The brain tumor gene negatively regulates neural progenitor cell proliferation in the larval central brain of Drosophila

In Drosophila, as in other holometabolic insects, neural stem cells called neuroblasts proliferate during two neurogenic periods (for a review, see Campos-Ortega and Hartenstein,1997). In the embryonic neurogenic period, individual neuroblasts delaminate in a stereotyped pattern from the neuroectoderm and divide repeatedly and asymmetrically to generate a new neuroblast and a smaller daughter cell, called a ganglion mother cell (GMC)(Campos-Ortega, 1997; Edenfeld et al., 2002; Skeath and Thor, 2003; Technau et al., 2006). Each GMC is a transient intermediate progenitor cell that generally divides once to produce two lineage-specific postmitotic ganglion cells (GCs), which subsequently initiate their differentiation processes. Towards the end of embryogenesis, most neuroblasts stop proliferating and enter a period of quiescence. During the second, larval neurogenic period, most neuroblasts resume proliferation and generate the majority of the cells that comprise the central brain and ventral ganglia of the adult(White and Kankel, 1978; Truman and Bate, 1988; Prokop and Technau, 1991; Ito and Hotta, 1992; Truman et al., 2004).

In contrast to embryonic neurogenesis, less is currently known about the genetic systems which control and limit the activity of neural progenitor cells during postembryonic development, notably in the developing brain, where over 90% of the neurons in the adult central nervous system are generated. A candidate gene for proliferation control in the fly brain is the tumour-suppressor gene brain tumor (brat) which encodes a member of the conserved NHL family of proteins that regulate growth and differentiation in a variety of organisms(Arama et al., 2000). Recessive mutations in brat lead to neoplastic overgrowth, characterized by a dramatic enlargement of the larval brain hemispheres and the ability of pieces of brain tissue to metastasize and form secondary tumours when injected into the abdomen of a wild-type host fly(Kurzik-Dumke et al., 1992; Woodhouse et al., 1998; Arama et al., 2000). However,the functional role of the brat gene in this process is not clear since the neural phenotype has not been characterized in detail and initial findings are contradictory. Thus, the brain overproliferation phenotype of brat mutants is thought to be primarily due to expansion of optic lobe neuroblasts (Kurzik-Dumke et al.,1992; Gateff et al.,1993), yet the major sites of brat expression in the CNS throughout development are not the optic lobes, but rather the central brain and ventral nerve cord (Arama et al.,2000). In addition to its involvement in brain development, the brat gene acts in translational repression during early embryogenesis(Sonoda and Wharton, 2001) and negatively regulates cell growth and ribosomal RNA synthesis in imaginal discs(Frank et al., 2002); whether brat acts by similar molecular mechanisms in the brain is not known.

In this report, we use molecular markers and mosaic analysis to study the role of brat in neural proliferation control and tumour suppression in the Drosophila brain. Our findings indicate that bratmutation leads to loss of proliferation control in the central brain. Mosaic analysis with a repressible cell marker (MARCM) analysis demonstrates a larval requirement for brat since induction of mutant clones in the first larval instar is sufficient to cause massive overproliferation in the central brain that continues into adulthood. Molecular analysis of brat MARCM mutant clones demonstrates that loss of brat function leads to loss of nuclear Prospero (Pros) expression. Both brat and prosmutant MARCM clones comprise supernumerary cells, which do not express neuron-specific markers such as Embryonic lethal abnormal vision (Elav). By contrast, the vast majority of cells in mutant clones strongly express molecular markers and cell cycle control genes which characterize neural progenitor cells of the wild-type postembryonic brain. Detailed clonal analysis indicates that loss of brat function, and loss of pros function, cell-autonomously affect the differentiation of ganglion cells. Targeted expression of wild-type brat or wild-type pros in brat mutant clones restores cell cycle exit and differentiation of brat mutant cells, which in turn abrogates central brain tumour formation. Taken together, our results demonstrate that the tumour suppressor brat negatively regulates cell proliferation during larval central brain development of Drosophila and identify prospero as a downstream effector of brat in cell fate specification and proliferation control.

