Notch signalling regulates veinlet expression and establishes boundaries between veins and interveins in the Drosophila wing

The wings of Pterigota insects have characteristic structures, the veins, that differentiate in precise two-dimensional patterns (Snodgrass, 1935). In Drosophila there are four longitudinal veins formed by proximo-distal stripes of cells that appear more compact and have higher pigmentation than intervein cells. Although veins are classified as dorsal or ventral, depending on the wing surface where they protrude, every vein has both dorsal and ventral components that are specified independently during imaginal development. These come into contact after the disc evaginates and the dorsal and ventral wing surfaces become apposed, when interactions between dorsal and ventral vein territories may contribute to the final differentiation of the vein (Garcia-Bellido, 1977).

The position of the veins is pre-figured in the third instar wing disc by the localised expression of veinlet (ve) in four dorsoventral stripes of cells (Sturtevant et al., 1993). ve encodes a membrane protein with seven trans-membrane domains (Bier et al., 1990), which facilitates signalling via Torpedo (Top), the Drosophila epidermal growth factor receptor homologue (Sturtevant et al., 1993; Price et al., 1989). High levels of Top signalling occur in places of ve expression, and this activity appears to direct wing cells into a vein differentiation pathway (Diaz-Benjumea and Garcia-Bellido, 1990a). Vein differentiation continues during pupal development and involves the restriction of specific gene products and cell adhesion proteins to veins or intervein territories (Fristrom et al., 1993; Montagne et al., 1996).

A major element involved in establishing the correct width of both dorsal and ventral vein components is the membrane receptor Notch. Loss-of-function alleles of Notch are characterised by vein-thickening phenotypes, whereas Notch gain-of-function alleles cause thinner and incomplete veins, both in Drosophila and in other species (de Celis and Garcia-Bellido, 1994; Davies et al., 1996). The establishment of vein thickness may be analogous to ‘lateral inhibition’ in proneural clusters (Artavanis-Tsakonas and Simpson, 1991), since it seems to involve the restriction of cell fate to a limited subset of the cells that have the competence to enter into a particular cell differentiation pathway. However, during vein formation several neighbouring cells acquire the same differentiation state, whereas in proneural clusters only single cells are able to follow a neural fate. The bases of these different outcomes of Notch signalling, namely single cells in proneural clusters and stripes of cells in vein territories, are relevant to understanding the mechanisms that confer versatility to Notch function during development.

The activation of Notch during vein differentiation depends on interactions with Delta (Dl), a transmembrane protein with extracellular EGF-repeats (Kopczynski et al., 1988; Vassin et al., 1987). Analysis of the phenotypes produced by tempera-ture-sensitive alleles of Notch and Delta indicates that they are required in vein differentiation from the end of larval development until at least 24 hours after puparium formation (hAPF) (Shellenbarger and Mohler, 1978; Parody and Muskavitch, 1993). Thus Notch activity is likely to participate during both the establishment of veins in the imaginal disc and the maintenance of vein territories throughout pupal development. Intracellular components of Notch signalling include Suppressor of Hairless (Su(H)) and the basic helix-loop-helix (bHLH) proteins encoded by the Enhancer of split complex (E(spl)) (Knust et al., 1992; Delidakis and Artavanis-Tsakonas, 1992; Schweisguth and Posakony, 1992; Fortini and Artavanis-Tsakonas, 1994). The expression of E(spl) genes depends on Notch activity, and involves direct transcriptional activation by Su(H) (Jennings et al., 1994; Fortini and Artavanis-Tsakonas, 1994; Lecourtois and Schweisguth, 1995). In the wing blade, one of the seven bHLH proteins of the E(spl) complex (E(spl)m β) is expressed in a pattern related to vein development, suggesting that it could be involved in mediating the effects of Notch on vein differentiation (de Celis et al., 1996a).

We have studied the expression and function of different components of the Notch pathway in both the wing disc and pupal wing to analyse the mechanism of Notch activation in vein territories. We show that the transcription of Notch is higher in broad domains that correspond to the interveins, whereas Dl expression is restricted to vein territories. The asymmetry in the distribution of Notch and Delta leads to Notch activation in vein/intervein boundaries and to the separation of vein from interveins. We find that the activities of Notch and the vein-promoting gene ve are linked: Ve is required to activate Dl expression, and Notch effectively represses ve transcription, therefore restricting ve to the domain where Notch is not activated. In boundary cells within the domain of Dl expression Notch activation results in the accumulation of E(spl)m β mRNA and in an increase in the levels of Notch itself, therefore maintaining the polarity of signalling and the separation of veins from interveins during pupal devel-opment.

