The tumour suppressor L(3)mbt inhibits neuroepithelial proliferation and acts on insulator elements.

In Drosophila, defects in asymmetric cell division often result in the formation of stem-cell-derived tumours. Here, we show that very similar terminal brain tumour phenotypes arise through a fundamentally different mechanism. We demonstrate that brain tumours in l(3)mbt mutants originate from overproliferation of neuroepithelial cells in the optic lobes caused by derepression of target genes in the Salvador-Warts-Hippo (SWH) pathway. We use ChIP-sequencing to identify L(3)mbt binding sites and show that L(3)mbt binds to chromatin insulator elements. Mutating l(3)mbt or inhibiting expression of the insulator protein gene mod(mdg4) results in upregulation of SWH pathway reporters. As l(3)mbt tumours are rescued by mutations in bantam or yorkie or by overexpression of Expanded, the deregulation of SWH pathway target genes is an essential step in brain tumour formation. Therefore, very different primary defects result in the formation of brain tumours, which behave quite similarly in their advanced stages.

Development of the Drosophila nervous system recapitulates many steps in mammalian neurogenesis. Neurons in the adult fly brain arise from stem cells called neuroblasts which undergo repeated rounds of asymmetric cell division during larval stages [1,2]. After division, one daughter cell remains a neuroblast while the other is called the ganglion mother cell (GMC) and divides just once more into two differentiating neurons. Most larval neuroblasts are inherited from the embryo [3] but the so-called optic lobe neuroblasts (NB) located laterally on each brain lobe (Fig. 1d) pass through a neuroepithelial (NE) stage (Fig. 1e) and are therefore a particularly suitable model for mammalian neurogenesis [4,5]. During early larval stages, the NE cells of the optic lobes (OL) proliferate and separate into the inner (IOA) and outer (OOA) optic anlagen (Fig. 1c, control, cross section) [6,7]. During late larval stages, NE cells switch to a neurogenic mode. On the medial side, they generate optic lobe neuroblasts (OL NBs), which generate the neurons of the medulla, the second optic ganglion [4,8]. OL neurogenesis is controlled by a wave of lethal of scute (l(1)sc) expression passing through the neuroepithelium from medial to lateral [5]. The activity of the Jak/STAT pathway inhibits neural wave progression and thereby controls neuroblast number. Differentiation of neuroepithelial cells also involves the Notch, Epidermal Growth Factor (EGF) and Salvador-Warts-Hippo (SWH) pathways [9-13].

Figure 1

l(3)mbt is necessary to prevent tumorous over proliferation of the larval CNS

(a) Brains of control and l(3)mbt/Df(3R)D605 larvae raised at 32°C stained for Deadpan (anterior). Scale bars represent 50 μm.

(b) Brains of control and l(3)mbt mutant larvae stained for Miranda and PH3 (for overview see Supplementary Fig. S1b). Scale bars represent 20 μm.

(c) Brains of control and l(3)mbt mutant larvae expressing GAL4 and stained for Deadpan and E-Cadherin. Upper panels show top views (anterior side) with optic lobe neuroepithelia (NE) outlined. Middle panels show cross section through brain, optic lobe tissues are outlined: neuroepithelia (NE) of inner optic anlagen (IOA), NE of outer optic anlagen (OOA) and optic lobe neuroblasts (OL NBs). White boxes outline close up cross section views that are shown in the lowest panel. White arrow points to folding of IOA. Scale bars represent 50 μm in top rows and 20 μm in bottom row (close up).

(d) Schematic drawing of mid-late third instar larval brain (lateral view). Optic lobes consist of outer (OOA) and inner optic anlagen (IOA). Only OOA are shown, which consist of neuroblasts (NB, dark blue), neuroepithelial cells (NE, red) and lamina cells (La, grey). Central brain (CB) and ventral nerve cord (VNC) neuroblasts in light blue (larger circles) and CB neurons in green.

(e) Schematic drawing of neurogenesis in outer optic anlagen. Neuroepithelial cells (NE, red) give rise to medulla neuroblasts (NB, blue) and lamina cells (grey). NBs give rise to medulla neurons (green).

(f) Quantification of optic lobe neuroepithelial (NE) volume for the following genotypes: l(3)mbt/TM3 (N=16), l(3)mbt (N=5) and Df(3R)D605/l(3)mbt (N=4). N is the number of brain hemispheres quantified. Error bars indicate standard error of the mean (SEM).

(g) First and second instar stage optic lobes of control and l(3)mbt mutant brains stained for PH3 (left panel only) and Actin. Note that in late L2 inner and outer optic anlagen (IOA, OOA) are clearly detectable in control brains whereas they are indistinguishable in the l(3)mbt mutants. Scale bars represent 20 μm.

