Signal Integration During Development: Insights From the Drosophila Eye

Transcription

1 DEVELOPMENTAL DYNAMICS 229: , 2004 REVIEWS A PEER REVIEWED FORUM Signal Integration During Development: Insights From the Drosophila Eye Matthew G. Voas and Ilaria Rebay* The Drosophila eye is a highly ordered epithelial tissue composed of 750 subunits called ommatidia arranged in a reiterated hexagonal pattern. At higher resolution, observation of the constituent photoreceptors, cone cells, and pigment cells of the eye reveals a highly ordered mosaic of amazing regularity. This relatively simple organization belies the repeated requirement for spatially and temporally coordinated inputs from the Hedgehog (Hh), Wingless (Wg), Decapentaplegic (Dpp), JAK-STAT, Notch, and receptor tyrosine kinase (RTK) signaling pathways. This review will discuss how signaling inputs from the Notch and RTK pathways, superimposed on the developmental history of a cell, facilitate context-specific and appropriate cell fate specification decisions in the developing fly eye. Lessons learned from investigating the combinatorial signal integration strategies underlying Drosophila eye development will likely reveal cell cell communication paradigms relevant to many aspects of invertebrate and mammalian development. Developmental Dynamics 229: , Wiley-Liss, Inc. Key words: Drosophila; Ras; receptor tyrosine kinase; Notch, signal integration; developmental timing; transcription factor; signal transduction; cell fate specification; patterning; development Received 28 July 2003; Revised 12 September 2003; Accepted 16 September 2003 INTRODUCTION During development, multicellular organisms must coordinate the growth, differentiation, and maintenance of many different cell types. To achieve this coordination, each cell must continually integrate a complex array of signals, including both inductive and inhibitory cues, and then translate these instructions into spatially and temporally appropriate developmental responses. Many of the signaling mechanisms regulating cell fate decisions are used repeatedly throughout development, generating distinct responses in different contexts. Thus, in addition to properly integrating all incoming signals, each cell must continually re-interpret the same set of signals as appropriate to the particular context in which they are received. This review will consider the question of combinatorial usage of the Notch and receptor tyrosine kinase (RTK) signaling pathways in directing distinct cell fates, focusing on recent advances in understanding derived from studies of Drosophila eye development. Literature explaining how the ommatidium is assembled will be summarized briefly, taking as a developmental starting point events that occur after the R8 cell is specified. This explanation will provide the backdrop for subsequent discussion of mechanisms of cross-talk between the Notch and RTK signaling pathways and how such events combine with the developmental history of a cell to define a combinatorial code that specifies distinct cell fates. A Primer of Drosophila Eye Development Drosophila has proven to be an extremely powerful model system to investigate developmental signaling strategies both because of the ease with which one can combine genetic, genomic, molecular, biochemical, and cellular approaches, and because it is increasingly clear that the signaling mechanisms con- Whitehead Institute and Department of Biology, Massachusetts Institute of Technology, Cambridge, Massachusetts Grant sponsor: National Institutes of Health, Eye Institute; Grant number: EY 12549; Grant sponsor: American Cancer Society; Grant number: RGP DDC. Dr. Voas present address is Department of Developmental Biology, Stanford University School of Medicine, Stanford, California. *Correspondence to: Ilaria Rebay, Whitehead Institute, 9 Cambridge Center, Cambridge, MA rebay@wi.mit.edu DOI /dvdy Wiley-Liss, Inc.

2 SIGNAL INTEGRATION IN THE DROSOPHILA EYE 163 Fig. 1. A: A scanning electron photomicrograph of a wild-type adult Drosophila compound eye reveals the highly regular honeycomb-like arrangement of the ommatidia. B: Tangential sections through a wild-type adult eye reveal the underlying cellular architecture. Ommatidia are arranged with lattice-like precision, creating a structure often referred to as the neurocrystalline lattice. C: The same tangential section at higher magnification reveals the organization of the photoreceptors for a single ommatidium. The dark circles, indicated by the numbers 1 7, are the light-sensing organelles of the photoreceptors, called rhabdomeres. Each photoreceptor extends a single rhabdomere, with the overall arrangement forming a trapezoidal pattern. Photoreceptors R1 R6 have larger rhabdomeres and are referred to as the outer photoreceptors, as they surround the smaller inner rhabdomeres of the R7 and R8 cells. In this plane of section, only the R7 rhabdomere is visible as that of R8 sits directly underneath. Pigment granules produced by the pigment cells form a hexagonal meshwork that supports the photoreceptors of each ommatidium. D: A third instar eye antennal imaginal disc, labeled with an antibody that recognizes all the photoreceptor neurons, is oriented posterior to the left. The position of the morphogenetic furrow (MF) is indicated with a black arrow. At this point, the MF has traversed approximately half of the eye field. Posterior to the MF, successive cycles of cell fate induction have led to the specification of the photoreceptors. Even at this stage and magnification, the regular lattice-like organization of the developing ommatidia is apparent. E: A single ommatidium from a pupal eye imaginal disc, stained with anti-armadillo to outline the cell membranes. Four centrally located cone cells (C), and the surrounding primary (1 ), secondary (2 ), and tertiary (3 ) pigment cells and bristles (B) are annotated. F: A schematic view of the cell fate specification events leading to assembly of a mature ommatidium. R8 is the initial photoreceptor. The remaining seven photoreceptors and four lens-producing cone cells are recruited during the late third instar. During pupal development, the pigment and bristle cells join the ommatidium at the periphery and any surplus undifferentiated cells are removed by apoptosis. The