EDL/MAE regulates EGF-mediated induction by antagonizing Ets transcription factor Pointed

In the development of a wide variety of animal tissues, a small number of pre-specified cells induce their neighbors to generate secondary cells with similar fates (Mangold and Spemann,1927). This two-step induction strategy, termed homeogenetic induction, has the potential to produce an endless cascade of induction,leading to either malignant transformations or hyperproliferative diseases(reviewed by Tang et al.,1997; Zwick et al.,2002). In normal development, the induction must thus be regulated both in cells that release the inductive signal and in the cells that receive it, so that cells that receive the inductive signal do not continue to induce other cells. Although recent studies examining induction have identified many evolutionarily conserved signaling pathways acting in the induced cells(reviewed by Edlund and Jessell,1999), much less is known about the distinct gene regulatory mechanisms that occur within the cells with inducing ability. We address two related questions: (1) `how is inducing ability regulated?'; and (2) `how do cells with inducing ability themselves respond to the inductive signals they release?'.

In Drosophila, two regions of the nervous system are known to use homeogenetic induction: (1) the specification of photoreceptor neurons during ommatidial assembly in the compound eye(Tomlinson and Ready, 1987),and (2) the formation of subepidermal stretch receptor precursors (chordotonal organ precursors: COPs) in the embryonic peripheral nervous system(Jarman and Jan, 1995). In both processes, cells already specified as photoreceptor neurons or COPs recruit surrounding cells to assume fates similar to their own. The inducing signal is a TGFα-like protein Spitz, that acts through EGF receptor to activate the Ras/MAPK signaling pathway in the induced cells (reviewed by Zipursky and Rubin, 1994). The outcome of the induction is remarkably constant: each ommatidium in the eye contains precisely eight photoreceptor neurons (R1-R8), and each hemisegment in the embryo generates exactly eight COPs (C1-C8). In order to achieve such constancy, the induction process must be exquisitely controlled.

Within the induced cells, the induction is regulated by phosphorylation of Ets transcription factors Pointed P2 (PNTP2) and YAN, which respectively acts as positive and negative regulators of neuronal differentiation(Brunner et al., 1994a; O'Neill et al., 1994; Rebay and Rubin, 1995). Both proteins possess an Ets-specific domain called the Pointed domain(Klämbt, 1993) that are phosphorylated by MAPK upon stimulation by the induction signal. This modification leads to the degradation of YAN and activation of PNTP2 function. Activation of PNTP2 leads to the production of another isoform produced by the pointed (pnt) gene, PNTP1, a constitutive transcriptional activator that is necessary for neuronal development of the induced photoreceptor cells and COPs (O'Neill et al., 1994).

Induction is also regulated by restricting the cells with inducing ability. Inducing cells are selected from groups of cells, called proneural cluster,that express a bHLH transcription factor Atonal (reviewed by Jan and Jan, 1995). In the larval eye imaginal disc, proneural cluster consists of a moving front of differentiation called the morphogenetic furrow. From a stripe of cells in the morphogenetic furrow, evenly spaced R8 photoreceptor neurons are selected through lateral inhibition. In the embryo, lateral inhibition likewise selects COPs C1-C5 from Atonal⁺ proneural clusters. In both tissues, Atonal expression is linked to inducing ability thorough expression of Rhomboid, a founding member of the Rhomboid family of intramembrane serine proteases,which is involved in the proteolytic activation of Spitz EGF (reviewed by Freeman and Gurdon, 2002). Rhomboid⁺ R8 and C1-C5 secrete Spitz EGF and act as founder cells,inducing their neighbors to assume photoreceptor neuronal fates and COP fates,respectively (Jarman et al.,1993; Jarman et al.,1994; Lage et al.,1997; Okabe and Okano,1997). R8 then induces the formation of R2 and R5, which in turn express rhomboid and serve as the secondary source of Spitz signal. Although in the eye Rhomboid paralog Roughoid plays a major role in induction,misexpression of Rhomboid causes recruitment of supernumerary photoreceptor neurons, implying that spatial regulation of Rhomboid is nevertheless essential for generating the correct ommatidium(Freeman et al., 1992; Wasserman, 2000). R8/R2/R5 and C1-C5 constitute EGF signaling centers, inducing R3/R4/R1/R6/R7 and C6-C8,respectively (Freeman et al.,1992; Tomlinson and Ready,1987; Lage et al.,1997; Okabe and Okano,1997). Thus, the transcriptional regulation of rhomboidconstitutes a key element in specifying the cells with inducing ability.

