Hairless induces cell death by downregulation of EGFR signalling activity

During development of multicellular organisms, tissues are sculpted by a finely tuned interplay of cell proliferation, cell differentiation and cell death. These processes are coordinated by a small number of genes and pathways that regulate developmental decisions and the cell death machinery (for a review, see Bangs and White, 2000). One such example is the Notch signalling pathway, which facilitates the cell-cell communication that precedes cellular differentiation, proliferation and apoptosis in a vast variety of developmental processes in multicellular eukaryotes (for a review, see Artavanis-Tsakonas et al., 1999). Several studies have demonstrated that the Notch signal supports global growth in different tissues. Depending on the context, the Notch signal promotes cell death or cell survival, for example during Drosophila pupal eye development (Chao et al., 2004; Dominguez et al., 2004; Miller and Cagan, 1998; Wech and Nagel, 2005). Moreover, an increased Notch signal is sufficient to suppress cell death, arguing for an anti-apoptotic function of Notch in this developmental context (Ye and Fortini, 1999; Wech and Nagel, 2005). In accordance with this anti-apoptotic capacity, a reduction or loss of Notch activity, for example by overexpression of its antagonist Hairless (H), is correlated with a reduction in tissue size and is accompanied by a pronounced increase in apoptotic cell numbers (Go et al., 1998; Müller et al., 2005).

Notch signal transduction centres around the DNA-binding protein Suppressor of Hairless [Su(H)], which acts together with the intracellular domain of the activated Notch receptor as a transcriptional activator of Notch target genes (for a review, see Artavanis-Tsakonas et al., 1999). H is a major antagonist of Notch signalling in Drosophila. It mediates transcriptional silencing of Notch target genes by assembling a repressor complex with DNA-bound Su(H) (Barolo et al., 2002; Morel et al., 2001). Thereby, H recruits two global co-repressors, Groucho (Gro) and C-terminal binding protein (CtBP) (Barolo et al., 2002; Morel et al., 2001), the binding of which is needed in combination to confer the full repressive activity (Nagel et al., 2005; Nagel et al., 2007).

The Epidermal growth factor receptor (EGFR) pathway is a second, equally important signalling pathway that is utilised in metazoans to control cell fate decisions as well as cell proliferation and apoptosis. EGFR signalling is essential for cell survival at many stages during Drosophila development (e.g. Kurada and White, 1998; Baker, 2001) (for a review, see Doroquez and Rebay, 2006). Activation of EGFR switches on a cascade of protein kinases that leads to the phosphorylation and activation of the Mitogen-activated protein kinase (MAPK; also known as Rolled). MAPK phosphorylates a number of targets and causes various effects, for example the downregulation of the proapoptotic protein Head involution defective (Hid; also known as Wrinkled), thereby preventing programmed cell death (Bergmann et al., 1998; Kurada and White, 1998). Accordingly, a constitutively activated form of the Ras kinase abrogates cell death during late eye development (Miller and Cagan, 1998). Manifold genetic interactions between Notch and EGFR signalling pathway components have been repeatedly uncovered by means of relevant mutants, strongly arguing for a complex interplay between the two pathways during Drosophila development (Müller et al., 2005) (for a review, see Doroquez and Rebay, 2006). The molecular basis underlying the cross-talk between the Notch and EGFR signalling pathways in the context of cell death regulation and survival, however, remains largely obscure.

Here, we have investigated the molecular mechanisms underlying H-mediated cell death. We show that in Drosophila S2 cells, as well as during imaginal development, a wild-type H form induces Caspase 3 (also known as Decay) activation and, subsequently, apoptosis. This proapoptotic activity of H requires the binding of either one, or both, of the co-repressors Gro and CtBP. Moreover, H overexpression does not repress the Drosophila inhibitor of apoptosis, Diap1. Instead, we observe reductions in rho-lacZ and activated MAPK levels that are largely restricted to early stages of photoreceptor cell specification. Most interestingly, we find a strong upregulation of the Runx transcription factor lozenge and its target genes klumpfuss and argos, both known to be involved in the downregulation of EGFR signalling as well as in the activation of programmed cell death in the Drosophila eye. Together, our data support the hypothesis that H mediates cell death by the transcriptional silencing of Notch target genes involved in cell survival, i.e. of negative regulators of lozenge, klumpfuss and argos, which in turn causes a downregulation of EGFR signalling. Hence, our data emphasise once more the importance of a balanced cross-talk between Notch and EGFR signalling for cell survival and differentiation during Drosophila development.