(h) Outer optic anlagen (OOA) (upper two panels) and inner optic anlagen (IOA) (bottom panels) of third instar brains expressing GAL4 at 29°C (top and bottom left panel) and 32°C (middle and bottom right panel) (see methods) stained for Miranda. Scale bars represent 50 μm.

Characterization of Drosophila genes identified in brain tumor suppressor screens has demonstrated that defects in neuroblast asymmetric cell division result in the formation of stem cell derived tumors that metastasize and become aneuploid upon transplantation [14,15]. These screens also identified lethal (3) malignant brain tumor (l(3)mbt) [16,17], a conserved transcriptional regulator [18] that is also required for germ-cell formation in Drosophila [19]. L(3)mbt binds to the cell cycle regulators E2F [20] and Rb [21] but the relevance of these interactions is unclear. We show that in Drosophila, L(3)mbt regulates target genes of the Salvador-Warts-Hippo (SWH) pathway (Fig. 2i) that are important in proliferation and organ size control [22-24]. The SWH-pathway is regulated by the membrane proteins Expanded (Ex) and Fat, which activate a protein complex containing the kinases Hippo and Warts to phosphorylate the transcriptional co-activator Yorkie [25-27]. Yorkie acts together with the transcription factors Scalloped [28-31] and Homothorax [32] to activate proliferative genes like Cyclin E and the microRNA bantam (ban) [33] and Drosophila inhibitor of apoptosis 1 (diap1, thread in Flybase). Upon phosphorylation, Yorkie is retained in the cytoplasm and its target genes are not activated. In Drosophila the main role of the SWH-pathway is to limit proliferation in imaginal discs and its absence leads to tumorous overgrowth [34]. In vertebrates, many homologs of key pathway members are tumor suppressors indicating that this function is conserved [34].

Figure 2

Candidate screen reveals function for SWH-pathway in optic lobe proliferation

(a) Optic lobes of control and l(3)mbt mutant brains expressing GAL4 and stained for aPCK and Cnn (same channel, surface and cross section views). Note the mitotic GAL4 positive cell within the mutant epithelium (white arrow). Scale bars represent 20 μm.

(b) Optic lobe epithelium in cross section view of control and l(3)mbt mutant brains stained for Actin. Scale bars represent 20 μm.

(c) Larval brains expressing a dominant active (DA) form of Egfr (λ-Top) display strong overproliferation of inner optic anlagen (IOA) neuroepithelia whereas a dominant negative (DN) form of Bsk has no phenotype. Brains express GAL4 and are stained for Actin, Miranda and Prospero. Scale bars represent 50 μm.

(d) Larval brains expressing a dominant active (DA) form of Jak (Hop) show overproliferation of outer optic anlagen (OOA, outline) whereas a dominant active (DA) form of FGF receptor lambda-Htl-H3 has no phenotype. All brains express GAL4 and are stained for Actin, Miranda and Prospero. Scale bars represent 50 μm.

(e) Larval brains expressing Hpo and anti-apoptotic p35 permanently activate the Salvador-Warts-Hippo (SWH) pathway and show underproliferation of optic lobe neuroepithelia (NE, outlined) whereas control brains show no phenotype. All brains express GAL4 and are stained for Actin, Miranda and Prospero. Scale bars represent 50 μm.

(f) Larval brains expressing dominant active (DA) Yki show strong overproliferation of optic lobe neuroepithelia (NE, outlined) but no overproliferation of neuroblasts (NB) whereas control brains show no phenotype. All brains express GAL4 and are stained for Actin and Miranda. Scale bars represent 50 μm.

(g) Quantification of optic lobe neuroepithelial (NE) volume of control (N=11) and Yki (N=4) expressing brains. N is the number of brain hemispheres quantified. Error bars indicate SEM.

(h) Larval brains expressing ban miRNA show overproliferation of optic lobe neuroepithelia (NE, outlined, arrows) whereas control brains show no phenotype. All brains express GAL4 and are stained for Actin and Miranda. Scale bars represent 50 μm.

(i) Schematic drawing of the Salvador-Warts-Hippo (SWH) pathway.

L(3)mbt contains three MBT domains which bind mono- or dimethylated histone tails [21,35]. Biochemical experiments in vertebrates have suggested a role in chromatin compaction [21] but whether this role is conserved is not known. Results published while this paper was under review have shown that germline genes are upregulated in l(3)mbt mutant brains and are necessary for tumor formation [36]. Our data indicate that L(3)mbt is bound to insulator sequences, which affect promoter-enhancer interactions and influence transcription [37,38]. In Drosophila, the proteins CTCF, CP190, BEAF-32, Su(Hw), Mod(mdg4) and GAF are found at insulator sequences but how these factors act is unknown [38].

