The Drosophila Medea gene is required downstream of dpp and encodes a functional homolog of human Smad4

A striking example of the conservation of molecular mechanisms throughout development has been the finding that closely related members of the Transforming Growth Factor-β (TGF-β? superfamily, DPP in Drosophila melanogaster, BMP-4 in Xenopus laevis, and BMP-2/4 in the Zebrafish Danio rerio, play essential roles in the specification of pattern along the embryonic dorsal-ventral axis (Dale et al., 1992; Ferguson and Anderson, 1992b; Hammerschmidt et al., 1996; Irish and Gelbart, 1987; Jones et al., 1992; Mullins et al., 1996). In Drosophila, dpp is transcribed at uniform levels in the dorsal 40% of the embryonic nuclei and is required within this expression domain for the production of all dorsal structures (Irish and Gelbart, 1987; St. Johnston and Gelbart, 1987). Genetic and embryological experiments have shown that dpp acts as an extracellular morphogen to promote dorsal cell fates in a dose-dependent fashion (Ferguson and Anderson, 1992a; Wharton et al., 1993). Similarly, in Xenopus, BMP-4 is expressed ventrally (Dale et al., 1992; Jones et al., 1992) and functions in a concentration-dependent fashion to pattern the ectodermal and mesodermal germ layers of the embryo (Dosch et al., 1997; Wilson et al., 1997). These observations raise a fundamental question regarding the conserved molecular mechanisms of DPP and BMP-4 action: how do the intracellular signal transduction pathways downstream of the DPP/BMP-4 ligands faithfully transduce small differences in DPP/BMP-4 activity to produce differential developmental outputs?

The intracellular signal transduction machinery that responds to the DPP/BMP-4 ligand is likewise highly conserved. In all species examined, TGF-β ligands signal through receptor complexes containing two types of transmembrane serine-threonine kinases, type I and type II. Biochemical experiments with mammalian TGF-β receptors have shown that upon ligand binding, the constitutively-active type II kinase phosphorylates and activates the type I kinase (Massagué, 1996; Wrana et al., 1994). In Drosophila, the products of the thick veins (tkv), saxophone (sax) and punt genes function as components of the dpp receptor complex (Brummel et al., 1994; Letsou et al., 1995; Nellen et al., 1994; Penton et al., 1994; Xie et al., 1994). The tkv and sax genes encode type I receptors, while the punt gene encodes a type II receptor.

Two additional genes involved in dpp signaling, Mothers against dpp (Mad) and Medea, were identified in a genetic screen for mutations that act as maternal-effect enhancers of a weak allele of dpp, dpp^hr4 (Raftery et al., 1995). Molecular cloning of Mad (Sekelsky et al., 1995), isolation of the related C. elegans genes sma-2, sma-3 and sma-4 (Savage et al., 1996), and identification of vertebrate Mad homologs (Baker and Harland, 1996; Graff et al., 1996; Hoodless et al., 1996) have shown that this family of genes, known as Smads, are extremely well conserved across metazoan phyla. Their defining structural features are two highly conserved ‘Mad homology’ domains, MH1 (N-terminal) and MH2 (C-terminal), separated by a linker region of variable sequence and length (Hoodless et al., 1996). Subsequent work has shown that members of the Smad family of proteins are essential cytoplasmic components of the TGF-β signal transduction pathway (Heldin et al., 1997).

Functional and structural analysis indicates that Smad family members can be subdivided into three major classes. The first class of Smads transduce signals downstream of specific TGF-β family members. For example, Smad1 and Smad5 signal downstream of the bone morphogenetic protein (BMP) family of ligands and receptors, whereas Smad2 and Smad3 transduce signals from the TGF-β and activin receptors (Eppert et al., 1996; Graff et al., 1996; Hoodless et al., 1996; Lagna et al., 1996; Macias-Silva et al., 1996; Zhang et al., 1996). The second class of Smads is defined by human SMAD4 and its presumptive orthologs (Hahn et al., 1996; Lagna et al., 1996; Savage et al., 1996). Human SMAD4 was isolated as a tumor suppressor gene (originally named DPC4 for Deleted in Pancreatic Carcinoma-4) that was mutated or absent in a majority of pancreatic carcinomas (Hahn et al., 1996; Schutte et al., 1996; Thiagalingam et al., 1996). Smad4 family members show significant structural differences from the receptor specific Smads in both the MH1 and MH2 regions. Furthermore, functional analysis indicates that they are involved in signaling downstream of all TGF-β superfamily members examined (Chen et al., 1997; Lagna et al., 1996; Zhang et al., 1997). A third class of Smads, including Smad6, Smad7 and Drosophila Dad, was recently identified as negative regulators of TGF-β signal transduction pathways (Hayashi et al., 1997; Imamura et al., 1997; Nakao et al., 1997; Tsuneizumi et al., 1997).

A paradigm for Smad function during TGF-β signal transduction has been developed from a variety of experimental systems (Heldin et al., 1997). Upon activation by a TGF-β superfamily ligand, a type I/type II receptor complex associates transiently with the appropriate receptor-specific Smad and phosphorylates carboxy-terminal serines that are necessary for Smad activation. The phosphorylated Smad then associates with Smad4 and translocates into the nucleus to regulate the expression of target genes. In Xenopus, Smad2 and Smad4 have been shown to associate together with the winged helix transcription factor, FAST-1, to form a tripartite complex that binds to an enhancer element upstream of an activin responsive gene (Chen et al., 1996, 1997). In Drosophila, the MH1 domain of MAD has been shown to bind specific DNA sequences necessary for the transcription of the target gene vestigial in the wing imaginal discs, while in tissue culture cells human SMAD4 binds sequences upstream of the TGF-β-responsive element 3TP-Lux (Kim et al., 1997; Yingling et al., 1997). Thus, the Smad family of proteins are components of a signal transduction pathway that extends directly from the receptor to the nucleus.

