Sanpodo and Notch act in opposition to Numb to distinguish sibling neuron fates in the Drosophila CNS

Development of the Drosophila central nervous system (CNS) is characterized by the sequential production of stem-cell-like precursors called neuroblasts, intermediate precursors called ganglion mother cells (GMCs), and finally sibling postmitotic neurons and glia (reviewed in Goodman and Doe, 1993). Neuroblast formation is regulated by the balance of proneural and neurogenic gene activity in the neuroectoderm (reviewed in Campos-Ortega, 1995). Proneural genes are expressed in clusters of about five neuroectodermal cells and promote neuroblast formation, whereas cell interactions mediated by Delta (ligand) and Notch (receptor) restrict the number of neuroblasts to one per cluster. In the absence of Delta or Notch function, there is an approximate 5-fold increase in all early forming neuroblasts (Brand and Campos-Ortega, 1988; Cabrera, 1990; Martin-Bermudo et al., 1995; Skeath and Carroll, 1992). In most cases, functional Delta/Notch signaling also requires the nuclear proteins Mastermind (Mam), Supressor of Hairless (Su(H)), Neuralized and Enhancer of split (E(spl)) (reviewed in Artavanis-Tsakonas et al., 1995).

The next step of neurogenesis is the repeated asymmetric division of neuroblasts to bud off GMCs. Recent evidence indicates that the asymmetric segregation of intrinsic determinants into the GMC during neuroblast division plays a major role in distinguishing neuroblast and GMC siblings. Within neuroblasts, the Inscuteable protein coordinates mitotic spindle orientation with asymmetric protein and RNA localization so that, when a neuroblast divides, a stereotyped set of gene products asymmetrically segregate into the GMC (Kraut et al., 1996; Shen et al., 1997; Li et al., 1997). These products include Numb, Miranda, Staufen and Prospero proteins, and prospero RNA (Rhyu et al., 1994; Hirata et al., 1995; Spana and Doe, 1995; Knoblich et al., 1995; Ikeshima-Kataoka et al., 1997; Shen et al., 1997; Li et al., 1997; Broadus et al., 1998). prospero encodes a transcription factor necessary for establishing GMC-specific gene expression (Doe et al., 1991; Vaessin et al., 1991); staufen encodes an RNA-binding protein capable of localizing prospero RNA (Li et al., 1997; Broadus et al., 1998); miranda encodes a Prospero-binding protein that regulates localization and release of Prospero into the GMC (Ikeshima-Kataoka et al., 1997; Shen et al., 1997) and numb encodes a membrane-associated protein whose function in GMCs has yet to be determined (Uemura et al., 1989; Rhyu et al., 1994). The combined action of these intrinsic determinants contributes to the specification of GMC fate, but is not sufficient to explain all the differences between neuroblast and GMC cell types.

The last step of neurogenesis is the division of each GMC to produce a pair of postmitotic neurons (or glia); in most cases, the two sibling neurons differ in gene expression and/or axon projections. The mechanism by which GMCs divide to produce sibling neurons with different identities is not well understood. Proteins involved in asymmetric neuroblast/GMC division are either not localized (Prospero), or have not been examined, during sibling neuron division. A role for cell interactions is suggested from cell ablation results in the related grasshopper embryo (Kuwada and Goodman, 1985).

Cell interactions and intrinsic determinants both regulate sibling cell fate in the adult peripheral nervous system (PNS) and in the MP2 CNS lineage. In the adult PNS, the external sense organ precursor produces the SOPIIa and SOPIIb daughter cells; SOPIIa generates a bristle and a socket cell and SOPIIb divides to make a neuron and a glial cell (Jan and Jan, 1995). Loss of Notch or Delta function at the time of sense organ precursor division duplicates SOPIIb at the expense of SOPIIa, showing that Notch signaling is required for SOPIIa cell fate (Hartenstein and Posakony, 1990; Schweisguth and Posakony, 1992; Parks and Muskavitch, 1993). Conversely, SOPIIb fate is due to the asymmetric localization of Numb into the SOPIIb cell, where it antagonizes Notch function (Uemura et al., 1989; Rhyu et al., 1994). In the MP2 lineage, Numb is partitioned into the dMP2 sibling neuron where it antagonizes Notch signaling to distinguish dMP2/vMP2 sibling cell fates (Spana et al., 1995; Spana and Doe, 1996).

