shortsighted acts in the decapentaplegic pathway in Drosophila eye development and has homology to a mouse TGF-β-responsive gene

Pattern formation in a developmental field requires cells to sense their position and develop appropriately. This is often accomplished by the use of intercellular signals which are interpreted to specify local cell fates. Three such signals have been shown to be involved in patterning the embryo and imaginal discs of Drosophila; they are encoded by the hedgehog(hh), decapentaplegic(dpp) and wingless(wg) genes. Homologues of all these exist in other species, forming the hedgehog (Riddle et al., 1993; Echelard et al., 1993; Krauss et al., 1993), transforming growth factor-β (TGF-β) (Kingsley, 1994) and Wnt (Nusse and Varmus, 1992) gene families; members of these families play crucial roles in the development of a variety of structures. The roles of these signaling molecules in Drosophila imaginal disc development vary according to the disc. In the leg imaginal disc, hh is expressed in the posterior compartment, and activates dpp along the dorsal half of the compartment boundary and wg along the ventral half (Basler and Struhl, 1994). The proximity of dpp- and wg-expressing cells leads to the expression of the homeobox genes distalless (dll) and aristaless (al) and to distal outgrowth of the leg (Diaz-Benjumea et al., 1994; Campbell et al., 1993). In the wing disc, hh is again expressed in the posterior compartment, activating dpp along the length of the compartment boundary (Basler and Struhl, 1994; Tabata and Kornberg, 1994), while wg has an early role in specifying the ventral compartment (Couso et al., 1993; Williams et al., 1993) and later controls the development of the wing margin (Couso et al., 1994). Ectopic expression of any of these three molecules can cause a long-range reorganization of disc pattern (Struhl and Basler, 1993; Basler and Struhl, 1994; Tabata and Kornberg, 1994; Diaz-Benjumea et al., 1994; Capdevila and Guerrero, 1994; Felsenfeld and Kennison, 1995; Pan and Rubin, 1995; Li et al., 1995; Jiang and Struhl, 1995; Lepage et al., 1995).

These genes also play a role in the development of the eye imaginal disc, which is unique among the imaginal discs in that cells within it develop in an asynchronous, repetitive process which does not involve stable compartment boundaries. A wave of differentiation moves across the eye disc from posterior to anterior; its front is marked by an indentation called the morphogenetic furrow (Ready et al., 1976). This furrow coincides with a stripe of expression of dpp (Masucci et al., 1990). During furrow progression, expression of dpp is dependent on hh expression and presumably secretion of the protein by the differentiated cells posterior to the furrow (Heberlein et al., 1993b; Ma et al., 1993). Anterior to the endogenous furrow, ectopic expression of hh in patches of cells can induce outwardly moving rings of dpp expression with differentiating photoreceptors within them (Heberlein et al., 1995). The same effect can be produced by loss-of-function mutations in the gene for protein kinase A (pka), the activity of which antagonizes the activity of hh (Strutt et al., 1995; Pan and Rubin, 1995). dpp expression thus correlates with the propagation of differentiation; it also seems to be required for propagation, as large clones mutant for dpp fail to differentiate (Heberlein et al., 1993b).

Cells are thought to respond to the hh/dpp signal by first arresting in the G₁ phase of the cell cycle, then undergoing apical constriction and apical-basal contraction to produce a furrow, and finally differentiating as photoreceptors in a regularly spaced array (reviewed in Thomas and Zipursky, 1994). At the molecular level, the proteins that have been shown to be expressed in response to hh (Heberlein et al., 1995) include, in addition to dpp, hairy, a negatively acting helix-loop-helix protein (Brown et al., 1991); atonal, a proneural helix-loop-helix protein (Jarman et al., 1994); string, a cell cycle regulator (Edgar and O’Farrell, 1989); scabrous, a secreted determinant of ommatidial spacing (Baker et al., 1990; Mlodzik et al., 1990); and further downstream, elav, a neuronal nuclear protein (Robinow and White, 1991). The mechanism of induction of these genes is not known, but they are induced at different positions relative to the dpp-expressing cells, suggesting the operation of either a graded signal or a cascade of signals. Once initiated, photoreceptor differentiation proceeds by the sequential recruitment of cells into clusters; at least in the case of the R7 cell, this involves local signaling between cells in contact (reviewed by Zipursky and Rubin, 1994).