Here we report the isolation of additional alleles of Medea and present a detailed phenotypic characterization of its role in embryonic dorsal-ventral patterning. Molecular characterization of Medea indicates that it encodes the Drosophila homolog of Smad4. We obtained embryos completely lacking maternal and zygotic Medea activity, and we used the technique of mRNA injection to demonstrate functional conservation between human SMAD4 and MEDEA.

zenf16 p^p, ry P720, ry P1575^ttk, P{hsFLP}¹, FRT^3R-82B, and FRT^3R-82BP{ovo^D1}^3R fly stocks were obtained from the Bloomington Drosophila Stock Center (FlyBase, 1997). zen^f16 is also known as zen¹. P720 and P1575^ttk refer to the P-element insertions P{ry^+t7.2=PZ}l(3)00720 and P{ry^+t7.2=PZ}ttk⁰²⁶⁶⁷ at cytological positions 100B5–7 and 100D1-2, respectively. P{hsFLP}¹, FRT^3R-82B, and P{ovo^D1}^3R are as described (Chou and Perrimon, 1996). Df(3R)E40 (Locke et al., 1988) was obtained from K. Tartof. Med⁴ (Raftery et al., 1995) and Df(3R)KpnA were obtained from L. Raftery. Mad¹² and Df (2L)C28 stocks were obtained from W. Gelbart (Sekelsky et al., 1995). Additional information about Drosophila stocks is available in FlyBase (1997).

The zen^f16p^p chromosome was subjected to unselected recombination with wild-type (Oregon R) chromosomes to eliminate extraneous recessive lethal loci. Hatching of homozygous zen^f16p^p embryos was reduced from 98% (n=346) to 23% by the introduction of one copy of a weak dpp mutant allele, dpp^hr56 (n=195). Thus, a very small decrease in dpp signaling greatly enhanced the penetrance of the lethal phenotype of zen^f16 (Wakimoto et al., 1984). We reasoned that a similarly small decrease in the efficiency of dpp signaling, as might be expected from a 50% decrease in the activity of a component in the dpp signaling pathway, might also be sufficient to enhance the penetrance of the zen^f16 mutant phenotype. Furthermore, we reasoned that, like the dpp receptors, other essential dpp signaling components were likely to be expressed maternally. Accordingly, we screened for maternal-effect mutations that dominantly enhanced the zen^f16 phenotype.

Homozygous zen^f16p^p males were starved for 8 hours, fed 25 mM ethylmethane sulfonate (EMS) in 1% sucrose for 24 hours, and mated for 6 days to zen^f16p^p females. Approximately 5000 mutagenized F¹zen^f16p^p females, each bearing a mutagenized haploid genome, were collected and mated individually to zen^f16p^p males in egg laying blocks. The viability of embryos from approximately 4000 productive matings was assessed by visual inspection with a dissecting microscope. F¹ females that produced 75–100% inviable embryos with cuticular phenotypes characteristic of partial ventralization were considered as candidate dominant maternal enhancer mutants. Secondary matings of F¹ female mutant candidates to zen^f16p^p/TM3 Sb e males allowed the recovery of the enhancer mutations from some of the viable zen^f16p^p/TM3 Sb e progeny (Fig. 1).

Eight enhancer mutations on chromosome 3 were recovered in trans to the balancer chromosome TM3 Sb e through matings of individual candidate females to zen^f16p^pe/TM3 Sb e males and assessment of the progeny of each female for dominant enhancement of the zen^f16 mutation. The Sb and e markers were used to follow the segregation of the zen^f16p^p chromosome bearing the enhancer mutation. One enhancer mutation was recovered in trans to the chromosome 2 balancer CyO from iterated matings of individual females to Sp/CyO; zen^f16p^p males.

The identities of the mutant genes were assessed both by multi-factorial recombinant mapping with respect to visible marker mutations and by complementation analysis. The enhancer mutation on chromosome 2 was localized to the chromosomal interval between the aristaless and dumpy loci. It was identified as an allele of Mad by non-complementation with the phenotypically null mutant Mad¹² and Df(2L)C28; it has been named Mad^Ez. The eight mutations on chromosome 3, numbered 11 through 18, were shown by complementation analysis to be allelic to one another. A deficiency of Medea, Df(3R)E40, failed to complement mutants 11, 15 and 16, and two mutants, 13 (L Raftery, personal communication) and 15, failed to complement Med⁴ (Raftery et al., 1995), identifying the eight enhancer mutations as alleles of Medea; they have been named Med¹¹ through Med¹⁸. The deficiency chromosome Df(3R)KpnA also failed to complement Med¹¹ and Med¹⁶, although Southern blot analysis has indicated that it contains Medea DNA sequences (not shown).

Embryonic cuticle preparations were performed as described by Wieschaus and Nüsslein-Volhard (1986). β-galactosidase expression in the amnioserosa from the P{Kr-lacZ} reporter construct was assayed as described by Ferguson and Anderson (1992a).

Wings dissected from adult flies were mounted in Gary’s magic mountant (Ashburner, 1989). Adult flies were fixed for scanning electron microscopy first in 70% ethanol and then in PBS with 2.5% glutaraldehyde. Fixed flies were returned to 70% ethanol, dissected as necessary, and dehydrated first in ethanol and then in hexamethyldisilazane. Samples were mounted onto stubs with either tape or a graphite bed, coated with gold in a sputter coater (Technics), and viewed with a JEOL JSM-840A scanning electron microscope at a working distance of 39 mm and an accelerating voltage of 15 kV.

A ru h FRT^3R-82Bsr e Medea¹³ chromosome was generated and placed in trans to FRT^3R-82BP{ovo^D1}^3R in virgin females that also carried P{hsFLP}¹ on the X chromosome. Expression of the FLP recombinase was induced by heating pupae or adults to 37°C for 2 to 3 hours. Females were mated after heat shock to Medea¹⁴, P{Kr-lacZ}/TM3 P{Ubx-lacZ} males to produce embryos for phenotypic analysis. Expression in the amnioserosa of β-galactosidase from the P{Kr-lacZ} construct was assayed 20 hours after egg laying at 18°C. Embryos that received a wild-type copy of Medea from the TM3 balancer were identified by Ubx-lacZ expression.

For mRNA injection experiments, 500-1000 females generated as above were mated with Med¹⁴P{Kr-lacZ}/TM3 P{Ubx-lacZ} males at 18°C in an egg-laying cup on apple juice plates. Eggs were collected for injections every 2-3 days on fresh apple juice plates smeared with yeast paste. Flies were kept in the dark during collections and kept in constant light on fresh apple juice plates without yeast between collections. Each 2 hour collection produced approximately 25–50 embryos for injection; a second successive plate was sometimes also productive. In practice, a mating cup became productive after a week of pre-mating and remained so for only approximately 10 days.

