Research ArticleDevelopmental Biology

Direct Response to Notch Activation: Signaling Crosstalk and Incoherent Logic

See allHide authors and affiliations

Science Signaling  27 Jan 2009:
Vol. 2, Issue 55, pp. ra1
DOI: 10.1126/scisignal.2000140

This article has a correction. Please see:


Notch is the receptor in one of a small group of conserved signaling pathways that are essential at multiple stages in development. Although the mechanism of transduction impinges directly on the nucleus to regulate transcription through the CSL [CBF-1/Su(H)/LAG-1] DNA binding protein, there are few known direct target genes. Thus, relatively little is known about the immediate cellular consequences of Notch activation. We therefore set out to determine the genome-wide response to Notch activation by analyzing the changes in messenger RNA (mRNA) expression and the sites of CSL occupancy within 30 minutes of activating Notch in Drosophila cells. Through combining these data, we identify high-confidence direct targets of Notch that are implicated in the maintenance of adult muscle progenitors in vivo. These targets are enriched in cell morphogenesis genes and in components of other cell signaling pathways, especially the epidermal growth factor receptor (EGFR) pathway. Also evident are examples of incoherent network logic, where Notch stimulates the expression of both a gene and the repressor of that gene, which may result in a transient window of competence after Notch activation. Furthermore, because targets comprise both positive and negative regulators, cells become poised for both outcomes, suggesting one mechanism through which Notch activation can lead to opposite effects in different contexts.


Signaling through the Notch receptor controls many different decisions in development, contributes to the maintenance of tissues in the adult, and is associated with a number of diseases, including cancers. Despite this multitude of roles, the basic Notch transduction pathway is relatively simple: Activation of Notch elicits a proteolytic cleavage, releasing the Notch intracellular domain (NICD), which enters the nucleus and interacts directly with the DNA binding proteins of the CSL [CBF-1/Su(H)/LAG-1] family to regulate transcription (1, 2). Thus, effects on gene expression are a primary consequence of activating the pathway. Nevertheless, relatively few targets of Notch are known and analysis of Notch activation has largely focused on end points assayed many hours or days after manipulations to the pathway, rather than on the immediate and direct consequences.

In elaborating distinct responses, Notch signaling frequently is integrated with the output from other pathways, such as pathways activated by growth factors signaling through receptor tyrosine kinases (RTKs) and those activated by the ligand Wnt (37). For example, mutual cross-regulation between RTK, Notch, and Wnt signaling results in periodic oscillations essential for somitogenesis (8) and antagonistic interactions between RTK and Notch (known as Lin12 in Caenorhabditis elegans) are well documented in C. elegans vulval development, where Notch targets include negative regulators of RTK signaling (9, 10). Similar antagonistic relationships between Notch and RTK signaling have been observed in other tissues, including Drosophila embryonic muscle precursors (11). However, the nature of pathway integration differs according to the context, in some cases showing synergistic rather than antagonistic interactions (4, 5). The mechanisms underlying these different outcomes and the extent to which these involve direct transcriptional cross-regulation remain, in many cases, unknown.

Many studies have focused on long-term or indirect effects, or effects that are both long term and indirect, of Notch activation because of technical limitations in exerting fine temporal control over pathway activation. To circumvent this problem, we have used an ectodomain shedding protocol to precisely time Notch activation in Drosophila cells (12) and monitored effects on transcription. With this approach, we observed rapid stimulation of the expression of the best-characterized Notch targets, the E(spl) genes, within 30 min of Notch activation and increased occupancy by the Drosophila CSL protein [Su(H)] over a similar time course (12). The latter suggests that, although CSL contributes to target gene repression in the absence of NICD, the stability of its association with the DNA may increase when it is complexed with NICD. Thus, analyzing Su(H) occupancy and mRNA expression changes within a short time of Notch activation provides an approach for obtaining a snapshot of the genome-wide response to Notch mediated by the canonical CSL-dependent pathway.

With the temporal activation protocol, we have investigated the short-term transcriptional response to Notch in DmD8, a Drosophila cell line that has characteristics similar to those of adult muscle progenitor cells. Adult muscle progenitor cells are specified during embryogenesis but do not differentiate until pupal stages and Notch activity is required to prevent premature differentiation (13). Although Notch has diverse functions, this outcome of preventing differentiation is one that has been observed in many contexts. To understand the full scope of the Notch response in such processes, we have used the DmD8 cell model to assess the global changes in mRNA expression and the genome-wide sites of occupancy of Su(H) within 30 min of activating Notch. The intersection of these data identifies genes that have a high probability of being direct targets, and we have confirmed the responses for several newly identified targets in vivo and shown that they contribute to the regulation of differentiation. The Notch direct targets are enriched in genes involved in cell morphogenesis and cell signaling. We also identify several examples where one response target acts in opposition to another. These so-called type 1 incoherent feed-forward loops have the potential to create a window of competence after Notch activation.


Genome-wide response to Notch

To determine the transcriptional response to Notch activation, we first set out to identify genes whose expression was increased within 30 min in DmD8 cells with the use of expression microarrays. mRNA was isolated from control cells, from cells 30 min after Notch activation was initiated [through ligand-independent EDTA-induced ectodomain shedding (12)], and from cells where Notch activation was carried out in the presence of a γ-secretase inhibitor (DFK-167) to inhibit the cleavage event that is required for Notch signaling. Probes prepared from the mRNA populations were then hybridized in pairwise combinations to Drosophila transcriptome microarrays to identify genes that were significantly different (P ≤ 0.05) between RNA populations (see Fig. 1A and fig. S1A for plots of the normalized data). From these analyses, we found 197 genes significantly up-regulated after Notch activation in DmD8 cells (P ≤ 0.05), of which 76 were sensitive to inhibition by DFK-167 (Fig. 1B). We note that the number of DFK-sensitive genes is larger than expected and that there were genes in both Notch-activated and DFK-sensitive data sets that were not overlapping. This may, in part, be due to the fact that other pathways that lead to changes in gene expression are regulated by γ-secretase (DFK sensitive, but not Notch) and that there are some false positives among the genes activated by Notch. However, we consider it likely that many of the nonoverlapping genes are also Notch targets, as they are associated with Su(H) binding (see below). Furthermore, there was also substantial overlap of 100 genes with Notch-responsive genes in another cell type, Drosophila Kc cells, which are related to Drosophila blood cells (table S1). To validate the array data, we tested 16 genes in independent assays, and 15 of 16 could be successfully validated with greater than twofold change in mRNA expression (fig. S1C). Furthermore, the best-characterized direct Notch target genes, the E(spl) genes, were among the genes whose expression increased the most, as were other previously reported targets (Fig. 1, C and E, Tables 1 and 2, and tables S1 to S3).