non-neuronal accessory cells provide structural support and isolate each ommatidium from its neighbors. trolling basic developmental processes have been highly conserved in evolution (Rubin, 1988; Halder et al., 1995). Thus, knowledge of the molecular circuitry of cell cell communication used in Drosophila is relevant to the study of mammalian development and its associated defects and diseases. The Drosophila compound eye provides a particularly ideal tissue in which to study the mechanisms of signal integration. The generation of marked mosaic lineages early in eye development has conclusively demonstrated the absence of an obligatory clonal relationship between the cells comprising each ommatidium (Ready et al., 1976). Instead, all available evidence supports a model whereby specification of each cell type occurs in a lineageindependent process whereby cells are recruited in a stereotyped sequence of inductive interactions that involves extensive networks of intercellular signaling events. Thus, a complex interplay between the RTK, Notch, Hedgehog, Wingless, and Decapentaplegic (Dpp/TGF- ) pathways is required for many aspects of retinal development, including eye fate specification, regulation of proliferative vs. differentiative choice points, and establishment of distinct cell fates. Finally, the eye is dispensable for viability or fertility of the fly, enabling one to experimentally manipulate these essential signaling pathways specifically in the eye and then analyze the effects in a whole animal at any developmental stage. The Drosophila compound eye is a complex neural tissue with precise cellular architecture. The adult eye is composed of several hundred reiterated subunits called ommatidia, each of which contains eight photoreceptors and a complement of non-neural support cells arranged in an invariant pattern (Fig. 1A C). Careful observations of the morphologic events underlying the assembly of this so-called neurocrystalline lattice have provided a backdrop for many studies aimed at unraveling the molecular details of cell fate specification strategies. The presumptive eye and antennal tissues of the fly, referred to as

3 164 VOAS AND REBAY the eye antennal imaginal discs, are allocated late in embryogenesis as two bilateral groups of approximately 20 cells (Garcia-Bellido and Merriam, 1969). Little is known about eye antennal disc development before the second instar larval stage except that cellular growth and proliferation is continuous. Beginning in the second larval instar stage, cells become restricted to either antennal or eye fates. This process uses the Notch and RTK signaling pathways, with Notch signaling promoting eye specification and Ras signaling antagonizing it (Kumar and Moses, 2001a). Cellular differentiation initiates during the middle of the third larval instar stage at the posterior edge of the eye disc and sweeps across in an anterior direction, taking approximately 2 days to traverse the entire field (Fig. 1D; reviewed in Wolff and Ready, 1993). An apical constriction of the disc epithelium called the morphogenetic furrow (MF) marks the passage of this wave of differentiation (Ready et al., 1976). Propagation of the MF involves a chain reaction of genetic events linked to the ensuing neuronal differentiation. Continuous reiterations of these signaling events produce a spatiotemporal gradient of photoreceptor recruitment and differentiation within the presumptive eye field, with younger ommatidia found at the anterior and older ones at the posterior. Intriguingly, while initiation and propagation of the MF was once thought to be an interesting but esoteric aspect of Drosophila development, studies in zebrafish suggest a similar strategy may be used in the developing vertebrate retina (Neumann and Nuesslein-Volhard, 2000). In the fish, photoreceptor differentiation initiates in the center of each prospective eye field and propagates radially outward (Neumann and Nuesslein-Volhard, 2000; Stenkamp et al., 2000; Russell, 2003). To what extent the full molecular complexity of MF initiation and propagation as described in the Drosophila retina is redeployed in the context of vertebrate photoreceptor differentiation remains to be determined. As the MF advances, changes in Fig. 2. A schematic depiction of the events that lead to recruitment and spacing of the R8 photoreceptor in the third instar eye disc. Individual cells are depicted as squares. Posterior is oriented to the top. Anterior to the morphogenetic furrow (MF, indicated with horizontal black arrow; posterior-to-anterior direction of propagation indicated with two short vertical black arrows) is the proneural region, shown in light gray. Proneural enhancement, involving Notch-mediated stimulation of Atonal expression (dark gray squares), occurs in the MF and subdivides the proneural region into discrete fields of cells. These proneural clusters, through a process of Notch-mediated lateral inhibition that restricts Atonal expression, are ultimately refined down to a single Atonal-expressing R8 precursor cell that will nucleate each ommatidium. These events set up the subsequent lattice-like structure of the eye by ensuring regular spacing between individual R8 cells. the expression pattern of the proneural gene Atonal reflect the twostep process that specifies R8, the first photoreceptor neuron to be determined (Fig. 2). The first step restricts Atonal to clusters of approximately 12 cells, referred to as the proneural clusters (PNCs; Jarman et al., 1993). The high level of Atonal expression in the PNCs results in Notch activation within the MF in a process termed proneural enhancement. The PNCs are staggered relative to newly born ommatidia, setting up the ultimate honeycomb arrangement of ommatidia in the eye. After proneural enhancement, Notch-mediated lateral inhibition within each PNC refines the Atonal pattern (Baker, 2000). The lateral inhibition mechanism that selects R8 cells is essentially the same as the one that selects neuroblasts in the embryonic epidermis and will not be discussed here, because it has been extensively reviewed (Artavanis-Tsakonas et al., 1999; Baker, 2000). It is interesting to note that, over a very short developmental time period, Notch pathway activation produces opposite results within the same set of cells (Fig. 