The Medea gene was localized between the two P[ry⁺] insertions P720 at 100B5–7 and P1575 at 100D1-2 by three-factor meiotic recombination mapping. Females were generated with st ry P720 and st ry P1575 chromosomes in trans to st ry ca Med¹⁵ and st ry Med⁺ recombinant chromosomes with breakpoints between the Medea gene and each P-element insertion were collected in trans to st ry Med¹¹ or st ry Med¹³. The genetic distance between Medea and P720 was 0.3 cM (21 st ry Med⁺ recombinants among 7021 st and st ry progeny) and that between Medea and P1575 was 0.06 cM (7 st ry Med⁺ recombinants among 11890 st and st ry progeny). These recombinant chromosomes, as well as an additional seven P1575 recombinant chromosomes, were used in RFLP analysis.

P1 phage clones with inserts from the 100D region (Berkeley Drosophila Genome Project) were obtained from T. Kaufman (clones DS00235, DS03999, and DS07056 are referred to here as 235, 3999, and 7056, respectively). Overlap among P1 genomic inserts was determined by restriction enzyme analysis and by Southern blotting with ³²P-labeled terminal subclones. Gaps between P1 inserts were partially filled with cosmids (including clone 91C5) obtained from I. Siden Kiamos (Foundation for Research and Technology - Hellas) or recovered from Drosophila genomic DNA libraries constructed in a cosmid vector (S. King library, provided by C.-I. Wu; clones K17 and K31) or in λEMBL3 (Tamkun et al., 1992), provided by A. Mahowald; clone λ7. The genomic inserts of these clones defined two contiguous regions (Fig. 5A). Subcloned restriction fragments from the genomic DNA clones were used as probes for RFLP analysis to align the physical and genetic maps and to localize Medea. Other probes included a tramtrack (ttk) cDNA (Xiong and Montell, 1993) and a genomic fragment from abnormal wing discs (awd) (Dearolf et al., 1988).

RFLPs (restriction fragment length polymorphisms) were identified by Southern blot hybridization of these probes to DNA prepared from Med¹⁵ and each of the two P-element insertion stocks, after digestion with a panel of eleven restriction enzymes having 4-base recognition sites (AluI, DdeI, HaeIII, HhaI, HinfI, HpaII or MspI, NlaIII, RsaI, Sau3AI or DpnII, ScrFI, and Taq^αI). Probes that revealed a RFLP between the P-element insertion chromosome and the Med¹⁵ chromosome were used to probe Southern blots with restriction digested DNA from the 14 (P1575) or 21 (P720) recombinant chromosomes. The Med¹⁵ lesion was shown by this analysis to lie in a 40 kb region defined by two such probes: a Taq^αI RFLP identified by the probe NN.1 and an NlaIII RFLP identified by the probe GN.1.

cDNAs from three distinct transcription units within the delimited region were isolated by screening embryonic and imaginal disc cDNA libraries (Brown and Kafatos, 1988) with ³²P-labeled cosmid K17 or subcloned fragments from P1 3999. The Medea cDNA clone O513 was obtained from a 0- to 4-hour embryonic cDNA library; four additional Medea cDNA clones (H12, H25, H36, and H44) were recovered from a 4- to 8-hour embryonic cDNA library. The two other transcription units were represented by the cDNA clone H31, isolated from the 4- to 8-hour embryonic library, and the imaginal disc cDNA clone I7114.

The sequence of the Medea cDNA clone O513 was determined in both directions from subcloned restriction fragments spanning the length of the cDNA and has been placed into GenBank (accession number AF039232). The cDNA sequence ends with fourteen consecutive adenosines, twelve of which represent the adenosines in the primer that was used to synthesize the cDNA (Brown and Kafatos, 1988).

Genomic Medea DNA sequence was determined from DNA isolated from the Medea homozygous mutants (Med¹⁵ and Med¹⁷) and from Medea/TM3 heterozygotes. After digestion with BssHII and NcoI, the 3 to 5 kb DNA fraction was excised from a 1% agarose gel, extracted with GlasPac/GS (National Scientific Supply Co.), and cloned into MluI- and NcoI-digested pGEM-5Zf(−) (Promega). Medea clones were identified by hybridization to a ³²P-labeled 550 bp BstXI fragment from the Medea cDNA, and wild-type clones derived from the TM3 balancer chromosome were excluded by way of a HinfI polymorphism. The Medea mutant DNA clones were sequenced in both directions with primers spaced 400 to 500 bp apart. Wild-type sequence of the Medea genomic region was determined as a consensus from the mutant sequences and has been placed into GenBank (accession number AF039233). This sequence begins at position 315 of the cDNA and ends 537 bp downstream of the cDNA 3′ terminus. Nine exons were identified by comparison of the cDNA and genomic sequences. The sizes and placement of the eight introns are: intron 1 after cDNA base 682, 236 bp; intron 2 after base 951, 86 bp; intron 3 after base 1306, 201 bp; intron 4 after base 1528, 156 bp; intron 5 after base 2143, 60 bp; intron 6 after base 2426, 61 bp; intron 7 after base 2595, 56 bp; and intron 8 after base 2722, 59 bp.

The (A)₁₄ at the 3′ terminus of the Medea cDNA is encoded within the genomic DNA and does not represent 3′ polyadenylation. A single AATAAA poly(A) specification consensus sequence identified within the genomic sequence, 142 bp downstream of the cDNA terminus, may encode the true poly(A) signal sequence. However, preliminary sequencing has indicated that the Medea cDNA clones H36 and H44 terminate in polyadenosines after a single G insertion at position 3180 and after position 3188, respectively, perhaps reflecting the partial use of the cryptic poly(A) site AATATA at position 3158.