Fig. 1

Identification of Notch direct targets by genome-wide expression and ChIP arrays. (A) Plot of average spot intensities from control and Notch-activated (Nact) DmD8 cells; significant differences in gene expression are indicated by dots that deviate from the red line. (B) Overlap between up-regulated genes (red), genes exhibiting sensitivity to DFK (yellow), and genes associated with ChIP peaks (green) in DmD8 cells. RPG represents Notch-regulated genes associated with peaks that are listed in table S1. (C) Heat map indicating increased or decreased expression or presence of ChIP peaks for the genes indicated in each condition. For expression analysis, colored boxes indicate AvgM (log2) scores according to the scale shown. ChIP peak scores are sum of AvgM for all probes defining a peak and are in excess of 5 except for twist, aop, cut, and hibris. Gray boxes indicate that no significant change in expression or Su(H) peak was detected. (D) Pie chart representing the percentage of Su(H) ChIP peaks associated with genes regulated by Notch in DmD8 cells (DmD8) or in Kc cells (Kc) or in both cell types. AP represents peaks assigned to genes; AP1, AP2, and AP3 are used to indicate genes belonging to each category in table S2; NP represents ChIP peaks that are not assigned to regulated genes. (E) Genomic region surrounding E(spl) gene complex [black lines and boxes (exons) represent transcribed regions], with graphs showing matches to Su(H) sites (red) and Su(H) conserved sites (gold) and oligonucleotides hybridizing to enriched fragments in ChIP [blue, only significant peaks identified by Tamalpais (P < 0.05) are plotted]. Height of bars indicates Patser score (7.5 to 9.79) or enrichment (AvgM log2, 0.3 to 2.3). Horizontal scale, large tick marks are spaced by 50 kb.

Table 1 Direct Notch targets in DmD8 cells: up-regulated in DmD8, DFK sensitive and associated with ChIP peaks. Dashes indicate not tested. “pupa” indicates myoblast expression was detected in pupa only. Genes from Table 1 are highlighted in gold (yellow) in tables S1 and S2.
View this table:
Table 2 AP targets with further validation. Dashes indicate not tested. “pupa” indicates myoblast expression was detected in pupa only.
View this table:

Although our analysis within 30 min of Notch activation gives a high probability of identifying direct targets, the time interval is sufficient that some secondary targets may be included. To distinguish direct from indirect targets, we performed chromatin immunoprecipitation (ChIP) to identify regions occupied by Su(H) in the Notch-activated DmD8 cells (12) and hybridized the bound DNA fragments to oligonucleotide tiling arrays covering the Drosophila genome (14). This approach is based on the assumption that all transcriptional changes involve recruitment by Su(H) so that any effects on transcription independent of this core transcription factor will be excluded from our target data sets.

In total, we identified 262 significant Su(H)-binding peaks across the genome [P ≤ 0.05, greater than five adjacent probes (15)] (Fig. 1B). Forty-nine percent of the Su(H) ChIP peaks were defined as “regulated peak genes” (RPGs), as they were located in or adjacent to genes that met the conditions for regulation by Notch (Fig. 1, B and D, and table S1). Of these, 18 genes were stimulated by Notch but were not sensitive to DFK (RPG2), 71 genes were sensitive to DFK but not stimulated by Notch (RPG3), and 37 genes were both stimulated by Notch and DFK sensitive (RPG1) in DmD8 cells (Table 1 and gold highlighting in tables S1 and S2). A further 30% of peaks were assigned based on their proximity to genes that were stimulated after Notch activation in Kc cells (Fig. 1C). Thus, 79% of Su(H) ChIP peaks were assigned to Notch-regulated genes, which we called “assigned peaks” (APs) (table S2). Of 16 loci tested, 14 could be validated independently (fig. S1D). For all subsequent analysis, we have focused only on those Notch-regulated genes that were associated with ChIP peaks (RPG and AP genes). In addition, 39 genes were stimulated in response to Notch and were DFK sensitive in DmD8 cells, but cannot be assigned to ChIP peaks. These 39 genes may be indirect targets and are listed in table S3 as “regulated genes” (RGs), along with other genes that were significant in two or more of the expression array experiments.

Analysis of the sequences defined by the AP subset of ChIP peaks showed that 95% contained matches to Su(H) binding sites (Patser score >5; fig. S2A) and that they were significantly enriched for high-scoring Su(H) sites (P < 0.001) and for multiple Su(H) sites (P < 0.001) (fig. S2, B and C). Despite this statistical enrichment, only 11 of 41 of Su(H) binding-site clusters identified computationally (16) and 1.08% of high-scoring (Patser score >7.5) Su(H) sites in noncoding sequences were occupied. We also analyzed the sequences defined by the 21% of peaks that were not assigned to up-regulated genes, which we call “nonassigned peaks” (NPs). As with the APs, most (93%) contained matches to Su(H) sites and were enriched for high-scoring and clustered Su(H) sites (fig. S2). Thus, the NP ChIP peaks have a high probability of identifying additional targets, and our failure to detect Notch-dependent stimulation of proximate genes could be either due to technical problems (false negatives) or because the associated targets are repressed through other mechanisms. Likewise, the APs that were associated with genes stimulated by Notch signaling in Kc cells but that were not Notch sensitive in DmD8 cells, despite having Su(H) bound, may represent repressed targets where Notch signaling is insufficient to overcome the repression in DmD8 cells or targets that require a cooperating activator in DmD8 cells.