2). Within the MF, during proneural enhancement, Notch strongly up-regulates atonal expression. Just posterior to the MF, during lateral inhibition in the PNC, Notch eliminates atonal expression from all cells but one. Such dramatic differences in developmental outcomes, in this case absence vs. extra R8 cells, is a common theme in Notch signaling and more broadly emphasizes the critical roles that timing and context play in the way a cell interprets a signal during development. Each newly specified R8 cell behaves as a founder cell for one ommatidium by initiating the sequential recruitment of other cells (Fig. 1F). From detailed morphologic studies, the exact time at which each cell joins has been established (Lebovitz and Ready, 1986; Tomlinson and Ready, 1987; Baker and Rubin, 1989; Wolff and Ready, 1993). Shortly after the R8 cell is specified, photoreceptors R2 and R5 are added, followed by R3 and R4 to form the five-cell precluster. At this point, all undetermined cells complete a coordinated S-phase and then undergo mitosis in a process referred to as the second mitotic wave. After the second mitotic wave, the R1 and R6 photoreceptors are recruited, followed by R7. After the photoreceptors are specified, the non-neural cell types join the ommatidium. First the four lens-secreting cone cells are added, followed by the primary, secondary, and tertiary pigment cells (1, 2, and 3 PCs). Finally, un-

4 SIGNAL INTEGRATION IN THE DROSOPHILA EYE 165 Fig. 3. A: The RTK/Ras/MAPK pathway is shown here using Drosophila epidermal growth factor receptor (EGFR) as the representative receptor tyrosine kinase (RTK). Upon binding to the secreted ligand Spitz, EGFR dimerizes and trans-phosphorylates to create a binding site recognized by Drk, which in turn recruits Sos and allows it to activate the GTPase Ras. Activated Ras recruits Raf, setting off the MAPK cascade. Activated MAPK translocates to the nucleus, where it phosphorylates the ETS DNA binding domain transcription factors Yan and PntP2. Phosphorylation of Yan abrogates its repressor activity, while phosphorylation of PntP2 increases its transcriptional activation potential. The net result is that transcriptional targets that had been repressed by Yan are activated by PntP2. B: The Drosophila Notch signaling pathway. Notch is cleaved upon binding the membrane bound ligand Delta, allowing the Notch intracellular domain (NICD) to move to the nucleus and interact with the transcription factor Su(H), converting it from a repressor to an activator. Other factors such as Mastermind (not depicted) may associate with the nuclear complex. Transcription of the Enhancer of split (E(spl)) locus is a common response to Notch activation. E(spl) proteins block expression of proneural genes such as atonal during the lateral inhibition step of R8 selection, or genes of the Achaete-Scute complex during embryonic neurogenesis. In this way, the Notch pathway restricts neural cell fate. Fig. 4. Combinatorial regulation of the prospero enhancer element, as modeled after the experiments by Xu et al. (2000). In undifferentiated cells, the absence of high levels of epidermal growth factor receptor (EGFR) signaling allows the transcriptional repressor Yan to block prospero expression. Ttk88 is also present, and presumably contributes to silencing of prospero, although not through the same enhancer element. Lozenge (Lz) is also expressed in these cells but is not sufficient to activate prospero in the presence of these two repressors. In cone cells, EGFR activation down-regulates Yan and activates PntP2, leading to induction of prospero expression. However, the presence of the Ttk88 repressor moderates transcriptional activation. In the R7 cell, activation of both the Sev and EGFR pathways results in sufficiently high levels of activation to down-regulate Ttk88, thereby removing the last check on prospero expression.

5 166 VOAS AND REBAY used cells are eliminated by apoptosis, at which point the fly has completed approximately half of pupal development (Fig. 1E, F). Signaling Machinery: A Brief Overview of the RTK and Notch Pathways The RTK signaling pathway has been highly conserved in evolution and is used reiteratively by all multicellular organisms, in many developmental contexts, to regulate a broad spectrum of events. How reiterative deployment of the same signaling cascade allows specific and context-appropriate cellular responses remains an area of intense investigation (Simon, 2000). In addition to a critical role in normal development, misregulation of the RTK pathway has been implicated as a causative event in many human cancers (van der Geer et al., 1994). This underscores the importance of understanding the intricacies of how this pathway is normally regulated as a prerequisite to understanding how compromised signaling can lead to pathogenic events. The cellular mechanics of RTK signaling have been extensively reviewed (van der Geer et al., 1994; Schlessinger, 2000) and will be summarized only briefly here (Fig. 3A). Although RTKs have extensive and critical roles throughout development, discussion of signaling events in these contexts is beyond the scope of this review. Instead, the focus will be on the general principles derived from studying RTK signaling during cell fate specification in the fly eye. Two RTKs, the epidermal growth factor receptor (EGFR) and Sevenless (Sev), contribute to Drosophila eye development. Sev is activated by a multipass transmembrane ligand called Bride of sevenless (Boss) and has the specific task of inducing the R7 photoreceptor cell fate (Rubin, 1991; Shilo, 1992; Zipursky and Rubin, 1994; Raabe, 2000). EGFR interacts with several different ligands, the most widely used of which is the TGF- like protein Spitz (Rutledge et al., 1992). Spitz activation of EGFR regulates a vast number of developmental processes, from embryogenesis to adulthood, including virtually every stage of eye development (Schweitzer and Shilo, 1997; Freeman, 2002). A second EGFR ligand called Argos is the only known naturally occurring inhibitory ligand for an RTK (Schweitzer et al., 1995; Vinos and Freeman, 2000). Developmentally, Argos is used in many, if not all, of the same times and places as Spitz (Freeman et al., 1992; Stemerdink and Jacobs, 1997; Wasserman and Freeman, 1998; Freeman, 