The Drosophila gene Serrate encodes an EGF-like transmembrane protein with a complex expression pattern in embryos and wing discs

. Cellular interactions play an important role during the development of multicellular organisms. Strikingly, individual elements necessary for cell–cell interactions, as signalling molecules, receptors or adhesion proteins, have been found in some cases to be conserved to a fairly high degree, suggesting that similar functions are mediated by similar proteins or domains of proteins. One group of proteins involved in cell–cell interactions is characterized by the presence of one or several repeated units, each of which shows similarity to the epidermal growth factor, EGF. Members of this family have been found in a variety of organisms, mainly in vertebrates, but also in invertebrates: in nematodes, sea urchins (Hursh et al. 1987; Suyemitsu et al. 1989) and Drosophila. They are either secreted, e.g. EGF itself, TGF-α(Lee et al. 1985; Marquard et al. 1984) or several factors of the blood coagulation system (see Furie and Furie, 1988, for a review); membrane-bound, e.g. the receptor for the low density lipoprotein, LDL (Russell et al. 1984; Südhoff et al. 1985); or membrane-associated, e.g. some proteins of the extracellular matrix (see Engel, 1989, for review). Two common characteristics of proteins containing EGF-like repeats are their participation in protein-protein interactions and their localization in the extracellular space. Several genes encoding EGF-like proteins have been described in Drosophila: Notch (N;Wharton et al. 1985a; Kidd et al. 1986), Delta (DI; Vâssin et al. 1987; Kopczynski et al. 1988), slit (sir, Rothberg et al. 1988) and crumbs (crb;TepaB et al. 1990) and in Caenorhabditis elegans (lin-12, glp-1; Greenwald, 1985; Yochem et al. 1988; Yochem and Greenwald, 1989). These genes participate in important developmental processes and, in some cases, mutations in the genes exhibit pleiotropic phenotypes, suggesting a multiplicity of interactions with other proteins.

Using probes of the neurogenic genes N and DI, several cross-hybridizing clones were isolated (Knust et al. 1987; Rothberg et al. 1988). Two of the clones could be associated with genes necessary for the proper organization of the central nervous system (CNS) (sli;Rothberg et al. 1988) or the epidermis (crb;TepaB et al. 1990; TepaB and Knust, 1990). Here, we present the molecular characterization of the gene Serrate (Ser), the DNA of which was isolated by cross-hybridization with an N probe. Two dominant mutations of this gene, which lead to defects in the morphogenesis of the wing, could be mapped molecularly to the transcription unit. Eight revertants of the dominant mutation, Ser^D, were induced by X-rays and molecular lesions were found in three of them within the Ser transcription unit. The expression pattern of the protein in wing imaginal discs is in good agreement with the observed phenotype. During embryogenesis, RNA and protein show a complex expression pattern in wild type, which is not, however, correlated with any obvious phenotype in embryos deficient for the Ser gene. Ser alleles display strong phenotypic interactions with mutations in the genes N and Dl, indicating interactions at the protein level.

Flies were grown on standard medium and crosses were performed at room temperature or at 25 °C unless otherwise stated.

The following alleles were used:

References a:Lindsley and Zimm, 1985, 1990; b: gift from C. Niisslein-Volhard; c: Vassin et al. 1985; d: Vhssin and Campos-Ortega, 1987; e:Lehmann et al. 1983.

Oregon-R was used as wild-type stock. All mutations were kept over appropriate balancer chromosomes.

For the induction of revenants of Ser^D, males of the genotypemwh red e Ser^D/TM1 were irradiated with 5000 rad and crossed immediately to females of the genotype In(3R)C Tb Sb e/TM2. Flies of the F₁ generation, which no longer exhibited the Ser^D-phenotypc (mwh red e Ser^Rx/ Balancer) were crossed to TM1/TM6 and balanced stocks were established.

cDNA clones were isolated from libraries made from embryonic RNA of different developmental stages (Poole et al. 1985; Brown and Kafatos, 1988). Genomic clones were isolated from two genomic libraries, derived from the Oregon-R wild-type strain and cloned either into the EMBL-4 phage vector (Pirrotta et al. 1983; kindly provided by Dr V. Pirrotta) or into λ-DASH (kindly provided by Dr H. Vassin). Screening of libraries, radioactive labelling of DNA probes, hybridizations, preparations of phage, plasmid and genomic DNA and Southern blot analysis were essentially as described by Maniatis et al. (1989).