The following alleles were used: at the Notch locus, the null allele N^55e11 (Kidd et al., 1983), the loss-of-function alleles fa^nd and N^MLz (a novel Notch loss-of-function allele generated by the insertion of a PlacZ element in Notch, data not shown) and the gain-of-function alleles Ax²⁸, Ax^59b and Ax¹⁶¹⁷² (Kelley et al., 1987); at the Dl locus, the lethal alleles Dl^M1, Dl^M2 and Dl^M3 (de Celis et al., 1991) and a Dl LacZ line Dl^P1171; at the Su(H) locus, the lethal allele Su(H)^AR9 (Schweisguth and Posakony, 1994); at Hairless, the lethal allele H₂ (Bang and Posakony, 1992) and at the E(spl) locus, the deficiency E(spl)^b32.2 that deletes the seven E(spl) bHLH genes (Schrons et al., 1992). As a vein-specific marker we used a Star-lacZ line (Heberlein et al., 1993). To analyse the consequences of the ectopic expression of E(spl)m β protein we used the Gal4 system (Brand and Perrimon, 1993). The UAS line UAS-E(spl)-M β (de Celis et al., 1996a) was combined with the GAL4 lines GAL4-179 (a gift from A. Brand) and GAL4-sal (Thomas et al., 1995). The UAS-ve was generated by cloning the coding sequence of ve into PUAST (Brand and Perrimon, 1993). The expression patterns of these lines were characterised in third instar disc and in pupal wings by combining each GAL4 line with the reporter UAS line UAS-IMPT (Sweeney et al., 1995).

The expression patterns of Notch, Delta and E(spl)m β were analysed in mutant discs or pupal wings of the following genotypes: a combination of the viable veinlet allele ve¹ (Sturtevant et al., 1993) and the viable vein allele vn¹ (Simcox et al., 1996), which results in the elimination of all longitudinal veins (ve vn; Diaz-Benjumea and Garcia-Bellido, 1990b); in the transgenic line rho³⁰, which results in ectopic ve expression (Noll et al., 1994), and in the Notch alleles fa^nd and Ax¹⁶¹⁷². The phenotypic analysis of genetic combinations was carried out in flies raised at 25°C, unless otherwise stated. Wings were mounted in lactic acid/ethanol (1:1) and photographed using a Zeiss axiophot microscope.

Clones were generated by X-ray-induced mitotic recombination. Larvae were irradiated (dose 1000R; 300 R/minute, 100 Kv, 15 mA, 2 mm aluminium filter) 48-72 hours after egg laying. Minute⁺ clones were scored in males of the following genotypes f^36a; Su(H)^AR9/M(2)Z P[f⁺]30B, f^36a; Dl/^M1M(3)w P[f⁺]87, f^36a; Dl/^M3M(3)w P[f⁺]87, f^36a; H²/M(3)w P[f⁺]87 and f^36a; E(spl)^b32.2/M(3)w P[f⁺]87. Mitotic recombination proximal to the f⁺ insertion produces homozygous mutant cells labelled with the cell marker forked (f). Clones in Dl+ background were induced in flies bld Dl^M1/Dp(3;3)bxd110, Dl⁺ and bld Dl^M2/Dp(3;3)bxd110, Dl⁺. Clones in ve vn background were induced in flies f^36a; ve¹vn¹Dl^M1 /ve¹vn¹P[f⁺]87F and ^{f36a ;}ve¹vn¹E(spl)^b32.2/ve¹vn¹P[f⁺]87F.

Whole-mount in situ hybridisation with digoxigenin-labelled DNA probes in imaginal discs were performed as described previously for both imaginal discs (Cubas et al., 1991) and pupal wings (Sturtevant et al., 1993). In situ hybridisation with digoxigenin-RNA-labelled probes was carried out using the same protocols, but the hybridisation step and washes were at 55°C. The following DNA probes were used: 3 kb EcoRI Dl cDNA clone (Vassin et al., 1987), 0.7 kb HindIII/(NotI) fragment from c-m β-14a (Delidakis and Artavanis-Tsakonas, 1992), 3 kb BglII/(KpnI) from a Notch cDNA clone (Kidd et al., 1983) and a ve RNA probe synthesised from a ve cDNA clone (Sturtevant et al., 1993). For Dl and m β similar antisense mRNA probes were also synthesised for some experiments.