Most fly stocks carrying transgenes and recombinant chromosomes were obtained from the Bloomington Stock Center and their detailed description can be found at FlyBase(http://flybase.bio.indiana.edu). Other stocks were kindly provided by colleagues: GAL4^NP3262 (GETDB database, http://flymap.lab), FRT40A, brat¹¹ (Frank et al., 2002) (a gift from D. Frank); FRT82B,pros¹⁷ (Reddy and Rodrigues, 1999) (a gift from V. Rodrigues); UAS-brat^FL1B (Sonoda and Wharton, 2001) (a gift from R. Wharton); UAS-pros^L(wt)9D (a gift from C. Doe). The following recombinant chromosomes other than described in Bello et al.(Bello et al., 2003), were generated for this study: brat¹¹, GAL4^NP3262;FRT40A, UAS-mCD8::GFP^LL5, UAS-nlslacZ^20b; FRT40A,UAS-mCD8::GFP^LL5; FRT40A, brat¹¹,UAS-mCD8::GFP^LL5, UAS-nlslacZ^20b. GFP labelling of wild-type and zygotic brat mutant larval nervous systems was achieved by crossing brat¹¹, GAL4^NP3261/CyO,actGFP^JMR1 to FRT40A; UAS-mCD8::GFP^LL6,UAS-nlslacZ^J312 or FRT40A, brat¹¹/CyO,actGFP^JMR1; UAS-mCD8::GFP^LL6,UAS-nlslacZ^J312. Heterozygous wild-type and homozygous brat mutant larvae lacking CyO, actGFP^JMR1 were selected under the fluorescence microscope. The detailed MARCM genotypes indicated in the figure legends were obtained from various stocks assembled using standard genetics.

For MARCM experiments, embryos of the appropriate genotype were collected on yeasted grape juice-agar plates over a 4- to 6-hour time window and raised at 25°C for 24-28 hours before heat shock treatment. Heat-shock induction of FLP in the newly hatched larvae was performed by immersing the plates at 37°C in a water bath for 60 minutes. Larvae were then collected and plated at low density on standard cornmeal/yeast/agar medium supplemented with live yeast. GFP-labelled MARCM clones were examined in mid (66-72 hours) to late(96-120 hours) third instar larval central nervous systems (CNSs).

Larval tissues were fixed and immunostained as previously described(Bello et al., 2003). To generate a neural progenitor cell-specific anti-grainy head (Grh) antibody, a synthetic peptide covering 19AA of Grh [aa 798-816 of the O-isoform(Uv et al., 1997)] was used for immunization of rabbits (carried out by Neosystems). Polyclonal rabbit anti-Grh 711 was used at a 1:200 dilution. Other primary antibodies used were:rat anti-Brat (Sonoda and Wharton,2001) 1:200; rabbit anti-β-galactosidase (β-gal; Cappel)1:7000; mouse anti-β-gal (Promega) 1:1000; rabbit anti-PH3 (Upstate Biotechnology) 1:400; mouse anti-Mira [mAb81(Matsuzaki et al., 1992); a gift from P. Overton] 1:50; mouse anti-Pros (mAbMR1A, Developmental Studies Hybridoma Bank, DSHB, Iowa University) 1:5; mouse anti-cyclin B (CycB)(mAbF2F4, DSHB) 1:10, mouse anti-CycE (mAb8B10; a gift from H. Richardson)1:50; rat anti-Elav (mAb7E8A10, DSHB) 1:30; and mouse anti-Grh (BF1; a gift from S. Bray) 1:3. All fluorescent images were captured using a Leica TCS SP scanning confocal microscope. Optical sections ranged from 0.5 to 2 μm,recorded in line average mode with picture size of 512×512 pixels. Captured images from optical sections were arranged and processed using IMARIS(Bitplane) or ImageJ software. Figures were arranged and labelled using Adobe Photoshop.