2000; Elstob et al., 2001; Carmena et al., 2002). The Spitz/Argos tandem ensures restricted local activation of EGFR (reviewed in Freeman, 2000; Rebay, 2002). Compared with the RTK/Ras/ MAPK pathway, in which a large number of proteins are required to relay the signal from cell surface to the nucleus (Fig. 3A), the Notch pathway (Fig. 3B) is deceptively simple in its molecular composition (reviewed in Greenwald, 1998; Artavanis-Tsakonas et al., 1999). As with EGFR signaling, Notch pathway activation is required repeatedly throughout development of many different organisms (Artavanis-Tsakonas et al., 1999). The reiterative requirement for both EGFR and Notch signaling makes the developing fly eye an ideal tissue in which to investigate how repeated use of common signaling machinery elicits spatially and temporally appropriate cellular responses. EGFR SIGNALING IS REQUIRED FOR ALL EYE CELL FATES EXCEPT R8: WHAT IS THE SOURCE OF SPECIFICITY? During Drosophila eye development, EGFR stimulates proliferation in undifferentiated cells (Xu and Rubin, 1993), specifies cells to antennal vs. eye fates (Baker, 2001; Kumar and Moses, 2001a), helps initiate (but not propagate) the MF (Kumar and Moses, 2001b), helps determine the spacing of R8 founder cells (Baonza et al., 2001), specifies multiple cell fates (Freeman, 1996), and suppresses apoptosis posterior to the MF (Dominguez et al., 1998). Such extensive pleiotropy raises the important question of how distinct and context-appropriate cellular responses are achieved upon redeployment of the same signaling machinery. This discussion will focus on the roles of EGFR in specifying multiple cell fates in the fly eye as the context in which to consider this fundamental issue. Genetic analyses have revealed an essential role for Egfr in specification of all cell fates in the fly eye except R8. For example, in eye clones that are homozygous for a null allele of Egfr, specification of R8 founder cells is unaffected, whereas all subsequent steps of ommatidial assembly are blocked (Xu and Rubin, 1993; Dominguez et al., 1998; Baker and Yu, 2001; Baonza et al., 2001). Similar results are seen by expression of a dominant negative (EGFR DN ) transgene or by shifting a temperature-sensitive (ts) allele to the nonpermissive temperature (Freeman, 1996; Kumar et al., 1998). Conversely, transgenic expression of a constitutively activated form of EGFR (EGFR ACT ) results in overproduction of all cell types in the ommatidium except R8 with the exact time of expression an important determinant of specific outcome (Freeman, 1996). These findings suggest the existence of temporal regulatory mechanisms that contribute to signaling specificity (see Perspectives section for further discussion). Thus, the response of an undifferentiated precursor cell to EGFR depends upon the time point of activation, with early stimulation generating photoreceptor neurons and late stimulation producing PCs. But what are the actual molecular mechanisms underlying specificity? One model would involve deferring the role of specifier to other signaling factors by proposing that the EGFR signal is entirely permissive rather than instructive. According to this hypothesis, the primary function of EGFR during ommatidial assembly might be to prevent apoptosis, thereby allowing cells to adopt appropriate cell fates in response to other signaling cues. Supporting this idea, it is well established that EGFR signaling inhibits apoptosis in the eye (Bergmann et al., 1998; Kurada and White, 1998; Baker and Yu, 2001). For example, no cells from Egfr null clones survive to adulthood, pre-

6 SIGNAL INTEGRATION IN THE DROSOPHILA EYE 167 sumably because these clones undergo extensive programmed cell death soon after the MF has passed (Xu and Rubin, 1993; Dominguez et al., 1998). Also, temperature sensitive inactivation of Egfr in pupae increases apoptosis in the eye, apparently because the choice between 2 and 3 PC fate or apoptosis in the mid-pupal retina hinges upon EGFR activation (Yu et al., 2002). In this view, the EGFR ligand Spitz can be considered a survival signal available in limited quantity, similar to a growth factor that grants survival to a mammalian cell. To resolve the question of whether EGFR signaling provides any instructional information with respect to differentiation decisions in the eye, further dissection of the Egfr loss-offunction phenotype was performed in a context in which the need for a survival signal was obviated. Specifically, the baculovirus caspase inhibitor p35, when expressed in the developing eye, almost completely prevents cell death (Hay et al., 1994). The resulting adult flies have excess PCs, but normal numbers of photoreceptors. This outcome reflects the fact that the majority of programmed cell death in the eye occurs during pupal development when only the PCs remain to specified (Wolff and Ready, 1991). Arguing against an exclusively permissive function, induction of Egfr null clones in a p35-expression background blocks differentiation of all photoreceptors except R8 (Wolff and Ready, 1991; Baker and Yu, 2001; Baonza et al., 2001). That inhibiting cell death is not sufficient to allow differentiation in the absence of EGFR signaling suggests that EGFR activation provides specific cell fate instructions. Given that EGFR activation appears to contribute more than just a permissive survival signal in the developing eye, the ability of undifferentiated cells to respond differently to the same signal has led to the suggestion of a combinatorial code that integrates EGFR-mediated inputs with other factors to specify all the necessary cell types (Freeman, 1997; Kumar and Moses, 1997; Hayashi and Saigo, 2001; Tsuda et al., 2002). Because the larval eye imaginal disc contains a temporal gradient of ommatidial maturity, reflecting the progression of the MF, perhaps the particular factors that make up the code are expressed at different positions relative to the MF. By using this logic to combine spatial and temporal patterning information, researchers have begun to construct combinatorial models to account for the specification of each cell type in the eye (discussed below). These models begin to incorporate the complexity of signaling inputs that are required to achieve even the most simple developmental