We have shown previously that the Notch antagonist Hairless (H) regulates cell death during eye development in Drosophila (Müller et al., 2005; Wech and Nagel, 2005). In fact, transient overexpression of H protein in S2 cell culture causes measurable apoptosis. We used this effect to identify those parts of the H protein that are relevant for cell death induction: different deletion constructs were transiently induced in S2 cell culture and cell death rates determined. In order to relate this to the transfection efficiency, we measured the activity of a co-transfected, constitutively active lacZ construct and compared the levels with or without induction of a given construct. Amongst the various H deletion constructs tested, we identified only three that failed to induce apoptosis: HΔS, which deletes the Su(H)-binding domain; HΔC, which deletes the very C-terminal 15 amino acids that contain the binding site for CtBP; and HΔG, which deletes a central domain that overlaps the binding site for Gro (supplementary material Fig. S1). We conclude that the ability of H to assemble a repression complex on Notch target genes is likely to be a major prerequisite for cell death induction.

Next, we investigated the involvement of Gro and CtBP in H-mediated cell death more directly. We generated stably transfected S2 cell lines in order to quantify and compare the effects of H with another well-defined cell death inducer. We used full-length H (HFL) and H variants with mutated binding sites for Gro, CtBP, or both (H^*G, H^*C, H^*GC, respectively) (Fig. 1) (Nagel et al., 2005). As a positive control, we used eiger, which encodes the Drosophila TNF-receptor homologue. Eiger is known to induce cell death in Drosophila S2 cells (Moreno et al., 2002). Upon induction, we examined DNA fragmentation, a hallmark of apoptosis, by performing dUTP nick-end labelling (TUNEL) staining of the cells and quantified the number of dead cells. In our testing system, expression of the full-length H form resulted in 25% TUNEL-positive cells, a value comparable to that observed after induction of Eiger (Fig. 1A). Mutation of the Gro-binding site (H^*G) or the CtBP-binding site (H^*C), or both (H^*GC), had similar effects on cell death induction: all mutant forms were severely reduced in their capacity to induce DNA fragmentation and the number of TUNEL-positive cells was similar to that in the wild-type S2 cell control (Fig. 1A,B). These observations show that H requires the two co-repressors CtBP and Gro for cell death induction in cultured Drosophila cells.

To assess the importance of the co-repressors for H-mediated cell death in the whole animal, transgenic flies were generated containing the different point mutants of H under the control of yeast upstream activating sequences, UAS-Gal4 (Brand and Perrimon, 1993). We have demonstrated previously that ectopic expression of the full-length H form (HFL) in the developing retina induces caspase-dependent cell death, resulting in a rough eye phenotype and in a size reduction of the adult eye by nearly 50% (Müller et al., 2005) (Fig. 2A-H; supplementary material Fig. S2). Accordingly, coexpression of the baculovirus caspase inhibitor protein p35 improved the HFL-induced small eye phenotype (Fig. 2B,C,H). Notably, p35 rescued eye size to nearly wild-type values, but did not rescue the irregular arrangement of the ommatidia. This observation is to be expected because a repression of Notch signalling activity during eye development by HFL overexpression provokes differentiation defects in addition to cell death (Baker, 2001; Chao et al., 2004; Miller and Cagan, 1998). Hence, we investigated the cellular architecture of the retina during early pupal development. During this stage, overexpression of H primarily causes the death of the cone cells (Wech and Nagel, 2005), such that on average two to three cone cells remain (Fig. 2A′-C′; supplementary material Fig. S3A-A″). Later in pupal development (∼60 hours after puparium formation), eye discs showed extensive signs of cell disorganisation and distinct cell types were no longer detectable (supplementary material Fig. S3B-B″). As predicted, p35 restored the normal complement of four cone cells in the H overexpression background (Fig. 2A′-D′,I). Significantly weaker effects were observed upon overexpression of the single mutants H^*G or H^*C, as compared with HFL (Fig. 2; supplementary material Fig. S2). This was not due to a change in the subcellular localisation of the H mutant forms, as all of them showed the expected nuclear localisation (data not shown). Closer inspection revealed that H^*G resulted in smaller eyes and a reduced number of cone cells, indicating that this construct retained proapoptotic potential, although to a lesser degree than did HFL (compare Fig. 2D,D′ with E,E′ and H,I). However, overexpression of H^*C caused defects very similar to the combination of HFL plus p35, suggesting that this construct had lost most of its capacity to induce cell death (compare Fig. 2A-D′ with F,F′ and H,I). Hence, the rough eyes of H^*C-expressing flies must be likewise attributed to differentiation defects. We note that in the cell culture assay, these two point-mutants behaved almost identically and both failed to induce cell death (Fig. 1A,B). Likewise, flies expressing the mutant H^*GC variant, which lacks the binding sites for both co-repressors, suffered little or no cone cell loss. In addition, they closely resembled wild-type with regard to both the size and architecture of the eye (Fig. 2G,G′,H,I; supplementary material Fig. S3C-C″). This observation is in accordance with the cell culture assay data and conforms to the notion that H requires the recruitment of both co-repressors for normal activity (Nagel et al., 2005; Nagel et al., 2007). Apparently, this not only holds true for its role in cellular differentiation, but for cell death induction as well.