To identify genes controlling sibling neuron cell fate (and other cell fate decisions in the CNS), we performed a saturation mutagenesis of the third chromosome. Here we describe mutations in five genes that result in the equalization of a wide variety of sibling neuron fates throughout the CNS. We find that sanpodo (spdo; Salzberg et al., 1994), Delta, Notch and mam are required to specify one sibling fate, whereas numb antagonizes the function of these genes to specify the other sibling fate. Dosage-sensitive interactions between spdo and numb indicate that these genes likely act in the same biochemical pathway. Genetic experiments show that spdo is downstream of numb. spdo has recently been shown to encode a homologue of tropomodulin (Dye et al., 1998), a vertebrate pointed-end F-actin-binding protein (Fowler, 1996). Our data suggest that Spdo may be a new member of the Notch signaling pathway and, together with the accompanying paper, raise the possibility of cytoskeletal regulation of Notch signaling.

Four independent numb stocks were used numb¹pr cn Bc/CyO; numb²pr cn Bc/CyO; numb³pr cn Bc/CyO and l(2)06740 from the BDGP which we call numb⁴ because it harbors a mutation in the numb gene. For a deficiency that removed the numb locus we used w; Df(2L)N22-3/CyO. Eleven spdo alleles were identified in our mutagenesis (J. B.S. and C. Q. D., unpublished data); all yielded identical CNS phenotypes and one (spdo^ZZ27) does not make detectable spdo transcript or protein (data not shown). This suggests that all spdo alleles are either null or strong hypomorphs. Two independent numb;spdo double mutant stocks were used and both gave identical results: numb¹/CyO, P{ftz-lacZ}; spdo^G104/TM3, P{ftz-lacZ} and numb²/CyO; spdo^ZZ27/TM3, P{ftz-lacZ}. In addition to the ten Delta alleles and one mam allele identified in our mutagenesis, we used the following fly stocks to analyze the loss-of-function phenotype of different neurogenic genes on CNS sibling cell fate: w^a; Df(1)N^81K1rb/y w f; +/SM1 Dp(1:2)51b, Dl³e/TM6 Tb, E(spl)^Drv202.96tx/TM6 Tb, neu^IN94/TM3, cn mam^IL115bw/CyO and l(2)04615/CyO from the BDGP P-element collection which is a hypomorphic allele of mam.

Immunohistochemical staining was carried out as described in Skeath et al. (1992) and immunofluorescent staining was carried out as described in Spana et al. (1995). The following dilutions were used: 1:10 monoclonal antibody 2B8 (anti-Even-skipped; Patel et al., 1994); 1:500 anti-phosphohistone H3 (Upstate Biotech.); 1:1000 anti-Zfh-1 mouse polyclonal sera (Lai et al., 1991); 1:10 monoclonal antibody 22C10 (Fujita et al., 1982); 1:3000 rabbit anti-β-galactosidase (Cappel); 1:1000 rabbit anti-Odd (Ellen Ward and Doug Coulter, personal communication).

We scored the number of Eve⁺ RP2 and U neurons in both thoracic and abdominal hemisegments. In wild-type embryos, there are six thoracic and ten abdominal Eve⁺ EL neurons; we only quantitated EL neurons in abdominal hemisegments. To quantitate d/vMP2 neurons, we stained appropriately staged embryos for 22C10 and scored the axonal projections from the cells in the d/vMP2 position. To quantitate mitotic Eve⁺ GMCs, we counted phosphohistone⁺/Eve⁺ cells at stage 12-14 (for ELs) and stage 11 for medial Eve⁺ cells.

We performed a large scale EMS mutagenesis of the third chromosome, screening for alterations in the CNS pattern of the Even-skipped (Eve) homeodomain transcription factor. Eve is detected in a small number of identified GMCs and neurons (Patel et al., 1989; Broadus et al., 1995) and changes in Eve can be used to detect altered neuroblast, GMC, or neuronal identity (e.g. Doe et al., 1988, 1991; Chu-LaGraff and Doe, 1993; Yang et al., 1993; Bhat et al., 1995; Chu-LaGraff et al., 1995). Here we describe mutations in five genes that equalize sibling neuron fate; results from the entire screen will be described elsewhere.