Genomic fragments and cDNAs were subcloned into the Bluescript vector (Stratagene). The protocol described by Hong (1982) was used to obtain progressive deletions of subcloned fragments, which were then sequenced using a modified chain termination method (Sanger et al. 1977) with bacteriophage T7 DNA polymerase. Computer analysis was carried out on an IBM PC/AT with the DNA/Protein Sequence analysis software of J. M. Pustell/Intemational Biotechnologies, Inc., New Haven. Computer homology search was carried out with the program of Lipman and Pearson (1985) and the NBRF protein bank.

Total RNA from staged embryos, third instar larvae, pupae, male and female flies were isolated according to the method described by Auffray and Rougeon (1980). Poly(A)⁺ RNA was enriched by oligo(dT)-cellulose chromatography. The RNA was separated on agarose gels, blotted to nylon membranes and hybridized as described by Vassin et al. (1987). In situ hybridizations to whole-mount embryos were performed with a digoxigenin-labeled probe essentially as described by Tautz and Pfeifle (1989).

A BamHI fragment encoding amino acids 642–1024 (see Fig. 1D) was subcloned into the pATH 11 expression vector (Spindler et al. 1984). After induction with 5 μg ml⁻¹ indolylacrylic acid, the trpE-fusion protein was isolated according to the method of Rio et al. (1986). About 1 μg of protein in Freund’s complete adjuvant was used to immunize Balb/c female mice intraperitoneally. After one boost with the same amount of protein in Freund’s incomplete adjuvant, the serum was tested on western blots and by immunohistochemistry. Antibody staining with the polyclonal anti-Ser serum was done essentially as described by Tepaß et al. (1990). Pictures were taken with a Zeiss microscope equipped with Nomarski optics.

The cuticle of embryos up to 48 h old was prepared according to the method of Van der Meer (1977) or that of Wieschaus and Niisslein-Volhard (1986). Wings were dehydrated in 80 % isopropanol and mounted directly in Hoyer’s medium.

One of the clones isolated by cross-hybridization with an N probe (N10–00) was mapped to the chromosomal position 97F on the third chromosome (data not shown). Using this clone, we isolated 11 additional, overlapping cDNAs, which yielded a composite length of 5.0kb. Restriction site mapping, blot hybridizations and sequencing permitted the unambiguous alignment of the different cDNAs and thus confirmed that all cDNAs derived from the same transcription unit (Fig. 1). Using these cDNA clones, we isolated several overlapping phage clones from genomic libraries, representing a total length of about 35 kb of genomic DNA. By hybridization of the different cDNAs to the genomic region and restriction site mapping, we mapped the transcribed part represented by the cDNAs to an interval of about 20 kb that contains at least nine major introns. Several of the intron–exon boundaries were verified by sequencing (Fig. 2).

The nucleotide sequence representing the entire length of the composite cDNAs was determined. As none of the cDNAs contained a translational start site, we sequenced part of a genomic fragment immediately 5′ to the cDNAs (g 9229 in Figs 1 and 2), and found the putative translation start site. The combined genomic and cDNA sequences reveal one long open reading frame of 4224bp (120bp genomic and 4104bp cDNA), which is followed by a 3′ non-translated region of 903 bp (Fig. 3). The primary gene product consists of 1408 amino acids and has a calculated relative molecular mass of 150×10³. A hydropathy profile (Kyte and Doolittle, 1982) revealed three hydrophobic regions, a C-terminal region of 25 amino acids with characteristics of a membrane-spanning domain and an N-terminal stretch of 10 hydrophobic amino acids as part of a signal peptide (Perlman and Halverson, 1983). The third hydrophobic domain (amino acid 535 to 579) corresponds to the insertion within the 6th EGF-like repeat (see below). Thus, the protein encoded by this transcription unit represents a putative transmembrane protein, which is schematically depicted in Fig. 1. No attempts were made to determine the transcriptional start site. However, a heptanucleotide located 1.1 kb upstream of the translation start matches well (6 out of 7 nucleotides) to the consensus of Drosophila transcription start sites (ATCAG/TTC/T; Hultmark et al. 1986).