Next, we investigated the consequences of H overexpression at the cellular level during larval development. Clones of cells overexpressing HFL or the double mutant H^*GC were generated using a somatic excision protocol. These clones are labelled with green fluorescent protein (GFP) and can be easily compared with their wild-type twin clones (Fig. 3). When HFL was ectopically induced during early larval stages, a strong upregulation of activated Caspase 3 activity was observed (Fig. 3). According to the cone cell loss phenotype, HFL gain-of-function clones often showed a reduced amount of Cut-positive cells owing to the death of this cell type (supplementary material Fig. S4A-D″′). By contrast, gain-of-function clones ectopically expressing H^*GC no longer induced Caspase 3 activity and showed no sign of cell death induction (Fig. 3; supplementary material Fig. S4E-E″′). These results demonstrate that H overexpression induces apoptosis in developing tissues and that it requires the co-repressors Gro and CtBP for this activity. Most likely, H mediates apoptosis by assembling a repression complex on Notch target genes that are relevant for cell survival.

Cell survival is ensured by a class of proteins that directly inhibit caspases. In Drosophila, these are named Drosophila inhibitor of apoptosis proteins (DIAPs) (for a review, see Palaga and Osborne, 2002). Because loss of Diap1 (also known as Thread) is sufficient to promote cell death, diap1 is a possible transcriptional target of H. However, local overexpression of full-length or mutant H constructs did not affect diap1 transcript levels as compared with those of wild type (data not shown), suggesting that diap1 is not a transcriptional target of Notch signalling. At the onset of apoptosis, Diap1 undergoes post-transcriptional modifications, leading to a proteasome-mediated degradation of the Diap1 protein (for a review, see Palaga and Osborne, 2002). Given the fact that H directly interacts with a component of the regulatory cap of the proteasome (Müller et al., 2006), we reasoned that H might be involved in targeting Diap1 for degradation. However, we could not detect any direct physical binding between H and Diap1 in yeast two-hybrid assays and in co-immunoprecipitation experiments, indicating that H and Diap1 are not in the same protein complexes in the tissues analysed (data not shown). In addition, endogenous Diap1 protein levels were not significantly decreased in eye discs in which HFL was overexpressed (supplementary material Fig. S5A-A″). Together, these findings suggest that H induces cell death neither by transcriptional repression of diap1, nor by the degradation of Diap1 protein.

Apart from inhibiting apoptosis, Notch signalling promotes cell proliferation by activating E2f, which promotes the G1–S transition (Baonza and Freeman, 2005), or by activating eyegone (eyg), which is known to be essential for proliferation in the eye disc (Dominguez et al., 2004). We reasoned that H overexpression might affect apoptosis indirectly by repressing these proliferation-related genes. However, neither E2f nor eyg expression was attenuated upon overexpression of HFL, when compared with wild-type sibling cells (supplementary material Fig. S5B-C″). We conclude that H-mediated cell death is not a consequence of reduced E2f or eyg activity.