Our data show that tumor formation in l(3)mbt mutants is initiated by the uncontrolled overproliferation of neuroepithelial cells in the optic lobes due to the upregulation of proliferation control genes normally repressed by the SWH-pathway. L(3)mbt is located at DNA sequences bound by chromatin insulators and we propose that the function of L(3)mbt as a chromatin insulator is essential for repressing SWH target genes and preventing brain tumor formation.

RESULTS

l(3)mbt tumors originate in the optic lobes

l(3)mbt mutants - like brat or lgl - showed a strong increase in neural stem cells positive for the neuroblast marker Deadpan (Fig. 1a and Supplementary Fig. S1a) resulting in an abnormal enlargement of the brain (Fig. 1a). However, in contrast to brat and lgl mutants [39-43], the asymmetric segregation of determinants was unaffected (Fig. 1b, Supplementary Fig. S1b and data not shown). Instead, we observed abnormal enlargement of the optic lobes. In wild type brains (Fig. 1c, control) the epithelial monolayer of the OOA forms OL neuroblasts at of its medial edge (Fig. 1d, e). In l(3)mbt mutants, the neuroepithelia of both IOA and OOA were massively expanded (Fig. 1c, middle and right panels). Initially, this led to a delay in OL neuroblast formation (Fig. 1c, top view) but later, OL neuroblasts were formed and their number was significantly increased (Fig. 1a, c). Unlike in wild type, neuroblast formation was also seen in the center of the OL epithelium (Fig. 1c, close up). To quantify the epithelial phenotype, we reconstructed optic lobe neuroepithelia (of IOA and OOA) in 3D (see methods). While OL epithelia were on average 1.2×105μm3 in wild type third instar larvae, their average size was significantly increased to 3×105μm3 in l(3)mbt mutants and 8.25×105μm3 in l(3)mbt larvae (Fig. 1f). During first larval instar (L1), cell number and mitotic pattern within the OL of l(3)mbt mutants were normal and during early second instar (L2 early), the IOA and OOA separated normally (Fig. 1g). However, starting in late second instar (L2 late) the mutant epithelia expanded and started folding. Since tumors did not form in l(3)mbt clones and available RNAi lines did not cause OL phenotypes, we generated a small hairpin micro RNA [44] to test cell autonomy of the overproliferation defect. Expression of l(3)mbt in central brain (CB) neuroblasts resulted in depletion of L(3)mbt protein (Supplementary Fig. S1d) however, this did not cause overproliferation (Supplementary Fig. S1c). In contrast, when expressed in neuroepithelia of the optic lobes l(3)mbt caused a strong overproliferation in the IOA and OOA (Fig. 1h). Epithelial expansion was also seen in l(3)mbt mutant wing imaginal discs (Fig. 3b, c, data not shown and [16]). In most adult wings, this resulted in a significant size increase (Supplementary Fig. S1e, f) but occasionally, very small deformed wings could also be observed. Taken together, these data suggest that overproliferation of neuroepithelial cells during late L2 stages initiates tumor formation in l(3)mbt mutants.

Figure 3

SWH target genes are misregulated in l(3)mbt mutants

(a) Simulated time course of optic lobe development from second to third larval instar stage in WT brains expressing diap1-GFP4.3 (outlined) and stained for Miranda and Actin. Note the changing expression pattern of diap1-GFP4.3 in optic lobes. Optic lobes consist of neuroblasts (NB), neuroepithelial cells (NE) and lamina cells (La). Scale bars represent 50 μm.

(b) diap1-GFP4.3 expressed in control and l(3)mbt mutant brains stained for Miranda (Mira, top panels, optic lobe neuroepithelia are outlined). diap1-GFP4.3 expressed in control and l(3)mbt mutant wing discs stained for DAPI (bottom panels, maximum intensity projections). Each genotype contains one copy of diap1-GFP4.3. Scale bars represent 50 μm.

(c) ex-lacZ (ex) expressed in control and l(3)mbt/Df mutant brains (top panels) stained for LacZ, optic lobe neuroepithelia are outlined. ex-lacZ expressed in control and l(3)mbt/Df mutant wing discs stained for LacZ (bottom panels, maximum intensity projections). Each genotype contains one copy of ex-lacZ. Scale bars represent 50 μm.

(d) bantam-Sensor-GFP expressed in control and l(3)mbt mutant wing discs. Note that bantam-sensor-GFP is reduced in l(3)mbt mutants reflecting an increase in bantam activity. Each genotype contains one copy of bantam-sensor-GFP. Scale bars represent 50 μm.