Spitz EGF acts not only as a paracrine inducer, but also has an autocrine function; inducing cells secreting Spitz themselves receive the Spitz signal and must respond to it for cell survival(Tio and Moses, 1997). Although all cells within the ommatidium require EGF receptor function(Freeman, 1996), it is not known whether or not the Ras signaling pathway downstream of the receptor is used in the same way in all cells. We found that hyperactivation of pnt function abrogates the inducing ability of the inducing cells. Hence, inducing cells must possess a mechanism to escape the inhibitory effect of pnt. We have identified a novel Ets-related factor EDL (ETS-domain lacking), containing the Pointed domain but not the ETS domain(Shilo, 1998), that may mediate this mechanism. EDL has also been identified as MAE (modulator of the activity of ETS), a protein that binds YAN and promotes its phosphorylation by MAPK (Baker et al., 2001). Although such activity suggests a role of EDL/MAE in the induced cells, we find that EDL/MAE is specifically expressed in cells that act as the induction center by producing Spitz EGF. Furthermore, we show that EDL/MAE abolishes transcriptional activation function of PNTP2 through direct binding, rather than promoting it as suggested by Baker et al.(Baker et al., 2001). This antagonistic action on PNTP2 blocks an autocrine pathway downstream of Spitz EGF, thereby allowing inducing cells to express their inductive function.

The P17 enhancer trap strain harbors an insertion of the P[ry⁺; lacZ] enhancer trap P-element PZ(Mlodzik and Hiromi, 1992). This insertion reproduces the early embryonic expression pattern of edl in the ventral neuroectoderm, but does not express lacZin the chordotonal precursor or in the eye imaginal disc. Local mobilization of the P-element was performed according to Tower et al.(Tower et al., 1993), and new insertions into the edl locus were screened by polymerase chain reaction (PCR). We identified two lines, edl^JSand edl^JV, harboring PZ element insertions at positions -38 and +4, respectively, from the putative transcription initiation site of edl. edl^JV insertion was again mobilized,and rosy^- excision lines were screened for those that are lethal over the deficiency Df(2R)P34 (55A; 55F). One such line, edl^L19, carried a deletion that includes the entire open reading frame of edl. Its 3′ breakpoint resides outside the region covered by our chromosomal walk. Excision lines were also generated from the P17 insertion. The U104 line, which extends farthest towards the edl ORF, fully complements edl^L19, and did not exhibit the edlmutant phenotype.

Somatic recombination clones were induced using the FRT technique(Xu and Rubin, 1993), over FRT43D, P[w⁺]47A. Embryos homozygous for edl^L19 were identified using the CyO,wg-lacZ balancer chromosome. FRT-mediated mitotic recombination clones of pnt^Δ88(Klämbt, 1993) were made over FRT82B, ub-GFP, P[w⁺]90D, RpS3² (also called M(3)w¹²⁴), and those for a null allele of rhomboid,rhomboid^PΔ5(also called ve^rho-PDelta5)(Freeman et al., 1992) were made over P[w⁺]70C, RpS17⁴ (also called M(3)i⁵⁵), FRT80A.

Ectopic expression of edl was achieved by the GAL4/UAS system(Brand and Perrimon, 1993),using the following drivers: elav-GAL4 C155(Lin and Goodman, 1994), sevE-GAL4 K25 (Brunner et al., 1994a), CY2 (gift of T. Schüpbach) and en-GAL4 (gift of A. Brand and N. Perrimon). UAS-pntP2 (Klaes et al.,1994), UAS-phl.gof (also called D-raf^F179)(Brand and Perrimon, 1994). Ras85D^v12.sev (also called sevE-Ras1^Val12)(Fortini et al., 1992) has been described.

In situ hybridization and antibody staining were performed as described(O'Neill and Bier, 1994; Tautz and Pfeifle, 1989; Tomlinson and Ready, 1987)with minor modifications. Confocal microscopy was done using BioRad MRC 1024 mounted on a Zeiss Axioplan2 microscope. For light microscopy, adult heads were fixed in 0.1 M cacodylate buffer or phosphate buffer (pH 7.4), 2.5%glutaraldehyde, 2% paraformaldehyde for 4 hours to overnight. Samples were postfixed in 1% OsO₄ in the same buffer for 3-4 hours, then dehydrated and embedded in Durcapan (Fluka). Sections (0.5 μm) were cut using a Reichert microtome, stained with Toluidine Blue and viewed in bright field microscopy. For scanning microscopy, flies were fixed in 0.1 M cacodylate buffer (pH 7.4), 2% OsO₄ overnight at room temperature. After dehydration, samples were dried using Peldry II (Ted Pella), coated with gold palladium in a Denton Desk II sputter coater and photographed in a JEOL 840 SEM.