Several genetic screens have revealed strong interactions between members of the EGFR and Notch signalling pathways (for a review, see Doroquez and Rebay, 2006). Moreover, EGFR signalling has been implicated in the inhibition of proapoptotic signals and to be directly required for the survival of cells (Bergmann et al., 1998; Kurada and White, 1998; Miller and Cagan, 1998). Thus, we considered the possibility that H interferes with EGFR signalling activity, thereby inducing cell death. To reveal a potential interplay between H and the EGFR signalling pathway, we altered the dosage of EGFR signalling components in a Gmr>H background. The ability to rescue or to enhance the H-mediated small eye phenotype was assessed. If H acts as a negative regulator of EGFR signalling activity in this context, then downregulation of EGFR signalling activity would be expected to enhance H-mediated cell death, whereas upregulation of EGFR signalling should rescue it. Indeed, reducing argos (aos) or anterior open (aop) activity, both known to antagonise EGFR signalling, partially suppressed the Gmr>H eye phenotype (Fig. 4A,C,D). Accordingly, halving the gene dose of Star, which is required to process the EGFR ligand, caused the expected phenotypic enhancement, whereas overexpression of rhomboid (rho), which acts in concert with Star, ameliorated the phenotype (supplementary material Fig. S6A,C).

Next, we evaluated the relationship between H and EGFR signalling at the cellular level. We overexpressed HFL and H^*GC in groups of cells and compared them with their wild-type twin clones for the expression of different EGFR signalling components. Cells overexpressing HFL showed significantly reduced expression levels of the rho-lacZ reporter as well as of the activated, di-phosphorylated form of MAPK (diP-Erk) (Fig. 5A-A″,C-C″). Notably, downregulation was restricted to the morphogenetic furrow, where primary photoreceptor cells are specified. This restriction suggests that H antagonises a role of EGFR signalling in the selection and/or differentiation of photoreceptor cells, rather than in suppressing cell death. This is in line with the known role of Notch signalling in these processes and agrees with numerous differentiation defects observed upon HFL overexpression. By contrast, repression of the EGFR components was no longer detectable in H^*GC-expressing clones. These results demonstrate unequivocally that Gro and/or CtBP are required for mediating the negative influence of H on EGFR signalling (Fig. 5B-B″,D-D″).

The effect of H on EGFR signalling raised the question as to the underlying molecular mechanisms. There exists ample evidence for an intensive cross-talk between both pathways during Drosophila development (e.g. Doroquez and Rebay, 2006; Hasson et al., 2005; Orian et al., 2007). Our results so far indicate that H mediates its effects by transcriptional silencing of Notch target genes that regulate cellular differentiation and survival. The former might involve a direct regulation of rho transcription by Notch signalling, as we have shown that a rho-lacZ reporter is downregulated by overexpression of HFL. Alternatively, an as yet unknown activator of rho might be repressed by H. However, the spatial restriction of H activity on EGFR signalling to the morphogenetic furrow cannot be reconciled with the rather general proapoptotic effects of H. More likely, H targets other factors involved in apoptosis. One candidate is klumpfuss (klu), which we found in a genetic screen as a modifier of H-mediated cell death during eye development (Müller et al., 2005). Originally, klu was identified as playing a role in the differentiation of a variety of tissues, including leg and bristles (Klein and Campos-Ortega, 1997), that are all Notch-regulated processes. Interestingly, klu was also shown to activate programmed cell death during retina development in Drosophila (Rusconi et al., 2004; Wildonger et al., 2005), and is therefore an excellent candidate for a target of H. In addition, klu was shown to reduce the levels of EGFR signalling (Rusconi et al., 2004), similar to what we observed in HFL gain-of-function clones (Fig. 5C-C″). Indeed, we found that removal of one copy of klu significantly ameliorated the small, rough eye phenotype of Gmr>H (Fig. 6A-D). We tested the idea that HFL is able to regulate klu activity by examining the expression of a klu-lacZ reporter construct. This klu-lacZ reporter is expressed in the differentiating ommatidia in late larval eye discs and during pupal stages of eye development (Wildonger et al., 2005). Ectopic expression of HFL, but not of H^*GC, caused a cell-autonomous induction of klu-lacZ outside of its normal expression domain (Fig. 7C-D″). To further explore the link between klu and H, we examined whether HFL also induces the expression of the Drosophila Runx protein Lozenge (Lz), which was recently shown to promote apoptosis by activating transcription of klu (Wildonger et al., 2005). Lz is expressed ubiquitously at low levels in the larval eye disc. Again, in HFL gain-of-function clones, ectopic expression of lz was observed, whereas H^*GC overexpression had no effect (Fig. 7A-B″). In accordance with this, removal of one copy of lz resulted in a strong suppression of the Gmr>H eye phenotype, producing a significantly larger eye than the mutant, with only mild roughness (Fig. 6A,C,D). Moreover, levels of the EGFR inhibitory protein Argos, whose gene is a direct transcriptional target of Lz in the process of apoptosis induction (Wildonger et al., 2005), were enriched in HFL-overexpressing cell clones as compared with wild-type and H^*GC cell clones (Fig. 7E-F″). Accordingly, overexpression of Argos enhanced the eye phenotype of Gmr>H flies (supplementary material Fig. S6). Together, these observations provide compelling evidence that H induces cell death by activating lz and its targets klu and aos.

