DNA-binding domain mutations in SMAD genes yield dominant-negative proteins or a neomorphic protein that can activate WG target genes in Drosophila

Signaling molecules of the Transforming Growth Factor-β (TGFβ)family regulate differentiation and proliferation in many species. Four TGFβ proteins have been characterized in Drosophila and each of these has a counterpart in mammals(Newfeld et al., 1999). For example, Decapentaplegic (DPP) shows significant amino acid similarity to mammalian BMP2 and BMP4. In addition, the function of these proteins appears to be conserved. In cross-species experiments, human BMP2 and BMP4 rescued dpp mutant phenotypes in Drosophila(Padgett et al., 1993), and recombinant DPP protein induced bone formation in mammalian cells(Sampath et al., 1993). The functional conservation revealed by cross-species experiments suggests that studies of TGFβ signaling in Drosophila will impact on our understanding of TGFβ signaling in mammals.

In brief, DPP signal transduction begins with a complex of transmembrane receptor serine-threonine kinases. Upon ligand binding, one of the receptors phosphorylates the cytoplasmic protein Mothers against DPP (MAD)(Newfeld et al., 1996). As a result of MAD phosphorylation, the MAD-related protein Medea (MED) is recruited to form a heteromeric complex with MAD(Wisotzkey et al., 1998). The MAD/MED complex then translocates to the nucleus and co-activates the transcription of DPP target genes (e.g. Xu et al., 1998).

All TGFβ family members use this mechanism, and MAD-related proteins(the SMAD family) are found in many species. SMAD family members contain two functionally distinct and highly conserved MAD homology (MH) domains. The amino-terminal MH1 domain appears to be responsible for transcriptional activation, whereas the carboxy-terminal MH2 domain appears necessary for forming multi-SMAD complexes (Lagna et al., 1996).

TGFβ1 was discovered through its anti-mitotic effects on fibroblast cells in culture (Barnard et al.,1990). However, TGFβ1 was unable to induce growth arrest in fibroblast-derived tumors (Fynan and Reiss, 1993). Subsequently, `loss of heterozygosity' studies have shown that human SMAD genes act as tumor suppressors in a wide variety of tissues. Homozygous mutations in SMAD2 or SMAD4 are detected in 20% of breast, 42% of colorectal, 17% of lung and 80% of pancreas tumors,as well as in the inherited cancer Autosomal Dominant Juvenile Polyposis(Eppert et al., 1996; Riggins et al., 1997; Howe et al., 1998). More recently, reporter gene assays of SMAD mutant alleles isolated from human tumors revealed a universal inability to activate transcription at wild-type levels (Dai et al., 1998; Xu and Attisano, 2000). These studies led to the hypothesis that all SMAD tumor mutations are loss-of-function mutations. One prediction of this hypothesis is that all mutations induce tumors via a single mechanism – the inability to transduce an anti-mitotic signal encoded by a TGFβ family member (e.g. Massagué et al.,2000).

However, the modular nature of SMAD proteins and our experience with mutations in Mad (e.g. Sekelsky et al., 1995) suggest that the situation is more complex. We propose the alternative hypothesis that there are multiple classes of SMAD mutation (loss of function and gain of function). A prediction of our hypothesis is that human SMAD mutations in different classes induce tumor formation via distinct mechanisms.

Here, we report a study, using the Gal4/UAS system, in which we generated phenotypes for mutant alleles of Mad and Med, as well as for mutant alleles of human SMAD2 and SMAD4 isolated from pancreatic and colon tumors. Our study establishes a set of principles for the transgenic characterization of human mutant alleles. In wild-type flies, the expression of a loss-of-function mutation does not generate a mutant phenotype, whereas the expression of a gain-of-function mutation generates a mutant phenotype. Furthermore, different classes of gain-of-function mutation generate different phenotypes. Phenotypes generated by dominant-negative alleles mimic genomic loss-of-function mutations. Phenotypes generated by neomorphic mutations are unrelated to the wild-type function of the gene.

The entire locus was amplified by PCR from heterozygous animals. PCR products were sequenced using the Thermosequenase cycle sequencing kit (USB). The lesion was identified by the presence of two bases at a particular position corresponding to the mutant allele and the wild-type allele on the balancer chromosome.