Inactivation of brat alleles in homozygous individuals results in the production of tumour-like neoplasm in the larval brain(Arama et al., 2000). The overgrowth of brain hemispheres is gradual during larval development and becomes prominent in whole mounts of late third larval instars(Fig. 1A,B). At earlier larval stages, brain hemispheres in brat mutants still have the same overall size as in wild type; in both cases, the larval brain hemispheres are characterized by a central brain region and laterally located developing optic lobes (Fig. 1C,D). Examination of cell morphology in the wild-type central brain reveals a limited number of large cells, which are the postembryonic neuroblasts, associated with smaller cells, which are their progeny (Fig. 1E). By contrast, the brat mutant central brain is characterized by dense cellular masses of numerous small and pleiomorphic cells in addition to a limited number of large cells(Fig. 1F).

In order to gain insights into the origin and genetic mechanisms of tumour formation caused by loss of brat function, we used MARCM clonal analysis to label individual brat mutant neural lineages in a wild-type heterozygous background (Lee and Luo, 1999; Lee and Luo,2001). Positively marked somatic clones mutant for brat¹¹, a strong loss-of-function allele of brat,were generated and examined 4 days later in late third instar larvae. Wild-type clones were induced in parallel as control, hence for each genotype(wild type and brat) we examined more than 300 specimens carrying 5-20 clones in each brain hemispheres. Our analysis indicated that brat mutant clones in the larval optic lobes are indistinguishable from wild-type clones. By contrast, brat mutant clones in the larval central brain populated large areas of the brain hemispheres(Fig. 2C), as compared to wild-type central brain clones (Fig. 2A). Whereas wild-type clones are composed of 50-100 cells in average (n>100), the majority of late third instar bratmutant clones (n>100) were up to ten times larger than control clones. Thus, less than 20% of brat mutant clones comprised <100 cells, whereas more than 80% of the clones comprised 200-1000 cells. Examination of mutant clones in early third instar larvae revealed clearly distinguishable and separated brat mutant clones (not shown), whereas in late third instars, we frequently observed hemispheres within which the central brain was almost entirely labelled with GFP(Fig. 2C). This suggested that multiple brat mutant clones had already merged into an indistinguishable labelled cell mass overgrowing the whole central brain.

By contrast to the central brain area, brat mutant clones in larval ventral ganglia appeared normal, both in size and shape, when compared to wild-type control clones (Fig. 2B,D). These data suggest that brat affects cell proliferation in a cell-autonomous manner specifically in the larval central brain. We, therefore, wondered whether we could identify, more precisely, the central brain area within which proliferation defects in brat mutant clones occur. Our clone induction protocol revealed that increase in size of mutant clones initiated frequently in the dorsal central brain, however, due to the lack of region-specific anatomical markers, we could not allocate this region in more detail. Nevertheless, we could exclude the possibility that the affected central brain area covers the mushroom body neuroblasts. In the wild-type central brain, mushroom body neuroblasts are exceptional with regard to frequency and temporal window of proliferation(Truman and Bate, 1988). Yet, brat mutant clones of mushroom body neuroblast lineages did not result in enlarged clones (data not shown).

To investigate the nature of the cells that overproliferate in brat mutant lineages, we first defined a set of molecular markers that enabled us to characterize the cell types in wild-type neural lineages. Wild-type MARCM clones examined in late third instar animals typically contain a single large neuroblast (∼10 μm in size) associated with a discrete packet of smaller cells that corresponds to its entire progeny generated during larval development (Fig. 3A) (Bello et al.,2003). The latter population is mostly composed of immature adult-specific ganglion cells which extend neurites towards the inner neuropile. Neuronal extensions associate in a common bundle known as cell body fibre tract (CBT) (Fig. 3A).