DNA sequences flanking the P-element insertion point in the P17 line were recovered by plasmid rescue, and were used to initiate a chromosomal walk. Genomic restriction fragments were used to screen a 4-6 hour embryonic cDNA library (Novagen). We isolated three classes of cDNAs, represented by N1, N4 and N9, that failed to hybridize with each other. Only N9 showed an expression pattern similar to the lacZ expression pattern of the P17 enhancer trap line in the ventral neuroectoderm. Screening of a 4-8 hour embryonic cDNA library (Brown and Kafatos,1988) using N9 as a probe resulted in isolation of nine additional cDNAs, of which clone 115-3A was the longest. These cDNAs define the edl transcription unit that produces a 1.6 kb transcript. N9, whose length is 2 kb, possesses an upstream exon not found in other cDNAs. As the signal produced by a probe unique to N9 was weak, this cDNA appears to represent a minor transcript of edl. Nucleotide sequences of cDNAs N9 and 115-3A were determined by the chain termination method using Sequenase v.2(US Biochemical Corporation), and was compared with the genomic sequence obtained by the Drosophila Genome Project(FlyBase, 1999). Within the 37 kb of DNA downstream of the edl gene, there was no potential exon capable of encoding an ETS domain. The genomic rescue transgene was made by subcloning a 18 kb XbaI genomic fragment into the pCaSpeR4 vector. To generate the UAS-edl effector construct, the edl open reading frame was PCR-amplified from cDNA clone 115-3A using primer 5′-TCAAGAACTCAAACGTTGCG-3′ and the T7 primer and subcloned into the EcoRI site of the pUAST vector(Brand and Perrimon, 1993). Sequences were verified by cycle sequencing according to the instructions of the manufacturer (ABI). GMR-GAL4 carries the GAL4 coding region in the pGMR vector (Hay et al., 1994).

Transfections and CAT activity measurements were performed essentially as described (Pascal and Tjian,1991; O'Neill et al.,1994) with minor modifications. In all transfections, 100 ng of each expression plasmid were cotransfected along with 2 μg of E₆BCAT and 3 μg of pBluescriptSK(-) (Stratagene) using the calcium phosphate method except various amount of EDL/MAE expression plasmid(12.5-200 ng) were used in Fig. 3E. For each plasmid, six (Fig. 3D) or two (Fig. 3E) transfections were performed in parallel and the resulting data were averaged. Expression plasmid pPacEdl was generated by amplifying the open reading frame of edl from the N9 cDNA clone using primers 5′-ACGGAAGCCATATGCAAGTGGAATC-3′ and 5′-GAATCCTCGAGATATGTACAAC-3′, and subcloning into pPacUbx+Nde after digestion by NdeI and XhoI. Other expression plasmids are described in O'Neill et al. (O'Neill et al., 1994), and were generous gifts from I. Rebay. The pPacEdl clone used in Fig. 3D has a single base mutation causing a conservative amino acid change (K to R) at position 159. Identical results were obtained in small scale experiments using a pPacEdl clone without this mutation.

PNT-derived GST fusion proteins were made by inserting appropriately digested fragments from pPacPntP2 or pPacPntP1(O'Neill et al., 1994) into the pGEX-KG vector (Guan et al., 1991). Other GST-fusion constructs have been described (Brunner et al.,1994a; Lai and Rubin,1992; Peverali et al.,1996). GST fusion proteins were purified from bacteria using glutathione-agarose beads (Sigma). Amounts of GST fusion proteins were estimated using CDNB assay as described in the manufacturer's protocol(Pharmacia Biotech). [³⁵S]methionine-labeled EDL/MAE was synthesized using an in vitro transcription/translation system (Promega). Agarose beads (10 μl) containing 0.5 μg GST fusion proteins were incubated with 5 μl of ³⁵S-labeled EDL/MAE for 2 hours at 4°C in 0.2 ml of binding buffer [12 mM HEPES (pH 7.0), 100 mM KCl, 0.5 mM EDTA, 5 mM MgCl₂, 1 mM DTT, 0.1% NP-40, 5% glycerol, 50 mg/ml BSA,1 mM PMSF], and were washed with binding buffer. Bound proteins were released in SDS sample buffer, and analyzed by SDS-PAGE and autoradiography.

Proteins Myc-tagged at the N terminus were made by in vitro transcription/translation using TNT T7 Quick kit (Promega) and templates generated by PCR. Primers for the 5′ end contained a T7 promoter sequence for in vitro transcription. Electophoretic mobility shift experiments were carried out using DIG Gel Shift Kit (Roche) with minor modifications. Oligonucleotides containing EBS sequences with or without a mutation (EBS* or EBS) (Albagli et al., 1996)were self-annealed and 3′-end labeled with DIG-11-ddUTP using terminal transferase. Each 12 μl reaction contained 0.5 μl of protein solutions(or the sequential dilutions for myc-Edl), 0.6 ng labeled EBS probe and 0.5μl anti-Myc monoclonal antibody (9E10). Maximum amount of unlabeled probes used for a competition assay was 12.5 ng per reaction. Solutions were mixed on ice and left for 20-25 minutes at room temperature. The order of mixing did not cause any significant changes in the results. Samples were then resolved on a 0.25× TBE, 2.5% glycerol, 5.25% polyacrylamid gel preelectrophoresed at 4°C for 1 hour at 80 V. The gel was electroblotted onto GeneScreen Plus membrane (NEN), crosslinked by UV and subjected to a chemiluminescent detection using CDP-Star (Roche) as a substrate. Signals were recorded by Lumi-Imager (Boehringer-Mannheim).