The Mad¹ allele contains an A615T mutation. By using PCR (Lorson et al., 1999),this mutation was copied into the Mad cDNA using a pair of complementary mutant primers. The Mad¹ reverse primer 5′-CTGGACGGACGATTACTGGTCTCCCATCGC-3′ (the mutant base is shown in bold) and the M13 forward primer were used to create the 5′ Mad¹ fragment. The Mad¹ forward primer 5′-GCGATGGGAGACCAGTAATCGTCCGTCCAG-3′ and the M13 reverse primer were used to create the 3′ Mad¹ fragment. A full-length Mad¹ cDNA was produced using the M13 forward and M13 reverse primers, and the annealed 5′ and 3′ Mad¹ fragments as a template. The Mad¹² allele contains a C1601T mutation(Sekelsky et al., 1995). The Mad¹² reverse primer 5′-GCGGAGTATCATCGCTAGGATGTGACCTCG-3′ and the M13 forward primer were used to create the 5′ Mad¹² fragment. The Mad¹² forward primer 5′-CGAGGTCACATCCTAGCGATGATACTCCGC-3′ and the M13 reverse primer were used to generate the 3′ Mad¹² fragment. Mad mutant cDNAs were cloned into the XbaI and KpnI sites in pUAST (Brand and Perrimon,1993). SMAD4 mutant cDNAs(Schutte et al., 1996) were excised by cutting with BamHI and EcoRV, and cloned into blunted XhoI and BglII sites of pUAST. An asymmetric SacI site was used to check orientation. Additional SMAD mutant cDNAs(Riggins et al., 1996) were excised by cutting with BamHI, blunting the ends and cutting with KpnI. cDNAs were then cloned into the XbaI (blunted) and KpnI sites in pUAST. Multiple independent fly lines were generated for five mutants.

Fly stocks were as described: In(2L)dpp^s6 and dpp^hr4 (St Johnston et al., 1990); Df(2L)C28, Mad¹, Mad¹¹and Mad¹² (Sekelsky et al., 1995); Df(3R)E40 and Med⁷(Raftery et al., 1995); zw3^M11 FRT101 and zw3^sggD127 FRT101(Siegfried et al., 1992); UAS.Dpp, UAS.CA-Arm, UAS.DN-TCF, UAS.Mad, UAS.Gbb, UAS.lacZ, 24B.Gal4,32B.Gal4, 69B.Gal4, A9.Gal4, C765.Gal4, MS1096.Gal4, T80.Gal4, ap.Gal4,dll.Gal4, dpp-blink.Gal4, en.Gal4 and ptc.Gal4(Drysdale and Crosby, 2005). Dominant enhancement of dpp^s6/dpp^hr4wing phenotypes by Df(2L)C28, Mad¹, Df(3R)E40 and Med⁷ was evaluated by examining at least 500 individuals of each genotype. Adult Gal4/UAS genotypes were generated using males from two independent lines homozygous for a UAS construct crossed to Gal4 bearing females at 25°C. Tables containing quantitative data for all Gal4/UAS genotypes are available upon request. For Gal4/UAS combinations involving SMAD2 alleles, wing size was calculated as described(Marquez et al., 2001). Clones of cells homozygous for the genetic null zw3^sggD127 or the protein null zw3^M11 were generated by standard methods(Siegfried et al., 1992).

Histochemical detection of β-galactosidase activity in embryos was conducted as described (Newfeld et al.,1996). Embryonic cuticles were prepared by standard methods. Vein primordia in pupal wing imaginal discs were detected with a monoclonal antibody recognizing Drosophila SRF(Marenda et al., 2004) or a riboprobe transcribed from a rhomboid cDNA(Wolff, 2000). Anterior margin bristle primordia in third instar larval wing discs were detected with a monoclonal antibody recognizing Achaete(Skeath and Carroll,1991).

The Mad¹ allele intrigued us for two reasons. First, it was the only Mad allele isolated from two screens (75,000 chromosomes in total) for modifiers of dpp adult viable phenotypes(Sekelsky et al., 1995; Su et al., 2001). By contrast,four alleles of Mad, including the Df(2L)C28 deletion, were identified in a pilot screen (1850 chromosomes) for dominant maternal enhancement of dpp embryonic lethal phenotypes(Raftery et al., 1995). Second, in an initial genetic analysis, Mad¹ was a stronger enhancer of two dpp recessive embryonic lethal mutations than was Df(2L)C28 (Sekelsky et al., 1995).