Signal transduction by dpp and by members of the TGF-β family in other species has not been well characterized beyond the level of the receptors, which are heterodimeric complexes with serine-threonine kinase activity (Massague, 1992; Xie et al., 1994; Nellen et al, 1994; Penton et al., 1994; Brummel et al., 1994). A subtractive hybridization approach has been used to identify early response genes, the transcription of which is rapidly induced by TGF-β; these include c-jun, junB, c-fos (Pertovaara et al., 1989) and the leucine zipper protein TSC-22 (Shibanuma et al., 1992). TSC-22 is also induced by follicle stimulating hormone (FSH) in rat Sertoli cells (Hamil and Hall, 1994) and by other stimuli including phorbol esters and dexamethasone (Shibanuma et al., 1992).

We report here the identification of a Drosophila homolog of TSC-22, shortsighted(shs). Mutations in shs cause a delay in photoreceptor neural differentiation relative to dpp expression, suggesting that shs may be required for the transmission of the dpp signal. shs shows genetic interactions with dpp as well as with other genes involved in eye disc patterning and photoreceptor differentiation, and its overexpression can partially suppress the eye phenotype of the dpp^d-blk allele (Masucci et al., 1990). In addition, shs has a role in inducing rotation of the second optic lobe, which appears to require its expression in the brain. shs encodes a cytoplasmic protein expressed anterior to the furrow in a hh-dependent manner.

The P element collections from which shs alleles were isolated were generated in the Rubin (L. Higgins and G. M. R., unpublished data), Spradling (Karpen and Spradling, 1992) and Kiss (Török et al., 1993) laboratories; their cytogenetic locations were determined by the Berkeley Drosophila genome project. hh^rJ413 (Ma et al., 1993) S⁵⁶⁷¹ (Heberlein et al., 1993a) and wg^P (Kassis et al., 1992) are P element alleles. Other alleles used were yan¹ and yan² (Lai and Rubin, 1992), dpp^d-blk (Masucci et al., 1990) and hh¹ (Heberlein et al., 1993b). To make the GMR-shs constructs, the complete class 1 or class 2 cDNA sequences shown in Fig. 4B were subcloned into the EcoRI site of pGMR (Hay et al., 1994). These were introduced into the germline of w¹¹¹⁸ flies by P element-mediated transformation. Clones mutant for shs^rI43, shs⁶⁹⁰³ and shs^rI43Δ¹⁵ were induced using the FLP-FRT system (Xu and Rubin, 1993).

To generate the ISB.8G4 monoclonal antibody against the longer form of the shs protein, a 1.2 kb StuI/BamHI fragment of the class 2 cDNA, covering amino acids 57–457, was end-filled with the Klenow fragment of DNA polymerase I and subcloned in frame into the SmaI site of pGEX-3 (Pharmacia). The fusion protein was purified on glutathione-agarose beads and injected into mice. Polyclonal sera were also raised against amino acids 735–1039 of class 2, using a 0.9 kb NheI/SpeI fragment subcloned into the SmaI site of pGEX-3, and against amino acids 142–224 of class 1, using a 1.2 kb PvuII fragment subcloned into the SmaI site of pGEX-3. These sera gave qualitatively similar staining to the ISB.8G4 monoclonal antibody, although their signal/background ratio was low.

Flies were prepared for scanning electron microscopy as described by Kimmel et al. (1990). Adult eyes were fixed, embedded and sectioned as described by Tomlinson and Ready (1987). Eye imaginal discs were stained with antibodies as described by Xu and Rubin (1993), except that the detergent used was 0.2% Triton for anti-elav and antiscabrous staining, and 0.3% Triton + 0.3% deoxycholate for ISB.8G4 staining. The polyclonal anti-scabrous antibody was a gift from Nick Baker and was diluted 1:50. Rat anti-elav monoclonal antibody and ISB.8G4 were diluted 1:1; other polyclonal sera were diluted 1:500. For double labeling with anti-elav and X-Gal, the antibody staining was performed first and followed by a wash in PBS and incubation in X-Gal staining buffer. The dpp-lacZ reporter used was BS3.0, described in Blackman et al. (1991). Adult heads were cryosectioned and stained with X-gal as described in Mismer and Rubin (1987). Whole pupae aged from the white prepupal stage at 20°C were sectioned by the same procedure.