In addition, we found a putative TATA box (GTATA-TAAAG) in a distance of 30 bp further upstream. Whether this and/or a further upstream promoter is used, or whether there are even additional introns in this region, is still under investigation.

The most characteristic features of the extracellular domain are two cysteine-rich regions, the first of which is composed of 14 EGF-like units. EGF-like repeats, especially those found in other Drosophila proteins, are usually composed of about 38 amino acids and are characterized by six cysteine residues with a regular spacing and by additional conserved amino acids. The 4th, the 6th and the 10th repeat differ from this type of repeat in that they carry insertions of 68, 46 and 36 amino acids, respectively (Fig. 3). The clustered appearance of a limited variety of amino acids may reflect their origin from repetitive elements, which is actually supported, at least for the first insertion, by a 53% similarity at the DNA level to the opa repeat (Wharton et al. 1985b). The first insertion contains 11 clustered, positively charged amino acids, which might be involved in electrostatic interactions, while the insertion in the 6th repeat would allow hydrophobic interactions with other proteins. The 10th EGF-like repeat contains 13 threonine residues, which is similar to a threonine stretch found in the Drosophila glutactin (Olson et al. 1990).

The second cysteine-rich domain between the EGF-like repeats and the transmembrane domain consists of 11 cysteine residues concentrated within a fragment of 66 amino acids:

where C is cysteine and X_n designates any other amino acid. Part of the spacing pattern of the cysteine residues is reminiscent of the pattern found in the N-terminal propeptide of the αl-chain of the human collagen I precursor (Chu et al. 1984) (Fig. 4A). However, the functional significance of this similarity is not known.

The predicted protein exhibits a region of striking similarity to the DELTA protein, encoded by one of the neurogenic genes (Vassin et al. 1987; Kopczynski et al. 1988), which extends from the putative signal peptide cleavage site to the insertion of the 4th EGF-like repeat (Fig. 4B). This region has a 43 % similarity to DELTA. It is also remarkable that the block of EGF-like repeats in both proteins starts with an incomplete repeat.

The putative intracellular part of the protein contains three possible phosphorylation sites, which fit the consensus sequence for phoshorylation by cyclic AMPdependent serine or threonine protein kinases (Krebs and Beavo, 1979; Cohen, 1985).

The transcription unit encoding the EGF-like protein described above was mapped by in situ hybridization to the chromosomal region 97F. Two mutations were known to map in this region: Serrate (Ser-, here named Ser^D) and Beaded (Bd), which have been reported to map meiotically at 3–92.5 and 3–93.8, respectively (Lindsley and Zimm, 1985,1990). Only one allele of Bd was available, Bd^G (Beaded of Goldschmidt), which has been reported to be lethal in homozygosity (Lindsley and Zimm, 1985). Using overlapping deficiencies, we could map the lethality of Bd to the chromosomal interval 97F1-98A1/2, which is uncovered by Df(3R)D605. SerD is viable over any deficiency of this region, but the close proximity to Bd suggests its localization in the same interval.