Neuroblasts express the coiled-coiled protein Miranda (Mira) at the cell cortex (Ceron et al., 2001; Akong et al., 2002). Cortical Mira becomes polarized during mitosis forming a crescent that segregates into the budding GMC (Fig. 3C). The transcription factor Grainy head (Grh) also labels the large neuroblast(Fig. 3H). The mitotic profile of the neuroblast is revealed by expression of the cell cycle markers phosphorylated histone H3 (PH3) (Fig. 3B,D) and CycE (Fig. 3D,E). Careful inspection of mitosis using the nuclear localisation of PH3 and visualisation of centrosomes further indicates that CycE protein is detectable in neuroblasts throughout the cell cycle.

Although GMCs are not distinguishable morphologically from the surrounding postmitotic ganglion cells, their identification in close proximity to the parental neuroblast is defined in wild-type clones by expression of the cell cycle markers CycE (Fig. 3D,E)and/or PH3 (Fig. 3E-G), as well as nuclear Grh expression (Fig. 3H) (Cenci and Gould,2005; Almeida and Bray,2005) and association of Mira at the cellular cortex(Fig. 3F,I). Beside their difference in size, two molecular characteristics make it possible to distinguish neuroblasts and GMCs. First, expression of CycE is detectable in GMCs only at interphase (PH3-negative nucleus, Fig. 3D) but not during mitosis(PH3-positive nucleus, Fig. 3E). Second, localisation of Mira at the cellular cortex is never polarized during GMC mitosis (PH3-positive nucleus in Fig. 3F).

GMCs are also characterized by nuclear localization of Pros, which is clearly detectable by co-staining with PH3(Fig. 3G) or Grh(Fig. 3H). In addition, nuclear Pros expression is also detectable in postmitotic ganglion cells(Fig. 3G,H), which also express the neuronal marker Elav (Fig. 3I). These data show that in wild-type neural lineages of the larval central brain, only neuroblasts and GMCs are actively engaged in the cell cycle, and that each cell type can be unambiguously identified by a unique combination of markers (Table 1).

We next used these molecular markers to characterize brat mutant lineages in the central brain. Previous genome-wide transcript profiling of adult brat^k06028 mutant brains had revealed high expression levels of cell cycle regulators(Loop et al., 2004). To determine whether the observed increase in cell number reflects changes in cell cycle regulation, we analyzed the distribution of cell cycle markers in brat¹¹ clones using antibodies against CycE, CycB and PH3. In sharp contrast to wild-type neuroblast lineages, most brat mutant cells within a labelled mass of clones expressed CycE(Fig. 4A) and CycB (not shown)and a large number of cells appeared engaged in mitosis as indicated by PH3 staining of numerous nuclei (Fig. 4B,E).

With few exceptions, all mutant cells within the clones lacked immunoreactivity for nuclear Pros (Fig. 4C,C′) and nuclear Elav(Fig. 4D), unlike the surrounding heterozygous wild-type cells devoid of GFP labelling. Moreover, we never observed any cellular extensions comparable to axon fascicles,suggesting an invariable lack of axonal processes. Since MARCM clones allow unambiguous identification of neuronal processes in both, wild-type and various mutant lineages, due to stable GFP labelling(Lee and Luo, 1999; Lee and Luo, 2001), these data suggest that brat mutant cells are impaired in differentiation. In support of this notion, we observed that the vast majority of cells in brat mutant clones showed both nuclear Grh(Fig. 4C,C′) and cortical Mira (Fig. 4E). Close inspection of clones revealed that a few large cells scattered at the surface showed polarized Mira distribution in a cortical crescent. As in wild-type neuroblasts, this asymmetric distribution of Mira was always associated with PH3 immunoreactivity, suggesting the presence of several asymmetrically dividing neuroblasts because of the merger of individual GFP-labelled brat mutant clones (see also Fig. 2C). In contrast to these large cells displaying Mira in a cortical crescent, the majority of mutant cells showed uniform cortical distribution of Mira even during mitosis(Fig. 4E). These data indicate that brat mutant clones comprise an excessive number of neural progenitor cells that are unable to exit the cell cycle but continue to proliferate.