We examined six pairs of sibling neurons with unequal fates (RP2/RP2sib, aCC/pCC, dMP2/vMP2 and three pairs of U/Usibs) and five pairs of presumptive sibling neurons with indistinguishable fates (EL/EL neurons) (Fig. 1). RP2/RP2sib develop from the Eve⁺ GMC 4-2a: RP2 is Eve⁺, expresses the Zfh-1 transcription factor and the 22C10 epitope, and extends a motor axon out the intersegmental nerve (ISN); RP2sib is smaller, downregulates Eve, does not express Zfh-1 or 22C10 (Figs 1, 2A-C, 3A-D; Broadus et al., 1995; Chu-LaGraff et al., 1995). aCC/pCC develop from the Eve⁺ GMC 1-1a: aCC is an Eve⁺, Zfh-1⁺, 22C10⁺ motoneuron projecting out the ISN and pCC is an Eve⁺, Zfh-1⁻, 22C10⁻interneuron (Figs 1, 2A-C, 3A-D; Broadus et al., 1995). The dMP2/vMP2 interneurons develop from the Odd-skipped (Odd)-positive MP2 precursor: dMP2 is Odd⁺ with an posterior axon projection, while vMP2 downregulates Odd and has an anterior axon projection (Figs 1, 4A; Spana et al., 1995).

The U/Usib neurons develop from three Eve⁺ GMCs that divide to yield a cluster of six initially Eve⁺ neurons. The three U neurons maintain Eve, whereas the Usib neurons rapidly downregulate Eve (Figs 1, 2A-C). Although we have not used lineage techniques to confirm the U/Usib relationship, we base our sibling assignment on the proximity of these Eve⁺ neurons and their apparent similarity to the Eve expression profile of RP2/RP2sib. In older embryos, two additional Eve⁺ neurons develop near the three Eve⁺ U neurons, but we are unable to determine the identity of their siblings and will not consider them in our analysis.

Each abdominal hemisegment has 9-10 Eve⁺ lateral (EL) neurons (Figs 1, 2C; Patel et al., 1989). Using anti-phosphohistone as a mitotic marker, we have never observed Eve⁺ mitotic GMCs at the EL position (0/290 hemisegments; stages 12-14). In contrast, it is common to observe Eve⁺ mitotic GMCs among the medial Eve⁺ cells (6/36 hemisegments; stage 11). Thus, the EL neurons develop from Eve⁻GMCs. The ELs are derived from a neuroblast lineage containing only 10-13 interneurons (Schmidt et al., 1997; A. Schmid and C. Q. D., data not shown), suggesting that the EL neurons are siblings. Thus, it appears that five Eve⁻GMCs produce five pairs of Eve⁺ EL sibling neurons (Fig. 1).

spdo was first identified as a gene controlling neuronal number during embryonic PNS development (Salzberg et al., 1994). We isolated 11 alleles of spdo based on dramatic alterations in eve expression in the CNS. Embryos homozygous for the null spdo^ZZ27 allele, subsequently called ‘spdo embryos,’ show normal Eve⁺ GMCs (Fig. 2D) but an equalization of sibling neuron identity as detected by Eve and other markers. The RP2 motoneuron is duplicated at the expense of the RP2sib, as shown by staining for Eve (Figs 2D,E, 3E-H), Zfh-1 (Fig. 3E-G) and 22C10 (Fig. 3H). The aCC motoneuron is duplicated at the expense of the pCC interneuron, as shown by staining for Zfh-1, 22C10 and by following axonal projections (Fig. 3E-H). The Usib fates are duplicated at the expense of the U neurons, as shown by Eve staining (Fig. 2D,F). Finally, dMP2 is duplicated at the expense of vMP2, as shown by Odd and 22C10 staining (Fig. 4B). We see no change in the EL neurons Eve⁺ expression (Fig. 2F); this is not surprising, because each sibling neuron is Eve⁺ (Fig. 1; see above). We have no markers to differentiate the siblings in this lineage.

Although the spdo sibling neuron phenotype is identical to the Notch sibling neuron phenotype (see below), none of the 11 spdo alleles show the excess neuroblast formation characteristic of Notch mutations. Furthermore, spdo germline clones yield an Eve CNS phenotype identical to embryos that lack only zygotic spdo function (data not shown). Thus, spdo does not appear to function during Notch-mediated lateral inhibition in the neuroectoderm.

We identified mutations in two other genes, Delta (10 alleles) and mam (1 allele), that yield similar equalization of sibling neuron fates. Because both genes are in the well-characterized Notch signaling pathway (Artavanis-Tsakonas et al., 1995), we tested null and hypomorphic alleles of several ‘Notch pathway’ genes: Delta, Notch, mam, neuralized and E(spl). As expected, mutations in all genes result in an excess of neuroblasts due to failure of lateral inhibition within the neuroectoderm (data not shown); however, we could still score sibling neuron fates due to the specificity of our markers.