In a mutagenesis screen, we isolated eight revertants of Ser°. One of these is a cytological visible deletion (Df(3R)Ser^RX3; 97E7–11;97F3–11) and is embryonic lethal, whereas the seven others die as second instar larvae or pupae. By Southern blots using genomic fragments representing the entire transcribed region, we detected restriction fragment length polymorphisms in the DNAs of Ser^D, Bd and three revertants as compared to wild-type or the parental DNA, respectively (Fig. 2). The polymorphisms detected in the DNAs of Ser^D and Bd after digestion with different restriction enzymes can be explained by insertions of unknown sequences into the last exon (Ser^D) or into the intron following the translational start (Bd). The revertant Ser^RX106 is associated with an internal deletion, which eliminates about 9 kb of the transcribed region. The polymorphisms detected in the DNA of Ser^RX82 are likely to be caused by a small deletion of about 0.5 kb, whereas the defects mapped in Ser^RX107 cannot be interpreted unambiguously.

Ser^Dand Bd are dominant gain-of-function mutations In the genetic crosses performed, both Ser^D and Bd behave as dominant gain-of-function mutations. These mutant phenotypes cannot be produced by simple hemizygosity. The phenotype of Ser^D is fully penetrant over the deficiency as well as over the duplication the locus. Bd is lethal in homozygosity and in heterozygosity over any deficiency of this region and its phenotype, although still recognizable, is reduced by the duplication.

The phenotype of Ser^D is characterized by notches at the tip of the wings, especially between the third and the fifth vein. In nearly all animals, hairs on the posterior margin of the wing and on the alula and some bristles at the anterior margin of the wing are missing (Fig. 5B). Penetrance and expressivity of the phenotype of the heterozygotes are slightly temperature dependent. At 18°C, the phenotypes described above are nearly fully penetrant, whereas at 28°C the phenotype is weaker, mainly with respect to the notches, and the penetrance is slightly reduced. In agreement with Belt (1971), we found that homozygous Ser^D flies are viable and fertile and that their phenotype is enhanced (Fig. 5C).

The phenotype of Bd heterozygotes shows a high degree of variability. The weak Bd phenotype is characterized by one or several small indentations in the wing, in most cases at the anterior margin (Fig. 5D) and resembles that of Ser^D. In the intermediate phenotype (Fig. 5E), the indentations are more pronounced and affect most of the first vein. In its strongest form, the first vein and most parts of the second vein are completely missing and the wing blade is reduced to about half its normal size (Fig. 5F). Bd also affects the development of the halteres, which are often reduced in size and, in the most extreme case, they are only rudimentary (not shown). In contrast to Ser^D, the hairs at the posterior margin of the wing and the alula are only slightly affected even in the most extreme Bd phenotype. The Bd phenotype is strongly temperature dependent, in that, at 18°C, nearly all animals show a strong phenotype. At 28°C, about 78% of the animals are wild type, most of the remaining ones exhibit a weak and only a few per cent express the intermediate phenotype. The enhancement of the Bd phenotype can be obtained by any condition that prolongs the time of development, e.g. reduced temperature or overcrowded cultures. This observation is in good agreement with the already reported sensitivity of Bd to the concomitant presence of Minute mutations, which are also known to delay development (Lindsley and Zimm, 1990).

The adult phenotype of transheterozygous Bd/Ser^D flies shows characteristics of both mutations (not shown). Due to the overlapping features of the Bd and Ser^D phenotypes, the phenotype of the transheterozygotes cannot be unambiguously ascribed to one or the other mutation. We could not observe an enhancement of any of the phenotypic traits described.

Whereas Ser^D is homozygous viable (see above), Bd is lethal in homozygosity as well as with a deficiency of this region. The embryonic lethality ranges between 4% (Bd/Bd) to 7% (Bd/ Df(3R)ro^82b) and the remaining homo- or hemizygous animals die as first or second instar larvae. Deletion of the 97F interval (Df(3R)D605/Df(3R)ro^82b) leads to embryonic lethality in homozygosity. The cuticle phenotype of the dead embryos did not show any major defects compared to wild-type embryos.