Our work allows two important conclusions: that overexpression of H induces cell-autonomous apoptosis, and that H requires the co-repressors Gro and CtBP for its proapoptotic activity. It is known that H assembles a repression complex together with the two co-repressors, resulting in transcriptional downregulation of Notch target genes. Hence, the ability of H to induce cell death is most likely a consequence of the repression of Notch target genes that are involved in cell survival. We note, however, that not every cell that receives an overdose of H dies. One simple explanation for this observation is that the only cells that die are those in which the relevant Notch target genes are normally active, as these cells require a Notch signal for survival. As H results in a repression of Notch activity, these cells would be driven into cell death, whereas those cells that do not depend on higher Notch levels for survival would be resistant to an H overdose. How is this effect of H realised at the molecular level? So far, we have been unable to narrow down our analyses towards one target gene, the repression of which by the H repressor complex induces apoptosis. The most straightforward idea, repression of the anti-apoptotic protein Diap1, is not supported by our data. Instead, we found that EGFR signalling activity is downregulated as a consequence of the upregulation of several negative regulators of EGFR.

The existence of a densely woven network of genetic interactions between the EGFR and Notch signalling pathways is well established. This intensive cross-talk harmonises many developmental processes, such as proliferation, differentiation, cell fate specification, morphogenesis and programmed cell death. Still, the molecular basis of this genetic interplay remains largely obscure. So far, few molecular intersections between the Notch and EGFR pathways have been revealed (for a review, see Doroquez and Rebay, 2006). For example, EGFR signalling causes phosphorylation of the co-repressor Gro, thereby negatively modulating the transcriptional outputs of Notch signalling via the Enhancer of split [E(spl)] genes (Hasson et al., 2005). Conversely, a myc-Gro complex was shown to inhibit EGFR signalling during neural development in the Drosophila embryo (Orian et al., 2007). Although mutual antagonism is probably the most prominent relationship in EGFR-Notch interactions, in some developmental situations both pathways cooperate to potentiate each other's signalling activities. One such example with regard to cell survival has been described in the retina of rugose mutant flies, where cell type-specific cell death could be reversed by an increase in Notch or EGFR signalling activity, indicating that both pathways adopt an anti-apoptotic function in this developmental context (Wech and Nagel, 2005). Also, R7 photoreceptor cell specification requires the combined input of both Notch and EGFR signals (for a review, see Doroquez and Rebay, 2006). Moreover, Notch defines the scope of rho expression in the Drosophila embryo, thereby activating the EGFR pathway required for early ectodermal patterning (Walters et al., 2005). Also, during the development of mouse embryonic fibroblast, the Notch receptor-processing γ-secretase presenilin acts as a positive regulator of ERK basal level activity (Kim et al., 2005).