Together, the combination of measuring mRNA expression and Su(H) occupancy shortly after Notch activation has identified genes that are specifically subject to transcriptional regulation, rather than having altered mRNA stability or microRNA regulation, and that have a high probability of being direct Notch targets. These encompass many of the known Notch targets [genes of the E(spl) family and Him], and more than a quarter (27% of AP genes) have been linked to Notch activation in genetic or molecular experiments (; table S2, “prev. N link”). Of the putative targets, 37 genes fulfill the “gold standard” of significance (P ≤ 0.05) in all three data sets from DmD8 cells (Table 1). Many of these (22 of 37) are previously unidentified targets and, in all cases tested, the regulation of these genes by Notch was substantiated.

Enrichment of Notch targets for genes involved in cell morphogenesis and signal transduction

To gain an overview of the cellular outcome of Notch activation, we searched for biased representation of gene function (biological process) among the putative targets with the use of Gene Ontology (GO) terms ( Although this approach has the shortcoming that uncharacterized genes from the target list are not included, it nevertheless reveals whether specific types of biological process are over- or underrepresented, particularly if compared with other data sets. A hypogeometric test on the AP genes revealed a significant enrichment in genes involved in cellular development and cell differentiation, including genes in the cell fate commitment and cellular morphogenesis categories (Fig. 2, A and D). The link between Notch and cell fate commitment is expected, especially as this group includes many negative regulators of cell fate. Because we started from myoblast-related cells, the enrichment in genes related to nervous system and eye development was unexpected. However, close inspection shows that the enriched genes are not neural specific: All are widely linked to cell fate regulation, including muscle differentiation. Indeed, several have mammalian orthologs that are important in preventing premature differentiation of myoblasts. The significant enrichment for genes associated with cell morphogenesis, including genes such as unc-5 (encoding a netrin receptor) (Fig. 2D), cornetto (corn, encoding a microtubule binding protein) (Fig. 2D), trio (encoding a Rho guanosine triphosphatase) (Fig. 2B), and hibris (hbs), roughest, kirre (encoding immunoglobulin domain proteins) (table S2), implies that Notch also directly controls cell shape and organization. Previously, most identified Notch target genes were nuclear effectors, which suggested that Notch primarily exerted its effects on cell differentiation through a transcriptional hierarchy. These data show that this is not the case.

Fig. 2

Notch targets are enriched in cell signaling genes and include multiple components of the EGFR/RTK pathway. (A) GO categories that are enriched in the AP targets, ranked by P values. Enrichment was calculated with a hypogeometric test (with Bonferroni correction) (52) and results were filtered for categories with threefold or greater enrichment and P ≤ 10−10. Categories associated with cell signaling (solid orange shading) and morphogenesis (open yellow outline) are indicated. Abbreviations: cell. morph. during differentiation, cellular morphogenesis during differentiation; cell surf. rec. linked sig. trans, cell surface receptor-linked signal transduction; instar larval or pupal develop., instar larval or pupal development. (B) Responses of signal transduction genes [includes some indirect (RG) Notch targets]; colored boxes illustrate responses or ChIP peaks for the indicated genes in each condition (colors and scales as in Fig. 1D). (C) Putative direct Notch targets include components of EGFR signaling, an RTK pathway. Genomic regions surrounding each indicated gene are shown with symbols and graphs as in Fig. 1E. Shaded boxes and numbers at the bottom of each panel indicate expression ratio (AvgM log2) in the conditions indicated (P ≤ 0.05, dark gray; P ≤ 0.1, pale gray). (D) Putative direct Notch targets include genes involved in morphogenesis (unc-5, corn), Notch [bigbrain (bib), Notch, numb] and Wnt [naked (nkd)] pathways. ChIP peaks associated with the indicated genes; genomic regions, symbols, and graphs are as in Figs. 1E and 2C.

The AP and RPG target sets are also enriched in genes associated with signal transduction and cell communication, specifically with cell surface receptor–linked signal transduction (Fig. 2, A and B). Similar enrichment is seen in the Notch up-regulated genes in Kc cells (fig. S3). In contrast, there was no enrichment for signal transduction among genes regulated by other signals in >10 genome-wide studies that we analyzed using the same criteria [including Ras, fig. S3 (17, 18), transforming growth factor–β (TGF-β) (19, 20), and Wnt (21, 22)]. As well as many components of the Notch pathway itself, the AP signal transduction targets include core genes of bone morphogenic protein (BMP), Wnt, and RTK, specifically the epidermal growth factor receptor (EGFR) pathways along with genes involved in integrin and G protein–coupled receptor (GPCR) signaling (Fig. 2B). Besides Notch, the most significantly enriched of these pathways is transmembrane receptor protein tyrosine kinase signaling pathway (6.28-fold enriched, P = 0.0001), including genes related to the EGFR pathway. The latter comprise positive and negative regulators that act at multiple levels of the pathway. Thus, genes encoding the receptor EGFR (Egfr), the extracellular inhibitor Argos (aos), and the nuclear Ets domain–lacking (edl, also known as mae) (Fig. 2, B and C) were all “gold” targets, up-regulated in DmD8 cells, inhibited by DFK, and associated with ChIP peaks (Tables 1 and 2). Signaling through RTKs involves the Ras mitogen-activated protein kinase (MAPK) pathway, and four other Ras pathway components or regulators are AP genes [anterior open (aop), pointed (pnt), sprouty (sty), corkscrew (csw)] (Fig. 2, B and C). Two others (Gap1, mkp3) are regulated genes in DmD8 cells, and one [rhomboid (rho)] was up-regulated in Kc cells. Most of the RTK pathway genes were also found to respond to inputs from Notch signaling in genome-wide studies of Ras and Notch integration in Drosophila embryos (23). These EGFR pathway targets indicate that Notch influences RTK signaling at points beyond the mkp3 phosphatases (known as lip-1 in C. elegans) identified previously (9, 10) and the overrepresentation of signal transduction genes shows that one major output of Notch activation is its modulation of other pathways.