Immunocytochemistry with the mAb323 antibody to detect E(spl)bHLH expression was performed as described in Jennings et al. (1994). We also used rabbit anti-β-galactosidase (Cappel), mouse monoclonal anti-Notch (Fehon et al., 1991), mouse anti-IMP (Sweeney et al., 1995) and rat-anti Ventral veinless (a gift from J. Casanova and M. Llimargas) antibodies. Secondary antibodies were from Jackson Immunological Laboratories (used at 1/250).

Loss-of-function alleles of Notch and Dl always result in the differentiation of thicker veins, indicating that Notch function is essential for the vein components to acquire their appro-priate width. To determine which other elements of the Notch pathway are required to regulate vein differentiation we have analysed the phenotypes produced by lethal alleles of Dl, E(spl), Su(H) and H in clones. We find that lethal alleles of these genes cause vein thickening (Dl, E(spl) and Su(H)) or vein loss (H) similar to that observed with Notch loss-and gain-of-function alleles respectively. Thus, Dl, E(spl) and Su(H) mutant cells differentiate thicker veins when they appear in vein regions, but they never produce ectopic veins when restricted to intervein territories (Fig. 1A-F,H). The strongest vein-thickening phenotypes are observed when mutant clones extend into two adjacent interveins. Conversely, clones of H lethal alleles cause strong phenotypes of vein loss when the clones cover both dorsal and ventral vein component (not shown), and weaker vein-loss phenotypes when mutant cells are restricted to only one vein surface (Fig. 1G). These observations indicate that modifications in the activity of the Notch pathway can only alter the differentiation of cells that already have the capability to form veins, as it is the case for Notch mutations (de Celis and Garcia-Bellido, 1994). The thicker veins that differentiate in Su(H) and E(spl) mutant clones are formed by mutant cells that lie within the normal vein region (Fig. 1E,F,H). However all three Dl lethal alleles analysed (Dl^M1, Dl^M2 and Dl^M3) also cause the differentiation of vein histotype in non-mutant cells abutting the clone when the boundary between mutant and non-mutant cells is close to the normal vein (Fig. 1C). This was analysed in more detail by generating Dl^M1 and Dl^M2 clones in a Dl⁺ duplication background, to eliminate any effects of Dl haploinsufficiency in neighbouring cells. In these experiments a similar phenotype of vein induction in wild-type cells was seen (Fig. 1D), indicating that Dl is required in vein cells to suppress vein differentiation in adjacent intervein cells.

Some aspects of the phenotypes produced by Su(H) and E(spl) are milder than those of Dl or Notch null alleles. Firstly the veins that differentiate in Su(H)^AR9 and E(spl)^b32.2 clones are narrower than those formed within comparable clones of Notch or Dl mutant cells (Table 1). Secondly clones restricted to only one wing surface have no effects in the differentiation of veins in the opposite surface (Table 1). Finally, the sizes of Su(H)^AR9 and E(spl)^b32.2M⁺ clones are comparable to those of Ccontrol M⁺ clones (data not shown), demonstrating that these alleles do not affect cell viability.

Mosaic analysis of different genes of the Notch pathway confirms that they participate in a cell interaction mechanism that restricts vein differentiation. The expression of both Notch and Delta proteins in the imaginal disc is modulated in vein versus intervein territories (Fehon et al., 1991; Kooh et al., 1993). Using a vein-specific marker, we confirmed that high levels of Notch protein are associated with intervein regions in late third instar and pupal wing discs (Fig. 2A-C and data not shown). Double staining to compare Notch and E(spl)bHLH proteins with Dl expression reveals that, although the patterns are largely complementary with Dl in regions where Notch and E(spl) are lowest, there is overlap at the vein/intervein boundaries where cells express all three genes (Fig. 2D-I).