Mutations in neuralized and E(spl) have no effect on the identity of the sibling neurons that we assayed, despite strong defects in the earlier process of neuroblast formation. In contrast, Delta, Notch and mam mutations all yield similar sibling neuron phenotypes in addition to excessive neuroblast formation; we illustrate these results using embryos homozygous for a hypomorphic mam allele in which neuroblast formation is essentially normal but sibling neuron fates are equalized. Loss of mam does not affect eve expression in GMCs (Fig. 2G), but leads to the duplication of RP2, Usib (Figs 2G-I, 3I-L), aCC (Fig. 3I-L) and dMP2 (Fig. 4D) fates at the expense of the RP2sib, U, pCC and vMP2 fates, respectively. We observe no change in the number of Eve⁺ EL neurons in embryos that lack Notch, Delta or mam function (Fig. 2I; data not shown); this is not surprising, because each sibling neuron appears to be Eve⁺. Thus, mutations in three genes (Delta, Notch and mam) have precisely the same sibling neuron phenotype as spdo mutations, suggesting that spdo, Delta, Notch and mam act together to specify asymmetric sibling neuron fate.

Numb is known to bind to the intracellular domain of Notch and antagonize Notch signaling but, with the exception of the dMP2/vMP2 neurons, it has not been reported to play a role in sibling neuron development in the CNS (Guo et al., 1996; Spana et al., 1995; Spana and Doe, 1996). However, due to the widespread role of Notch in specifying asymmetric sibling neuron identity, we re-investigated the CNS function of numb. We assayed sibling neuron development using four different numb alleles and a deficiency that uncovers the numb locus (Table 1). We find that two of these alleles, nb² and nb⁴, are stronger alleles than those used in previous studies. The CNS phenotypes indicate that the four alleles fall into an allelic series (nb² > nb⁴ > nb¹/nb³; Table 1). In embryos homozygous for the strongest numb allele (nb²), we observe an equalization of sibling neuron phenotype for all siblings tested, with the exception of aCC/pCC (Fig. 3M-P). RP2 is transformed into RP2sib approximately 50% of the time (Fig. 2J-L; Table 1); three Usibs are transformed into three U neurons (Fig. 2L; Table 1); and dMP2 is transformed into vMP2 (Fig. 4C; Spana et al., 1995). The numb phenotypes for RP2, Usib and dMP2 neurons are reciprocal to those observed in spdo, Delta, Notch or mam embryos. This is consistent with studies showing that Numb antagonizes Notch function (Guo et al., 1996; Spana and Doe, 1996) and extends this interaction to a diverse array of CNS sibling neurons. In addition, we observe a strong decrease in the number of Eve⁺ EL neurons in numb mutant embryos (Fig. 2L; Table 1).

There is clear evidence of maternal numb function during CNS development (Table 1), which may account for the lack of a fully penetrant numb sibling neuron phenotype. For example, when females heterozygous for the weak nb¹ allele are crosses to males heterozygous for the strong nb² allele, the nb²/nb¹ embryos have an intermediate CNS phenotype; however, when females heterozygous for the strong nb² allele are crossed to males heterozygous for the weak nb¹ allele, the nb²/nb¹ embryos have a more severe CNS phenotype (Table 1). Thus, changing the dose of maternal numb product directly affects CNS development and suggests that numb may have earlier CNS functions in addition to sibling neuron specification.

spdo and numb have opposite sibling neuron phenotypes and so we determined the epistatic relationship between the two genes by examining the phenotype of a numb;spdo double mutant. We find that the numb;spdo double mutant phenotype is identical to embryos lacking spdo alone (Figs 2M, 4E). Thus, spdo is genetically downstream of numb, just as has been observed for Notch pathway mutations in other lineages (Spana and Doe, 1996; Guo et al., 1996).

We next assayed whether sanpodo and numb exhibit dosage-sensitive interactions, as gene products that act in the same biochemical pathway often do. We find that the sibling neuron phenotype in numb embryos is sensitive to the level of spdo. For example, homozygous nb² embryos show a loss of EL and RP2 neurons, but reducing the dosage of spdo by one-half in nb² embryos leads to a recovery of Eve⁺ EL and RP2 neurons (Fig. 2N,O; Table 2). We observed similar results using independently isolated alleles of both numb and spdo (Table 2). Thus, halving the dosage of spdo strongly suppresses the numb CNS phenotype. These results show that the numb phenotype is extremely sensitive to the dosage of spdo, consistent with the two proteins acting in the same biochemical pathway.