In re-examining the initial data on Mad¹, it became clear that certain classes of gain-of-function mutation (e.g. dominant negative) could not be detected in an assay for enhancement of lethality. The original data showed that heterozygosity for an allele with a complete loss of Mad function, such as Df(2L)C28, resulted in essentially maximum enhancement. Only 0.5-3.0% of the expected progeny survive(Sekelsky et al., 1995). In this assay, Mad mutant alleles with dominant-negative effects would appear to be loss-of-function alleles, as less than 0% of the expected progeny is not possible – 0% was seen for Mad¹ enhancement of dpp^hr56.

We conducted a less restrictive test of the relationship between Mad¹ and dpp to determine whether Mad¹ was a dominant-negative allele. For the test, we exploited a dpp adult-viable mutant phenotype(Nicholls and Gelbart, 1998). The test is based upon a transheterozygous dpp mutant genotype that is adult viable with occasional wing vein defects(Fig. 1A). Truncation of Longitudinal vein 5 (L5) is observed in 11% of flies bearing the dpp^s6/dpp^hr4 genotype, while 2%display truncations of L4 and L5.

When we placed a single copy of Df(2L)C28 into dpp^s6/dpp^hr4 individuals, the frequency of L5 defects increased to 95%, 20% had defects in L4 and L5(Fig. 1B), and 5% had defects in L5 and L2, or L5 and the posterior crossvein. When we placed the Mad¹ allele into dpp^s6/dpp^hr4 individuals, the frequency and severity of vein defects increased beyond that seen with Df(2L)C28. In flies carrying Mad¹, the frequency of wings with defects in L4 and L5 increases to virtually 100%. Furthermore,10% of wings with L4 and L5 defects have missing crossveins (anterior,posterior or both) and small margin notches(Fig. 1C). We also noted that a statistically significant amount of lethality was associated with the dpp^s6/dpp^hr4 Mad¹ genotype. Only 71%of the expected dpp^s6/dpp^hr4 Mad¹progeny were recovered (P<0.005). No lethality was associated with either the dpp^s6/dpp^hr4 or the dpp^s6/dpp^hr4 Df(2L)C28 genotypes. Mad¹ has a greater effect on dpp^s6/dpp^hr4 vein phenotypes and survival than does a deletion of Mad, indicating that Mad¹ is a gain-of-function allele with dominant-negative activity.

We then identified the lesion in Mad¹ and in six additional Mad alleles (Table 1). The Mad¹ mutation is an amino acid substitution (Q90L) due to an A to T transversion at nucleotide 615 of the cDNA. Of thirteen sequenced Mad mutant alleles(Sekelsky et al., 1995; Chen et al., 1998) (and this report), Mad¹ is the only mutation in the MH1 domain. Given that Mad¹ is a dominant-negative allele with a lesion in the transcriptional activation domain, we generated the following two-part hypothesis for Mad¹ dominant-negative activity. First, we propose that the MAD¹ protein is unable to activate transcription, but is capable of receptor phosphorylation and of forming complexes with its partner SMAD protein Medea (MED). Second, as a result,MAD¹ dominant-negative activity derives from the formation of non-functional complexes that deplete the pool of MED to the point that wild-type MAD proteins are unable to find sufficient partners for normal activity.

Circumstantial evidence for the first part of this hypothesis is provided by two sources. First, biochemical studies of human SMAD4 show that MH1 mutations prevent DNA binding and transcriptional activation(Dai et al., 1998; Xu and Attisano, 2000). Second, an examination of aligned SMAD MH1 sequences(Newfeld et al., 1999)revealed that the amino acid affected in the MAD¹ protein, Q90, is conserved in all fly and vertebrate SMADs that participate in DNA binding(receptor-associated SMADs and co-SMADs), but is not conserved in any antagonist SMADs. Furthermore, in the structure of the SMAD3 MH1 domain bound to DNA, Q90 is one of three residues that make direct base contact(Shi et al., 1998). Nematode SMAD sequences show the same relationship between Q90 and DNA binding. Q90 is present in the unusual antagonist DAF-3. DAF-3 antagonizes TGFβ signal transduction by binding to DNA and repressing gene expression(Thatcher et al., 1999), a mechanism not used by other antagonist SMADs. In addition, Q90 is not present in the atypical receptor-associated SMADs DAF-8 and DAF-14. Both DAF-8 and DAF-14 stimulate the expression of TGFβ target genes by inhibiting DAF-3 function (Inoue and Thomas,2000). Drosophila SMAD2 is the single exception. The amino acid corresponding to Q90 in Drosophila SMAD2 falls in a run of nine amino acids unlike any other SMAD protein.

