 1:1 dilution in PBS. Images were taken by a Nikon Spinning Disk Confocal microscope (UMass Medical School Imaging Core Facility).

High-density oligonucleotide GeneChip Drosophila Genome 2.0 Arrays (Affymetrix 900531) containing 18,880 probe sets/chip were used to analyze expression profiles of wild-type and experimental genetic backgrounds (Raf^GOF, Raf^GOF+Notch^DN and Raf^GOF+Notch^IC). In these microarrays, each probe set contains 14 pairs of oligonucleotide to measure the level of transcription of each annotated sequence. In total, the array uses more than 500,000 data points to measure the expression of 18,500 transcripts and variants. Four hybridization controls (bioB, bioC and bioD from Escherichia coli, and cre from P1 bacteriophage), five poly-A controls (dap, lys, phe, thr and trp from Bacillus subtilis) and three housekeeping gene controls (actin 42A, glyceraldehyde-3-phosphate dehydrogenase 2, eukaryotic initiation factor 4a) were included for quality control purposes. cRNA sample preparation, array hybridization and scanning followed protocols as recommended by the manufacturer (http://www.affymetrix.com/index.affx). The hybridization intensity signal of each transcript was determined using the Affymetrix Microarray Suite 5.0 software package. Intensity values were scaled such that the overall fluorescence intensity of each array was equivalent. The GeneSpring (version 7.30; http://www.agilent.com) and GENECLUSTER (version 2.0; http://www-genome.wi.mit.edu/cancer/software/genecluster2/gc2.html) software packages were used to identify transcripts that were differentially expressed, and to conduct hierarchical two-dimensional clustering and self-organizing map analyses. The criteria for significant changes are: first, the signal values (after the initial flooring to an arbitrary value of 11 units) of a specific transcript in the wild-type and experimental conditions had to have a relative change of at least twofold; second, the absolute signal change must have at least 50 units; third, these changes had to be reproducible in all four independent replicate experiments; fourth, these changes had to demonstrate statistically significant P values ≤0.05. As a result, RNAs represented by 1127 probe sets showed significantly changed expression in at least one of the genetically modified backgrounds. Microarray data are available in the NCBI Gene Expression Omnibus (GSE58351).

We acknowledge the Bloomington Drosophila Stock Center and FlyTrap for providing fly stocks. Y.T.I. is a member of the UMass DERC (DK32520), a member of the UMass Center for Clinical and Translational Science (UL1TR000161) and a member of the Guangdong Innovative Research Team Program (No. 201001Y0104789252). L.T. thanks Dr Lim Young-Mi for preparing DNA and antibody. We thank Juerg Straubhaar for help with bioinfomatics.