(e) Wing disc expressing GAL4; diap1-GFP4.3; UAS-l(3)mbt and stained for L(3)mbt (maximum intensity projection). Scale bars represent 50 μm.

(f) Wing disc expressing GAL4; bantam-sensor-GFP; UAS-l(3)mbt and stained for L(3)mbt (maximum intensity projection). Scale bars represent 50 μm.

(g) Wing disc expressing GAL4; ex-lacZ; UAS-l(3)mbt and stained for L(3)mbt and LacZ (maximum intensity projection). Scale bars represent 50 μm.

Signaling pathways controlling optic lobe proliferation

As staining of l(3)mbt mutants for aPKC and Actin did not reveal any change in apical basal polarity (Fig. 2a, b) we used the GAL4 driver line to express dominant active, dominant negative and RNAi constructs for the major signaling pathways, epigenetic complexes and epithelial polarity genes (Supplementary Fig. S2e). The GAL4 driver is expressed in the IOA and OOA from first to third larval instar and in imaginal disc epithelia (Supplementary Fig. S2a, b). We analyzed epithelial tissue size in the IOA, the OOA and the imaginal discs and estimated the number of OL neuroblasts (Supplementary Fig. S2c, d, e and see methods).

Activation of the epidermal growth factor (EGF) pathway promoted epithelial growth mainly in the IOA (Fig. 2c, S2e). Activation of the Jak/STAT pathway increased epithelial size (Fig. 2d, S2e) [5] whereas deregulation of Dpp or over activation of FGF pathways did not cause any visible phenotypes (Fig. 2c, d).

In contrast, inhibition of the SWH-pathway (Fig. 2i) resulted in a phenotype similar to l(3)mbt mutants (Fig. 2f, h, Supplementary Fig. S2e; see also [11]). expanded (ex) RNAi in the optic lobes (Supplementary Fig. S2c) caused epithelial overproliferation similar to what has been described for ex mutants [11]. Overexpression of Hippo together with the apoptotic inhibitor P35, strongly reduced the size of optic lobe epithelia (Fig. 2e). Upon expression of non-phosphorylatable Yorkie [26,27], the size of neuroepithelia in the IOA and OOA was five to ten fold increased (Fig 2f, g) a phenotype also observed by Reddy et al. (2010) [11]. A similar albeit milder phenotype was observed upon expression of the Yorkie target ban (Fig. 2h, S2e). Thus, inhibition of the SWH-pathway or overexpression of its target genes can recapitulate the increased OL proliferation seen in l(3)mbt mutants.

L(3)mbt tumor formation involves the SWH-pathway

The SWH-pathway inhibits expression of diap1, ban, and – as part of a negative feedback loop – ex (Fig. 2i) [45]. To investigate SWH-pathway activity we used a diap1-GFP4.3 reporter, in which the second transcriptional start site of diap1 (TSS2) controls GFP expression (see below for a map of the diap1 locus and Reference [30]). In wild type brains diap1-GFP4.3 is expressed in the neuroepithelium of the optic lobes during second and early third larval instars but becomes restricted to three small stripes during mid third instar (Fig. 3a). However, in l(3)mbt mutants, diap1-GFP4.3 remained expressed throughout the overgrowing neuroepithelium (Fig. 3b, note that the high GFP expression in the control is in lamina cells). Moreover, in wing imaginal discs, diap1-GFP4.3 was upregulated in the entire wing pouch in l(3)mbt mutants (Fig. 3b). The ex-lacZ reporter was only moderately expressed in wild type epithelia of the OOA but was upregulated in l(3)mbt mutants (Fig. 3c). A similar upregulation was seen in the wing imaginal disc (Fig. 3c). To test ban-miRNA activity we used a negative GFP-sensor carrying multiple ban binding sites [46]. When ban is active this GFP-sensor is downregulated. In control optic lobes the ban-GFP-sensor was not detectable indicating strong ban expression. The limited dynamic range of the ban-GFP-sensor did not allow us to detect further ban upregulation in the optic lobes (data not shown). In l(3)mbt mutant wing discs, however, GFP was almost completely lost indicating a strong increase in ban activity (Fig. 3d). Thus, l(3)mbt inhibits the expression of SWH target genes. These effects were cell autonomous and not a consequence of tumor formation since expression of l(3)mbt in the posterior compartment of the wing disc using GAL4 (engrailed-GAL4) increased expression of ban (Fig. 3f), diap1-GFP4.3 (Fig. 3e) and ex-lacZ (Fig. 3g) only in this compartment.