In order to experimentally test our hypothesis that the MAD¹protein is transcriptionally inactive, we generated UAS.Mad¹ and UAS.Mad¹². The Mad¹² allele has a nonsense mutation (Q417Stop) in the MH2 domain that removes the three serines phosphorylated in response to DPP signaling. The Mad¹² allele behaves exactly like a deletion of Mad in both genetic and biochemical assays(Sekelsky et al., 1995; Hoodless et al., 1996). We used UAS.Mad¹² as a control for overexpression of a loss-of-function mutation. We also examined UAS.Mad, as a control for overexpression of the wild-type protein.

In an initial characterization, we expressed these transgenes in wild-type wing discs and assayed their effect on vein formation in adult wings. We have shown, using Mad¹² mutant clones, that vein formation is dependent on Mad activity(Marquez et al., 2001). Expression of UAS.Mad (Fig. 2B) (Marquez et al.,2001) produces excess vein tissue. Alternatively, Madgenomic loss-of-function genotypes (transheterozygous combinations of hypomorphic alleles) show reduced venation and growth defects(Fig. 2D)(Sekelsky et al., 1995). Expression of the loss-of-function allele UAS.Mad¹² with 69B.Gal4 had no effect on any aspect of wing development(Fig. 2C). Expression of UAS.Mad¹ with 69B.Gal4 led to vein truncations and incomplete wing outgrowth (Fig. 2E). Expression of UAS.Mad with 69B.Gal4 did not lead to defects in vein formation or wing outgrowth (data not shown). The UAS.Mad¹ phenotype resembles the Madloss-of-function phenotype (compare Fig. 2D with 2E).

In addition to effects on venation, many UAS.Mad¹genotypes showed significant reductions in viability. For example, 52% of the expected UAS.Mad¹/T80.Gal4 flies, 55% of the UAS.Mad¹/dll.Gal4 flies and 69% of the UAS.Mad¹/24B.Gal4 flies were observed(P<0.005). Expression of UAS.Mad or UAS.Mad¹² had no effect on viability. These studies are consistent with our genetic analysis of Mad¹, supporting its identification as a dominant-negative allele.

To further test our hypothesis that the MAD¹ protein is transcriptionally inactive, we examined two molecular markers for vein formation in wing imaginal discs. We analyzed Drosophila Serum response factor (SRF; Blistered – FlyBase) protein expression and rhomboid (rho) transcript accumulation. DrosophilaSRF expression is widespread in third instar wing discs but its expression is downregulated in vein primordia by rho in pupal wing discs(Fig. 3A)(Biehs et al., 1998). rho transcription in vein primordia may be directly dependent upon DPP signaling via MAD (Fig. 3E)(Yu et al., 1996). Thus, in vein development, the expansion of Drosophila SRF expression indicates a lack of rho activity and the absence of rhoactivity indicates that it is not being transcribed by MAD.

In UAS.Mad/ptc.Gal4 discs, Drosophila SRF expression is reduced in the primordia of L3, L4 and L5(Fig. 3B). This is consistent with the excess vein phenotype of UAS.Mad/ptc.Gal4 wings(Fig. 2B). In UAS.Mad¹²/69B.Gal4 discs, Drosophila SRF expression is essentially unaffected (Fig. 3C). This is consistent with the wild-type appearance of UAS.Mad¹²/69B.Gal4 wings(Fig. 2C). In UAS.Mad¹/69B.Gal4 discs, Drosophila SRF expression is expanded in the primordia of L3, L4 and L5(Fig. 3D). This is consistent with the reduced venation of UAS.Mad¹/69B.Gal4 wings(Fig. 2E). These results suggest that MAD¹ is incapable of activating rhoand, furthermore, that MAD¹ protein can block the ability of endogenous MAD to activate rho.