Although not reaching the threshold for inclusion in Fig. 2A, there is significant enrichment for genes associated with programmed cell death (fivefold enrichment, P = 0.00003). Notch has been linked to apoptosis in several contexts (2429), but no direct mechanistic connection has been established. Two putative targets are head involution defective (hid, also known as Wrinkled) and reaper (rpr) (fig. S4), both of which are key cell death regulators (30). This implies that Notch regulates apoptosis directly, rather than indirectly, through genes involved in cell fate determination. In agreement, expression of an rpr::lacZ reporter gene, containing 11 kb of 5′ regulatory sequences fused to lacZ (31), was strongly increased in clones of cells expressing NICD in vivo (fig. S4). In the same tissue, NICD elicited increased programmed cell death as measured by activated caspase3 (fig. S4).

In contrast, the AP genes are not enriched for genes involved in cell cycle control or mitosis, despite the link between Notch and cell proliferation in many contexts (including the adult muscle progenitors). Indeed, many GO categories associated with mitosis and cell division are calculated to have lower-than-expected frequency (although P values are >0.05). This is strikingly different from the gene profile after exposure to activated Ras, where genes associated with mitotic cell cycle were highly enriched, both in mammalian and Drosophila cells (fig. S3) (17, 18). Nevertheless, our putative Notch targets include two proliferation-related genes—string (stg, the Drosophila homolog of cdc25), which encodes a phosphatase, and diminutive (dm, the Drosophila homolog of myc)—that were identified previously as direct targets of Notch in leukemia cells (32). Those studies also found no significant enrichment for cell division–related genes and showed that Myc was necessary for the Notch-dependent proliferation. Similarly, we find that the expression of dm is stimulated by Notch in vivo and that mutations in dm interfere with the ability of Su(H) to promote overgrowth in vivo (fig. S5), suggesting that Dm is an important intermediary in the Notch-induced proliferation in Drosophila. Thus, effects of Notch on cell proliferation may involve a small number of direct targets, including dm and stg, in contrast to cell differentiation where Notch appears to act through multiple targets.

Regulation and expression of newly identified Notch targets in muscle progenitors

To confirm the Notch regulation of the putative targets, we focused initially on the previously unknown EGFR and RTK targets and took two approaches. First, we tested whether fragments encompassing the regions identified by ChIP function as Notch responsive enhancers (NREs) in transient transfection assays. The fragments from egfr, edl, argos, pnt, or rho were used to drive luciferase expression, and cotransfection with NICD stimulated the production of the luciferase reporter (Fig. 3A). This response was abolished when the Su(H) sites in the gene fragments were mutated. Second, we tested whether three of the NREs conferred positive Notch regulation in vivo (Fig. 3, B to J). The in vivo correlate of DmD8 cells are the adult muscle progenitors, which are specified in embryonic stages, persist and proliferate during larval stages, then fuse to form the adult muscles in the pupa (Fig. 4A). The NREs from argos (aosNRE::lacZ), edl (edl(D)NRE::lacZ), and Egfr (EgfrNRE::LacZ) all conferred expression in the adult muscle progenitors (Figs. 3, C and H, and 4E). This constituted a subset of the endogenous expression patterns; for example, argos is expressed in the wing vein primordia as well as the muscle progenitors, but the aosNRE directs expression specifically in the muscle progenitors (compare Fig. 3, B and C). The NREs also differed slightly in the time of expression; aosNRE::lacZ and edl(D)NRE::lacZ were detected in larval stages (Fig. 3, C and H) and EgfrNRE::lacZ was seen in the unfused myoblasts in the pupa (Fig. 4E). Nevertheless, expression from all three enhancers was increased by NICD (in transient transfection assays). Expression of Su(H)VP16, in which Su(H) is converted to a constitutive activator by fusion with the viral VP16 activation domain, also stimulated the expression of the reporter from the NREs in the adult muscle progenitors (Fig. 3, D and I). Conversely, reduction of Notch activity by RNA interference (RNAi) decreased the expression from the NRE reporters (fig. S6). Furthermore, when the Su(H) binding sites in argosNRE, edl(D)NRE, or EgfrNRE were mutated, the expression in muscle progenitors was lost and the reporter was no longer stimulated by Su(H)VP16 (Fig. 3, F, J, and G, and fig. S6). Thus, these three EGFR pathway genes, which encode both positive (EGFR) and negative (Argos) regulators of the EGFR pathway, are all stimulated by Notch in vivo in the myogenic progenitors.

Fig. 3

EGFR pathway genes are Notch targets in vivo. (A) Response of the indicated enhancers to NICD (dark and light blue) in transient transfection assays; light blue bars represent results when the Su(H) sites were mutated. Averages of more than three biological replicate experiments; error bars are standard error of the mean. NME is negative control (Notch mutant enhancer); m3 is NRE from E(spl)m3. (B to D, F to J) β-Galactosidase expression from the indicated transgenes in the absence (B, C, F, H, and J) or presence (D, G, and I) of Su(H)VP16. “mut” indicates enhancers where Su(H) sites have been mutated and discs in F and J were stained for three times longer. “argos” (B) is expression from a lacZ P-element inserted at the argos locus. MPs in (B) to (D) and (F) to (J) indicate muscle progenitors, and asterisks in (H) to (J) mark expression in the air sac, which is not regulated by Notch. (E) aosNRE::lacZ expression is significantly increased with Su(H)VP16 (P < 0.001), mean values for gray levels (arbitrary units) in images from more than five discs per genotype. (K) Model summarizing EGFR pathway genes that are regulated by Notch activation. Gold and blue boxes: direct targets identified in three or more (gold) or two (blue) conditions, including ChIP peaks. Gray boxes: Notch-sensitive genes in two or more expression array experiments (solid line) or in one expression array experiment (dashed line) that are not associated with Su(H) peaks.