We observed a significant decrease in the levels of activated MAPK (diP-ERK), which provides a good assessment of EGFR pathway activation, upon induction of H. Activated MAPK directly phosphorylates two transcription factors, Aop (Yan) and Pointed (Pntp2). Phosphorylation inactivates Aop, which in the unmodified state, represses EGFR targets. At the same time, phosphorylation activates Pointed, which then causes EGFR target gene transcription (for a review, see Shilo, 2005; Doroquez and Rebay, 2006). As H is a well-defined transcriptional repressor of Notch target genes, it is most unlikely that it impedes EGFR activity at the level of phosphorylation. Moreover, we do not think that H acts at the level of transcriptional regulation of EGFR target genes, even though combinatorial and antagonistic activities of the nuclear effectors of the EGFR and Notch signalling pathways have been described during eye development (for a review, see Doroquez and Rebay, 2006). Instead, we favour the hypothesis that H represses the transcription of EGFR activators, or might indirectly provoke the activation of EGFR repressors that affect, for example, the production of EGFR ligands or signal transduction.

Rho activity is required for a timely and spatially regulated release of EGFR ligands. Accordingly, the expression of rho is highly dynamic during Drosophila development, and precedes the appearance of EGFR-induced activated MAPK (for a review, see Shilo, 2005; Urban, 2006). Hence, downregulation of rho by H would eventually result in lower levels of activated MAPK (diP-Erk). In contrast to other components of the EGFR signalling pathway, ectopic expression of rho results in EGFR activation in a wide range of tissues, indicating that Rho is an essential and limiting factor (for a review, see Shilo, 2005). So far, transcriptional control is the only known means of rho regulation (for a review, see Urban, 2006). The complex array of enhancers regulating rho expression reflects the dynamic pattern of EGFR activation throughout Drosophila development (for a review, see Urban, 2006).

Interestingly, we observed a transcriptional repression of rho-lacZ in H gain-of-function clones that was dependent on the co-repressors Gro and CtBP. This effect might very well be direct, because it was shown previously that rho transcription is regulated by Su(H) in the neuroectoderm as well as in the gut of the Drosophila embryo (Fuss and Hoch, 2002; Markstein et al., 2004). As mentioned above, Notch signalling has also been shown to regulate rho expression in the embryonic ectoderm (Walters et al., 2005). Moreover, during egg development, a band of Notch activity establishes the boundary between the two dorsal appendage tube cell types, whereby Notch levels are high in rho-expressing cells (Ward et al., 2006). In accordance with this, potential Su(H)-binding sites are present in the regulatory regions of rho¹ and rho³, making a direct regulation of rho during eye development via the Notch-Su(H)-H complex very likely (A.C.N., unpublished). We note, however, that the downregulation of rho-lacZ and of activated MAPK were focussed at the morphogenetic furrow, where primary photoreceptor cells are specified and ommatidia are founded. Regulation of rho by H would then be expected to interfere with photoreceptor formation rather than with cell survival, which is in agreement with the disturbed cellular architecture of H gain-of-function flies.

Most interestingly, upon H overexpression, we observed ectopic induction of lz, klu and aos. All three genes are known to be involved in cell death induction during pupal eye development (Wildonger et al., 2005). There it was shown that the Runx protein Lz binds to the regulatory regions of klu and aos, resulting in the direct transcriptional activation of these target genes. Therefore, one might speculate that H executes its effect on klu and aos activity via the activation of lz. Moreover, as klu and aos are well-known inhibitors of EGFR signalling activity, this in itself suggests that H impedes EGFR signalling activity via these factors. This interpretation helps to explain why aos expression is induced in H gain-of-function clones, although it is well known that aos is triggered by EGFR signalling, thereby forming an inhibitory loop that acts on EGFR activity (for a review, see Shilo, 2005). The high levels of Lz still activate aos in H gain-of-function clones, keeping activity of the EGFR pathway low. Alternatively, aos and klu levels might be increased as a consequence of the downregulation, by H, of an as yet unknown repressor. Since H behaves as a kind of `multi-adaptor protein', which not only recruits the transcriptional silencers Gro and CtBP to Notch targets but also binds other proteins such as Pros26.4 (Müller et al., 2006), it is also possible that H interacts with positive regulators of lz, klu and aos.