decision and provide a critical foundation for further exploration of the mechanisms underlying signaling specificity in all animals. DOWNSTREAM TRANSCRIPTIONAL INTEGRATION: THE CASE OF PROSPERO It is becoming increasingly clear that signaling specificity is mediated in large part by means of complex interactions between the transcriptional effectors of the major signaling pathways, such as Yan and PntP2 in the case of EGFR, and other tissue specific transcription factors. In ways that are not yet clear, appropriately shuffled combinations of nuclear regulators lead to expression or repression of distinct subsets of target genes, resulting in context-appropriate developmental responses. An elegant study by Carthew and colleagues (Xu et al., 2000) provides an example of this emerging paradigm (Fig. 4). The model chosen for these studies was the prospero (pros) gene, which encodes a transcription factor required for proper axonal projection by the R7 photoreceptor (Chu-Lagraff et al., 1991). The R7 precursor forms an equivalence group with the four presumptive non-neural cone cells but is distinguished from the cone cell precursors by strong Prospero expression. This difference appears to reflect the elevated level of RTK activation that results from exclusive activation of both Sev and EGFR in the R7 precursor (Chu-Lagraff et al., 1991; Doe et al., 1991; Vaessin et al., 1991; Kauffmann et al., 1996). Consistent with this interpretation, increasing the levels of Ras pathway activation in the presumptive cone cells increases Pros expression and transforms them into ectopic R7s (Fortini et al., 1992; Kauffmann et al., 1996). Conversely, loss of Sev reduces Pros expression in the presumptive R7 (Kauffmann et al., 1996). Based on these data, differential transcriptional regulation of pros expression appeared likely to provide a good model for investigating RTK signaling specificity in the eye. To address this issue, the regulatory region of the pros locus that recapitulates its endogenous expression pattern when fused upstream of a lacz reporter was identified (Xu et al., 2000). The responsiveness of this reporter to changing RTK signaling conditions was demonstrated by showing that pros-lacz expression is reduced when RTK signaling levels are lowered and expanded when pathway activation is increased. Two further experiments demonstrated that these effects are mediated by the RTK transcriptional effectors PntP2 and Yan: first, both PntP2 and Yan can bind to ETS binding sites (EBSs) within the pros enhancer in vitro; second, mutation of the EBSs in the pros enhancer obliterates reporter expression in the eye, even in the presence of increased EGFR pathway activation. Thus, pros expression is transcriptionally regulated in response to RTK activation. However, as mentioned above, all eye cells receive stimulation from the EGFR. So why then is pros expression only induced in the R7 precursor? Hints as to the answer come from studying mutants in the Runt domain transcription factor encoded by lozenge (lz; Daga et al., 1996; Flores et al., 1998). Lz is normally expressed in all undifferentiated cells in the eye and then specifically in the R1, R6, R7, cone and pigment cells. The finding that, in lz mutants, pros expression is abolished, led to the identification and characterization of Lz binding sites in the pros enhancer (Xu et al., 2000). Lz binds specifically to these sites in vitro, and mutation of the consensus sequence abolishes both in vitro binding and in vivo reporter expres-

7 168 VOAS AND REBAY sion. Furthermore, broad expression of EGFR ACT results in ectopic Pros expression wherever Lz is expressed. Thus, Lz appears to provide a spatially restricted signal that is integrated with inputs from the RTK transcriptional effectors Yan and PntP2 to induce cell-type specific expression of the target gene pros. Presumably only in the R7 precursor does Ras/ MAPK pathway activation reach sufficient levels to fully displace the Yan repressor from the EBSs and allow synergistic Lz-PntP2 activity to induce pros expression. As described below, inputs from the Notch pathway also appear to be superimposed on and/or integrated with Lz- and RTKmediated signals to effect these cell fate decisions. Full validation of this model will require detailed analysis of target gene promoter binding site occupancy in the different cell types at different time points in their development. Such studies should be greatly facilitated by the microarray-based technologies now possible in the postgenomic era (Wyrick and Young, 2002). NOTCH HAS MULTIPLE ROLES IN OMMATIDIAL ASSEMBLY Like EGFR, the Notch receptor is used throughout Drosophila eye development (Cagan and Ready, 1989). Although its function in and around the MF has been intensely studied, less is known about its roles after R8 has been established. This section will summarize the current state of understanding of how Notch functions in photoreceptor fate specification. Because Notch null eye clones that lack R8 cells never initiate the ommatidial assembly process (Baker and Yu, 1997; Li and Baker, 2001), studies of Notch function posterior to the MF have relied upon genetic tricks to avoid interfering with the R8 selection process. Two studies, one that used a temperature sensitive (ts) allele of Notch (Cagan and Ready, 1989), and another that used a combination of ts Delta alleles (Parks et al., 1995), produced remarkably similar results. In both studies, inactivation of the pathway immediately posterior to the MF resulted in massive overproduction of neurons in the larval eye disc and excess cone cells in the larval and pupal discs. These defects manifest themselves in adults as a scar that contains highly disorganized ommatidia. Of interest, in ommatidia that are approximately seven ommatidial rows posterior to the scarred region (and thus approximately 14 hr more mature), inactivation of Notch or Delta results in the loss of R7 and one or more cone cells. Similar biphasic influences on cell fate were seen for other cell types. From these experiments, it is obvious that the Notch pathway, like the EGFR pathway, is used reiteratively throughout the process of ommatidial assembly. Among Notch s many roles in ommatidial assembly, only a few have been studied in detail. This discussion will focus on