Fig. 4

Notch regulation and myoblast expression of previously unknown targets. (A) Diagram summarizing stages in myoblast differentiation. Gray shading indicates abundance of Mef2 in late larval myoprogenitors (adepithelial cells) and in mid-pupa where cells with higher Mef2 fuse to form myofibers; flanking unfused myoblasts exhibit lower abundance of Mef2. Notch is required to prevent differentiation both before the increase in Mef2 and at fusion stages. (B to H) Expression of indicated targets in myoprogenitors or myoblasts at larval (B to D) or pupal (E to H) stages detected by β-galactosidase activity (B and C, gray), Gal4 insertion (D, green, CG9650::Gal4/UAS::H2B-YFP; α-cadherin, purple, reveals disc outline), or with antibodies against β-galactosidase (E and F, green; E′ and F′, gray), CG18446 (G, green; G′, gray), and Hnt (H, gray). For those marked NRE, expression is from a reporter gene driven by a fragment encompassing the ChIP peak associated with the indicated gene. In 21-hour pupae (E to H), myofibers are detected by the increased abundance of Mef2 (E and F, purple, arrowheads) or 1151::Gal4/UAS::H2B-YFP (G, purple, arrowheads) and Notch targets (green, E to G; gray, E′ to G′ and H, arrows) are expressed at highest abundance in the flanking myoblasts with lower Mef2. Fluorescent channels in E′ to G′ and H have been inverted. (I) Response of the indicated enhancers to NICD in transient transfection assays. Averages of more than three biological replicate experiments; error bars are the SEM.

We performed similar analyses on a range of other targets. The ChIP-identified Su(H) binding regions from 16 other genes were tested in transient transfection assays and in all cases gave a robust response to NICD (Tables 1 and 2 and Fig. 4I). Control fragments, containing matches to the Su(H) consensus sequence that were not enriched in the ChIP, showed less than twofold stimulation. Subsequently, we investigated whether the corresponding genes were expressed in the muscle progenitors in vivo, consistent with a function downstream of Notch in these cells. Of the 24 genes tested (with the use of transposon insertions, enhancer-lacZ transgenes, or available antibodies), all were expressed in the adult muscle progenitors in the late larva or early pupal stages (Tables 1 and 2 and Fig. 4, B to H). Finally, we assessed the response to Notch in vivo by analyzing, for a subset of genes, whether expression was increased in the presence of NICD or Su(H)VP16, was decreased when Notch activity was ablated by RNAi, or met both conditions. All of the 19 genes tested showed a response in these assays (Tables 1 and 2 and figs. S4 and S5), confirming that they are regulated by Notch.

EGFR and Notch cooperate in muscle progenitors

Previous studies have shown that Notch is required to maintain the adult muscle progenitors and prevent their premature differentiation (13). We therefore examined whether the disruption or hyperactivation of the previously unidentified Notch targets phenocopied the defects caused by aberrant Notch signaling in this context. In wild-type animals, the adult muscle progenitors express the basic helix-loop-helix (bHLH) protein Twist and the abundance of the muscle transcription factor Mef2 is low. The abundance of Mef2 increases during the transition to pupal stages (33) and subsequently increases further in nuclei of cells that have fused to form myofibers (Fig. 4A). Persistent Notch activity results in unfused myoblasts at pupal stages (13). Conversely, Notch inhibition results in premature differentiation. This could be detected by premature increase in Mef2, which is accompanied by the precocious appearance of muscle-like cells that are positive for myosin heavy chain and actin (Fig. 5, A to C, E and G), in larvae where Notch function was ablated by RNAi. We also observed a significant decrease in proliferation [as measured by the mitotic marker phosphorylated histone H3 (pH3)] and in cell death (as measured by activated caspase3) in the RNAi-treated larvae (Fig. 5K), consistent with the fact that proliferation- and apoptosis-promoting genes are among the direct Notch targets.

Fig. 5

Notch, EGFR, and other targets regulate differentiation of adult muscle progenitors. Ablation of Notch or EGFR causes premature muscle differentiation in adult muscle progenitors. (A) Wild-type wing disc; green indicates muscle progenitors (MPs) as shown in (B) to (F). (B to D) Mef2 is absent from nuclei of wild-type MPs (B), but is present in nuclei of MPs expressing Notch RNAi (N-RNAi) (C) or dominant-negative EGFR (DN-EGFR) (D). (E and F) Phalloidin staining of actin shows that small ectopic muscles are present when MPs express Notch RNAi (E) or DN-EGFR (F). Boxed regions in E and F are shown at higher magnification and lower intensity to reveal striations. % indicates frequency of discs exhibiting striated muscle structures. (G) Myosin heavy chain (mhc) staining is detected in MPs expressing DN-EGFR. (H to J) Hth-EnR and Zfh-1 expression prevents myoblast differentiation. In wild-type pupa (21-hour), differentiating muscles have aligned nuclei and Mef2 is elevated compared to adjacent cells (arrows, H). Expression of Hth-EnR (I) or Zfh-1 (J) perturbs differentiation, fewer (J) or no (I) ordered nuclei are detected, and there is no differential in Mef2 abundance. (K) Effects of Notch and Hnt on proliferation and cell death. Number of cells with phosphorylated histone H3 (pH3) or activated caspase 3 (casp3*) were significantly (P < 0.005) reduced in N-RNAi– or Hnt (light gray)-expressing myoprogenitors compared to wild type (dark gray).

The relationship of EGFR to Notch function has not been examined in the adult muscle progenitors, but extensive studies in the embryo have shown that Notch and EGFR act antagonistically to regulate the selection of muscle founder cells (11). If the same relationship holds in adult muscle progenitors, inhibition of EGFR signaling should have effects opposite to those produced by knockdown of Notch. Conversely, if EGFR signaling contributes to Notch function, as suggested by the fact that it is a direct target, blocking EGFR should produce effects similar to those produced by blocking Notch signaling. Our results support the latter model. Inhibition of EGFR activity (with dominant-negative EGFR) had outcomes similar to those produced by knockdown of Notch with RNAi: Both led to elevated abundance of Mef2 (Fig. 5, B to D) and the differentiation of striated muscle-like cells in a high percentage of specimens (Fig. 5, E to G). Thus, in the adult muscle progenitors, Notch and EGFR act in the same direction to prevent premature differentiation, despite the fact that Notch targets here include genes encoding EGFR pathway inhibitors, such as Argos, which inhibits ligand-receptor interactions.