The lesions in six Medea mutants were identified by double stranded sequencing of the complete coding region of each mutant chromosome. The Med¹¹ mutation is a 405 bp deletion, starting in intron 4 and continuing into exon 5 through cDNA position 1888, into which the 5 base duplication GTCTT has been inserted; the presence of this deletion has been confirmed by Southern blot analysis. Med¹²,Med¹⁴ and Med¹⁶ have the single base substitutions G2481A, G2145A and C1804, respectively. Med¹⁵ has the missense mutations C2732T and C2837A. Med¹⁷ has the missense mutation C2839A and the silent substitution G2874A that ablates a PstI restriction site (confirmed by Southern blot analysis). Sequencing was performed on an ABI Prism 377 or 377XL DNA sequencer at the University of Chicago Cancer Research Center DNA Sequencing Facility. Sequence was analyzed with Sequencher (Gene Codes Corp.) and Lasergene (DNASTAR) software.

The full-length Medea cDNA of plasmid O513, consisting of a HindIII-NotI fragment and a NotI fragment, was cloned into the pGEM-11Zf(−) vector (Promega) at its HindIII and NotI sites to generate p11Zf-Medea; p11Zf-Medea was digested with XbaI at a unique vector site 3′ of the cDNA insert and transcribed with the SP6 Message Machine kit (Ambion). The cDNAs H31 and I7114 in the pNB40 vector were linearized after the 3′ UTR with NotI and transcribed with the SP6 Message Machine kit. dpp mRNA was synthesized as described by Holley et al. (1996). pSP35T-tkv-a was constructed from tkv-a cDNA (Holley et al., 1996) that was PCR-amplified from pBluescript KS-tkv-a and cloned into pSP35T via NcoI and XbaI sites in the 5′ and 3′ PCR primers, respectively (J. Neul, personal communication); pSP35T-tkv-a was linearized with EcoRI and transcribed using the SP6 Message Machine kit. pSP35T-Mad was constructed from Mad cDNA (Sekelsky et al., 1995) that was PCR-amplified from pBluescript-Mad and cloned into pSP35T via NcoI and XbaI restriction sites in the PCR primers; pSP35T-Mad was linearized with SacI and transcribed with the SP6 Message Machine kit. Human SMAD4 cDNA (Hahn et al., 1996) in the pGEM-3Z vector (Promega) was digested with BamHI and transcribed with the SP6 Message Machine kit.

Capped mRNAs were injected into the dorsal side of pre-blastoderm stage embryos, and the resulting phenotypes were analyzed as previously described (Ferguson and Anderson, 1992a; Holley et al., 1996).

Southern blotting, northern blotting, restriction digestion, and subcloning techniques were conducted according to standard procedures (Sambrook et al., 1989).

Photomicroscopy was performed with an Axiophot microscope (Zeiss) with DIC Nomarski, transmitted light, dark field, or phase contrast optics. Photographic images were digitized using a Polaroid Sprint Scan 35 slide scanner or UMAX Powerlook flatbed scanner. Some images were obtained with a Progres 3012 digital camera (Kontron Elektroniks). Figures were assembled using Adobe PhotoShop 3.0 and printed using a Kodak XL dye-sublimation printer.

To identify genes that participate in dpp signal transduction, we conducted a genetic screen for dominant maternal enhancers of a weak allele of zerknüllt (zen), a downstream target of dpp (Ferguson and Anderson, 1992a; Rushlow and Levine, 1990). The zen gene encodes a homeodomain protein expressed at cellular blastoderm stage in the dorsal 10% of nuclei, which are fated to become amnioserosa (Rushlow et al., 1987). Mutant embryos lacking zen activity do not differentiate amnioserosa, and they die with a partially ventralized cuticular phenotype (Arora and Nusslein-Volhard, 1992; Rushlow and Levine, 1990). The allele we chose for this screen, zen^f16, confers a subliminal phenotype: 98% of zen^f16 mutant embryos hatch (Wakimoto et al., 1984) (Fig. 1 and Materials and Methods). We recovered nine mutations that act as dominant maternal-effect enhancers of zen^f16. In females carrying any of these mutations, 75–100% of zen^f16 progeny die as embryos with a partially ventralized cuticular phenotype. Eight of the newly induced mutations were mapped to the third chromosome and shown to be alleles of Medea. These have been named Med¹¹ through Med¹⁸. The remaining mutation mapped to the second chromosome and proved to be an allele of Mad; it has been named Mad^Ez.

Phenotypic analysis indicates that Mad^Ez is a hypomorphic allele of the gene. Homozygous Mad^Ez flies are viable and fertile at 25°C. However, flies carrying Mad^Ez in trans to the phenotypically null allele Mad¹² display imaginal disc defects (data not shown) similar to those observed in other partial loss-of-function allelic combinations of Mad mutations (Sekelsky et al., 1995). When rare egg-laying competent females of genotype Mad^Ez/Mad¹² are mated with wild-type males, all their progeny die as embryos. One-half (55%, n=533) of the embryos, of presumptive genotype Mad¹²/+, have a weakly ventralized phenotype (compare Fig. 2A and Fig. 2E). The remaining embryos, of presumptive genotype Mad^Ez/+, have a variably expressive dorsal-open phenotype (Fig. 2F), similar to the zygotic phenotypes caused by mutations in the dpp receptor genes punt and thick veins (Affolter et al., 1994; Brummel et al., 1994; Letsou et al., 1995; Nellen et al., 1994) or in the gene encoding the zinc-finger transcription factor schnürri (Arora et al., 1995; Grieder et al., 1995).

Even though all eight Medea alleles recovered in our screen enhance zen^f16 to a similar degree, they result in a variety of phenotypes. Six strong alleles (Med¹¹, Med¹², Med¹³, Med¹⁴, Med¹⁶ and Med¹⁸) are lethal in trans to each other and to a deficiency of the locus, Df(3R)E40 (data not shown). Trans-heterozygous larvae arrest with a phenotype identical to that described by Raftery et al. (1995). In contrast, flies homozygous for either of the remaining two mutations, Med¹⁵ or Med¹⁷, are viable without obvious phenotypic abnormalities.

Phenotypic analysis of embryos from homozygous Med¹⁵ females reveals the direct role of Medea in patterning the embryonic dorsal-ventral axis. When Med¹⁵ females are mated with wild-type males, all of their progeny die with a partially ventralized phenotype (compare Fig. 2A with Fig. 2C) that is nearly identical to that of embryos lacking zen activity. Consistent with their cuticular phenotypes, we found that these embryos do not differentiate amnioserosa as assayed by β-galactosidase staining of a Kr-lacZ transgene (compare Fig. 2B with Fig. 2D).