The edl/mae gene (hereafter referred to as edl)was identified through enhancer trap lines that harbor P-element insertions at 55E (Fig. 1A,B; see Materials and Methods). edl encodes a 177 amino acid polypeptide that contains a region similar to the Pointed domain found in many Ets proteins(Klämbt, 1993) (reviewed by Graves and Petersen, 1998)(Fig. 1C,D). In contrast to all other proteins that contain the Pointed domain, EDL/MAE lacks the conserved DNA-binding domain, the ETS domain. Because of the potential function of EDL/MAE in Ras/MAPK signaling, we examined the expression pattern of edl in two tissues where Ets proteins function as downstream targets of Ras/MAPK signaling. In the eye imaginal disc edl mRNA was expressed in clusters of cells in two rows in the morphogenetic furrow(Fig. 1E). Expression was seen in a small number of cells in each cluster, with a spacing roughly corresponding to that of the ommatidial clusters. To examine edlexpression at the cellular level, we used an edl enhancer trap line edl^JS that expresses lacZ in the eye imaginal disc. Expression of this edl-lacZ reporter initiated in R8 cells within the morphogenetic furrow, corresponding to the stage in which R8 induces R2 and R5 (Fig. 1F). Subsequently, R2/R5, which act as the secondary source of induction, also initiated edl-lacZ expression at lower levels. During the development of the embryonic chordotonal organs, edl mRNA was present in COP C1-C5, but was undetectable in C6-C8(Fig. 1G). As in the eye imaginal disc, edl expression was transient and disappeared from the COPs before they started dividing. Thus, in both the ommatidium and the chordotonal organ, edl expression is detectable only in cells with inducing ability.

To address the role of edl in inducing cells, we identified loss-of-function mutations in edl. The edl^JV line contains a P-element insertion in the vicinity of the presumptive transcription initiation site of edl(Fig. 1A) and has viability of 5% in trans to a deletion of the 55E/F region, Df(2R)P34. As this effect on viability was reverted upon excision of the P-element and was completely suppressed by a transgene containing the entire edl coding region, edl^JV represents a reduction of function allele of edl. In addition, we generated a lethal allele, edl^L19, that removes the entire edlgene. Both edl^L19 homozygotes and edl^L19/Df(2R)P34 animals die as late embryos or early larvae.

Analysis of edl mutants revealed that in both the eye and chordotonal organ, the loss of edl reduced the efficiency of Spitz-mediated induction. In retinal sections of edl^JV/Df(2R)P34 and edl^JV/L19 animals, about 3% of ommatidia showed loss of photoreceptor cells, of the R1-R6 and R7 photoreceptor subtype(Fig. 2A,L). The R8 cell, which most strongly expresses edl expression within the ommatidium, was always present, even in ommatidia where other photoreceptor cells were missing(Fig. 2A, inset). A similar phenotype was seen in edl^L19 mutant clones, which entirely lack edl function. This phenotype was almost completely rescued by an edl⁺ transgene(Fig. 2L). The requirement of edl was more pronounced when the level of the inducing signal was compromised. Star is a dosage-sensitive component of Spitz-mediated induction in the eye, and is required for the transport of Spitz EGF to the Golgi apparatus (Heberlein and Rubin,1991; Kolodkin et al.,1994; Lee et al.,2001). In Star^-/+ animals, 30% of ommatidia show a reduction in the number of photoreceptor neurons, with the average number of R1-R7 cells reduced per ommatidium of 0.39(Fig. 2B,D,M). When edl^JV/L19 mutation was placed in the Star^-/+ background, 65% of ommatidia lacked at least one neuron, with 1.71 photoreceptor cells missing per ommatidium on average (Fig. 2C,E,M). Similarly, the edl mutation enhanced the reduction in the number of photoreceptor neurons in a hypomorphic allele of spitz (Fig. 2F,G,M). These synergistic effects of edl and Star/spitzsuggest that edl participates in the induction of R1-R7 by Spitz EGF.

The effect of the edl mutation in Spitz-mediated induction was also observed in the chordotonal organ. In each abdominal hemisegments of wild-type embryos, COP C1, C2, C3, C6 and C7 each generate a scolopidium that consists of a neuron and four support cells, and form a lateral chordotonal organ (Lch5) composed of five scolopidia. The loss of edl function caused a loss of one or two scolopidia from Lch5 in about 25% of the hemisegments (Fig. 2H,I,N).