Several of the identified Notch targets regulate myogenesis in Drosophila [twist (twi), Him, zfh-1 (3436)] and orthologs of these and others [homothorax (hth), HDAC4] have been linked to myogenesis in mammalian cells (37, 38). We tested whether three other targets, hth, zfh-1, and hindsight [hnt, also known as pebbled (peb)] have effects on differentiation in the adult myoprogenitors. Overexpression of Zfh-1 or Hnt or introduction of a dominant-repressor form of Hth [Hth-EnR (39)] produced phenotypes similar to excessive activation of Notch in the myoblasts. Rare surviving adults were flightless and there were many unfused myoblasts in pupae overexpressing Zfh-1 and Hth-EnR (Fig. 5, H to J). Hnt overexpression also resulted in flightless adults, but the disruptions at pupal stages were milder. Therefore, the in vivo effects are consistent with those of the previously unknown Notch targets functioning to regulate differentiation of the adult muscle progenitors.

Incoherent logic in the Notch response

The induction of both positive and negative components of the EGFR pathway is an example of so-called incoherent network logic, where the primary response includes two paths that have opposite effects. Incoherent feed-forward loops have been characterized in the context of transcriptional networks, where the inhibitors are transcription factors or RNA binding proteins that directly repress expression of another primary target, giving rise to a biphasic response (40, 41). With Egfr and argos, the interactions are predicted to occur at the level of the proteins, as Argos is a secreted antagonist of the EGFR ligand Spitz. However, the Notch targets also include pairs that may conform to conventional transcriptional incoherent feed-forward loops (Fig. 6A), including twi-E(spl)bHLH, stg-hnt, and dm-brat. The hnt gene encodes a zinc finger transcription factor with homology to RREB1, which has been shown to inhibit stg expression in follicle cells (42), and Brat is an RNA binding protein that affects dm mRNA stability in neuroblasts (43). Here, we have analyzed further the relationships between some of these genes [twi-E(spl)bHLH, stg-hnt, and to a lesser extent, Egfr-argos] in muscle progenitors.

Fig. 6

Incoherent feed-forward loops in the Notch response. (A) Examples of two incoherent feed forward loops in the Notch response. (B) Notch targets include stg, hnt, and twi. Genomic regions surrounding each indicated gene are shown with symbols and graphs as in Figs. 1E and 2C. (C) Levels of mRNAs for the indicated genes relative to mRNA for ribosomal protein 49 (rp49) in unstimulated cells. (D and E) Temporal profiles of mRNA expression after Notch activation. Cells were treated for 5 min with Notch-activating or control conditions and harvested immediately (t = 0) or at the indicated times. For 30C, cells were incubated continuously in Notch-activating conditions for 30 min and then harvested. The fold difference in mRNA expression between Notch-activated cells and control-treated cells at each time point was quantified for the genes indicated (standard error of the mean). (F) twi::lacZ is inhibited in myoprogenitors expressing N-RNAi or E(spl)mβ in comparison to wild type (wt). (G) Graphs show quantification of lacZ expression from multiple discs of each genotype as indicated; in both experiments, expression is significantly inhibited (P < 0.015). (H) stg::lacZ expression is inhibited in pupal myoblasts by Hnt overexpression (>hnt) compared to wild type (wt).

Because the outcome of the incoherent loops depends on the relationship in the expression of the genes encoding the two components, we analyzed mRNA expression before and after a short (5 min) pulse of Notch activation in DmD8 cells (Fig. 6, C, D, and E). Three of the genes (twi, stg, argos) had relatively abundant expression before Notch activation (Fig. 6C), and were augmented by a twofold increase that peaked at about 25 min after activation (Fig. 6E). In contrast, the reciprocal members of each putative loop [E(spl), hnt, Egfr] had little expression before the pulse of Notch activation (Fig. 6C) and were all strongly induced after Notch activation (Fig. 6, D and E). With E(spl) genes (for example, E(spl)m3 in Fig. 6D), the levels peaked at 25 min and subsequently declined. With hnt and Egfr the increase was more sustained, especially in the case of Egfr (Fig. 6E). In each case, therefore, the two components have different profiles, with one present at high levels from the outset and the other showing more marked (and sometimes more sustained) induction.

We next investigated whether the regulatory interactions between twi-E(spl) and stg-hnt fit those predicted by an incoherent feed-forward loop. The first prediction, that all four genes should be positively regulated by Notch, is already evident from the ChIP and expression arrays where they all fulfilled the criteria of significance in all data sets (Fig. 6B and Table 1). Furthermore, Su(H) binding regions from E(spl), stg, and hnt respond to NICD in transient transfection assays (Fig. 4I), and expression of twi::lacZ in myoblasts was significantly reduced when Notch function was reduced by RNAi (Fig. 6F). To test the second prediction, that E(spl) and Hnt repress twi and stg, respectively, we overexpressed the putative inhibitors in the muscle progenitors and monitored the consequences on the lacZ reporters. Elevated expression of E(spl)m7 or E(spl)mβ was sufficient to reduce twi::lacZ expression (Fig. 6, F and G) and elevated Hnt was sufficient to repress stg::lacZ (Fig. 6H). In addition, when Hnt was expressed earlier than normal in the myoblasts, it resulted in a decrease in mitosis, similar to effects of Notch RNAi, as expected from its effects on stg, which encodes a positive regulator of mitosis (Fig. 5K). The regulatory interactions, therefore, support the model of incoherent feed-forward loops in the Notch response and suggest that the likely outcome is a limited window of Twi and Stg activity.