To test whether the SWH-pathway is important for tumor formation in l(3)mbt mutants, we investigated genetic interactions. In l(3)mbt and ban double mutants, neuroepithelial size was significantly reduced and tumors did not form (Fig. 4a, compare Supplementary Fig. S4a). Neuroepithelial size was reduced from 4.5×105μm3 in ban to less than 1×105μm3 in ban larvae at the same developmental stage (Fig. 4b). Overexpression of Expanded (Ex) rescued the overproliferation in l(3)mbt mutants (Fig. 4c, d; OL neuroepithelial size 3.3×105μm3 in l(3)mbt, 1.8×105μm3 in GAL4, l(3)mbt/+ or in GAL4, l(3)mbt). Finally, tumor size in l(3)mbt mutants was reduced by removing one copy of yorkie (Fig. 4e and Supplementary Fig. S4b, c, d; OL neuroepithelial size 3×105μm3 in l(3)mbt and 2.2×105μm3 in l(3)mbt/+). Thus the SWH-pathway and its target genes are important for tumor formation in l(3)mbt mutants.

Figure 4

SWH-pathway shows genetic interaction with l(3)mbt

(a) Genetic interaction of ban and l(3)mbt: ban/+, l(3)mbt mutant brain and ban double mutant brain stained for Actin, Miranda and Prospero. Double mutants display the ban phenotype (see also Fig. S4a). Scale bars represent 50 μm.

(b) Quantification of optic lobe neuroepithelial (NE) volume of ban/TM6 (control) (N=5), ban/+, l(3)mbt (N=5) and ban (N=6) brains. N is the number of brain hemispheres quantified. Error bars indicate SEM.

(c) Genetic interaction of Ex and l(3)mbt: GAL4/+ control brains, UAS-Ex, l(3)mbt mutant brains and GAL4 rescued brains stained for Actin, Deadpan and Prospero (Pros). Scale bars represent 50 μm.

(d) Quantification of optic lobe neuroepithelial (NE) volume of GAL4, l(3)mbt/+ control brains (N=5), UAS-Ex, l(3)mbt mutant brains (N=6) and GAL4 rescued brains (N=5). N is the number of brain hemispheres quantified. Error bars indicate SEM.

(e) Genetic interaction of l(3)mbt and yki: l(3)mbt single mutant brain and yki/+; l(3)mbt double mutant brain stained for Miranda, E-Cadherin and Prospero. Double mutants show suppression of l(3)mbt phenotype by halving Yki levels (see also Supplementary Fig. S4 b-d for further double mutant brains as well as quantification). Scale bars represent 50 μm.

L(3)mbt is a nuclear protein

The SWH-pathway regulates gene expression by excluding Yorkie from the nucleus [26,27]. Surprisingly, neither the subcellular localization of Yorkie (Supplementary Fig. S3b and data not shown) nor the expression of the associated transcription factor Scalloped [28-31] (Supplementary Fig. S3a) were changed in OL epithelia or wing discs of l(3)mbt mutants, indicating that L(3)mbt does not influence SWH signaling activity.

To analyze the subcellular localization of L(3)mbt we generated a specific antibody (Fig. 5a). In Drosophila embryos and in larvae, L(3)mbt was nuclear in interphase and dispersed in the cytoplasm during mitosis (Fig. 5a and data not shown). This staining was specific since it was lost from l(3)mbt mutant larvae (data not shown and Western Blot analysis in Supplementary Fig. S5c) and upon l(3)mbt expression (Supplementary Fig. S1d and Fig. 3e, f, g). The localization was confirmed in flies expressing functional RFP-L(3)mbt or GFP-L(3)mbt fusions (see also Supplementary Fig. S5a, b). Live imaging of larval neuroblasts expressing RFP-L(3)mbt revealed that the protein accumulates in nuclear dots in interphase cells (Fig. 5b). When expressed in salivary glands, L(3)mbt localized to multiple bands on polytene chromosomes that were often characterized by reduced DAPI staining (Fig. 5c). Thus, L(3)mbt is a nuclear protein that most likely exerts its function by associating with chromatin.

Figure 5

L(3)mbt localizes to the nucleus

(a) Larval brain (anterior) stained for L(3)mbt and Miranda. Overview and close up on central brain neuroblasts (NB, outlined) in inter- and anaphase are shown. Note dispersion of L(3)mbt in mitotic NBs. Scale bars represent 50 μm in overview and 10 μm in close up (panels). (Neuroepithelium = NE, outlined in overview)

(b) Stills of time lapse imaging of mitotic central brain NB expressing GAL4 and RFP-L(3)mbt. Note that RFP-L(3)mbt is reduced and dispersed in the cytoplasm during mitosis. Arrow points to RFP-L(3)mbt dots in interphase nuclei. Scale bars represent 10 μm.