The function of Medea in patterning the imaginal discs was revealed by analysis of transheterozygotes between Med¹⁵ and any of the lethal Medea alleles. While some trans-heterozygotes die at the pharate adult stage, many escapers eclose with a range of phenotypes similar to those exhibited by dpp mutants defective in imaginal disc patterning (Spencer et al., 1982). The observed defects include split nota (Fig. 3A,B), intercalary and terminal gaps in L4 wing veins, absent or gapped posterior crossveins (Fig. 3C,D), and foreshortened legs that lack tarsal claws and often lack several tarsal segments (Fig. 3E,F).

Med¹⁷ confers a less severe phenotype than does Med¹⁵. 98% (n=550) of the embryos from homozygous Med¹⁷ females hatch. In addition, transheterozygotes between Med¹⁷ and any of the lethal Medea alleles are viable and do not display any disc patterning defects. However, when these trans-heterozygous females are mated with wild-type males, between 28 and 90% of their progeny die with a weakly ventralized phenotype (Table 1).

We performed a series of genetic crosses to compare the effect of each lethal mutation with that of a known deficiency of Medea. Like the alleles identified in the screen of Raftery et al. (1995), all of our lethal Medea alleles acted as strong dominant enhancers of the moderate dpp^hr4 mutation (Table 1). However, the percentage enhancement of a weaker dpp allele, dpp^hr56, varied among alleles. Approximately 50% of the dpp^hr56/+ progeny from females heterozygous for the Medea deficiency Df(3R)E40 or for the alleles Med¹³, Med¹⁴ or Med¹⁶ survived to adulthood. However, proportionally fewer dpp^hr56/+ progeny survived from females heterozygous for the other three mutations, Med¹¹, Med¹² and Med¹⁸. A similar phenotypic series was observed by measuring the percentage of ventralized embryos arising from females carrying each lethal Medea allele in trans to the weak allele, Med¹⁷. This antimorphic or dominant negative character of the Med¹¹, Med¹² and Med¹⁸ mutations suggests that they may produce polypeptides that interfere with the signaling activity of the wild-type Medea gene.

The Med¹⁵ and Med¹⁷ alleles confer homozygous viability; therefore they must retain some Medea activity. However, Med¹⁵ and Med¹⁷ can enhance other dpp-pathway mutants as strongly as does a deficiency of the Medea locus. Specifically, Med¹⁵ and Med¹⁷ enhance zen^f16 and dpp^hr4/+ as strongly as do null Medea alleles, and Med¹⁵ enhances dpp^hr56 as strongly as do null Medea alleles (Table 1). These results are consistent with the hypothesis that the Med¹⁵ and Med¹⁷ compromise one of two mutations independently mutable functions of Medea, while leaving a second function intact. To test this hypothesis, we determined whether a further reduction in maternal and zygotic Medea activity would worsen the dorsal-ventral patterning defects caused by Med¹⁵. Specifically, we compared the phenotypes of embryos from Med¹⁵ females mated with wild-type males to the phenotypes of embryos from Med¹⁵/Med¹⁶ females crossed to Med¹³/TM3 males (Med¹⁶ and Med¹³ are null alleles of Medea). All of the embryos from this latter cross are thus derived from females that contain a two-fold reduction in the dose of Med¹⁵, and one fourth of the progeny embryos, of genotype Med¹³/Med¹⁶, completely lack zygotic Medea activity. We found that this further reduction in Medea activity resulted in only a slight increase in the phenotypic severity of the mutant embryos: while all embryos from a Med¹⁵ mother mated to wild-type males have phenotypes similar to those caused by a weak allele of dpp, dpp^hr56, some embryos from the latter cross had phenotypes similar to those caused by a slightly stronger dpp allele, dpp^hr4 (Wharton et al., 1993; data not shown). Both of these phenotypes are much weaker than the phenotype caused by complete loss of Medea activity (see below) and support the hypothesis that Med¹⁵ severely compromises, but does not totally eliminate, one of two independently mutable activities of the Medea protein.

The ventralized phenotype of embryos from Med¹⁵ homozygous females indicated that Medea is involved in embryonic dorsal-ventral patterning. To determine the phenotype caused by the complete loss of maternal and zygotic Medea activity, we used the FLP-FRT system coupled with ovo^D1 (Chou and Perrimon, 1996) to induce mitotic recombination within the female germline and generate germ cells homozygous for the Med¹³ mutation. When mated to Med¹⁴/TM3 males, these germline clone (GLC) females laid eggs of two phenotypic classes. 48% (n=219) of the embryos had a partially ventralized phenotype (Fig. 4C) and were of presumptive zygotic genotype Med¹³/TM3. These cuticular phenotypes were characterized by the presence of dorsal hairs, internalization of filzkörper, and elimination of dorsolaterally derived structures of the head. To determine whether any embryos of this genotype formed amnioserosa, GLC females were mated with males carrying a P{Kr-lacZ} marker. 97% (n=152) of such embryos failed to differentiate any amnioserosa as assayed by β-galactosidase activity from P{KrlacZ}, (Fig. 4D), indicating a partial ventralization of the embryonic pattern. Thus, in the absence of maternal Medea activity, zygotic Medea function can only partially rescue the embryonic pattern.

The second class of embryos from the above cross, of presumptive zygotic genotype Med¹³/Med¹⁴, had cuticular phenotypes indistinguishable from those of dpp null mutant embryos (Fig. 4A). These embryos did not differentiate any dorsal structures, but instead differentiated ventral denticle bands around the entire embryonic circumference. In addition, no embryos of this genotype differentiated amnioserosa (Fig. 4B). These results indicate that Medea, like dpp, is required for the specification of all dorsal structures in the Drosophila embryo. In the remainder of the text, we refer to embryos from GLC females that lack both maternal and zygotic Medea activity as ‘Medea GLC-null’ embryos.