Notch-mediated inputs that contribute to specification of the R7 photoreceptor and the nonneuronal cone cell fates. The experiments described above establish that loss of Notch or Delta can lead to an R7-less phenotype (Cagan and Ready, 1989; Parks et al., 1995). In these eyes, the R7 precursor becomes an R1/R6 cell type as evidenced by the presence of an extra nucleus expressing the R1/R6 marker BarH1 (Tomlinson and Struhl, 2001). Conversely, constitutive activation of the Notch pathway leads to ectopic R7 cells and the loss of BarH1-positive nuclei (Fortini et al., 1993; Cooper and Bray, 2000; Tomlinson and Struhl, 2001). Thus, it appears that, when R1/R6 and then R7 join the ommatidium, one way that they are distinguished is by activation of Notch in the presumptive R7. Notch activation leads to repression of BarH1 in R7, steering it away from the R1/R6 fate. The source of Delta ligand in this case is R1 and R6 (Tomlinson and Struhl, 2001). Perhaps their slightly earlier integration into the ommatidium explains why R1/R6 activate Notch in R7 but do not themselves receive sufficient Notch activation to become R7s. As previously mentioned, activation of the Ras/ MAPK cascade by two RTKs, EGFR and Sevenless, is also essential for assigning R7 cell fate (Freeman, 1996). However, it is currently unclear whether EGFR/Sevenless and Notch work sequentially or in parallel in this context. The role of Notch in cone cell development has been explored in detail in an elegant study by Banerjee and colleagues that initiated with the observation that loss of Notch signaling results in elimination of the cone cell marker D-Pax2 (Flores et al., 2000). The link between Notch activation and cone cell expression of D-Pax2 appears direct, because Su(H) can bind to a D-Pax2 control element (the sparkling minimal enhancer, or SME) in vitro and cone cell-specific transcription of D-Pax2 requires functional Su(H) binding sites in vivo. In addition to Notch, the EGFR pathway is also necessary for expression of D-Pax2 in cone cells (Flores et al., 2000). Temperaturesensitive inactivation of EGFR, expression of EGFR DN, expression of a constitutively active Yan repressor (Yan ACT ), or loss of pointed function all lead to the loss of D-Pax2 expression in the larval eye disc. At the molecular level, Yan and PntP2 both bind to EBSs within the SME in vitro, and mutation of these sites eliminates cone cell expression of the SME. Finally, Lz is also required for cone cell expression of D-Pax2. Together, these data define the combinatorial input required for cone cell specification. While the cell fate specification strategies just described focus on examples derived from extensive investigations of Drosophila eye development, the paradigms presented will likely provide more general models for understanding how signaling inputs from multiple pathways are integrated at the level of combinatorial control of transcription factor expression and activity. A SIGNALING DUEL: NOTCH VS. EGFR Antagonism between the Notch and RTK pathways has been observed in many different developmental contexts in Drosophila (de Celis and Bray, 1997; Price et al., 1997; zur Lage and Jarman, 1999; Culi et al., 2001) and other organisms (Berset et al., 2001; Shaye and Greenwald, 2002). While EGFR and Notch function predominantly an-

8 SIGNAL INTEGRATION IN THE DROSOPHILA EYE 169 TABLE 1. Examples of Notch RTK Antagonism in Drosophila Development Context Cell fate promoted by RTK/Ras Notch function Eye-Antenna Antennal Specify eye fate Eye All cell fates after R8 Maintain undifferentiated state Eye Secondary/tertiary pigment cell Apoptosis Trachea Fusion cells (tip of tracheal branch) Restrict fusion cell fate Muscle Myoblast founder Specify fusion-competent cells Notum bristles Sensory mother cell (SMC) Restrict SMC fate Chordotonal organ Sensory organ precursor (SOP) Restrict SOP fate Wing Vein Specify vein boundary tagonistically, it is important to note that in certain contexts, the two pathways also appear to synergize (Price et al., 1997; Carmena et al., 2002), further emphasizing the importance of the spatial and temporal context on developmental outcome. Generally speaking, RTKmediated signaling tends to promote the adoption of a specific fate, while Notch restricts this fate by preventing differentiation and/or promoting a competing cell fate (Table 1). Although the molecular details of this relationship vary depending upon context, this section will describe the general trends that have emerged. First, Notch tends to inhibit production of RTK ligands, either directly or indirectly. For example, in the developing trachea, Notch activation leads to transcriptional repression of branchless, which encodes an FGFlike ligand that activates the Breathless RTK (Ikeya and Hayashi, 1999). In other developing tissues, such as the embryonic muscle, the pupal wing, and the pupal thorax, Notch inhibits transcription of rhomboid, thereby indirectly blocking production of the active, secreted form of the EGFR ligand Spitz (de Celis and Bray, 1997; Culi et al., 2001; Carmena et al., 2002). The relative contribution that such regulation makes with respect to signaling specificity remains to be determined. A second strategy involves Notchmediated regulation of the RTK effector MAPK. Based on the tendency of Notch to prevent RTK ligand production, it is perhaps not surprising that loss of Notch or Delta function can lead to an increase in activated MAPK, while expression of Notch ACT reduces MAPK activation (de Celis and Bray, 1997; Ikeya and Hayashi, 1999). However, Notch-mediated signaling may also directly interfere with the activation state of MAPK. In Caenorhabditis elegans, the RTK LET-23 promotes anchor cell fate in the vulva by activating a canonical Ras/MAPK signaling cassette (Aroian et al., 1990; Wang and Sternberg, 2001). This function is opposed by signaling through the Notch homolog LIN-12 (Greenwald, 1998; Wang and Sternberg, 2001). LIN-12 antagonizes the LET-23 pathway by activating lip-1, which encodes a MAPK phosphatase (Berset et al., 2001) that dephosphorylates MAPK, returning it to its inactive state. This mechanism may also be at work in Drosophila, because loss of Notch function in the developing eye increases MAPK activation in PNCs (Kumar et al., 1998, 2003). Efforts are