(c) Close up of polytene chromosome expressing GAL4 and stained for DAPI. Arrow heads point to DAPI rich bands at which GFP-L(3)mbt is absent. Scale bars represent 1 μm. (See also Supplementary Fig. S5.)

L(3)mbt binds to SWH-pathway target loci

To test chromatin binding we performed chromatin immunoprecipitation (ChIP) followed by quantitative PCR analysis from third instar larval brains and imaginal discs (see methods). diap1 can be transcribed from three promoters, TSS1, TSS2 and TSS3 [28,30]. L(3)mbt bound moderately to TSS2 and strongly to TSS1 (Fig. 6a). Consistent with this, diap1-GFP5.1 (contains TSS1) was expressed in OL neuroblasts and neurons and strongly upregulated in l(3)mbt mutants (Fig. 6b). This is surprising because previous reports concluded that TSS1 is not expressed in wing or eye imaginal discs [28,30]. For TSS3, we detected weak binding of L(3)mbt, however, the relevance of this interaction is unclear (Fig. 6a). L(3)mbt did not bind to the bxd Polycomb response element (PRE) in the Ubx locus (Fig. 6a) [35] but bound to other SWH target genes like Cyclin E (Fig. 6a) [47].

Figure 6

L(3)mbt binds at the TSS and regulates SWH target genes and Jak/STAT pathway activity

(a) ChIP analysis at the th/diap1, CycE, Ubx and control loci in WT brain and imaginal disc tissues performed with antibodies against dSfmbt and L(3)mbt. ChIP signals at PREs (green boxes) and other regions are presented as percentage of input chromatin (distances from transcription start site (TSS) indicated in kb). Error bars correspond to standard deviation of three ChIP experiments.

(b) diap1-GFP5.1 (including the first TSS of diap1) expressed in control and l(3)mbt mutant brains stained for Miranda (single plane in top panels, maximum intensity projections in bottom panels). Scale bars represent 50 μm.

(c) Histogram of L(3)mbt binding frequency relative to the closest TSS.

(d) KEGG pathway analysis of L(3)mbt target genes (3% FDR). Purple bars represent the observed percentage of target genes in a particular KEGG pathway. Grey bars represent the percentage expected on the basis of all annotated genes. The significance (posterior probability) of this enrichment is based on a Bayesian model-based gene set analysis (MGSA) [48].

(e) 10xStat92E-GFP expressed in wing discs (top panels, maximum intensity projections) and brains (bottom panels, single planes) of l(3)mbt/+ control and l(3)mbt mutant larvae stained for Miranda (bottom panels only). Scale bars represent 50 μm.

(f) ChIP-Seq tracks for L(3)mbt (purple) and ChIP-chip tracks for the PcG proteins Ph, Pho and dSfmbt (grey) at the th/diap1 and bantam loci. Purple and grey boxes represent bound regions. For L(3)mbt bound regions at 0.5% and 3% FDR are indicated. Annotated genes and genome coordinate positions correspond to the Drosophila melanogaster BDGP Release 5 (UCSC dm3) assembly. Green lines indicate L(3)mbt low occupancy peak at −23kb, and Hth and Yki binding site at −14kb [32] relative to the bantam miRNA.

(g) Overlap of genes specifically deregulated in l(3)mbt mutants [36] (also called MBT signature (MBTS) genes) with genes bound within +/− 2 kb and +/− 5 kb by L(3)mbt. The number of L(3)mbt bound genes versus total number of genes in each category as well as corresponding percentages are shown.

(h) ChIP-Seq tracks for L(3)mbt (purple) at germline genes vas, bam and bgcn. Rectangles outline L(3)mbt binding sites at the TSS. Boxes indicate L(3)mbt bound regions at 0.5% and 3% FDR.

To determine binding sites of L(3)mbt on a genome-wide level, we used Solexa-sequencing. Results from two highly correlated independent ChIP experiments (Pearson correlation 0.8835, Supplementary Fig. S6a) showed that L(3)mbt bound close to TSS (Fig. 6c and see methods). We assigned each bound region to the closest gene (see methods) and conducted a biological pathway (KEGG; Fig. 6d) and gene ontology (GO; Supplementary Fig. S6c) analyses of the predicted target genes [48]. As the SWH-pathway is not yet annotated in these databases it was manually added to our analysis. SWH target genes are among the top 10 pathways overrepresented among L(3)mbt bound genes (data not shown). We identified seven out of eleven known and predicted SWH target genes (ban, CycA, CycB, CycE, E2f, diap1, fj, ex, Mer, wg, Ser) and found a significant enrichment (p-value 0.025) of these targets among genes bound by L(3)mbt (Fig. 6f, Supplementary Fig. S6e; 3% FDR). Importantly, the functional binding site at TSS1 of the diap1 locus (Fig. 6a) was confirmed (Fig. 6f). In addition, we found a low occupancy peak at a conserved site 23kb upstream of the ban transcription unit (Fig. 6f). Thus, L(3)mbt binds to multiple SWH target genes.