Our genome-wide analysis within 30 min of Notch activation reveals a snapshot of the direct Notch response. In so doing it identifies many previously unknown Notch targets that, from our in vivo validation, are likely involved in the maintenance of adult muscle progenitors (Fig. 7). There is also significant enrichment for genes associated with cell morphogenesis, implying that Notch activity has a direct role in coordinating cell architecture, as well as cell fate. In addition, two important messages emerge. First, direct targets of Notch are highly enriched in signal transduction genes, including those encoding core components of RTK (EGFR), Notch, TGF-β, and Wnt pathways. In particular, these reveal a complex regulation of EGFR pathway, involving both negative and positive components. This signal transduction gene enrichment was seen in both cell types DmD8 and Kc and shows that through this complex transcriptional regulation, Notch has the potential to exert fine-scale tuning on other signaling pathways. Second, within the cohort of targets, several generate type 1 incoherent feed-forward loops, where two arms of a response act in opposition (Fig. 7) and are likely to result in a transient window of competence after Notch activation. The fact that targets comprise both positive and negative regulators also suggests a mechanism through which Notch activation can lead to opposite effects in different contexts, depending on which is activated more rapidly or at lower signal strength.

Fig. 7

Diagram summarizing outputs in the Notch response. Targets identified through genome-wide analysis in DmD8 cells imply that Notch acts at multiple levels to regulate differentiation in muscle progenitors. Included are several incoherent feed-forward loops where two targets act in opposition (orange and green); in each regulatory loop, one gene is expressed at higher levels early (orange shading).

It is well known that Notch integrates with other major signaling pathways, but it is unexpected to discover that such extensive pathway cross-talk occurs at the transcriptional level. Five developmental pathways have more than one Notch-regulated component, and both EGFR pathway and the Notch pathway itself are highly represented. Despite the fact that Notch often acts antagonistically to other pathways, the targets include positive, as well as negative, regulators. For example, besides the phosphatases of the Mkp3 family that were shown previously to be involved in antagonistic effects of Notch on RTK signaling (9, 10), we find that genes that act positively in the pathway, such as those encoding EGFR itself and Pnt, are also Notch targets. Furthermore, genome-wide studies in Drosophila embryos revealed extensive mutual regulation of core components of RTK signaling by Notch and Ras (23). These results suggest that transcriptional regulation of RTK signaling is also likely be important for cooperation between the two pathways, and, in agreement, we find that the two act cooperatively in the adult muscle precursor cells where EGFR is a Notch-responsive target. Similarly, the diversity of Notch pathway components that are putative targets implies that autoregulation may either augment or inhibit signaling, depending on the relative effects on each of the components. One possible outcome of negative autoregulation is an oscillatory activity of the pathway, as occurs in somitogenesis where Notch and RTK components were found to coordinately oscillate, along with the Wnt pathway antagonist Naked (Nkd) (8). Many of the oscillating components are orthologs of the genes that we identify as Notch targets in Drosophila, which potentially explains their co-regulation and suggests that genes such as nkd could be common nodes of pathway crosstalk.

Examples of incoherent logic among the Notch targets include the twi-E(spl)bHLH, stg-hnt, myc-brat, and Egfr-argos gene pairs. In each case, a gene (twi, stg, myc, Egfr) and its inhibitor [E(spl), hnt, brat, argos] are putative direct targets. The consequences for each example will depend on many parameters, including the production rates and thresholds required for activity of each factor. Outcomes observed for synthetic type 1 incoherent feed-forward loops in transcription networks are pulselike or biphasic when the two components exhibit different temporal dynamics (one accumulates faster than the other) or when they respond to different thresholds of signal (40, 41). The expression profiles for stg-hnt and twi-E(spl)bHLH suggest that the positive output target (Twi, Stg) is more abundant earlier. The likely outcome therefore is that Notch activation would result in a window of Twi and Stg activity that would be shut down if the inhibitors [E(spl), Hnt] accumulated sufficiently. In contrast, the profiles for Egfr and argos suggest the converse relationship because expression of the gene argos, encoding the inhibitor, is abundant early and transcription increases for Egfr at later times (and is expressed at later stages in vivo). In this case, therefore, the most likely consequence of the regulatory loop is a delayed onset or higher threshold for EGFR activity. Also different in the EGFR-Argos interaction is that instead of a transcriptional regulation, Argos acts directly to control EGFR activity.

The difference between the proposed relationship for argos-Egfr and for the other gene pairs highlights the difficulty of predicting the output from incoherent regulatory interactions. This is further illustrated by differing consequences of gain or loss of Notch activity on twi expression in the embryo and adult muscle progenitors. In both situations, twi and E(spl) have now been linked with Notch in an incoherent feed-forward loop. However, in the embryo, loss of Notch results in an increase in twi expression (34), whereas in the adult progenitors loss of Notch results in a decrease. The ChIP in DmD8 cells identifies a different Su(H) binding region from that mapped in the embryonic studies. Thus, it is likely that the specific enhancer used, and its organization, will be important in determining the output of the regulatory interactions. The deployment of different enhancer modules may also explain how individual genes can exhibit different responses in other tissues, and we note that many of the identified targets contain more than one peak of Su(H) occupancy.

Amongst the mammalian orthologs of the AP genes, many have been implicated in myoblast maintenance in mammalian cells, but not previously linked to Notch. These include orthologs of HDAC4, Hth (Pbx in mammals), and Zfh1 (ZEB1 in mammals), which all contribute to the repression of Mef2 or its targets (37, 38, 44). Likewise, Him and Zfh-1 have been shown to inhibit Mef2 function in Drosophila (35, 36). In addition, although EGFR acts antagonistically to Notch and promotes muscle formation in the Drosophila embryo, here we have found that at larval stages EGFR acts cooperatively with Notch in the adult muscle progenitors and is required to prevent premature differentiation. A similar relationship may exist in mouse, where reductions in Notch signaling caused accelerated differentiation of myoblasts (45) and a Ras antagonist promoted myotube formation (46).