Standard procedures were used for DNA analysis (Sambrook et al., 1989). DNA surrounding the l(2)06903, l(2)04230, l(2)01550, l(2)07692, l(2)02687, l(2)k02903 and l(2)k103018 P elements was isolated by plasmid rescue (Mlodzik et al., 1990). To isolate DNA surrounding the rI43 P element, a genomic library in λFIXII (Stratagene) was made by partial digestion of DNA from rI43 flies with Sau3A. Phage hybridizing to a probe covering the lacZ sequence present within the P element were isolated and the end fragment from one of these, which did not contain repetitive sequences, was subcloned. All these probes were used to screen a cosmid library (Tamkun et al., 1992) and the entire genomic region was assembled by a cosmid walk. Thirty cDNAs were isolated from a λgt10 third instar eye-antennal disc cDNA library (constructed by Alan Cowman) using probes adjacent to the l(2)06903 and rI43 elements. Eight were subcloned into pBluescript(II)SK+ (Stratagene); three full-length clones were sequenced on both strands, using an Automated Laser Fluorescent DNA sequencer (Pharmacia), and were mapped onto the genomic structure by hybridization and sequencing of junctional regions. Three other cDNAs were partially sequenced. Sequencing of the P element rescue fragments was used to define their precise insertion positions. Imprecise excisions of the rI43 and l(2)06903 P elements were generated by crossing to Δ2-3 and mapped by Southern blotting. None of them were found to delete into coding exons common to both transcripts. Sequences were analysed using Staden software and homology searches were done using the blastp program.

As an approach to identify novel genes acting in eye development, we have characterized mutations caused by the insertion of enhancer trap elements with expression in the eye disc. The reporter gene in such P elements is presumed to respond to enhancers that normally direct the expression of a nearby gene, which in some cases is disrupted by the P element. One enhancer trap line isolated in this search, rI43, had a P element inserted at chromosomal position 33E7-8. This insertion is semilethal; homozygous escapers have slightly small, rough eyes (Fig. 1B). The roughness is predominantly due to missing photoreceptors in about 30% of the ommatidia (Fig. 1H; Table 1), suggesting that the gene is required in photoreceptor differentiation. Consistent with this role, the phenotype is enhanced by mutations in Star (S) (Fig. 1E,K). S is required in photoreceptors R8, 2 and 5 for their differentiation and may act as a receptor for an inductive signal (Heberlein et al., 1993a; Kolodkin et al., 1994). In addition, the P element when heterozygous has no phenotype alone but enhances the very slight roughness caused by heterozygosity for S (data not shown). Conversely, the loss of photoreceptors in rI43 eyes is suppressed by mutations in yan (Fig. 1F,L; Table 1). yan acts in the nuclei of undifferentiated cells to prevent their differentiation prior to induction, so that mutations lead to extra photoreceptor differentiation (Lai and Rubin, 1992; Rebay and Rubin, unpublished data). Other P elements that failed to complement rI43 were later isolated (see below) and were shown to be inserted in the same gene. Heterozygosity for several of these P elements suppresses the extra photoreceptor formation seen in yan mutants (Table 1 and data not shown). We have named the gene mutated by these P elements shortsighted (shs), and will refer to the rI43 insertional mutation as shs^rI43. We have made clones in the eye mutant for shs^rI43 and other shs alleles caused by P element insertions and imprecise excisions. In these clones, extra or missing photoreceptors are found at an average frequency of only 6% in a total of 130 mutant ommatidia, rather than the 30% frequency of abnormalities seen in shs^rI43 mutant eyes. Two alleles also cause a decrease in rhabdomere size (data not shown). Very rarely, at a frequency of about 2%, an abnormal ommatidium was seen in the wild-type tissue adjacent to the clone. These results could indicate that the shs protein or a protein downstream of it acts non-autonomously. However, the evidence that we have suggests that none of these alleles are nulls (see Fig. 4A and Materials and Methods), and it is possible that remaining gene function could account for the results.