Spitz EGF acts through the EGF receptor, resulting in the activation of the Ras/MAPK signaling leading to the phosphorylation of Ets proteins YAN and PNTP2. Baker et al. (Baker et al.,2001) have reported that EDL/MAE promotes the MAPK-mediated phosphorylation of the repressor protein YAN, thus leading to its inactivation, and is also required for the transcriptional activation by PNTP2. Although the reduction in Spitz-mediated induction observed in edl mutants appears consistent with the role of EDL/MAE in promoting MAPK signaling, two lines of evidences argues against such model of EDL/MAE action. First, the expression of edl in chordotonal organs is detectable only in COPs C1-C5, which form using the proneural activity of atonal and do not require EGF receptor function for their specification (Okabe and Okano,1997). It is also unlikely that edl acts solely by regulating YAN activity, because loss of edl function had an effect in the absence of YAN; while a null allele of yan(aop¹)has increased number of scolopidia in Lch5,introduction of edl mutation caused a clear reduction in the number of scolopidia formed (Fig. 2J,K,O,P). EDL/MAE thus has targets other than repressor protein YAN. These results indicate that the mechanism by which EDL/MAE participates in Spitz-mediated induction is different from the one proposed by Baker et al.(Baker et al., 2001).

As our genetic analysis indicated that EDL/MAE has targets other than YAN,we investigated the effect of EDL/MAE on PNTP2 activity. To test whether EDL/MAE directly binds PNTP2, we performed pull-down assays on bacterially produced GST-fusion proteins. EDL/MAE bound to the N-terminal region of the PNTP2 protein (PNTP2 1-276), which contains the Pointed domain and is specific to the PNTP2 isoform, consistent with the result obtained by Baker et al.(Baker et al., 2001)(Fig. 3A,B). Neither the region common to the PNTP1 and PNTP2 isoforms (PNTP1 223-623), nor the PNTP1-specific region (PNTP1 1-223) captured EDL/MAE (Fig. 3B). PNTP2 bound EDL/MAE much more efficiently than did YAN(Fig. 3B). MAPK, which is encoded by the rolled (rl) gene(Brunner et al., 1994b), and D-jun (Jra – FlyBase), which has been shown to act synergistically with PNT in transcription assays (Treier et al., 1995), showed only background levels of binding(Fig. 3B).

To address the functional significance of the binding of EDL/MAE to PNTP2,we tested the effect of EDL/MAE on the DNA binding activity of PNTP2. Mobility shift experiments revealed that PNTP2 can bind DNA in the presence of EDL/MAE,making a ternary complex with EDL/MAE and DNA(Fig. 3C). Such a ternary complex was not detected with PNTP1 protein, consistent with the finding that EDL/MAE does not bind PNTP1 (Fig. 3A,B). We then asked whether EDL/MAE affects transcriptional activation function of PNTP2 in a culture cell transfection assay(O'Neill et al., 1994). When PNTP2 was expressed in Drosophila Schneider cells, it activated transcription of a reporter gene harboring Ets-binding sites(Fig. 3D). Transcriptional activation by PNTP2 was enhanced by co-transfection with a plasmid encoding an activated form of Ras. However, in both cases, co-expression of EDL/MAE completely suppressed the transcriptional activation by PNTP2(Fig. 3D). This suppression was dose dependent, and was specific to PNTP2, as EDL/MAE did not have such an inhibitory effect on PNTP1 (Fig. 3D,E). We conclude that EDL/MAE can bind PNTP2 and inactivate its function as a transcription activator. This activity is opposite to the one proposed by Baker et al. (Baker et al.,2001), who reported that EDL/MAE potentiates transcriptional activation by PNTP2 in monkey Cos-7 cells.

To further study the role of EDL/MAE activity in vivo, we examined the effect of ectopic expression of edl. In the developing eye, pnt is required for the neuronal differentiation of photoreceptor cells, whereas yan is a negative regulator(O'Neill et al., 1994; Rebay and Rubin, 1995). If EDL/MAE inactivates PNTP2 activity, we would expect that misexpression of EDL/MAE in presumptive neurons would inhibit their neuronal differentiation. However, if EDL/MAE promoted MAPK-mediated phosphorylation of PNTP2 and YAN,misexpression of EDL/MAE should produce extra neurons. When edl was expressed in all neurons, the size of the compound eye was severely reduced(Fig. 4A,B). Hardly any photoreceptor neurons were present, and most of the retina was occupied by pigment cells (Fig. 4C). In the eye imaginal disc, a massive reduction of differentiating neurons was seen from the earliest stages of ommatidial assembly(Fig. 4D,E). The effect on neuronal specification was cell-type specific; we found little or no expression of markers for the recruited cells R1-R7, but R8 specification was largely unaffected (Fig. 4F,G,and data not shown). This phenotype was indistinguishable from the pnt mutant phenotype (Fig. 4H,I), consistent with the idea that ectopic EDL/MAE blocks pnt function. However, misexpression of edl in non-neuronal cone cells did not transform these cells towards a neuronal fate, as seen following the activation of Ras signaling(Fig. 4J-M). Furthermore, the suppressive effect of EDL/MAE on neuronal differentiation could be seen even in the presence of activated Ras; when edl was co-expressed with activated Ras, edl completely suppressed the ectopic neurons produced by Ras activation (Fig. 4J-Q). Thus, EDL/MAE can inhibit neuronal differentiation either downstream or parallel to Ras activation, consistent with EDL/MAE having an inhibitory effect on the transcriptional activation by PNTP2.