The fact that so many of the direct targets are associated with the same regulatory function implies that Notch micromanages the regulation of cell fate commitment, acting through multiple targets. Similarly, the significant enrichment for genes with functions in cytoskeletal regulation, cell junctions, and cell adhesion argues that Notch signaling also impinges directly on cell shape and morphogenesis, rather than exerting all its control indirectly through effects on genes involved in cell fate determination. Included among the cytoskeletal regulators are two genes, disabled and trio, which have been linked genetically with Notch function, but previously postulated as Su(H) independent effectors in axon guidance (47), and talin, which encodes a positive factor for integrin adhesion (48) and which could explain the effects on integrins after Notch activation in endothelial cells (49, 50).

In summary, our genome-wide analysis in DmD8 cells has uncovered previously unknown aspects of the Notch response that are important for its role in preventing differentiation of muscle progenitors. As we find a considerable overlap with targets regulated in Drosophila blood-related Kc cells and orthologs of several genes have been identified as direct Notch targets in human cells (such as cdc25, myc, and mbs), it is likely that many of the mechanisms and targets identified will be widely relevant for Notch function in other signaling paradigms.

Materials and Methods

Methods summary for expression arrays and chromatin immunoprecipitation on genomic arrays

For expression analysis, labeled complementary DNAs were prepared from the mRNA populations and hybridized in pairwise combinations (control versus N; N+DFK versus N) with long-oligonucleotide microarrays representing 18,000 expressed genes from release 4 of the Drosophila genome. To activate Notch, cells were treated with EDTA as described previously (12). For ChIP, cross-linked chromatin was prepared from Notch-activated cells, sonicated, and precipitated with an antibody against Su(H) [α-Su(H), Santa Cruz]. Precipitated fragments were amplified and labeled, then hybridized to Nimblegen long-oligonucleotide tiling arrays covering the Drosophila genome [release 4; excluding repetitive sequences (14)]. For both expression and ChIP arrays, the average difference between samples after normalization is expressed as AvgM (log2). CyberT (51) was used to identify statistically significant differentially expressed genes and Tamalpais (15) to identify significant ChIP peaks. Peaks were assigned to genes according to position: 0 or 0.5 indicates peak lies within gene, 1 indicates peak assigned to the nearest gene on one or other strand, distance is indicated in base pairs (table S2). Further details are provided in supplementary material. GO analysis was performed with GoToolBox (52) and FuncAssociate (53).

Binding site analysis

A position-weighted matrix was generated from 56 Su(H) binding sites (belonging to 18 genes) that have been shown to bind Su(H) by gel shift assay experiments and Patser (54) was used to search the whole genome of Drosophila melanogaster for matches to the matrix. Details of the matrix and statistics are provided in supplementary methods.

Luciferase experiments and lacZ reporters

Fragments were amplified from Drosophila genomic DNA and cloned into a luciferase vector containing a minimal promoter from the hsp70 gene (pGL3::Min). Cell culture conditions and transfections were as described previously (55, 56). At least three biological replicates were performed in all experiments. For in vivo enhancer analysis, fragments were cloned into a transformation vector containing lacZ downstream of a minimal hsp70 promoter (pBlueRabbit; details available on request) and more than three independent transgenic lines were analyzed. Su(H) sites were mutated using oligonucleotides with 3-bp mismatch (introducing T at positions 4, 6, 7) in conjunction with Pfu turbo polymerase (Strategene) to amplify the mutated fragment and Dpn 1 cleavage to remove the template DNA.

Drosophila experiments

Alleles and stocks are described in Flybase ( unless otherwise indicated and are detailed in supplementary methods. Immunofluorescence staining of larval discs and pupal muscles was performed as described previously in Cooper and Bray (57) and Bernard et al. (58) with the following primary antibodies: antibody to Mef2 (1:500, gift of B. Patterson), guinea pig antibody to Myc (1:1000, gift of F. Martin and G. Morata), mouse antibody to β-galactosidase (1:20, Developmental Studies Hybridoma Bank). Increased or decreased β-galactosidase expression was detected histochemically; images were captured and quantified with Adobe Photoshop.

Accession numbers

The array data have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus (GEO, and are accessible through GEO Series accession numbers GSE9964 and GSE9965.


We thank B. Fischer, R. Russell, and S. Russell (FlyCHIP) for facilitating the expression array experiments and analysis; T. Southall and A. Brand (Gurdon Institute) for access to and help with genomic tiling arrays; R. Lyne and K. Rutherford (Flymine) for help with ChIP analysis; G. Hurlbut and S. Artavanis-Tsakonas for communicating results before publication; J. Silber, A. Lalouette, and A. Salzberg for fly stocks; and B. Patterson and G. Morata for antibodies. This work was supported by a program grant from the Medical Research Council. A.K. and F.B. were European Molecular Biology Organization long-term fellows. B.H. was supported by a grant from Association for International Cancer Research and S.C. by a UK Biotechnology and Biological Sciences Research Council studentship.

Supplementary Materials



Fig. S1. Array data plots and validation assays.

Fig. S2. ChIP peaks are enriched for high affinity and multiple Su(H) sites.

Fig. S3. Enrichment of GO categories amongst Ras responsive genes in Drosophila hemocytes and Notch responsive genes in Kc cells.

Fig. S4. Notch targets include apoptosis inducing genes.

Fig. S5. dm/myc is a Notch target and required for Notch induced proliferation.

Fig. S6. Enhancers from aos and Egfr are regulated by Notch in vivo.

Table S1. Regulated Peak genes (RPG): Genes regulated by Notch in DmD8 cells (EDTA stimulated and/or DFK inhibited) that are associated with peaks of Su(H).

Table S2. Assigned peaks (AP): Genes associated with Su(H) ChIP that are regulated by Notch in DmD8 and/or Kc cells.

Table S3. Regulated genes (RG): Genes stimulated by Notch activation and/or inhibited with DFK with no associated peak of Su(H) binding in DmD8 cells.

References and Notes

Stay Connected to Science Signaling

Navigate This Article