Previous studies have shown that cessation of proliferation in the developing Drosophila brain occurs during puparium formation and metamorphosis (Ito and Hotta,1992). We, therefore, wondered whether brat mutant clones cease proliferation or whether they are able to escape cell cycle stop signals during later stages of development. MARCM brat¹¹ clones were induced in first instar larvae and immunostained for PH3 in adult brains at 1 day, 2 weeks and 3 weeks after eclosion. Control wild-type clones induced in parallel and examined in the brain of adult flies comprised a limited number of cells that never showed PH3 immunoreactivity(Fig. 5A,B), consistent with the general notion that the adult brain is composed of postmitotic cells. In contrast, brat mutant clones appeared dramatically enlarged in size and cell number in the adult brain (Fig. 5C). Moreover, a large number of mutant cells displayed PH3 immunoreactivity even in 3-week-old adult flies, suggesting that these mutant cells were still mitotically active (Fig. 5D). These data indicate that loss of larval bratfunction results in mutant cells that are unable to respond to cell cycle termination signals during metamorphosis.

The features of brat mutant cells reminded us of the CNS phenotype observed in pros mutant embryos. During embryonic CNS development,Pros is required for both cell fate specification and proliferation control of neural progenitor cells. In the absence of Pros function, differentiation genes are not activated and progenitor cell-specific genes are not repressed(reviewed by Lu et al., 2000). Loss of Pros function results in continued mitotic activity that correlates with de-repressed, ectopic transcription of cell cycle genes(Li and Vaessin, 2000). This prompted us to investigate whether loss of nuclear Pros expression in brat mutant clones could be causally related to the overproliferation phenotype observed.

We therefore generated wild-type MARCM clones and clones mutant for the loss-of-function allele pros¹⁷. When assayed in third instar larvae, nuclear β-gal or GFP-labelled pros¹⁷clones appeared to have increased in size and be populating large areas of the central brain (Fig. 6B) when compared to wild-type control clones (Fig. 6A). The majority of pros mutant central brain clones were five to ten times larger than wild-type control clones(n>100). Thus, less than 5% of pros mutant clones comprised less than 100 cells, whereas more than 90% of these clones comprised 200-1000 cells. Occasionally we also observed brain hemispheres in which the central brain was almost entirely labelled with nuclear β-gal or GFP,suggesting that multiple pros mutant clones had already merged to an indistinguishable labelled cell mass. Close inspection of such a nuclearβ-gal-labelled cell mass revealed the presence of numerous pleiomorphic cells including large neuroblast-like cells(Fig. 6C).

In order to correlate Pros expression and function to those of Brat, we carried out anti-Brat immunostaining on third instar larval CNS consisting of either wild-type or pros mutant central brain clones(Fig. 6A,B). In both cases, we observed cytoplasmic Brat immunolabelling in all cells of the larval CNS(Fig. 6A,B), regardless of their size and morphology (Fig. 6C,C′). These data suggest that Brat is ubiquitously expressed in the developing wild-type CNS consistent with earlier reports on Brat RNA in situ hybridization and enhancer trap expression studies(Arama et al., 2000). Importantly, we did not observe any difference in anti-Brat immunoreactivity between GFP-labelled pros mutant cells and surrounding wild-type heterozygous cells (Fig. 6C,C′). These data suggest that Brat expression is unaffected in larval central brain pros mutant clones.