The regulation of Notch, Dl and E(spl) expression occurs at the transcriptional level, Dl mRNA being detected in the vein and Notch in broad stripes that correspond to the interveins in late third instar discs (Fig. 3A-C). The expression of Dl is maintained in pupal wings 24 hAPF (APF) in dorso-ventral stripes 6-8 cells wide, with those cells at vein-intervein boundaries accumulating maximal levels of Dl (Fig. 3E,J). In contrast, the expression of Notch evolves during pupal development; it is gradually lost from intervein territories during the first 12 hAPF (Fig. 3G-H), becoming restricted in pupal wings 24 hAPF to stripes of 2-3 cells wide localised at the veinintervein boundaries (Fig. 3D,I). At this stage, the cells that accumulate high levels of Notch correspond to those in which Dl expression is maximal (Fig. 2H-I; Fig. 3I,J). The expression of E(spl)m β is detected in the same positions as Notch in pupal wings (Fig. 3F,K) indicating that Notch activation is maintained at vein/intervein boundaries during pupal development. The expression patterns detected in 24 hAPF pupal wings are maintained at later stages (e.g. 30-35 hAPF, data not shown).

The expression of Dl, Notch and E(spl)m β in the developing veins was studied in several genetic backgrounds in which vein differentiation is altered. In ve vn wings all longitudinal veins are absent (Diaz-Benjumea and Garcia-Bellido, 1990b), presumably due to the reduction in Torpedo signalling caused by lower levels of the putative ligand Vn (Simcox et al., 1996; Schnepp et al., 1996) and the absence of Ve (Sturtevant et al., 1993). In the corresponding wing discs expression of Dl is eliminated from the veins (Fig. 4B), indicating that the transcriptional activation of Dl depends on the previous specification of veins by Top activity. In agreement, ectopic expression of ve in pupal wings, as in rho³⁰ mutants, leads to ectopic expression of Dl in similar regions (Fig. 4F-G). Expression of Notch in ve vn wing discs appears unchanged, as the two broad domains where Notch transcripts accumulate at higher levels are still present (Fig. 4A), suggesting that the modulation of Notch expression in the wing pouch is independent of the establishment of the veins per se. However expression of E(spl)m β is severely reduced in the wing pouch of ve vn discs (Fig. 4C), demonstrating that Notch is not activated in vein/intervein boundaries. The lack of Dl and E(spl) expression associated with the developing veins in ve vn discs is compatible with the observation that clones of Notch (de Celis an Garcia-Bellido, 1994), or clones of Dl or E(spl) (Fig. 1I,J), cannot rescue vein differentiation in ve vn mutant wings. In 24 hAPF ve vn pupal wings, the expression of Dl and E(spl)m β is also absent in the wing blade (Fig. 4F and data not shown). Furthermore, there is no accumulation of Notch mRNA in vein/intervein boundaries (Fig. 4E), indicating that this depends on an independent mechanism from that used to establish the initial intervein expression in the disc. We observed a similar failure to accumulate Notch in vein/intervein boundaries when Notch signalling is strongly reduced, as in fa^nd/NM^Lz pupal wings (Fig. 5E-F), suggesting that this late expression of Notch depends on Notch signalling.

The expression of Notch, Dl and E(spl)m β also changes when Notch signalling is modified using other Notch alleles (fa^nd and Ax¹⁶¹⁷²) in a manner that relates to the final pattern of veins in the mutant wings. In fa^nd pupal wings, the expression of Dl is detected in broader stripes of cells, and in addition Notch and E(spl)m β accumulation are displaced to novel positions that limit the broader mutant veins (Fig. 5A-C). Thus, a reduction in the efficiency of Notch signalling leads to a lateral displacement of the boundary between each vein and their adjacent interveins. In the gain-of-function allele Ax¹⁶¹⁷² we observe that Notch, Dl and E(spl)m β expression is absent in the regions corresponding to the vein-stretches eliminated by this mutation (Fig. 5D and data not shown), suggesting that the competence to form veins has been suppressed in these regions (see below).