All revertants are wild type in the heterozygous condition and homozygous lethal. Ser^RX3, which is a deficiency of several chromosomal bands, is embryonic lethal and exhibits no obvious phenotype, whereas all others, which do not show any cytological abnormalities, are larval lethals. All revertants are lethal with a deficiency of this region and do not complement each other. With one exception (Ser^RX119), all revertants are lethal with Bd.

Bd, Ser^D, revertants of Ser^D and the deficiency of the 97F region exhibit pronounced interactions with mutations of the neurogenic genes N and DI. The adult phenotypes of N and DI mutations have been described by Welshons (1965), Vassin et al. (1985), Vassin and Campos-Ortega (1987) and Alton et al. (1988), and are shown in Fig. 6A–C. Ser^D affects the phenotypes of N and of DI differently. In N/+;Ser^D/+ animals, both phenotypes seem to be enhanced (Fig. 6D). This effect is even more pronounced in combination with the recessive mutation notchoid (nd), in which nd becomes dominant in the presence of Ser^D and the phenotype is enhanced in hemizygous males (Fig. 6E). In contrast to this, Ser^D reduces the phenotype of the deficiency Dl^FX3 and. of other DI alleles (Fig. 6F).

The effects of Bd on N and DI are, like the Bd phenotype itself, strongly dependent on temperature. Flies heterozygous for N or nd in combination with Bd show a strong enhancement of probably both phenotypes (Figs 6G, H). On the other hand, Bd is embryonic lethal with amorphic DI alleles at room temperature, without showing any obvious cuticle phenotype. How-ever, at 29°C, a small per cent of transheterozygous flies were recovered that exhibit a very strong wing phenotype (Fig. 61). In addition, the tarsal segments are fused, the ocelli are enlarged and sometimes fused, and the compound eyes are smaller and rough (not shown). These phenotypic traits have also been described for antimorphic Dl alleles (Vassin and Campos-Ortega, 1987).

The deficiency of the 97F region does not change the adult N phenotype nor does nd become dominant. In contrast, the deficiency of the 97F region is lethal with the deficiency of Dl or other amorphic Dl alleles. About 4% of the progenies of these crosses die as embryos. The revenants of Ser^D behave differently with respect to Dl, in that six of them are viable as transheterozygotes and two are embryonic lethal. Interestingly, one of the latter two, Ser^RX3, is a cytological visible deficiency, which eliminates the whole gene and the other, Ser^RX106, carries a large deletion within the gene (see Fig. 2C).

Besides minor defects in the tracheal system observed in some of the dead embryos, which were obtained by different crosses described above, we detected no other major embryonic phenotype even after staining with tissue-specific antibodies, e.g. neural- or epidermal-specific antibodies.

The temporal expression pattern of Ser was analyzed by northern blot hybridization of cDNA clones to poly(A)⁺ RNA isolated from different developmental stages. In the RNA isolated from embryonic, larval and pupal stages, two transcripts of 5.6 and 5.9kb can be detected. As 8 out of a total number of 12 cDNAs isolated from embryonic and larval cDNA libraries have the same 3′ end, we suggest that the difference in size is not caused by differential termination. In RNA of adult flies, a transcript of 2.5 kb is present in both sexes, whereas in males two additional transcripts of 4.5kb and 5.2kb RNA can be detected (data not shown).

The spatial expression pattern of Ser was analyzed both by in situ hybridization of digoxigenin-labelled probes to RNA in whole-mount embryos, and by staining whole-mount embryos and imaginal wing discs with a polyclonal mouse serum that was directed against part of the extracellular domain (Fig. 1). The specificity of the serum was tested by staining an egg collection from a stock heterozygous for a deficiency of the 97F region. A total of 25 % of these embryos did not stain with the serum (data not shown).