In our KEGG analysis we also found enrichment for Jak/STAT signaling genes (Fig. 6d and Supplementary Fig. S6e). Jak/STAT signaling previously has been shown to regulate optic lobe neuroepithelial growth [5]. Indeed, the 10xSTAT92E-GFP sensor was significantly upregulated in both OL neuroepithelia and wing discs of l(3)mbt mutants [49] (Fig. 6e). Finally, we found “oocyte maturation” among the top 20 processes enriched in the KEGG and GO analysis (data not shown and Supplementary Fig. S6c). L(3)mbt bound close to the TSS of vasa (vas), bag of marbles (bam) and benign gonial cell neoplasm (bgcn) (Fig. 6h), which may explain the defects in germ cell formation in l(3)mbt mutants [19,50].

Microarray analysis of various Drosophila brain tumors has identified a set of genes that are specifically deregulated in l(3)mbt mutants and are termed L(3)mbt signature (MBTS) genes [36]. 63% of these MBTS genes and 85% of the MBTS genes with a described germline function had L(3)mbt bound within the next +/−2kb (Fig. 6g, Supplementary Fig. S6d), suggesting that the identified L(3)mbt bound regions have in vivo relevance.

L(3)mbt binds to insulator elements

To analyze L(3)mbt binding specificity, we searched for DNA motifs enriched among L(3)mbt binding sites. Among seven DNA consensus motifs, four matched the consensus for the chromatin insulators CP190, BEAF-32, CTCF and Su(Hw) (Fig. 7a) [51-54] whereas, three did not match any known consensus motif (Supplementary Fig. S7a). Using published ChIP-chip data [51] we determined the overlap in binding sites (Fig. 7b). We found a strong overlap of L(3)mbt binding sites with class I chromatin insulators CP190, BEAF-32 and CTCF (Fig. 7b) whereas the overlap with the class II insulator protein Su(Hw) is smaller.

Figure 7

L(3)mbt binds at insulator-bound regulatory domains and influences Abd-B expression

(a) De novo identified sequence motifs significantly enriched in L(3)mbt bound regions (0.5% FDR) correspond to motifs for the insulator-associated proteins CP190, BEAF-32, CTCF and Su(Hw) [51,52]. Discovered motifs are depicted as sequence logos generated from a position weight matrix (PWM). Histograms show motif enrichment in L(3)mbt bound regions, computed as the ratio of the PWM match frequency in the dataset (purple) and the background (grey) frequency (significance of the enrichment is assessed using Fisher’s exact test, ***P<0.001).

(b) Venn diagrams showing the overlap of regions bound by L(3)mbt (0.5% FDR, purple) and regions bound by insulator-associated proteins (grey) CP190, BEAF-32 and CTCF (class I insulator proteins), and Su(Hw) (class II insulator protein) [51].

(c) ChIP-Seq tracks for L(3)mbt (purple) and ChIP-chip tracks for insulator-associated proteins (grey) CP190, BEAF-32, CTCF and Su(Hw) [51] at the Bithorax complex (BXC) locus. Purple and grey boxes represent bound regions. For L(3)mbt bound regions at 0.5% and 3% FDR are indicated. Annotated genes and genome coordinate positions correspond to the D. melanogaster R5/dm3 assembly. (See also Figure S7)

(d) Ventral nerve cord (VNC) of control (Canton S) and l(3)mbt mutant brains stained for Deadpan (Dpn) and Abdominal-B (Abd-B). Scale bars represent 50 μm. Graph shows quantification of the Abd-B mean fluorescence intensity normalized to Dpn in control (N=4) and l(3)mbt (N=6) animals. Error bars represent SEM, p-value (one-tailed) = 0.041 (N is the number of VNCs quantified).

(e) Wing disc expressing GAL4; bantam-sensor-GFP; UAS-mod(mdg4)-IR and stained for Mod(mdg4). Scale bars represent 50 μm.

The best-studied locus for insulator proteins is the Hox gene cluster of the Bithorax complex (BX-C) [37,55]. Remarkably, L(3)mbt binding sites within this locus correlated strongly with CP190 and CTCF but less with Su(Hw) (Fig. 7c). To test whether L(3)mbt binding sites in the Hox-cluster are functional, we analyzed the expression of the homeotic gene Abdominal-B (Abd-B). CTCF mutants show a characteristic downregulation of Abd-B at the posterior end of the larval ventral nerve cord (VNC) [55]. We found a significant reduction of Abd-B in l(3)mbt mutant larval CNS (Fig. 7d) that closely resembled the change seen in CTCF mutants. Notably, a strong correlation between L(3)mbt and chromatin insulator binding was also observed at the SWH target genes ban and diap1 (Supplementary Fig. S7b, c).