Additional genetic evidence supported the idea that EDL/MAE interferes with PNTP2 function in vivo. As expected from the model that EDL/MAE acted by titrating PNTP2, halving the gene dosage of pnt enhanced the rough eye phenotype caused by a weak misexpression of edl(Fig. 4R,S), and the effect of ectopic edl was suppressed by co-expression of PNTP2 (data not shown). Furthermore, ectopic expression of EDL/MAE in other tissues generated phenotypes that mimicked loss of pnt function(Fig. 4T-W)(Scholz et al., 1993; Morimoto et al., 1996). These results strongly suggest that EDL/MAE interferes with transcriptional activation by PNTP2 in vivo.

The inhibitory effect of EDL/MAE on transcriptional activation by PNTP2 suggests that EDL/MAE normally functions by suppressing PNTP2 function. As edl expression is most prominent at inducing cells in both chordotonal organ and the eye, EDL may exert its effect on PNTP2 function in inducing cells that produce Spiz. Inducing cells not only secrete Spitz EGF,but they also receive Spiz and thereby activate the downstream Ras/MAPK signaling pathway, which could result in PNT activation. Although the activation of PNTP2 is an obligatory step of neuronal specification of induced cells, the consequence of PNT activation in inducing cells has not been studied. EDL expression in inducing cells raises a possibility that these cells need to lower PNTP2 activity to ensure normal development. We thus examined the effects of hyperactivation of pnt on COP formation and photoreceptor recruitment.

When the PNTP2 isoform and an activated form of Raf (MAPKKK) was expressed in the posterior compartment of each segment in the embryo, the number of scolopidia in Lch5 was reduced from the normal number of five to three or four in 71% of hemisegments (n=60)(Fig. 5G,H). This phenotype is similar to the edl mutant phenotype, supporting the idea that the role of edl is to repress PNT activity in inducing cells. We also hyperactivated pnt in all cells posterior to the morphogenetic furrow in the eye imaginal disc. Most ommatidial clusters in the early stages of ommatidial assembly contained fewer than the normal number of neurons, an effect that is opposite to the known role of pnt in promoting neuronal identity in induced cells (Fig. 5A,B). Cells that failed to initiate neural differentiation were of specific cell types; the specification of R3/R4/R1/R6 was severely disrupted, whereas the R8 cell was still present(Fig. 5C-F). Although a general disruptive effect on differentiation cannot be completely ruled out, these results suggest that hyperactivation of PNT may reduce the ability of inducing cells to recruit additional photoreceptors and COPs.

The production of the inducing signal requires the cleavage of Spitz by the Rhomboid family of intramembrane serine proteases(Freeman et al., 1992; Lee et al., 2001; Urban et al., 2001). In both the eye and the chordotonal organ, rhomboid expression marks cells with inducing ability. To test whether hyperactivation of PNT affects inducing ability by reducing rhomboid expression, we examined rhomboid mRNA during COP specification. Among the five COPs (C1-C5)that express rhomboid in wild-type embryos, C3 is particularly important for induction because one to two rhomboid-negative COP is recruited adjacent to COP C3 (Okabe and Okano, 1997). When PNT was hyperactivated, in many hemisegments rhomboid expression was reduced in COP C3(Fig. 5I,J). It is thus likely that the reduction in the number of scolopidia of Lch5 is due to a non-autonomous effect through the loss of rhomboid expression in C3. If EDL/MAE acts by lowering PNTP2 activity, loss of edl function is expected to cause a similar effect on rhomboid expression as hyperactivation of PNT. Indeed, in edl^L19 embryos rhomboid mRNA was undetectable in COP C3 in 25% of hemisegments(Fig. 5K), which is essentially identical to the penetrance of the segments with reduced number of scolopidia(Fig. 2). Similarly, in the eye both hyperactivation of PNT and reduction of EDL/MAE activity caused a reduction in the level of rhomboid expression(Fig. 5L-N). We propose that the role of EDL/MAE is to suppress PNTP2 activity in inducing cells, thereby allowing them to express rhomboid and produce the inducing factor Spitz EGF.