The temporal and spatial expression patterns of the Ser RNA and protein are the same during embryonic development and will therefore be presented together. The pattern comprises domains of expression in the fore- and hindgut, the epidermis, the tracheal system, the salivary glands and the CNS. The first expression is visible at stage 11 (all stages are according to Campos-Ortega and Hartenstein, 1985) in the anlage of the clypeolabrum (Fig. 7A) and shortly after this also in two longitudinal rows of cells in the hypopharyngeal lobe (Fig. 7B). These regions will later form the roof and floor of the pharynx, respectively. Expression in the roof continues until the end of embryogenesis, while expression in the epithelium of the floor ceases after germ-band retraction (Fig. 7D). At late stage 11, expression starts in a ring of cells surrounding the stomodeum (Fig. 7B), which finally come to lie in the anterior part of the proventriculus (Fig. 7B, D and G).

In the hindgut, two defined regions of expression can be distinguished from stage 12 onward: one region lies just caudalwards from the insertion sites of the Malphighian tubules, and the other region lies in the posterior-most 20% of the hindgut, caudalwards to a small restriction of the hindgut (Figs 7D, G and 8A).

Expression in the epidermis starts at stage 11 in a metameric pattern, which is maintained until after germ-band retraction (Fig. 7B, Cand F). In each segment, a dorsal and a ventral stripe can be distinguished (Fig. 7C and F), which are not in register with each other. In the thoracic segments, only dorsal stripes can be detected. After germ-band retraction, RNA is very abundant in the first thoracic segment. Furthermore, RNA can be detected in the gnathal segments and in the anlagen of the anal pads (Fig. 7C, D and G).

In the tracheal system, RNA and protein are expressed from stage 13 onward in the two main lateral trunks (Figs 7E, F and 8A), except a short region just between the ends of the tracheal stems and the anlagen of the posterior and anterior spiracles. The latter start expression in stages 11 and 13, respectively, and continue until the end of embryogenesis (Fig. 7C, F and G). Expression can also be detected in the secretory ducts of the salivary glands (Fig. 7D and G) and at the ventral side of the frontal sac (from stage 14 onward; Fig. 7D and G).

The protein is also clearly visible from stage 15 onward in the anterior and posterior commissures of each segment as well as in the roots of the segmental nerves (Fig. 7G). In addition, several axons within the brain hemispheres can be stained with this antibody (not shown).

In all tissues where the SERRATE protein is expressed it is associated with the membranes. In some tissues, most obvious in the hindgut, the trachea and the proventriculus at late stages of development, the SERRATE protein is localized on the apical surface of the respective cells (Fig. 8A). In some regions, e.g. in the roof of the pharynx or in the epidermis, the protein is often found in vesicle-like structures (Fig. 8B, C).

In wing imaginal discs of third instar larvae, the SERRATE protein is expressed in a row of cells located across the wing pouch and in three stripes crossing the former line perpendicularly and in some regions at the border of the disc (Fig. 7H). According to the fate map established by Bryant (1978), the stained regions correspond to the future wing margin and the anlage of the alula. It is not clear whether the expression in the three stripes reflects parts of the wing vein pattern.

At the beginning of our analysis, only two dominant mutations of this region were available, Ser^D and Bd. Both show restriction fragment length polymorphisms within the transcription unit at 97F presented in this paper (Fig. 2) and it is most likely that the mapped differences are brought about by insertions into an intron (Bd) or an exon (Ser^D). Since the parental chromosomes of these mutations are not available, we cannot verify that these defects cause the mutant phenotypes. In order to isolate more alleles of Ser, we induced new mutations by irradiating Ser^D flies, and screened the offspring for the absence of the dominant wing phenotype. Three out of eight revertants show restriction fragment length polymorphisms in the transcription unit under discussion (Fig. 2). The revertants genetically behave in the same way as the deficiency with respect to Ser^D, Bd and, in two cases, Dl, and can be regarded as hypomorphic or amorphic Ser alleles. Taken together, the genetic results, the expression pattern of the protein in the wing imaginal disc of third instar larvae, which is in agreement with the observed wing phenotype of both dominant mutations, and the molecular data confirm the identity between the described transcription unit at 97F and the genetically defined locus. However, final proof for the identity can only be obtained by transformation experiments.