To formally determine the order of action of the Medea gene relative to that of dpp and its receptor tkv, we injected mRNAs encoding DPP or a constitutively active form of the tkv receptor (tkv-a) into syncytial blastoderm embryos of various Medea mutant genotypes. Injection of either dpp or tkv-a mRNA into dpp null embryos induces amnioserosa, indicating that amnioserosa formation can result directly from the activity of the injected mRNA and not from any subsequent induction of dpp transcription (Holley et al., 1996). We first injected 4 μg/μl dpp or tkv-a mRNAs into embryos from Med¹⁵ mutant females. Injection of either mRNA resulted in the restoration of amnioserosa (dpp, 89%, n=166; tkv-a, 92%, n=52) in these mutant embryos (compare Fig. 5A,B with Fig. 2B). Furthermore, the majority of embryos differentiated a greater amount of amnioserosa than is present in the wild type. Thus, an increase in dpp signaling caused by the injection of either mRNA is sufficient to bypass a partial loss of Medea activity.

We then injected the same concentration of dpp or tkv-a mRNAs into embryos completely lacking maternal and zygotic Medea activity. Phenotypic analysis of the injected Medea GLC-null embryos demonstrated that Medea function is required for the dorsalizing activity of both injected mRNAs. Specifically, 0 of 54 Medea null embryos injected with 4 μg/μl dpp mRNA (Fig. 5C) and 0 of 31 injected with 4 μg/μl tkv-a mRNA (Fig. 5D) differentiated any amnioserosa tissue. Furthermore, the lack of any phenotypic effect of the tkv-a mRNA was not limited to the absence of induced amnioserosa; cuticle preparations from 25 of 29 Medea null embryos injected with 4 μg/μl tkv-a mRNA did not differentiate any dorsal structures (Fig. 5E). These embryos resembled uninjected embryos, indicating the complete lack of response to the constitutively active receptor. The remaining four embryos differentiated only a single small patch of cuticle containing a few (2-12) dorsal hairs (Fig. 5F). We postulate that this low level of response was due either to residual gene activity from the Med¹³ or Med¹⁴ alleles or to perdurance of wild-type Medea mRNA or protein from before the mitotic recombination event in the GLC females. We therefore conclude that Medea is absolutely required for signal transduction downstream of dpp and its receptor, thick veins, in the specification of embryonic dorsal-ventral pattern.

The Medea gene has been mapped distal to the eye color locus claret at meiotic map position 104 on the third chromosome, and it is deleted by Df(3R)E40 which removes polytene bands 100C5 through 100F1-5 (Raftery et al., 1995). To undertake a molecular analysis of Medea, we first localized the gene more precisely by mapping it relative to two P-element insertions, P720 located in 100B and the P1575 insertion in the tramtrack (ttk) gene at 100D3 (Fig. 6A). Genetic recombination experiments indicated that Med¹⁵ was located 0.3 cM distal to the P720 insertion and 0.06 cM proximal to ttk. We obtained P1, cosmid, and λ clones with inserts of Drosophila genomic DNA from this region, and we used subclones from them as probes to identify restriction fragment length polymorphisms (RFLPs) between the Med¹⁵ chromosome and each of the two P-element chromosomes. We correlated our physical and genetic maps of this region by mapping each RFLP with respect to Med¹⁵ and the two P-element insertions, identifying a 40 kb genomic region expected to contain the Medea gene (Fig. 6A).

By screening cDNA libraries with DNA probes from the cosmid and P1 clones, we identified three transcripts from the 40 kb region (Fig. 6A). To determine whether any of these transcripts were derived from the Medea gene, mRNA from each cDNA was injected into embryos from Med¹⁵ females. mRNA from the O513 cDNA, but not from the cDNAs H31 and I7114, rescued the loss of amnioserosa (data not shown). Moreover, injection of 3 μg/μl mRNA from the O513 cDNA was sufficient to restore normal amnioserosa pattern (79%, n= 94) to Medea null embryos (compare Fig. 7A to Fig. 4B), demonstrating that the O513 cDNA encodes the Medea gene.

Sequence analysis of the Medea cDNA revealed an insert of 3252 bp, and northern blot analysis of mRNA from 0- to 2-hour embryos identified a Medea transcript of 3.1–3.3 kb (Fig. 6B), indicating that this cDNA is nearly full length. Comparison of the genomic DNA sequence with that of the cDNA indicated that the Medea transcript is contained within 4.2 kb of genomic DNA and is composed of nine exons. The cDNA sequence (Fig. 6C) contains a large open reading frame (ORF). The 5′ end of the ORF contains four in frame methionine codons. The sequences upstream of each methionine codon share approximately equal similarity with the consensus Drosophila translational initiation site (Cavener and Ray, 1991); however, we have chosen to represent the protein as initiating at the third methionine codon, based on sequence similarity with its human homolog (see below). Assuming that this AUG is used as the start codon, translation of the ORF would produce a 745 amino acid protein of predicted molecular mass 79×10³.

The amino acid sequence of the deduced MEDEA protein was compared with sequences in the translated GenBank database. A high overall similarity was found between MEDEA and members of the Smad family of proteins. The highest degree of similarity was found between MEDEA and the murine and human Smad4 proteins (Fig. 6D). Smad family members are characterized by the presence of two domains of high amino acid similarity, MH1 and MH2. The MEDEA protein is 97% identical to human SMAD4 over the entire MH1 domain, and is 82% identical to SMAD4 within MH2. The MH2 domains of MEDEA and SMAD4 differ from those of the other Smads in that they each contain an insert of 30 amino acids, which is 69% identical between the two proteins, in place of eight amino acids conserved among other Smads. The high degree of sequence similarity between MEDEA and SMAD4, coupled with the conservation of the SMAD4-specific domain within MEDEA, strongly suggests that MEDEA and hSMAD4 may be related by direct evolutionary descent, and thus they may be true orthologs. MEDEA and SMAD4 are largely unrelated in the linker region between the MH1 and MH2 domains, which includes two contiguous runs of glutamine residues in MEDEA that are not contained in the SMAD4 sequence.