presumably under way to determine whether this effect on MAPK is mediated by the Drosophila lip-1 homolog mkph. A third molecular link underlying the antagonism between the Notch and RTK pathways may be the transcriptional regulation of yan (Rohrbaugh et al., 2002). Loss of Su(H) or Notch function in the larval eye disc results in loss of yan expression from undifferentiated progenitor cells posterior to the MF (Rohrbaugh et al., 2002). Because Yan antagonizes RTK signaling, Notch-mediated transcription of yan would seem an effective way to raise the threshold limit for RTK activation (reviewed in Rebay, 2002). Supporting such a model, an enhancer element upstream of the yan locus was identified that recapitulates the wild-type expression pattern posterior to the furrow. The Notch pathway effector Su(H) was shown to bind to multiple sites within this enhancer in vitro. However, the effects of genetic perturbation on the yan enhancer s expression pattern proved difficult to predict, suggesting more complicated regulatory mechanisms will need to be considered (Rohrbaugh et al., 2002). A fourth mechanism derives from the observation that loss of the RTK ligands Spitz, Vein, and Branchless results in a local reduction in Delta expression in various developmental contexts (de Celis et al., 1997; Ikeya and Hayashi, 1999). Conversely, activation of the RTK/Ras/MAPK pathway increases Delta production (Ikeya and Hayashi, 1999; Carmena et al., 2002). Because enhanced Delta expression results in increased Notch activation, these results suggest that the two pathways do not always act antagonistically but may at times influence each other positively. However, it is possible that the end result of this initial synergism could be the customary antagonism. For example, in the context of lateral inhibition, perhaps activation of the Ras pathway is one way in which asymmetry is initially generated in a group of otherwise equivalent cells. According to the theory of lateral inhibition, up-regulation of Delta in a cell increases the likelihood that this cell will initiate differentiation, while at the same time maintaining the undifferentiated state in neighboring cells (Muskavitch, 1994). Thus, by increasing Delta production, RTK signaling might influence the lateral inhibition mechanism, with the resulting local increase in Notch activation serving as a negative feedback loop to ensure that the effects of RTK ligand expression are restricted. As our knowledge of signal inte-

9 170 VOAS AND REBAY TABLE 2. Examples of Prepatterned Gene Expression in the Larval Third Instar Eye Disc a Cell fate Expression pattern in larval third instar eye imaginal disc determinant Undifferentiated cells Specified cells Atonal Anterior and within MF, proneural clusters Briefly in R8 Glass All cells posterior to MF Photoreceptors, pigment cells Lozenge All cells posterior to MF except R2, 3, 4, 5, 8 R1, 6, 7, cone, pigment cells Rough All non-atonal cells until ommatidial row 3 R2, 5, weakly in R3, 4 Tramtrack88 All cells posterior to MF Cone cells a MF, morphogenetic furrow. TABLE 3. A Combinatorial Code for the Specification of All Eye Cell Fates a Cell fate Signals Prepatterned determinants Factors necessary for celltype specific differentiation R8 Lack of Notch Atonal Atonal, Spalt R2, 5 EGFR Rough Rough R3, 4 EGFR??? Svp R1, 6 EGFR Lozenge Lozenge, Svp, BarH1,2 R7 EGFR, Sev, Notch Lozenge Lozenge, Prospero, Spalt Cone cells EGFR, Notch Lozenge, Ttk88 Lozenge, D-Pax2, Cut, Ttk88 1 PCu EGFR, Notch Lozenge?, Ttk88? Lozenge, BarH1, D-Pax2 2 /3 PCu EGFR, Notch Lozenge?, Ttk88? Lozenge? a A prepatterned determinant is expressed in undifferentiated cells when the indicated cell type is specified. Mutation of the determinant must also negatively affect the cell type in question. This list includes only those determinants whose expression pattern is somewhat restricted and, thus, can account for the generation of different types of cells in the eye. The factors in the right-hand column are necessary for the differentiation that is unique to the indicated cell type but not necessarily for the specification of that cell type. Svp, Seven-up; Sev, Sevenless; Ttk88, Tramtrack88; PC, pigment cell; EGFR, epidermal growth factor receptor. gration strategies expands, it is highly probable that additional regulatory links between these two highly intertwined signaling networks will be found. These mechanisms will likely include clustering and/or sequestering of the receptors into specific membrane domains, regulated endocytosis and recycling of receptor-ligand complexes, further crosstalk between cytoplasmic and nuclear effectors, as well as complex combinatorial interactions at the level of promoters of common target genes. COMBINATORIAL CODE SPECIFYING PHOTORECEPTOR FATE Although the RTK and Notch signaling pathways appear to provide some instructional information required to specify eye cell fates, the reiterative use of both pathways suggests that they are also somewhat generalized signals. Somehow cells must already know the appropriate response when stimulated by a particular combination of such semipermissive instructions. To explain this, researchers have proposed that the exact response of undifferentiated cells to a particular combination of RTK and Notch signals changes over time (Freeman, 1997; Kumar and Moses, 1997; Hayashi and Saigo, 2001; Tsuda et al., 2002). If this is true, then there must be distinct competence zones posterior to the MF, with changes in cellular competence governed by the expression of different transcription factors. For the purpose of this discussion, early expression patterns that encompass undifferentiated cells will be referred to as prepatterns. This section will summarize the available data in an attempt to reconstruct the combinatorial code of signals and prepatterns that specifies each eye cell fate (Tables 2, 3). Photoreceptor R8 As described above, R8 founder cells are selected by refinement of an Atonal prepattern. As Atonal expression is gradually lost from cells of the PNC, there is a complementary increase in expression of the homeobox protein Rough (Dokucu et al., 1996). In rough mutants, resolution of the Atonal expression pattern by lateral inhibition is delayed and incomplete, resulting in the production of 2 3 R8 cells per ommatidium (Heberlein et al., 1991). Conversely, heat-shock induced overexpression of rough prevents the initiation of Atonal expression (Dokucu et al., 1996). These data demonstrate that Rough is a negative regulator of Atonal. The failure of lateral inhibition in rough mutants also suggests that rough expression is activated by Notch. Consistent with this hypothesis, at the nonpermissive temperature, a Notch t.s. allele reduces the