We have analyzed the biological role of EDL/MAE, a protein that contains an Ets-type Pointed domain but no DNA-binding domain. Although the role of the Pointed domain as the target of the MAPK phosphorylation is well established(Wasylyk et al., 1998), the Pointed domain of EDL/MAE does not contain the consensus phosphorylation site and thus is unlikely to be regulated by the upstream signal. Emerging evidences indicate that this domain is also the site of protein-protein interaction, mediating homo- or hetero-oligomerization among Pointed domain-containing proteins (reviewed by Dittmer and Nordheim, 1998). The in vivo significance of such oligomerization, however, has not been demonstrated. Baker et al. (Baker et al.,2001) showed that the binding of EDL/MAE to YAN is required for MAPK-mediated phosphorylation of YAN, leading to inactivation of YAN function as a repressor of Ets target genes. As EDL/MAE has activities in the absence of YAN (Fig. 2N-P), EDL/MAE must have targets other than YAN. Our results show that the binding of EDL/MAE to the Pointed domain of PNTP2 (Baker et al., 2001) (this work) causes a profound effect on the activity of PNT; expression of EDL/MAE abrogated the activity of PNTP2 as a transcription activator in culture cell transfection assays(Fig. 3D). This effect is supported by misexpression studies in vivo, which showed that EDL/MAE misexpression causes phenotypes that mimic the loss of PNTP2 function(Fig. 4). Phenotypes of edl loss of function were also similar to the consequences of PNT hyperactivation (Figs 2, 5). We propose that EDL/MAE acts by antagonizing PNTP2 protein in photoreceptor neuronal differentiation and chordotonal organ development.

These results contrasts with that of Baker et al.(Baker et al., 2001) who showed that EDL/MAE promotes transcriptional activation by PNTP2 protein in monkey Cos-7 cells. Our EDL/MAE misexpression experiments in vivo support the idea that EDL/MAE antagonizes, rather than promotes, PNTP2 activity. The effects of EDL/MAE misexpression cannot be explained by the promotion of phosphorylation of YAN, because phosphorylation causes the inactivation of YAN(O'Neill et al., 1994), and loss of yan produces effects that are opposite to what we have observed by EDL/MAE misexpression (Lai et al., 1992; Tei et al., 1992; Okabe and Okano, 1997). The opposite effects of EDL/MAE on PNTP2-mediated transcription may be due to the difference in the cell lines employed in the transfection assays. It is also possible that EDL/MAE activity is used differently in diverse tissues; for example, the effect seen on the ventral denticle belts in the embryonic cuticle (Baker et al., 2001)may be due to the promotion of YAN inactivation within the ventral neuroectoderm, allowing PNTP1 to function in the specification of medial fates(Kim and Crews, 1993).

Within the developmental contexts examined in this study, edlexpression appear to be confined to cells with the ability to induce other cells using Spitz EGF (Fig. 1D,E). This suggests that EDL may have a role in regulating induction by Spitz. Secreted Spitz acts not only on the induced cells, but is also received by the inducing cells themselves. Although the molecular events leading to the activation of PNT within the induced cells is well established,whether the same regulatory cascade operates within the inducing cells had not been studied. We found that hyperactivation of PNT in inducing cells has a deleterious effect on induction; in the embryo, COP C3 loses expression of rhomboid, a factor that is essential for the production of Spitz EGF. Although inducing cells are positioned so that they receive highest levels of Spitz EGF that they produce, they may possess a mechanism to prevent hyperactivation of PNT. The phenotypes of the edl loss-of-function mutants and the effect of PNT hyperactivation are similar in both ommatidial and chordotonal organ development (Figs 2, 5). EDL/MAE is thus likely to be a part of the machinery that antagonizes PNTP2 to prevent the negative effect of PNT on induction in the inducing cells(Fig. 6).

A major challenge to our proposal that EDL/MAE acts in inducing cells by antagonizing PNT is that the loss of EDL/MAE function produces a rather mild effect on induction; most ommatidia are constructed normally in the edl null clone, and the loss of scolopidia is observed in only 25% of hemisegments in edl^- embryos. As this phenotype is weaker than that which can be achieved by an artificial activation of pntusing the GAL4/UAS-mediated overexpression (Figs 2, 5), it can be argued that the role that EDL/MAE plays in repressing PNT function might be minor. For example, inducing cells may possess multiple mechanisms to inhibit PNT activation, and deleting EDL alone may not lead to full activation. However,it is likely that our overexpression paradigm results in such a high level of PNT activation that cannot be achieved under physiological conditions. It is also possible that EDL does not completely block PNT activation in inducing cells, but just need to keep the level from reaching the state that results in interference of induction.