In numb embryos, there is a striking decrease in the number of Eve⁺ EL neurons (Fig. 2L; Table 2). Notch, Delta, mam and spdo single mutants do not alter the number of Eve⁺ EL neurons (Fig. 2; data not shown for Notch and Delta). Importantly, numb;spdo double mutant embryos show a complete rescue of Eve⁺ EL neurons (Fig. 2M; Table 2), suggesting that Numb acts to prevent Spdo-mediated downregulation of eve expression (i.e. in the absence of Spdo, the loss of Numb is irrelevant). These data are consistent with a model in which Numb blocks Notch/Spdo-mediated downregulation of eve in the neurons of the EL lineage (see Discussion).

We have shown that the Notch/Numb/Spdo pathway regulates asymmetric cell fate between many sibling neurons in the CNS. Our model (Fig. 5) is that Numb protein is asymmetrically segregated during GMC mitosis into one neuronal sibling where it blocks Notch/Spdo signaling, resulting in ‘B’ cell fate. In contrast, the neuronal sibling lacking Numb protein has an active Notch/Spdo signaling pathway, resulting in the ‘A’ cell fate. In addition, the Notch/Numb/Spdo pathway regulates binary sibling cell fate in the embryonic PNS (Dye et al., 1998), suggesting that it plays a fundamental and widespread role in establishing asymmetric sibling cell fates in multiple Drosophila tissues. Because Notch controls such a diverse array of cell fates, its activity is unlikely to specify directly a particular cell fate, but rather to allow sibling cells to respond differently to a shared environment of intrinsic or extrinsic cues. A current model is that Notch activity delays cell determination, allowing two adjacent cells to respond differently in an environment of extrinsic signals that are spatially uniform but temporally distinct (e.g. Fortini et al., 1993; Artavanis-Tsakonas et al., 1995; Dorsky et al., 1997). It is likely that sibling neurons use the Notch/Numb/Spdo pathway in a similar manner, but to generate two distinct cell fates in combination with intrinsic factors present in each GMC (and different in each GMC). This could occur by two mechanisms: (1) Notch signaling might actively specify cell fate in conjunction with existing intrinsic factors, or (2) Notch signaling might delay differentiation of one sibling until the intrinsic factors have changed (due to cell intrinsic or extrinsic events). In either case, the result is two different sibling cell fates.

Loss of numb affects all sibling neurons assayed except aCC/pCC. However, maternal numb product may control aCC/pCC fate as well, because there is a clear maternal numb contribution to CNS development (Table 1) and aCC/pCC are the earliest sibling neurons to form among those assayed here (see Fig. 1). Alternatively, numb might have no function in this lineage and binary fates could be determined by restricted activation of Notch in the future pCC due to restricted distribution of a Notch ligand, or to competitive ‘lateral inhibition’ between the aCC/pCC siblings (as ablation experiments suggest for the aCC/pCC siblings in the grasshopper embryo; Kuwada and Goodman, 1985).

During mammalian neurogenesis, proliferating cells within the ventricular zone of the brain can produce siblings with different fates: one cell remains in the ventricular zone as a proliferating precursor, while the sibling migrates away and may differentiate as a neuron (Chenn and McConnell, 1995). Vertebrate homologues of Notch, numb and spdo are known (Zhong et al., 1996; Fowler, 1996; Chenn and McConnell, 1995), raising the possibility that the Notch/Numb/Spdo signaling pathway may be an evolutionarily conserved mechanism for establishing asymmetric sibling cell fates in the CNS.

The ten Eve⁺ EL neurons appear to be five pair of siblings that develop from five Eve⁻GMCs: we never detect Eve⁺ mitotic GMCs within this lineage, the EL lineage contains only 10-13 neurons and there is no evidence for apoptosis in this lineage (Schmidt et al., 1997; A. Schmid and C. Q. D., data not shown). numb is required for eve expression in the EL neurons, but loss of spdo, Notch, Delta or mam does not affect the number of Eve⁺ EL neurons. Importantly, the concomitant removal of both spdo and numb completely restores the normal number of Eve⁺ EL neurons. These data are consistent with a model in which Notch/Spdo signaling represses eve expression in the EL neurons but, during wild-type development, Numb is equally distributed to both siblings at mitosis and thereby blocks Notch signaling in both EL sibling neurons (Fig. 5). Determining the distribution of Numb in the mitotic GMCs of the EL lineage awaits the development of a marker for these GMCs.