10 SIGNAL INTEGRATION IN THE DROSOPHILA EYE 171 Rough expression domain. However, the Atonal and Rough patterns never overlap, so Rough retains its ability to repress atonal in the absence of activated Notch (Dokucu et al., 1996). In summary, a Notchmediated lateral inhibition mechanism is used to refine prepatterns of Atonal and Rough. The result is evenly spaced Atonal-expressing R8s in a field of Rough-positive cells. Photoreceptors R2 and R5 Subsequent production of secreted Spitz by R8 and expression of Rough in cells other than R8 serve as the determinants for photoreceptors R2 and R5. In rough mutant eyes, presumptive R2 and R5 photoreceptors express a marker normally seen in R1, R3, R4, and R6 (Heberlein et al., 1991). These rough mutants also show non cell autonomous defects in ommatidial assembly, presumably because R2 and R5 are important sources for secreted Spitz (Tomlinson et al., 1988). Not far behind the MF, strong Rough expression is maintained only in R2 and R5, and weak expression can be seen in R3 and R4 (Kimmel et al., 1990). Thus, Rough is the key to establishing R2 and R5 cell fates. Photoreceptors R3 and R4 Unlike R2 and R5, loss of rough does not affect R3 and R4 in a cell autonomous manner. A different protein, the nuclear hormone receptor Seven-up (Svp), appears necessary to maintain R3 and R4 (Mlodzik et al., 1990), but is not expressed in undifferentiated cells and, therefore, is unlikely to be a critical determinant for R3 and R4 fates. It is known that signaling through the transmembrane receptor Frizzled lies upstream of Notch pathway signaling between R3 and R4, but these events occur after the R3/R4 fates have already been specified (Fanto et al., 1998; Fanto and Mlodzik, 1999). Perhaps the only source of R3/R4 specificity is the presentation of secreted Spitz by R2 and R5. More likely, there is a requirement for other as yet unknown transcription factors. Photoreceptors R1 and R6 Lozenge (Lz) is initially expressed posterior to the MF in all undifferentiated nuclei with the exception of those cells that give rise to the five-cell precluster. After the second mitotic wave, newly differentiating cells maintain Lz expression, where it participates in establishing cell type specific transcriptional programs (Flores et al., 1998, 2000). For the presumptive R1 and R6, Lz is necessary for BarH1 expression (Daga et al., 1996). Genetic mosaic analysis shows that BarH1 and BarH2 are necessary for R1, R6, and 1 PC fates (Higashijima et al., 1987), indicating that Lz acts as a determinant for R1 and R6. Photoreceptor R7 The R7 photoreceptor fate requires inductive signals from EGFR, Sev, and Notch. As described above, activation of Notch prevents R7 from assuming the R1/R6 fate. One important function of RTKs in the presumptive R7 is to activate an E3 Ligase complex composed of Seven in absentia (Sina), Phyllopod (Phyl), and Ebi (Li et al., 1997; Tang et al., 1997; Dong et al., 1999; Boulton et al., 2000; Li et al., 2002). This E3 complex ubiquitinates Tramtrack88 (Ttk88), a transcriptional repressor, targeting it for proteolysis. Without Sev, Ttk88 is stabilized and prevents neural differentiation, causing the presumptive R7 to become a cone cell. Ttk88 is expressed in all undifferentiated cells posterior to the MF, so presumably R1 and R6 need to down-regulate Ttk88 as well (Lai et al., 1996). This conclusion is supported by the observation that, in phyl null eye clones, R1, 6, and 7 become cone cells (Chang et al., 1995; Dickson et al., 1995). Why is it that R7 needs two RTKs to accomplish a task that R1/R6 perform with only one? Perhaps the need for Notch pathway activation in R7 has the side effect of reducing RTK/Ras/MAPK activity, possibly by transcriptional activation of a MAPK phosphatase as seen in C. elegans (Berset et al., 2001) or else by upregulating yan expression (Rohrbaugh et al., 2002). An alternative explanation is that Notch signaling up-regulates Ttk88 as has been demonstrated in the developing PNS (Guo et al., 1996; Okabe et al., 2001; Pi et al., 2001). Thus, the necessity of using Notch to specify the R7 fate may result in high concentrations of Ttk88 in the presumptive R7. To counteract this, two RTKs may be necessary to achieve sufficient RAS/MAPK activation to down-regulate the antineural function of Ttk88. Cone Cells As discussed above, Lz is necessary for the specification of cone cells and for the expression of cone cell markers like D-Pax2 and Cut. Cone cells are members of the R7 equivalence group and are readily transformed into R7 cells when the Ras pathway is inappropriately activated (Fortini et al., 1992). Conversely, loss of the Sev RTK transforms R7 into a cone cell (Hafen et al., 1987). A critical difference between R7 and cone cells is that the latter maintain Ttk88 expression (Lai et al., 1996). Loss of ttk function is sufficient to transform cone cells to R7 (Lai et al., 1996). Thus, Ttk88 stability is the critical readout of varying degrees of Ras pathway activation that distinguishes cone cell fate from R7 fate. The Notch pathway is also required for cone cell specification (Cagan and Ready, 1989; Parks et al., 1995; Tsuda et al., 2002). Part of this function is related to the transcription of differentiation factors like D-Pax2 (Flores et al., 2000). As noted above, activation of Notch may also affect Ttk88 levels. This possibility seems likely for cone cells, because they express Ttk88 at higher levels than undifferentiated cells (M.V. and I.R., unpublished observation). This observation suggests that some factor involved in the specification of cone cells up-regulates Ttk88. Analysis of the ttk88 promoter region may prove highly informative in this context. PCs There have been fewer efforts to uncover the mechanisms of PC fate determination. Like all other cell types in the eye, PCs need EGFR activation (Freeman, 1996). Also, Notch