Next, we utilized the molecular markers used to characterize wild-type and brat mutant lineages to further analyze pros mutant clones. Almost all cells in the clones were found to express CycE(Fig. 6D,D′), CycB(Fig. 6E), and numerous cells displayed PH3 immunoreactivity (Fig. 6D″). Co-labelling with PH3 and CycE showed that CycE is present at significant levels in most mitotic cells, regardless of morphology,size or position within these clones (Fig. 6D′,D″). GFP-labelled pros mutant clones did not show any cellular extensions comparable to axon fascicles, suggesting an invariable lack of axonal processes. Moreover, pros mutant clones lacked Elav immunoreactivity (Fig. 6F,F′), suggesting the lack of differentiated ganglion cells. By contrast, most pros mutant cells showed nuclear Grh(Fig. 6G,G′) and cortical Mira (Fig. 6F-G′). As for wild-type neuroblasts, few large cells scattered at the surface of clones showed polarized Mira distribution in a cortical crescent(Fig. 6F′,G′,arrows), suggesting the presence of several asymmetrically dividing neuroblasts. However, the vast majority of pros mutant cells showed uniform cortical distribution of Mira (Fig. 6F′,G′). These findings indicate that prosmutant clones phenocopy larval central brain brat mutant clones.

The remarkable similarities between brat and pros mutant clones suggested that both genes might act in the same pathway regulating proliferation in the larval central brain. Hence, we used MARCM to express full-length wild-type pros specifically in brat mutant clones. As a control we tested the ability of a full length wild-type brat cDNA (Sonoda and Wharton,2001) to rescue the brat mutant brain phenotype. Targeted expression of wild-type brat in brat mutant clones(n>50) resulted in a significant reduction of clone size in the central brain of third instar larvae (Fig. 7A-B′). Thus, more than 90% of late third instar central brain clones comprised <100 cells, whereas less than 10% of these clones comprised 100-500 cells. We frequently observed that individual clones merged to form a nest of GFP-labelled cells (Fig. 7A,B), however clones were still distinguishable because only a few superficially located cells showed nuclear Grh(Fig. 7A-B′) and cortical Mira (data not shown). Moreover, most cells within clones expressed Elav(Fig. 7A,A′) and Pros(Fig. 7B,B′) in their nuclei and they formed axonal processes contributing to individual fibre tracts (Fig. 7A,B), indicating the presence and proper differentiation of postmitotic ganglion cells. These data suggest that UAS-driven expression of the brat transgene in brat mutant clones results in clones that are comparable in size,number and marker gene expression to wild-type controls. In addition, even when examined in the adult brain, these clones appeared similar to wild-type control clones and lacked PH3 staining (data not shown).

Targeted expression of wild-type pros in brat mutant clones (n>50) resulted in a very similar phenotype(Fig. 7C-D′). More than 90% of late third instar central brain clones comprised <100 cells, whereas less than 10% of these clones comprised 100-500 cells. Similar to the brat rescue, a significant decrease in mitotic activity occurred, as exemplified by clone size and by PH3-immunoreactivity restricted to a few large cells and to a limited number of smaller associated cells, both of which expressed cortical Mira and nuclear Grh(Fig. 7C-D′). The presence and proper differentiation of postmitotic ganglion cells could be assessed by the detection of Pros (Fig. 7D,D′) and Elav (not shown) in the vast majority of cells within the clones. Moreover, most cells formed axonal processes assembled in wild-type-like fibre tracts (Fig. 7C,D). Thus, as for brat, targeted expression of pros in brat mutant clones results in clones that are comparable to wild-type lineages in size, number, and marker gene expression. These results demonstrate that pros can promote cell cycle exit and differentiation of brat mutant cells in a wild-type-like pattern,which in turn abrogates central brain tumour formation.

Previous studies suggested that brat loss-of-function mutants lead to massive cellular overgrowth and tumour formation in larval optic lobes of Drosophila. These studies also indicated an embryonic requirement for brat to suppress tumour formation(Kurzik-Dumke et al., 1992; Gateff et al., 1993; Woodhouse et al., 1998; Arama et al., 2000). By contrast, our analysis showed that the brat overproliferation phenotype is due to loss of proliferation control in the larval central brain;the optic lobes initially appear wild-type-like but subsequently are overgrown by neoplastic central brain brat mutant tissue. This conclusion is further supported by MARCM clonal analysis which demonstrated that loss of brat function causes overproliferation in the larval central brain only.