We have localized DNA lesions in four of our six strong Medea alleles. One mutation, Med¹¹, is a 405 base pair deletion internal to the Medea gene, starting in intron 4 and extending into exon 5 (Fig. 6A). The remaining three mutations are single base pair changes that result in stop codons. Med¹⁶ and Med¹⁴ truncate the protein within the linker domain separating MH1 and MH2, at amino acid positions 370 and 483, respectively. Med¹² truncates the protein at amino acid 595 within the MH2 domain. We note that Med¹², but not Med¹⁶ or Med¹⁴, displays antimorphic characteristics in some genetic crosses (Table 1). This phenotypic difference could be due to differential stability of the three truncated polypeptides. Alternatively, it is possible that the presence of MH2 sequences are necessary for the non-functional protein-protein interactions that confer the antimorphic phenotype.

The homozygous viable Medea alleles, Med¹⁵ and Med¹⁷, contain base pair changes that cause missense mutations within MH2. The Med¹⁵ chromosome has two such changes: a conservative amino acid change A679V and a non-conservative change T714K. The Med¹⁷ chromosome has the non-conservative change P715T. The adjacent amino acids affected by the Med¹⁵ and Med¹⁷ mutations, 714 and 715, map to the loop 3 (L3) region of the Smad4 MH2 domain crystal structure (Shi et al., 1997). The L3 region protrudes from the planar disk of the Smad4 trimer and has been hypothesized to mediate heteromeric interactions between Smad4 and other Smad family members.

We tested the ability of human SMAD4, as well as Drosophila Mad, to substitute for Medea function and restore dpp signaling in various Medea mutant embryos. Injection of 3 μg/μl of SMAD4 mRNA was sufficient to restore amnioserosa in 39% (n=77) of embryos from Med¹⁵ females and 68% (n=53) of Medea null embryos from GLC females (Fig. 7C). While injection of 2 μg/μl Mad mRNA was sufficient to restore amnioserosa in 63% of embryos from Med¹⁵ females (n=191), injection of the same concentration of Mad mRNA did not restore amnioserosa in any Medea null embryos from GLC females (n=76) (Fig. 7B). Because hSMAD4, but not Mad, was able to restore normal dorsal-ventral pattern to embryos completely deficient for Medea activity, we conclude that hSMAD4 function can substitute for MEDEA in this patterning process.

A variety of experiments in Xenopus embryos have shown that overexpression of various Smads after mRNA injection has phenotypic consequences for dorsal-ventral patterning as assayed by morphology or gene expression (Baker and Harland, 1996; Graff et al., 1996; Suzuki et al., 1997; Thomsen, 1996). To determine whether overexpression of Mad and Medea have similar phenotypic consequences in Drosophila, we co-injected equimolar amounts of Mad (3 μg/μl) and Medea (6 μg/μl) mRNAs into wild-type embryos and observed the effects on embryonic patterning. Ventral injection of these mRNAs had no phenotypic consequences on amnioserosa production (n=68). Furthermore, none of the cuticles from injected embryos displayed any alteration in the dorsal-ventral extents of either the dorsal epidermis or neurogenic ectoderm (n=106). Rather, many cuticles displayed a variable alteration in the anterior-posterior specification of the ventral denticles at the presumptive site of injection. Specifically, 14% of embryos differentiated normal cuticles, 42% of embryos differentiated cuticles with a truncation or fusion of one or two denticle bands, 32% of embryos differentiated cuticles in which three or more denticle bands were truncated or fused, and 12% of embryos differentiated cuticles with large ventral holes. Co-injection of these mRNAs on the dorsal side of the embryos had no phenotypic consequences, either at the level of amnioserosa production or cuticular patterning (data not shown). The lack of dorsalventral pattern defects in these injected embryos is in marked contrast with the strong dorsalization of the embryonic pattern observed after injection of dpp mRNA (Ferguson and Anderson, 1992a). Moreover, in embryos that lack endogenous dpp signaling (‘lateralized’ embryos from females of genotype snk^rm4Tl^9Q/snk²²⁹; Ferguson and Anderson, 1992a), co-injection of the same concentrations of Mad and Medea mRNAs was not sufficient to promote amnioserosa formation (n=155), as assayed by Kr-lacZ expression. Thus, over-expression of the Mad and Medea genes in the Drosophila embryo does not appear to be sufficient either to confer a ligand-independent response leading to amnioserosa formation or to perturb the embryonic dorsal-ventral pattern.

The conservation of the mechanisms of TGF-β signal transduction across metazoan phyla has facilitated rapid progress in the analysis of this signaling system since the identification of mutations in the Mad and Medea genes of Drosophila (Raftery et al., 1995; Sekelsky et al., 1995). We report here a genetic and molecular characterization of the Medea gene. We show that Medea, like the dpp receptor genes tkv and punt, is absolutely required for the dpp-dependent specification of dorsolateral and dorsal cell fates in the embryonic ectoderm (Irish and Gelbart, 1987; Letsou et al., 1995; Nellen et al., 1994; Ruberte et al., 1995; Terracol and Lengyel, 1994). Furthermore, we show formally that Medea, like Mad (Newfeld et al., 1997), is required downstream of dpp and the tkv receptor for the specification of dorsal cell fates. However, other results suggest that Medea may not be absolutely required to transduce all dpp-dependent signals during Drosophila development (Wisotzkey et al., 1998). We have cloned the Medea gene and shown that it shares significant sequence similarity with human SMAD4. The striking conservation of these signaling pathways is further illustrated by our finding that SMAD4 mRNA can substitute for Medea activity in embryonic dorsal-ventral patterning.

The conservation of sequence and function between Medea and human SMAD4 indicates that their protein products share essential structural features that mediate interactions with other signal transduction components. In particular, we expect that MEDEA and the hSMAD4 protein share the ability to interact productively with the Drosophila MAD protein and possibly other cofactors to control the activation and/or repression of downstream target genes in this patterning process. The sequence similarity between MEDEA and Smad4 is primarily restricted to the MH1 and MH2 domains. The MH1 domains of Drosophila MAD and human SMAD4 have DNA binding activity in vitro, and the MH2 domains of human SMAD1 and SMAD4 can activate transcription at a heterologous promoter (Kim et al., 1997; Liu et al., 1996; Yingling et al., 1997). We therefore anticipate that the MH1 and MH2 domains of MEDEA will be the essential domains in the transmission of dpp signals from receptor to nucleus, although the regulation of target gene expression by the dpp signaling system remains poorly understood. No function has yet been ascribed to the SMAD4 linker domain. The observations that MEDEA and SMAD4 diverge considerably within this region yet share all functions indicate that the linker may not carry out essential functions.