The first observable phenotype in shs^rI43 mutants, in the third instar larval eye imaginal disc, is a delay of neuronal differentiation relative to the morphogenetic furrow. In wild-type discs, cells begin to express the neuronal protein elav immediately following their expression of decapentaplegic (dpp) in the morphogenetic furrow (Fig. 2A). However, in discs homozygous for shs^rI43, there is a gap the width of one row of ommatidia between cells expressing dpp and those expressing elav (Fig. 2C); this would represent a delay of approximately 2 hours. The early ommatidial clusters also appear quite disordered. The same delay in differentiation is reflected in the expression pattern of scabrous (sca). The sca protein is expressed by clusters of cells in the morphogenetic furrow, and its expression is then refined to a single cell, the presumptive R8 photoreceptor (Mlodzik et al., 1990). This expression persists for three rows of ommatidia (Fig. 2B). In shs^rI43 mutant discs, sca expression persists in a single cell for four to five rows (Fig. 2D).

The delay in differentiation observed in shs mutant discs suggested that shs could act in the differentiation pathway triggered by the signaling molecules hedgehog (hh) and dpp. Since photoreceptor differentiation in turn results in hh production, this cycle leads to the movement of a morphogenetic wave across the eye disc (Heberlein et al., 1993b; Ma et al., 1993). In support of this hypothesis, the shs^rI43 phenotype is enhanced by loss of one copy of either hh (Fig. 1C,I) or dpp (data not shown). The eye becomes smaller and rougher, due to an increase in the number of missing photoreceptors and also to a more severe disorganization of the ommatidial lattice (Fig. 1I). Double mutants of shs^rI43 and the eye-specific allele hh¹, which blocks morphogenetic furrow movement (Heberlein et al., 1993b), all die as pupae, although hh¹ itself is viable. When shs^rI43 is combined with the eye-specific mutation dpp^d-blk (Masucci et al., 1990), the frequency of survivors is decreased relative to the single shs^rI43 mutant and the eye is extremely reduced (Fig. 5D). This reduction reflects the differentiation of a very small number of ommatidia in the eye disc (Fig. 5I). In contrast, the shs^rI43 phenotype is suppressed by the loss of one copy of wingless (wg; Fig. 1D), which restores the wild-type complement of photoreceptors (Fig. 1J); wg inhibits advancement of the morphogenetic furrow from the dorsal and ventral edges of the eye disc (J. E. T. and G. M. R., unpublished data). In the wing, shs^rI43 causes the loss of parts of the wing margin, a phenotype similar to some wg alleles (data not shown; Couso and Martinez Arias, 1994).

Differentiating photoreceptors extend axons through the optic stalk into the brain, where they form retinotopic projections to the first two optic ganglia (reviewed by Kunes and Steller, 1993). The axons of the outer photoreceptors R1-6 terminate in the lamina, while the inner photoreceptors R7 and R8 project to the medulla. The lamina neuroblasts are induced to enter mitosis by the incoming photoreceptor axons (Selleck and Steller, 1991), and the lamina is thus absent in mutants in which the eye is missing or disconnected from the brain (Fischbach and Technau, 1984; Steller et al., 1987). Cell division in the medulla is independent of the photoreceptor axons, but medulla structure has been shown to depend on the function of the rough, glass and Glued genes in the eye (Meyerowitz and Kankel, 1978). During pupal development, the medulla rotates from a position perpendicular to the lamina to lie parallel to it, by a mechanism that is not understood (Hofbauer and Campos-Ortega, 1990).

Another phenotype associated with shs is an effect on the rotation of the medulla. In the heads of adult shs^rI43 homozygotes, the R7 and R8 axons often project anteriorly relative to their wild-type position (Fig. 3B). The cell bodies of the medulla neurons are also present in an abnormally anterior position (data not shown). This phenotype reflects the failure of the medulla to undergo its normal developmental rotation between 39 and 48 hours of pupal development (Fig. 3E-H). The medulla also fails to rotate in hh¹ flies (Fig. 3C); since hh is not expressed in the optic lobes (Lee et al., 1992; data not shown), this suggests that rotation of the medulla may be influenced by the progress of the morphogenetic furrow in the eye disc. Interestingly, the shs^rI43 enhancer trap is normally strongly expressed in the larval optic lobes, but its expression there is absent in hh¹ mutants (Fig. 3I,J).