Our in vivo mosaic analysis reveals a cell-autonomous, larval requirement for brat to limit cell proliferation in the brain. Although brat is expressed in all parts of the nervous system both in the embryo (Arama et al., 2000) and larva (our study, Fig. 6A-C),induction of brat mutant clones in the first larval instar is sufficient to cause massive overproliferation in the central brain but not the ventral ganglia. This may suggest that either unknown compensatory mechanisms actively suppress a brat mutant phenotype in the larval ventral ganglia, or that this reflects region-specific differences in cell cycle control. Indeed, transcriptional activity of the mitotic regulator string/Cdc25 is regulated by a plethora of cis-acting elements, most of which are devoted to differential control of cell proliferation during embryonic and larval neurogenesis (Lehmann et al., 1999).

During postembryonic neurogenesis, intense proliferation takes place in the brain (Truman and Bate, 1988; Ito and Hotta, 1992). Our analysis shows that central brain brat mutant clones display sustained cell cycle marker expression, indicating that mutant cells are unable to withdraw from the cell cycle. This is further supported by the presence of enormous brat mutant clones with pronounced proliferative activity even in 3-week-old adult brains, an observation that contrasts with the postmitotic adult wild-type brain. Previous studies have shown that cessation of proliferation in the developing Drosophila brain occurs during metamorphosis (Ito and Hotta,1992), although the underlying genetic mechanisms are currently unknown. The elevated and aberrant cell cycle activity of central brain brat mutant cells suggests that these cells are either able to escape or that they lack cell cycle termination signals(Peterson et al., 2002; Bello et al., 2003).

Our mosaic analysis demonstrates that enlarged brat mutant clones comprise cells that display sustained expression of neural progenitor cell markers, and simultaneously lack marker gene expression specific for differentiating ganglion cells. Indeed, lack of axonal processes suggests that brat mutant clones comprise an excessive number of mutant cells that are unable to exit the cell cycle and hence do not differentiate into ganglion cells but rather continue to proliferate. These data indicate that brat mutation impairs proliferation control of neural progenitor cells, namely either neuroblasts and GMCs or only one of these progenitors,since in the wild-type central brain only these two cell types are actively engaged in the cell cycle. Based on our analysis we cannot distinguish unambiguously between the two possibilities and the underlying mechanisms. We do, however, exclude the possibility that differentiating ganglion cells de-differentiate due to brat mutation, because we consistently observed lack of differentiation right after clone induction and also at any later stages of mutant clone development. This was especially exemplified by the lack of nuclear Pros expression, which in the wild type is unambiguously detectable in differentiating progeny of larval neuroblast lineages, namely GMCs as well as ganglion cells (see Fig. 3) (see also Ceron et al.,2001; Akong et al.,2002).

Moreover, our loss-of-function analysis indicates that brat mutant MARCM clones lack Pros and also phenocopy pros mutant clones. Thus,enlarged pros mutant clones consist of cells that are devoid of Elav expression, that lack axonal processes but display sustained expression of Grh and Mira as well as cell cycle markers such as CycE, CycB and PH3. These data suggest that mutant clones are essentially devoid of terminally differentiating postmitotic ganglion cells, indicating that Pros functions like Brat in terminating neural progenitor cell proliferation and inducing ganglion cell differentiation. In the embryonic CNS, Pros functions to terminate cell proliferation by repression of cell-cycle activators and simultaneously to induce a differentiation program, effectively coupling the two events. This Pros function appears to be warranted by its localization in the basal cortex of asymmetrically dividing neuroblasts and hence its distribution to only one daughter cell, the GMC. Upon completion of mitosis,Pros translocates from cytoplasm into the nucleus where it executes its transcriptional program ensuring both terminal division of the GMC and cell differentiation of its progeny (for a review, see Prokopenko and Chia, 2005). In the larval CNS we observe nuclear localisation of Pros in GMCs and ganglion cells but not in the neuroblast, suggesting that Pros has comparable functional features in larval central brain neurogenesis.