According to the convention of Drosophila geneticists (Lindsley and Grell, 1968), the gene should be named Beaded, because this was the first name given to a mutation of this gene (Lindsley and Zimm, 1985, 1990). However, since many balancer chromosomes are marked with Ser^D, the name Serrate is more widely used. Thus, we would propose to make an exception to the rule and give the name Serrate to this transcription unit and the designation Ser^D and Ser^Bd for the dominant alleles.

According to the sequence analysis, the putative SERRATE protein is a transmembrane protein. Staining with specific antibodies shows localization of the protein in the membrane and thus confirms this assumption. The hydrophobic core of the putative signal peptide is preceded by an unusual long stretch of 68 amino acids, among them 18 basic residues. This situation is similar to the structure of the CRUMBS protein, where 67 amino acids, among them 19 basic residues, precede the hydrophobic core (TepaB et al. 1990). Despite its length, it could act as a functional signal peptide (Rottier et al. 1987), which is further confirmed by the presence of a potential cleavage site (Fig. 3). A second hydrophobic region of 25 amino acids, the putative membrane-spanning segment, is flanked at its carboxyl side by positively charged amino acids, which are presumed to interact with the polar groups of membrane lipids at the junction between the membrane and the cytoplasmic portion of the protein.

The 14 EGF-like repeats of the SERRATE protein exhibit a striking similarity to those found in other EGF-like proteins of Drosophila, i.e. the NOTCH, DELTA, SLIT and CRUMBS proteins (Wharton et al. 1985a, Kidd et al. 1986; Vassin et al. 1987; Kopczyinzki et al. 1988; Rothberg et al. 1988; TepaB et al. 1990), and to the blood coagulation factors among the vertebrate proteins (Rees et al. 1988, and references therein). Seven of the EGF-like repeats (repeat 3, 5–8, 11 and 14) carry three conserved AsX residues, which can be regarded as putative Ca²⁺-binding domains. The corresponding residues in the first EGF-like repeat of the blood coagulation factor IX have been shown to be essential for Ca²⁺-dependent interactions with other factors of the clotting cascade (Handford et al. 1990). Thus, the corresponding repeats of the SERRATE protein are likely to be involved in protein-protein interactions, as has been shown for other members of this family, such as EGF itself (Komoriya et al. 1984) or laminin (Graf et al. 1987).

A second cysteine-rich region, located between the EGF-like repeats and the transmembrane domain, has also been described for the NOTCH protein and the C. elegans proteins LIN-12 and GLP-1 (Greenwald, 1985; Yochem et al. 1988; Yochem and Greenwald, 1989). However, except for the high amount of cysteine residues, the SERRATE protein does not show any sequence similarity to the corresponding regions of these proteins. At its C terminus, the cysteine-rich region contains the tripeptide arginine-glycine-aspartic acid (RGD). This sequence occurs also in several proteins of the extracellular matrix, and seems to be crucial for their attachment to membrane receptors, the integrins (see Ruoslahti and Pierschbacher, 1987, for review). However, the possible function of this motif in the SERRATE protein remains to be tested biochemically.

The SERRATE protein is expressed in wing discs of third instar larvae as well as during embryogenesis. Strikingly, some of the regions stained in the wing disc correspond to parts that are affected in the adult wing of Ser^D and Ser^Bd, suggesting a direct correlation between the expression of the SERRATE protein and the development of parts of the wing. However, the cellular basis of the mutant phenotype is not yet known. One explanation would be cell death, similar to the process described to occur in nd wing disc development (Lindsley and Zimm, 1990).

The finding that expression begins relatively late in embryogenesis, that no major defects are detectable in embryos without any SERRATE protein (in a deletion, for example) and that embryos that only lack Ser (in the revertants) survive and only die as larvae indicate that the SERRATE protein may be mainly required during later stages of development. The lack of a major embryonic phenotype could also be explained by redundancy of functions, such that a deficit in the SERRATE protein can be compensated by another protein. An example to this is the fasciclin I protein, which is expressed on a subset of axons (Zinn et al. 1988). Whereas a deletion of this gene by itself does not produce any major defects in the embryo, the concomitant absence of the abl gene leads to severe defects in CNS development (Elkins et al. 1990).