However, we favour a model whereby H influences EGFR signalling activity on two levels. On the one hand, through transcriptional repression of rho, H causes a loss of EGFR signalling output that interferes with cell specification. On the other hand, by interfering with their repressor(s), H relieves the restriction on lz, klu and aos expression, causing their accumulation (Fig. 8). In consequence, the survival/death balance is tipped towards apoptosis in those cells that are susceptible to the effects of a lowered EGFR signal. Those cells that do not depend on high Notch and EGFR activity levels for survival would be resistant to an H overdose.

Finally, one can envisage that a downregulation of Notch and EGFR signalling activities, resulting from the overexpression of H, might leave a cell in a state of `uncertainty' that does not allow any further differentiation towards a certain cell type, but leaves the cell vulnerable to the apoptotic programme.

For transient transfection of S2 cells, cells were grown on Shields and Sang M3 insect medium (Sigma) containing 10% calf serum at 25°C. S2 cells (1×10⁶) were transfected by the calcium phosphate procedure. Briefly, S2 cells were transfected with pMT-Gal4 (0.4 μg) plus pUAST-reaper (0.26 μg) or with pMT-HFL, pMT-H-C2, pMT-HCD, pMT-HC6 or pMT-empty (mock) (each 0.66 μg) [for constructs see Maier et al. and Nagel et al. (Maier et al., 2002; Nagel et al., 2005)] and hs-lacZ reporter vector (1.33 μg). The latter was used to deduce the number of living transfected cells by measuring β-galactosidase activity with or without induction of the respective construct. The metallothionein promoter was induced by adding CuSO₄ (0.7 mM) to the cells with incubation for 24 hours. Cells were lysed by sonication, centrifuged and the supernatant assayed for β-galactosidase activity.

For stable transfection of S2 cells, cells were transfected with pMT-Gal4 (4.29 μg) and any one of the following constructs: pUAST-H, pUAST-H^*C, pUAST-H^*G or pUAST-H^*GC (0.71 μg each). For the control, S2 cells with pMT-Gal4 plus pUAST-eiger or pMT-Gal4 plus pUAST were established. The pUAST-eiger DNA was kindly provided by K. Basler (Moreno et al., 2002). We co-transfected pCophy (5 μg) and cultured the cells in Shields and Sang M3 insect medium in the presence of 10% foetal calf serum, 50 μg/ml gentamycin and under permanent selection with hygromycin (200 μg/ml). Expression of the pMT-Gal4 was induced by adding 0.7 mM CuSO₄ for 48 hours. Cells were analysed by TUNEL staining 48 hours after induction using the Cell Death Detection Kit (Roche). Briefly, cells were fixed in 4% paraformaldehyde for 1 hour, followed by a washing step with phosphate-buffered saline (PBS). The reaction mix from the supplier was added and incubated for 1 hour at 37°C in a humidified atmosphere in the dark. Cells were washed twice with PBS and several drops mounted with Vectashield (Vector Labs) on a microscope slide. The experiments were repeated three times and a total of 900 cells were analysed for each genotype. Cells were analysed on a Zeiss Axioscope linked to a BioRad MRC1024 confocal microscope. The fraction of dying cells was determined by counting the total and the TUNEL-positive cells; differences in the intensity of the fluorescent signal were not taken into account. Integrity and expression levels of the different H variants and controls were checked on western blots using anti-NTH (Maier et al., 2002) and anti-Actin (Sigma) antibodies as loading controls.

Flies were cultured on standard fly food. Details of fly stocks can be found in FlyBase. Crosses were performed at 25°C. Eye-specific expression of respective transgenes was induced with the Gal4 upstream activation sequence (UAS) system (Brand and Perrimon, 1993) using the Gmr-Gal4 driver and the relevant UAS-H lines and UAS-lacZ as control. UAS-H was as described previously (Maier et al., 1999). Generation of the H^*C, H^*G and H^*GC point mutations is described by Nagel et al. (Nagel et al., 2005). UAS-H^*G, UAS-H^*C and the double mutant form UAS-H^*GC were cloned by shuttling the respective coding sequences into pUAST. Several transgenic lines were generated for each construct and tested for their expression in vivo. At least three independent insertions showed similar levels of cell death and phenotypic consequences. Representative lines used in this work have similar expression levels as determined by western blot analysis.