Since we did not observe any deregulation of the diap1-GFP4.3 reporter upon RNAi mediated knockdown of CTCF, CP190, BEAF-32 and Su(Hw), we tested the insulator protein Mod(mdg4) that binds to CP190 and Su(Hw) but not directly to DNA [56,57]. Upon RNAi mediated knockdown of mod(mdg4) bantam-sensor-GFP was strongly downregulated (Fig. 7e, indicating an increase in bantam activity) whereas the diap1-GFP4.3 reporter was only mildly upregulated (data not shown). Thus, insulator protein function is necessary to control SWH target gene expression.

DISCUSSION

A new mechanism for tumor formation in Drosophila

brat, lgl and dlg were identified as Drosophila brain tumor suppressors. In all cases, defects in asymmetric cell division cause a huge expansion of the neuroblast pool [39-42,58,59]. In l(3)mbt mutants, however, the neuroblast pool is expanded because an upregulation of SWH target genes results in a massive expansion of neuroepithelial tissue. Why those neuroblasts proliferate indefinitely upon transplantation [15] is currently not understood for any of those mutants.

While the SWH-pathway is essential for tumorigenesis in l(3)mbt mutants, its overactivation can not recapitulate the neuroblast tumor phenotype seen in l(3)mbt mutants (this study and Reference [11]). Similar to the multifactorial origin of mammalian tumors, therefore, the combined deregulation of several signaling pathways could be required. The Notch pathway could be involved as it regulates the formation of OL neuroblasts from neuroepithelia [10,12,13] and Notch pathway genes are bound by L(3)mbt (Table S1 and data not shown). We also observe increased activity of the Jak/STAT pathway, a major regulator of OL development [5]. Finally, the deregulation of germline genes in l(3)mbt mutants that has been described while this manuscript was under review [36] could provide another exciting explanation.

L(3)mbt acts on insulator elements

Our results indicate that L(3)mbt acts on insulator elements, which isolate promoters from the activity of nearby enhancers acting on other genes [37,38]. Our analysis showed that L(3)mbt binding sites overlap with CP190, CTCF and BEAF-32, placing the protein into what has been called the class I of chromatin insulators [51].

The identification of a DNA consensus motif for a histone binding protein like L(3)mbt is highly unexpected as insulators are typically nucleosome free [51]. Currently, the activity of these important transcriptional regulators could be explained in several ways [37,38]. Either, they form physical barriers blocking the interaction between enhancers and promoters. Alternatively, they mimic promoters and compete with endogenous promoters for enhancer interaction. Finally, they could interact with each other or nuclear structures to form loop domains that regulate transcriptional activity. Our data suggests another model in which insulators interact with histones on nearby nucleosomes and influence the structure of higher order chromatin. Importantly, in the regions flanking CTCF binding sites nucleosomes are enriched for histones that are mono- and di-methylated on H3K4 or mono-methylated on H3K9 or H4K20 [60], the variants to which MBT domains can bind in vitro [18,21,35,61]. As the human L(3)mbt homolog L3MBTL1 was shown to compact nucleosome arrays in vitro [21], a model becomes feasible in which simultaneous binding to insulators and the surrounding nucleosomes reduces flexibility and thereby restricts the ability of nearby enhancers to interact with promoters on the other side of the insulator. However, our data could equally well be worked into the other prevalent models for insulator activity. Since L(3)mbt is currently the only chromatin insulator besides CTCF that is conserved in vertebrates, analysis of its homologs will certainly allow to distinguish between those possibilities.

A conserved role for L(3)mbt

OL development resembles vertebrate neurogenesis [4]. Both processes consist of an initial epithelial expansion phase followed by neurogenesis through a series of asymmetric divisions [62]. Together with previous findings [11], our data demonstrate that l(3)mbt and the SWH-pathway are crucial regulators of the initial neuroepithelial proliferation phase. Interestingly, the SWH-pathway has been implicated in regulating neural progenitors in the chicken embryo [63] and it will be exciting to test the role of mammalian L(3)mbt in this process. It is remarkable that YAP is upregulated [64] and L3MBTL3 is deleted in a subset of human medulloblastomas [65]. Medulloblastoma is the leading cause of childhood cancer death and investigating the role of the SWH-pathway might contribute to the progress in fighting this disastrous disease.