The localised expression of ve in the veins is a critical component of vein development. In agreement with previous reports (Sturtevant and Bier, 1995), we find that mutations in Notch affect ve expression. Thus hyper-activation of Notch signalling using strong Ax mutations results in the repression of ve transcription in the imaginal disc (Fig. 6A,D). Conversely, reductions in Notch signalling, as in fa^nd pupal wings, result in an increased number of ve-expressing cells in vein territories (Fig. 6H). Taken together, these observations suggest that ve is a target of Notch signalling, and a candidate to mediate these effects on ve is E(spl)m β. This was tested using the GAL4 system to direct ectopic expression of this protein in vein territories (Fig. 6B,F). We find that ectopic expression of E(spl)m β effectively represses ve transcription both in discs and in pupal wings (Fig. 6C,G) and consequently changes the vein pattern. Thus all combinations between UAS-E(spl)m β and GAL4 lines expressed in imaginal and pupal veins result in the elimination of vein stretches (Fig. 6J,K). The strongest phenotypes are observed when E(spl) expression is maintained at high levels in the developing veins during both larval and pupal development (GAL4-179/UAS-E(spl)m β Fig. 6F,J). Vein suppression by ectopic E(spl)m β expression in vein territories is rescued by the simultaneous presence of ectopic Ve (Gal4-sal UAS-E(spl)m β/UAS-ve flies, Fig. 6L), suggesting that most effects of E(spl)m β on vein formation are exerted through repression of ve transcription.

Notch activity is required during multiple developmental processes, both in Drosophila and in other organisms, where many elements of the pathway have been found to be conserved (Artavanis-Tsakonas et al., 1995). The basis of Notch versatility appears to derive from its ability to regulate different downstream genes in a context-dependent manner, and therefore to understand the relevance of Notch signalling it is important to compare different processes in which Notch is involved. We have studied the mechanism of Notch function during vein formation, where Notch activity appears to separate two populations of cells so that they follow distinct differentiation programs, vein and intervein. This involves defining the extent of vein-competent territories in the imaginal disc and restricting vein differentiation to the central domain of each competent territory during pupal development.

A critical component in the formation of veins is the tyrosine kinase receptor Top. Reductions in Top function result in the elimination of veins, whereas ectopic activation of Top produces ectopic veins (Brunner et al., 1994; Diaz-Benjumea and Hafen, 1994), and it has been postulated that the activity of Top is increased in vein regions through localised expression of ve (Sturtevant et al., 1993). Alterations in Notch function manifest their effects on ve expression in third instar imaginal discs, with reductions in Notch activity resulting in thicker stripes of ve-expressing cells, and ectopic activation of the Notch pathway repressing ve expression. Furthermore, ve activity is required for Notch, Dl or E(spl) mutant cells to differentiate as vein, and ectopic expression of ve rescues the vein loss characteristic of ectopic expression of E(spl)m β, indicating that Notch activity during vein differentiation is mediated through regulation of ve expression. Although ve could be a direct target for the Notch pathway, the fact that ve expression does not evolve from initial broad regions to narrower stripes, but is restricted to the veins from the outset, suggests that the early effects of Notch on ve are indirect. Thus Notch activity could negatively regulate transcription activators, analogous to the proneural genes, which promote expression of ve and therefore high levels of Top activity. Antagonism between Top and Notch signalling has also been noted in the specification of the photoreceptor cell R8 (Baker et al., 1990) and between the homologous pathways during vulval development in C. elegans (Horvitz and Sternberg, 1991).

The initial activation of Notch depends on the complementary distribution of Notch and its ligand Dl. Thus in the third instar imaginal disc high levels of Dl coincide with the developing veins whereas Notch transcription is most prominent in broad domains that separate adjacent veins, with maximal expression between LIII and LIV (Fig. 7A). These complementary expression patterns depend on separate spatial cues, because in discs with reduced Top activity (ve vn) there is no accumulation of Dl in the developing veins, but the pattern of Notch expression is unchanged. The dependence of Dl expression on Ve, and the regulation of ve by E(spl)m β, would ensure coordination between Notch activity and ve transcription, therefore linking Notch and Top signalling pathways (Fig. 7B).

The formation of veins of normal width also requires Notch and Dl functions throughout pupal development (Shellenbarger and Mohler, 1978; Parody and Muskavitch, 1993). Notch activity at this stage appears to be restricted to vein/intervein boundaries, as suggested both by the behaviour of cells mutant for Notch, Dl, Su(H) or E(spl) and the expression pattern of E(spl)m β. Thus, mutant clones for these genes cause thicker veins but never ectopic veins, and the strongest phenotypes of vein thickening are observed when mutant clones extend into two adjacent interveins, presumably because in these cases both vein/intervein boundaries have failed to form. Similarly, the elimination of vein/intervein boundaries observed in strong ticipate in maintaining the separation between veins and interveins during pupal development. Consistent with this model, we find that strong reductions in the level of Notch activity (fa^nd/N^Mlz) eliminate the expression of Notch in vein/intervein boundaries but do not perturb the early pattern of Notch expression, such as that between LIII and LIV veins. A positive feed-back loop regulates transcription of the Notch-related gene lin12 during vulvar development in C. elegans (Wilkinson et al., 1994), indicating that it may be a general mechanism through which high levels of Notch expression and function can be maintained in a cell population.