Several additional P elements allelic to rI43 were isolated from the collection of lethal P insertions maintained by the Berkeley Drosophila genome project (Karpen and Spradling, 1992; Török et al., 1993). Most of these were completely lethal, and viability could be restored in each case by transposase-induced excision of the P element. DNA surrounding these elements was isolated by plasmid rescue and used to screen a cosmid library to obtain a walk covering a large genomic region (Fig. 4A). The rI43 element proved to be inserted in a 6 kb region of middle repetitive DNA not present in the cosmid sequence; nonrepetitive sequences surrounding this were used to place it on the genomic map. Probes from this region were used to screen an eye-antennal disc cDNA library, and one set of cDNAs surrounding all the P elements was isolated. These fell into two classes representing alternatively spliced forms. The l(2)06903 and l(2)k02903 elements were inserted within the 5′-untranslated region of the class 1 cDNA, while the other P elements were within introns. The l(2)07692, l(2)01550 and l(2)04230 elements were in the middle of a 50 kb intron, suggesting that their mutant phenotype may be due to splicing of the upstream exon to sequences in the P element, leading to termination of shs translation.

Several of these cDNAs were completely or partially sequenced; the sequences of the class 1 (2.8 kb) and class 2 (5.8 kb) forms are shown in Fig. 4B. They encode proteins of 1212 and 225 amino acids respectively, with a common region at the C terminus. This region includes a 99 amino acid stretch that is 58% identical to the protein TSC-22 (Fig. 4C). TSC-22 was originally identified as a transcriptional target of TGF-β in mouse osteoblasts (Shibanuma et al., 1992) and as a target of FSH in rat Sertoli cells (Hamil and Hall, 1994). The region of homology includes a stretch of amino acids predicted to form a leucine zipper; the sequence upstream of the zipper is highly conserved but is not basic, as it is in the bZIP family of transcription factors (Busch and Sassone-Corsi, 1990). While TSC-22 itself is approximately the same size as the smaller shs transcript, a homologous 5 kb transcript is also present in certain rat tissues, predominantly the brain (Hamil and Hall, 1994).

To show that the TSC-22 homologs indeed represented the shs transcripts, we used a vector that directs high levels of expression in the morphogenetic furrow and in all cells posterior to it (pGMR; Hay et al., 1994) to express either the long or short form of the protein. As shown in Fig. 5E, one copy of the construct carrying the long form is sufficient to completely rescue the eye roughness of shs^rI43 homozygotes and to restore the normal complement of photoreceptors. Since the promoter is eye-specific, the wing phenotype is not rescued. This construct also fails to completely rescue the rotation of the medulla, which is still sometimes abnormal (Fig. 3D), suggesting that correct rotation requires shs expression in the brain. Consistent with this, shs mutant clones in the eye were never seen to cause a misrotation phenotype in the underlying medulla (data not shown). Overexpression of the short form of the shs protein in the pGMR vector did not rescue the roughness of shs^rI43 eyes (data not shown).

The TSC-22 homology and the interactions with hh and dpp mutations suggested that shs might act downstream of dpp in the pathway leading to photoreceptor differentiation. In support of this, the pGMR-shs class 2 construct slightly suppresses the small eye phenotype of the dpp^d-blk mutation (Fig. 5F). Thus overexpression of shs can partially compensate for a reduction in dpp expression.