Through our analysis of the two homozygous viable Medea alleles, Med¹⁵ and Med¹⁷, we have suggested that the MEDEA protein contains at least two independently mutable activities in the transduction of dpp signals. In particular, we propose that Med¹⁷ and especially Med¹⁵ are compromised in the dosage-sensitive specification of amnioserosa, but that both mutant proteins retain a separable function required for the specification of dorsolateral cell fates in the embryo.

What could the two separately mutable activities of Medea represent? One possibility is that each activity represents a differential capacity to transduce a signal downstream of each of the two type I DPP receptors, TKV and SAX. Embryos that lack both maternal and zygotic tkv activity differentiate no dorsal structures, similar to the complete loss of Medea (Nellen et al., 1994; Terracol and Lengyel, 1994). In contrast, although the phenotypes of embryos completely lacking sax activity have not been reported because of a requirement for sax during oogenesis (Twombly et al., 1996), existing mutations in sax result only in the loss of amnioserosa, similar to the phenotype caused by the Med¹⁵ mutation (Brummel et al., 1994; Nellen et al., 1994; Penton et al., 1994; Xie et al., 1994). These parallels suggest that Med¹⁵ and Med¹⁷ mutants may be defective in the response to signals downstream of the SAX receptor, while still transducing signals from the TKV receptor.

In light of this proposal, we note that both Med¹⁵ and Med¹⁷ have amino acid substitutions in loop 3, an element of the Smad4 crystal structure that is implicated in productive heteromeric interactions with activated receptor-specific Smad proteins (Shi et al., 1997). The mutant MEDEA proteins might therefore have a diminished capacity to form particular heteromeric complexes with MAD in response to signaling by one receptor but not another. Alternatively, the mutant MEDEA proteins could have selective disruptions in interactions with other components of the signaling system, such as factors that may collaborate with MAD and MEDEA to regulate expression of specific target genes. Full evaluation of this proposal awaits biochemical characterization of signaling downstream of the TKV and SAX receptors in vivo.

Observations from a series of mRNA injection experiments suggest that MAD and MEDEA activities are tightly restricted by the spatial distribution and level of dpp signaling. Although injected Mad or Medea mRNA can restore essentially normal dorsal-ventral pattern to embryos that lack some or all endogenous Medea activity, neither induces further dorsalization of the embryo. Furthermore, while injection of dpp or activated tkv mRNA induces ectopic amnioserosa in a concentration-dependent fashion (Holley et al., 1996), overexpression of Mad and Medea in wild-type embryos has no effect on dorsal-ventral pattern. Overexpressed MAD and MEDEA proteins therefore require additional positive input for activity, such as MAD phosphorylation by the DPP receptors or the activation of other proteins required for dpp signaling. We note that the Drosophila embryo differs in this regard from Xenopus and cell culture systems in which Smad overexpression elicits responses (Baker and Harland, 1996; Chen et al., 1996; Graff et al., 1996; Liu et al., 1997; Suzuki et al., 1997; Thomsen, 1996); responsiveness in these systems to ectopic Smad expression could reflect amplification of a low level of ambient signaling rather than de novo generation of a response.

Alternatively, the different responses of these systems to Smad overexpression may reflect inherent differences in the distributions of negative regulators such as Drosophila DAD and vertebrate Smad6 and Smad7, which are structurally related yet functionally opposed to the signaling Smads (Hayashi et al., 1997; Imamura et al., 1997; Nakao et al., 1997; Tsuneizumi et al., 1997). Such intracellular negative regulation of dpp signaling in the Drosophila embryo is also suggested by the observations that although genetic experiments demonstrate Mad activity is present in limiting quantity (Raftery et al., 1995), endogenous levels of dpp signaling in the embryo do not lead to detectable nuclear localization of MAD (Newfeld et al., 1996, 1997). Taken together, these observations suggest that a pool of cytoplasmic MAD may be needed to titrate a stoichiometric cytoplasmic inhibitor. Because Medea, like Mad, is easily rendered dosage sensitive, it may be subject to a similar negative regulation in the cytoplasm.

With this report and that of Raftery et al. (1995), two independent genetic screens have been conducted to identify maternally supplied, dosage-sensitive components of the dpp signaling pathway. Together these screens led to the recovery of five alleles of Mad and eleven alleles of Medea from among 6000 mutagenized haploid genomes. Additional work will be required to identify dpp signaling components that are less dosage sensitive than Mad and Medea and to identify negative regulators of dpp signaling. The recovery in this work of Medea alleles with a range of mutant phenotypes affords new opportunities for such efforts. Furthermore, our finding that human SMAD4 can replace the function of the Medea gene will allow the dissection of SMAD4 functions within a developmental context.

We thank William Gelbart, Stuart Newfeld, Eugene Xu, Chung-I Wu, Anthony Mahowald, Christopher Schonbaum, Laurel Raftery, Kenneth Tartof, Jeffrey Neul, Allen Shearn, Craig Montell, Joan Massagué, Janice Fischer, I. Siden Kiamos, Thomas Kaufman, and Roger Karess for fly stocks, cloned DNAs or genomic libraries. We thank Ed Williamson for assistance with scanning electron microscopy; Nipam Patel for advice and use of equipment for photomicroscopy; Terrence Banks, Christian Denes, Adriane Stewart, Deborah Dorsett for fly work, and Joseph June and Jim Miller for assistance in the preparation of genomic DNA. Genomic sequencing was performed at the University of Chicago Cancer Research Center DNA Sequencing Facility. We are grateful to Laurel Raftery and Richard Padgett for communication of results prior to publication. We thank Judith Austin, Jeffrey Neul, Eva Rosen and Harinder Singh for helpful comments on the manuscript. This work was supported by grants from the Mallinckrodt Foundation and from the American Cancer Society (DB-79090). J. B. H. and S. D. P. were supported by the NCI Training Program (5T32-CA09594), S. D. P. was supported by a N. I. H. postdoctoral fellowship (HD07959), and K. K. was supported by N. I. H. training grant GM07183. E. L. F. is a Pew Scholar in the Biomedical Sciences.