The activation of Notch at vein/intervein boundaries depends on the presence of Dl in vein cells: clones of Dl mutant cells induce neighbouring wild-type cells to differentiate as vein so that wild-type cells appear at both sides of novel vein/intervein boundaries. As expected, expression of Dl persists in the vein territories during pupal development, in a domain that includes the vein and its vein/intervein boundaries (Fig. 7A). This Notch loss-of-function alleles (fa^nd /N^Mlz) results in the differentiation of veins whose thickness is comparable to that of Notch null clones. After puparium formation Notch expression is restricted to vein/intervein boundaries and coincides with that of E(spl)m β, suggesting that the accumulation of Notch here is a consequence of Notch activation. This implies a positive feedback loop on Notch transcription that could par-domain is broader than that of ve and coincides with the expression of other genes associated with vein development, such as ventral veinless (de Celis et al., 1995a; and data not shown), and will be referred to as ‘pro-vein’ (Fig. 7B). Interestingly, the highest levels of Dl expression are detected in provein cells which also accumulate Notch and E(spl)m β. We do not know if Dl expression here is also required for Notch activation, but the fact that E(spl)m β expression/Notch activation occurs in more that one row of cells suggests that cells in vein/intervein boundaries have simultaneously the capability to receive Notch signals and to activate Notch in neighbouring cells (Fig. 7B). The overlap in the expression of Notch and Dl helps to explain the vein phenotypes produced when the gene dosage of Notch and Dl is altered (de la Concha et al., 1988), which indicates that the relative levels of Notch and Dl expression are critical for normal Notch signalling.

The extent of extra vein differentiation detected in mosaics differs depending on which component of the Notch pathway is mutant. Thus, Dl and Notch clones consistently cause stronger phenotypes than similar Su(H) or E(spl) clones. Although in the case of Su(H) it is possible that these differences could be the result of residual activity, the similar phenotypes observed between a lethal Su(H) allele and a E(spl) deficiency suggest that additional Notch intracellular components are required in vein differentiation. One candidate to participate with E(spl) in the repression of vein differentiation in response to Notch activation is the gene extramacrochaetae (emc), whose expression is also increased in boundary intervein cells (de Celis et al., 1995b). It is possible that Notch, in addition to activating the expression of E(spl)m β, also contributes to this increase in the levels of emc, and the combination of high levels of both emc and E(spl)m β is more effective in the repression of vein differentiation.

The analysis of loss-of-function alleles of Notch and of several other members of the Notch pathway indicates that they all participate in a signalling mechanism to limit the number of cells that differentiate as vein. The requirement for Notch in vein development shows similarities to that observed during the singling out of neural precursors: both processes appear to involve the establishment of a territory of competence and, subsequently, the restriction of cell differentiation to a subset of the competent cells. In other respects, vein differentiation reveals differences in the mode of Notch function. Thus, during neural precursor segregation, signalling operates transiently in all the cells surrounding the emerging neural precursor, leading to singling out of individual cells. In contrast, during vein formation stripes of cells acquire the same cell fate, presumably because Notch is not operative in the vein itself. Furthermore, Notch activity is required over a prolonged period of time to maintain the separation between the vein and the adjacent interveins, and its activation occurs in the cells that form the boundary of the developing vein. The basis of Notch activation here appears to be the heterogeneous distribution of the ligand and receptor of the pathway. In this respect, the establishment of vein thickness is similar to the formation of the wing margin, where the asymmetrical distribution of two Notch ligands, Dl and Serrate, result in persistent Notch activation in the dorso-ventral boundary (de Celis et al., 1996b; Doherty et al., 1996). Complementary patterns of ligand and receptor of the pathway could be a general aspect of Notch signalling in processes where Notch activity delimits cell populations that follow different fates.

We thank Professor Michael Ashburner in whose laboratory part of this work was carried out. We also thank Andrea Brand and Elisabeth Knust for providing fly stocks; Jordi Casanova and Marta Llimargas for the anti-Ventral veinless antibody; and Nick Brown for comments on the manuscript. This work was supported in part by a project grant to J. dC. and S. J. B. from the Welcome Trust.