The complex phenotypic interactions of Ser alleles with N and Dl indicate a close functional relationship between these genes. The similar adult phenotypes produced by Ser^D and N mutations lead us to speculate that they may represent aspects of the same phenotype. In particular, the enhancement of the N phenotype and the dominant character of nd in the presence of Ser^D could indicate that Ser^D reduces the amount of functional N product. Thus, in the presence of two copies of N⁺, the phenotype of Ser° is similar to N/+ and, in the presence of only one N⁺ copy, the mutant phenotype is enhanced. This assumption is further supported by the observations that the phenotype of the duplication of the N gene, Confluens, is reduced by Ser^D and vice versa (data not shown).

It has been previously proposed that N and Dl may interact in a dose-dependent manner and that their phenotypes are produced by an imbalance of their gene products (Alton et al. 1989; Godt, 1990). An N to Dl ratio of 1 (+/+;+/+ or +/N-, DI/+) will thus yield a wild-type or nearly wild-type phenotype, a ratio of less than 1 (e.g. in +/N; +/+) gives the N phenotype and a ratio of more than 1 (e.g. in +/+; DI/+) determines the Dl phenotype. According to the proposed model, the combination of Ser⁰ with the latter genotype would reduce the functional amount of N product and thus lower the ratio, which results in the reduction of the Dl phenotype, and this was indeed observed (Fig. 6F).

So far, only interactions during postembryonic development can be described by this model. The involvement of Ser during early neurogenesis is unlikely, as the expression of Ser starts well after the decision between neuroblasts and epidermoblasts, which is controlled by the neurogenic genes (e.g. see Knust and Campos-Ortega, 1989, for review) has been made. Afterwards, however, the expression domains of Ser partly overlap those of N (Johansen et al. 1989; Kidd et al. 1989) and Dl (Vassin et al. 1987; Kopczynski and Muskavitch, 1989), which would allow interactions between these proteins to occur. Particularly, the lethality of double heterozygotes of Dl and Ser^Bd or Dl and amorphic Ser alleles indicates considerable interactions between the two genes.

The observed phenotypical interactions may reflect interactions between the NOTCH, DELTA and SERRATE proteins at the molecular level. Recently, biochemical analysis has demonstrated that direct interactions between the NOTCH and the DELTA protein, expressed on cell membranes of Schneider cells, lead to the formation of specific aggregates of the otherwise non-adhesive cells (Fehon et al. 1990; T. Lieber, J. F. Krane, B. Hassel, J. A. Campos-Ortega and M. Young, personal communication). Sequence analysis of some point mutations in N (Hartley et al. A98T, Kelley et al. 1987) and Dl (B. Hassel and J. A. Campos-Ortega, personal communication) implies that at least some of the different EGF-like repeats have individual functions and seem to be involved in specific protein-protein interactions, which do not allow single amino acid exchange. The mutant phenotypes can be explained by disrupting specific interactions mediated by individual EGF-like repeats with different proteins at different times of development. The results obtained from our analysis indicate that the SERRATE protein is a good candidate for a molecule involved in interactions with the two other EGF-like proteins.

We thank J. A. Campos-Ortega for many helpful discussions, F. Grawe for excellent technical assistance and C. Niisslein-Volhard and the Bowling Green Stock Center for fly stocks. We thank J. A. Campos-Ortega and J. Deatrick for critical comments on the manuscript. U.T. was supported by a graduate studentship of the Graduiertenfôrderung of Nordrhein-Westfalen. This research was supported by grants and a Heisenberg fellowship to E.K. from the Deutsche Forsch-ungsgemeinschaft (Kn 250/1–3 and Kn 250/2–1).