We have generated a mouse monoclonal antibody specific to the longer form of the shs protein, using amino acids 57-457 as an antigen. This antibody weakly stains wild-type eye discs, with the strongest staining seen in a stripe just anterior to the morphogenetic furrow (Fig. 6A). This strong stripe is missing from shs^rI43 mutant discs (Fig. 6C), confirming that it represents the shs protein. In addition, discs carrying the pGMR-shs class 2 construct show the expected high level of staining in all cells posterior to the morphogenetic furrow (Fig. 6B). It is difficult to assay directly the effect of dpp mutations on shs expression, as no allele is known to remove dpp function completely from the eye; however, mutations that arrest the morphogenetic furrow, such as hh¹, have been shown to lead to the loss of dpp expression (Heberlein et al., 1993b). In a hh¹ mutant disc, the strong stripe of shs expression anterior to the furrow was not seen (Fig. 6D); this expression therefore requires either hh or dpp.

At a higher magnification, it can be seen that the shs antibody stains the apical tips of the photoreceptor cells, the cell bodies surrounding the nuclei (Fig. 6E) and the axons (data not shown). However, the nuclei themselves are not stained above a background level (Fig. 6E). shs is therefore unlikely to act directly as a transcription factor. The absence of any potential membrane-spanning domain in the shs sequence makes its subcellular location likely to be cytoplasmic.

We have identified a novel gene, shortsighted(shs), which acts in the pathway leading to photoreceptor differentiation and appears to be downstream of dpp expressed in the morphogenetic furrow. Partial loss-of-function mutations in shs lead to a delay in neuronal differentiation in the eye disc. shs shows genetic interactions with hh, dpp and wg, and its expression anterior to the furrow requires hh. It is homologous to a leucine zipper protein shown to be expressed in response to TGF-β in mouse osteoblasts (Shibanuma et al., 1992).

Although the progress of differentiation in the eye disc has been shown to require the activity of hh, probably acting through dpp, the mechanism by which these signaling molecules exert their effects is unclear. As ectopic expression of hh or loss of function of either the membrane protein patched (ptc) or protein kinase A (PKA) can induce dpp expression (Heberlein et al., 1995; Li et al., 1995; Jiang and Struhl, 1995; Pan and Rubin, 1995; Strutt et al., 1995; Lepage et al., 1995), these genes appear to act upstream of dpp. However, expression of a constitutively active form of PKA does not interfere with normal hh or ptc signaling and it is therefore likely that hh and ptc counteract the inhibitory effect of a basal level of PKA, perhaps by regulating a phosphatase (Li et al., 1995; Jiang and Struhl, 1995). Apart from the saxophone and thick veins receptors, both thought to be type I subunits with serine-threonine kinase activity (Xie et al., 1994; Nellen et al, 1994; Penton et al., 1994; Brummel et al., 1994), the elements of the dpp signaling pathway have not been identified. Three pieces of evidence suggest that shs may be such an element: first, the shs^rI43 and dpp^d-blk mutant phenotypes are synergistic; second, shs expression in the eye disc is dependent on either hh or dpp; and third, shs overexpression can partially compensate for a reduction in dpp activity. The delay in differentiation seen in shs^rI43 mutants could be due to a less effective induction of differentiation by dpp due to reduced levels of shs protein. However, we cannot exclude alternative models in which shs^rI43 mutants actually have an enhanced rate of furrow movement relative to photoreceptor differentiation, or in which shs is downstream of hh but upstream of dpp. In the embryo, shs RNA is expressed on the dorsal side in a pattern similar to that of dpp (data not shown) and its expression there is under the control of dpp (L. Dobens, personal communication).

The growth of mouse osteoblasts is inhibited by TGF-β, and rat Sertoli cells respond to FSH by ceasing to grow and initiating differentiation; in both cases, TSC-22 is induced (Shibanuma et al., 1992; Hamil and Hall, 1994). Similarly, the response of cells in the eye disc to dpp and/or hh is to arrest in G₁ and to differentiate, as well as to induce shs expression. The homology of shs to TSC-22 suggests that the signaling pathways used by members of the TGF-β family may be conserved between flies and mammals.