This raises the question when and where PNT uses the activity to curb induction. During both ommatidial assembly and the development of the chordotonal system, PNT promotes neuronal development in the induced cells. We suggest that PNT may also suppress inducing ability in such cells. This would create a negative feedback loop so that cell, once induced, does not itself acquire inducing ability. Although such mechanism would be effective in preventing uncontrolled spread of homeogenetic induction, the need for such regulatory system arises only if induced cells also have the opportunity to acquire inducing ability. This is indeed the case for R2/R5; these cells form via induction by R8, and then express rhomboid and become a secondary source of Spitz EGF. Other cells, such as R3/R4 could also potentially become inducers, because they have probably resided within the proneural cluster prior to the onset of induction and experienced Atonal expression, which promotes rhomboid expression(Lage et al., 1997) M.O.,unpublished). Repressive effect of PNT on rhomboid would thus be a mechanism to safeguard against the potential activation of rhomboidby Atonal within the proneural cluster. PNT may cause this repression via activating expression of a repressor or by acting as a repressor itself.

The inhibition of rhomboid expression is not the only way that PNT negatively regulates induction. In the eye, a rhomboid paralog roughoid plays a critical role in generating mature Spitz EGF(Urban et al., 2001). It is possible that roughoid may also be regulated by PNT to control induction. Furthermore, upon activation of Ras signaling induced cells produce negative regulators of the Ras pathway, such as Sprouty, Argos and Kekkon,generating negative feedback loops (Casci et al., 1999; Ghiglione et al., 1999; Golembo et al.,1996; Kramer et al.,1999; Schweitzer et al., 1995a). Because Argos is a secreted antagonist of Spitz EGF, its production by inducing cells could be detrimental for induction. The inhibition of PNT function by EDL/MAE may also serve to reduce Argos production in the inducing cells, allowing efficient induction.

Although induction in Drosophila eye and the chordotonal organ discussed here is `homeogenetic' in the sense that both the inducing cell and the induced cell are of the same cell type (photoreceptor neurons or COPs),they differ in genetic and molecular properties. Although neuronal specification of founder cells R8 and C1-C5 requires atonal function but not pnt, induced cells R1-R7 and C6-C8 depend on PNT activation and need atonal only indirectly. In addition, the induction itself generates a dichotomy between cells with inducing ability and those without,because induced cells acquire a different character (lack of inducing ability)from the inducing cell. Inducing cells, however, are prevented from expressing these characteristics through the repression of PNT function by EDL/MAE. Other instances of homeogenetic induction may also possess such properties, in order to generate cellular diversity, rather than equivalence. During the development of muscle progenitors in Drosophila, the size of the inductive field is defined by a group of cells similar to the proneural cluster; a small number of founder muscles are selected based on the activity of the bHLH transcription factor lethal of scute(Carmena et al., 1995) and EGF-mediated induction (Buff et al.,1998). Because edl is also expressed in a subset of muscle progenitors (data not shown), it may act in founder muscle selection in a similar way as it does in the eye and chordotonal organs.

Previous phylogenetic analysis revealed that the Ets protein family originated early during metazoan evolution and most of the functional diversity was already established prior to the separation of protostomes and deuterostomes (Degnan et al.,1993; Laudet et al.,1993; Laudet et al.,1999; Lautenberger et al.,1992; Price and Lai,1999). Although it is likely that such an ancestral Ets protein already contained a Pointed domain, the Pointed domain of EDL/MAE could not be classified as similar to any of the previously known Ets protein subclasses(Fig. 1G). This suggests that EDL/MAE-like protein may have already existed before the divergence of Ets proteins. It is tempting to speculate that EDL/MAE or EDL/MAE-like proteins may regulate inductive processes in other developmental processes in Drosophila and vertebrates.

We thank A. Brand, M. Fortini, U. Heberlein, C. Klämbt, D. Lin, T. Schüpbach and Kathy Matthews at the Bloomington Stock Center for fly strains; and E. Bier, N. Brown, C. S. Goodman, E. Hafen, K. Klämbt, H. Krämer, K. Moses and I. Rebay for clones, antibodies and libraries. We acknowledge J. Goodhouse, S. Ray, E. Suzuki, T. Takano, K. Mita and D. Welsh for assistance. We are grateful to S. Butler, G. Deshpande, S. Hayashi, M. Jindra and S. Kramer for critical reading earlier versions of the manuscript,and K. Bhat, G. Deshpande, M. Go, D. Watson, E. Wieschaus and all members of the Hiromi laboratory for helpful discussions. T.Y. received postdoctoral grants from Human Frontier Science Project, Princeton University and JST(project leader Yoshiki Hotta). This work was supported by grants from the Ministry of Education, Culture, Sports, Science, and Technology of Japan and JSPS (Research for the Future Program).