Gain-of-function clones of HFL and H^*GC were induced using the flp-out technique (Struhl and Basler, 1993) and the following stocks: y w hs-flp; act<CD2>Gal4, UAS-GFPnls/TM6B and UAS-H, UAS-H^*GC constructs. First- to second-instar larvae of all genotypes were subjected to a heat shock for 30 minutes at 37°C and discs were dissected after 72 hours. As reporters, we used eyegone-lacZ, klumpfuss-lacZ and rhomboid-lacZ. Mutant alleles employed in this study were aop¹; argos^rlt; klu⁰⁹⁰³⁶; lz¹; Star^IIN and UAS transgenes were EP-rho, UAS-aos, UAS-DER^DN1-7, UAS-klu and UAS-p35.

Yeast two-hybrid protein interaction assays were performed as described (Müller et al., 2006) using pEG-HFL and pJG-DIAP1. The latter was generated by cloning the coding region of diap1 in frame into the EcoRI/XhoI sites of pJG. Immunoprecipitations were according to Nagel et al. (Nagel et al., 2005) using protein extracts from 500 Oregon-1 embryos and 100 Oregon-1 eye imaginal discs. For immunoprecipitations, we used rabbit anti-Hairless A antibodies (Maier et al., 2002) and for detection anti-mouse Diap1 (Yoo et al., 2002) or anti-rat Hairless A antibodies (Maier et al., 1999).

Immunostainings on larval discs were according to Müller et al. (Müller et al., 2005). We used mouse anti-Argos, mouse anti-β-galactosidase, mouse anti-Cut, mouse anti-Elav and anti-Lozenge (developed by J. Sanes, G. Rubin and U. Banerjee; obtained from DSHB, University of Iowa, Department of Biological Sciences, Iowa City, IA 52242, USA), mouse anti-Diap1 (Yoo et al., 2002), guinea pig anti-E2f (gift from Terry Orr-Weaver, Whitehead Institute, Boston, MA), rabbit anti-NTH (Maier et al., 2002), rabbit anti-cleaved Caspase 3 (NEB Cell Signaling Technology) and mouse anti-activated MAPK (anti-diP-Erk) (Sigma). Secondary antibodies coupled to alkaline phosphatase, fluorescein, Cy3 and Cy5 were purchased from the Jackson Laboratory. Samples were embedded in Vectashield. Cobalt sulphide staining was performed as described (Wolff and Ready, 1991). Images were taken with a Zeiss Axioscope; confocal Images with a BioRad MRC1024 confocal system using BioRad Laser Sharp 3.1 software.

Samples of about 100 flies of each genotype were inspected by low-magnification microscopy. For closer examination, fly heads were dehydrated in ethanol, covered with a mixture of gold and palladium and analysed with a digital scanning microscope (Zeiss) and Orion version 6 software. Alternatively, heads are examined by light microscopy and digitally photographed (Optronics, Pixera). Eye size was determined using ImageJ software for pixel measurements, using at least 20 female eyes. To obtain numbers of ommatidia, all facets from the right eye of each of three different females of each genotype were counted. Statistical analyses included calculation of the mean, s.d. and of significance by Student's t-test. Figures were compiled using Corel Photo Paint and Corel Draw software.

We acknowledge K. Basler, B. Hay, R. Mann, H. Sun and Exelixis for fly lines, D. Maier for anti-H, B. Hay for anti-Diap1 and T. Orr-Weaver for anti-E2f antibodies. We thank I. Beck for introducing C.E.P. to the secrets of S2 cell culture, T. Stößer for excellent technical assistance, W. Ulrich for taking SEM pictures and the Botany Institute, University of Hohenheim, for providing us with the microscope. We thank A. Preiss for generous support and for critical reading of the manuscript. This work was supported by the German Science Foundation (DFG) to A.C.N. (NA 427/1-2).