Mutations in spdo, Notch, Delta and mam all yield the identical sibling neuron phenotypes, suggesting that Spdo may be involved in the Notch signaling pathway. spdo encodes a homologue of vertebrate tropomodulin (Dye et al., 1998). Tropomodulin caps the pointed ends of microfilaments and is thought to regulate their length (Fowler, 1996). Spdo could regulate the distribution of either Delta ligand or Notch receptor in the membrane, for example, to facilitate ‘capping’ observed when these two proteins interact in cell culture (Fortini and Artavanis-Tsakonas, 1994). Alternatively, Spdo could regulate the intracellular trafficking or processing of Notch protein from the endoplasmic reticulum to the plasma membrane required to produce a functional receptor (Pan and Rubin, 1997; Blaumueller et al., 1997).

Another interesting possibility is that Spdo may play a role in Notch signaling from the membrane to the nucleus. Notch signal transduction is not fully understood. The Notch intracellular domain is necessary and sufficient for Notch signaling, it can be detected in the nucleus in cell culture or when overexpressed, its nuclear localization is essential for its function and it can be released from a membrane tether via proteolytic cleavage (Coffman et al., 1993; Rebay et al., 1993; Struhl et al., 1993; Lieber et al., 1993; Kopan et al., 1994, 1996). The Su(H) transcription factor (Furukawa et al., 1991; Schweisguth and Posakony, 1992) can bind to the intracellular domain of Notch and is translocated into the nucleus upon activation of Notch in tissue culture assays (Fortini and Artavanis-Tsakonas, 1994). Taken together, these results suggest that Notch signal transmission may occur via ligand-induced cleavage of the intracellular domain of Notch followed by its nuclear translocation, where it functions with Su(H) to activate transcription. The endogenous Notch intracellular domain has never been observed in the nucleus, however, and thus the exact mechanism of Notch signal transduction remains unresolved. To elucidate the role, if any, Spdo plays during Notch signaling, it will be important to determine where it acts within the Notch pathway and whether it physically associates with any members of the Notch pathway.

Notch is known to function in many different tissues to control cell fate. Notch signaling mediates ‘lateral inhibition’ in the ectoderm and mesoderm which controls the number of neural and muscle precursors (Cabrera, 1990; Corbin et al., 1991; Skeath and Carroll, 1992; Martin-Bermudo et al., 1995), as well as cell interactions during oogenesis, eye development and limb patterning (e.g. Cagan and Ready, 1989; Ruohola et al., 1991; Kim et al., 1996; for review see Artavanis-Tsakonas et al., 1995). It is possible that the Notch pathway utilizes different components in each tissue. E(spl) and neuralized have no role in sibling neuron specification (data not shown), although they are essential for the earlier process of Notch-mediated lateral inhibition (Martin-Bermudo et al., 1995; Skeath and Carroll, 1992). Conversely, we have shown that spdo regulates Notch-mediated sibling cell fate decisions but is not involved in Notch-mediated lateral inhibition. In addition, cell clones homozygous for numb³ do not affect lateral inhibition in imaginal discs (Rhyu et al., 1994), but this is not a null allele of numb. To test whether numb is an obligate member of the Notch signaling pathway, it will be necessary to make germline and imaginal disc clones using null alleles of numb. Nonetheless, one interesting possibility is that spdo and numb regulate Notch signaling in non-epithelial cells (e.g. sibling neurons) but do not affect Notch-based signaling in epithelia (e.g. neuroectoderm or imaginal discs).

We thank Eric Spana for Fig. 4C; Karen Lolans, Tiffany Hor, Pablo Perez and Jennifer Liedel for help with fly stocks and phenotype analysis; Kathy Matthews, Gary Struhl, Julie Gates, Carl Thummel, Todd Laverty, Gerry Rubin, the Bloomington Stock center and the Berkeley Drosophila Genome Project for fly stocks. We thank Aloisia Schmid, Hugo Bellen and Catherine Dye for sharing unpublished data and reagents, and Eric Spana and Kate O’Connor for comments on the manuscript. This work was supported by a fellowship from the Cancer Research Fund of the Damon Runyon-Walter Winchell Foundation (DRG-1279) by an NIH grant (NS36570-01) and by a grant from the American Cancer Society (ACS-IRG 36-38) to J. B. S. and by an NIH grant (HD 27056) to C. Q. D., who is an Associate Investigator of the Howard Hughes Medical Institute.