In addition, our results provide evidence that Pros acts downstream of Brat in neural proliferation control. The following points support this notion: (1) brat mutant clones lack nuclear Pros; (2) brat and pros mutant clones are indistinguishable both at the morphological and at the molecular level; (3) Brat expression is unaltered in prosmutant clones, which together with point no. 1 strongly suggests that Brat is epistatic over Pros; and (4) trans-activation of wild-type pros in brat mutant clones is sufficient to promote both cell cycle exit and differentiation. Our experiments, however, do not provide any evidence about the direct or indirect nature of their interaction. Since overexpressed Pros is detected specifically in brat mutant clones in a wild-type-like pattern (Fig. 7D′), we exclude the possibility that brat acts as a translational repressor of Pros, comparable to its role in hunchback repression during embryonic abdominal segmentation(Sonoda and Wharton, 2001). In addition, brat mutation apparently does not affect prostranscription, since pros RNA in situ hybridization in zygotic brat mutants produced a pattern indistinguishable from wild-type controls (data not shown). Thus, Brat and Pros may act indirectly in the same pathway, regulating progenitor cell proliferation control in the brain. Alternatively, Brat may act in a process required to cargo Pros, comparable to the function of its mammalian homolog BERP(Yan et al., 2005).

Our in vivo mosaic analysis demonstrates that a single mutation in either brat or pros is sufficient to cause brain tumour formation in a cell-autonomous manner, suggesting that indefinite proliferation of brat and pros mutant cells is a cell intrinsic property(Caussinus and Gonzalez, 2005). GFP-labelled MARCM cells each derive from a common precursor cell, implying that brat and pros mutant cells all descend from individual tumour cells of origin and hence lead to brain tumour formation in a clonally related manner. Moreover, our data indicate that pros and brat mutant clones in the larval central brain are composed of an excessive number of mutant progenitor cells that are unable to differentiate into ganglion cells but rather continue to proliferate (summarized in Fig. 8). In this sense our results provide in vivo support for the notion that the initiating event in the formation of a malignant tumour is an error in the process of normal differentiation (Harris,2005).

In addition, the unlimited capacity to generate undifferentiated,proliferating progeny suggests that cells mutant for brat or pros retain self-renewing capacities. In human, brain cancers are thought to arise either from normal stem cells or from progenitor cells in which self-renewal pathways have become activated (reviewed in Al-Hajj and Clarke, 2004; Oliver and Wechsler-Reya,2004), however the underlying mechanisms are elusive. Our results in Drosophila may therefore provide a rationale and genetic model for the origin of brain cancer stem cells. Although parallels to human tumour formation are speculative, it is noteworthy that TRIM3, a human homolog of brat is located on chromosome 11p15(El-Husseini et al., 2001), a region frequently deleted in brain tumours(Fults et al., 1992; Sonoda et al., 1995; Schiebe et al., 2001). Moreover, functional studies have shown that the pros homologue Prox1 regulates proliferation and differentiation of neural progenitor cells in the mammalian retina(Dyer et al., 2003). These data may indicate that brat and pros function in cell differentiation and tumour suppression in an evolutionarily conserved manner.

While this manuscript was under review, Betschinger et al.(Betschinger et al., 2006) and Lee et al. (Lee et al., 2006)reported comparable findings on the function of brat in larval central brain tumour suppression.

We thank Y.-A. Barde, A. Merlo, K.-F. Fischbach, M. C. Ionescu, and J.-L. Boulay for critical review; T. Loop and L. Kammermeier for initial contributions; U. Stiefel for technical assistance; and M. Akam, S. J. Bray,C. Q. Doe, B. Edgar, D. Frank, P. Overton, H. Richardson, R. P. Wharton, the Bloomington Stock Center and the Developmental Studies Hybridoma Bank for strains and reagents. This work was supported by grants from the Swiss NSF (to H.R.), the Roche Research Foundation, Oncosuisse, and the Krebsliga beider Basel (to F.H.).