The function of the shs protein is not obvious from its sequence, since it contains a leucine zipper but no upstream basic region. In the bZIP family of transcription factors, the leucine zipper acts as a dimerization domain and the upstream basic region as a DNA-binding domain (Busch and Sassone-Corsi, 1990). However, the region upstream of the zipper in shs is almost identical to the mouse TSC-22 sequence; this evolutionary conservation suggests it may be a distinct functional domain. TSC-22 was reported to be present in both cytoplasmic and nuclear fractions (Shibanuma et al., 1992); however, our antibody detects shs predominantly in the cytoplasm. Although this antibody is specific to the longer form of the protein, polyclonal sera reacting weakly to the common region also show cytoplasmic staining (data not shown). Within the helix-loop-helix (HLH) protein family, members with the dimerization domain but no basic region, such as mouse Id and Drosophila extramacrochaetae, act by dimerizing with proteins containing the basic region and inhibiting their DNA-binding activity (Benezra et al., 1990; Van Doren et al., 1991). If shs similarly dimerizes with bZIP transcription factors to inhibit their activity, inhibition might be achieved by sequestering them in the cytoplasm in analogy to the NF-κB–I-κB interaction (Nolan and Baltimore, 1992; Naumann et al., 1993). There are no clear candidates for such factors; the Drosophila jun and fos homologs act positively in photoreceptor differentiation (Bohmann et al., 1994), so failure to inhibit them should lead to an increased number of photoreceptors rather than the reduction seen in shs^rI43. Alternatively, shs may have a cytoplasmic function unrelated to transcription. Identification of the proteins with which it interacts should allow us to distinguish between these possibilities.

There may be functional differences between the two forms of the shs protein; overexpression of the long form, but not the short form, posterior to the furrow in the eye disc can rescue the eye phenotype caused by the rI43 insertion, which is upstream of a common exon (Fig. 4A). The long form is approximately twice as abundant as the short form in a cDNA library derived from eye-antennal discs but, without an antibody specific to the short form, we cannot tell whether their expression patterns differ. The positions of the l(2)04230, l(2)01550 and l(2)07692 P elements suggest that they would affect only the class 2 transcript, which encodes the long form of the protein, while l(2)06903 and the other elements nearby might affect only class 1. All these insertions are homozygous lethal, indicating that both forms are likely to be essential for development.

In addition to its role in the eye disc, shs is expressed in the developing optic lobes and appears to function there to induce rotation of the medulla into its final position. Input from the eye is also required for this rotation, as shown by the analysis of rough and Glued mosaics (Meyerowitz and Kankel, 1978); shs expression in the brain may be sensitive to the progress of differentiation in the eye, providing the link between the eye and the brain. The loss of shs enhancer trap expression in the hh¹ brain supports this conclusion, since hh itself is not expressed in the brain at this stage (Lee et al., 1992; data not shown). However, the quality of our antibody does not allow us to determine how closely the rI43 enhancer trap reflects expression of the endogenous shs protein in this instance. dpp is expressed in the developing optic lobes (Kaphingst and Kunes, 1994), and it is possible that it could also influence shs expression there.

Additional functions of shs include a role in wing margin development and a requirement for both male and female fertility. Its effect on female fertility may be due to its action in the dorsal follicle cells of the egg chamber (L. Dobens, personal communication). The stronger alleles are lethal early in development; although a null allele has not yet been isolated, shs probably has an essential role in embryogenesis. Its expression pattern and its regulation by dpp (L. Dobens, personal communication) suggest that it could act downstream of dpp in the pathway leading to dorsoventral patterning of the embryo (Ferguson and Anderson, 1992). Further study of the functions and mechanism of action of shs will give us insight into a variety of developmental pathways, and perhaps by analogy into processes mediated by TGF-β family members in other organisms.

We are grateful to Todd Laverty for chromosome in situ hybridization, Noah Solomon for oligonucleotide synthesis, Lila Skrinska for help with embryo injections and Kelli Lopardo for assistance with antibody production. We thank Linda Higgins, Ulrike Heberlein, Ron Blackman and Nick Baker for fly stocks and reagents. We also thank Leonard Dobens and Fotis Kafatos for communication of results prior to publication. The manuscript was improved by the critical comments of Françoise Chanut, Ulrike Heberlein, Ilaria Rebay and Tanya Wolff. J. E. T. was supported by a Jane Coffin Childs Memorial Fund postdoctoral fellowship. G. M. R. is a HHMI investigator. This work was supported in part by NIH grant GM33135 to G. M. R.

The GenBank accession numbers for shs are L42511 (class 1) and L42512 (class 2)