Examination of rho transcript accumulation in pupal wing discs expressing UAS.Mad¹ confirms this interpretation. In UAS.Mad¹ discs, rho expression is reduced in the primordia of L3, L4 and L5. L4 expression is the most severely affected(Fig. 3F). This is consistent with the vein phenotype of UAS.Mad¹/69B.Gal4 wings(Fig. 2E). Because rhomay be a direct transcriptional target of MAD, this result supports our hypothesis that the MAD¹ protein cannot activate transcription.

We then noted that of the 18 sequenced Med mutant alleles only Med⁷ is an MH1 domain mutation(Das et al., 1998; Hudson et al., 1998; Wisotzkey et al., 1998). Furthermore, the C99S mutation in Med⁷ alters one of three cysteine residues that is predicted to coordinate a zinc atom(Chai et al., 2003). We tested Med⁷ and the Df(3R)E40 deletion of Medin our assay for enhancement of the dpp^s6/dpp^hr4 wing phenotype. In this assay, Df(3R)E40 increased the frequency of dpp^s6/dpp^hr4 wings with L5 truncations from 11%to 28%. When we placed the Med⁷ allele into dpp^s6/dpp^hr4 individuals, the frequency and severity of vein defects increased beyond that seen with Df(3R)E40. In flies carrying Med⁷, the frequency of wings with defects in L5 and L4 was 42%. Furthermore, 15% of these wings also have defects in L2, or the posterior crossvein or margin notches(Fig. 1D). Med⁷ is also a gain-of-function allele with dominant-negative activity.

Perhaps dominant-negative activity explains the relative infrequency of detectable MH1 mutations in Mad and Med. Just two out of 33 sequenced mutations (6%) occur in the MH1 even though this domain covers roughly 23% of the amino acids in these proteins.

Our identification of Mad¹ and Med⁷as gain-of-function alleles with MH1 mutations suggested that SMAD MH1 mutations identified in human tumors might also generate gain-of-function alleles. We tested this hypothesis with a set of five SMAD mutant cDNAs derived from pancreas and colon tumors(Table 2).

In studies with transgenes expressing wild-type SMAD2 or SMAD4, we showed that their phenotypes mirrored those of their fly homologs, Drosophila SMAD2 and MED, respectively(Marquez et al., 2001). This suggested to us that human SMAD proteins expressed in flies function in the same manner as their fly homologs do. By extending this idea to SMAD tumor alleles, we predict that the manner in which a mutation affects the activity of a human SMAD protein expressed in flies will inform us about the effect of that mutation on the human SMAD protein.

When SMAD alleles with mutations in the MH2 domain(UAS.SMAD4^524ST,UAS.SMAD2^Δ344-358 and UAS.SMAD4^493H) are expressed in flies, their wings are wild type in size and appearance like those expressing the loss-of-function allele UAS.Mad¹². Alternatively, wings expressing the MH1 missense mutation UAS.SMAD4^130S(Fig. 2F) are similar to those expressing the dominant-negative allele UAS.Mad¹(Fig. 2E). Both UAS.SMAD4^130S and UAS.Mad¹ engender defects in vein formation and wing outgrowth. Furthermore, UAS.SMAD4^130S expression results in reduced viability like UAS.Mad¹ does. For example, UAS.SMAD4^130S/dll.Gal4 flies were recovered at 36% and UAS.Mad¹/dll.Gal4 flies were recovered at 55% of the expected frequency. Results for SMAD4^130S suggest that it too is a gain-of-function allele with dominant-negative activity.

The expression of the MH1 missense allele SMAD4^100Tgenerated a wing phenotype never before reported in any study of TGFβsignaling in flies. Most wings expressing UAS.SMAD4^100Tstrongly throughout the wing blade (e.g. MS1096.Gal4, C765.Gal4 or T80.Gal4) have ectopic mechanosensory bristles on the blade(Fig. 4A). In wild type, two rows of mechanosensory bristles (stout and thin) are normally found in the triple row region of the proximal anterior wing margin(Couso et al., 1994). The ectopic mechanosensory bristles seen in wings expressing UAS.SMAD4^100T are largely derived from the transformation of normally `bristle-less' mechanosensory receptors called campaniform sensilla (Held, 2002).

Two campaniform sensilla (Twin Sensilla One and Two) are located on the dorsal surface of L1 (Fig. 4B). The most frequent transformation seen in UAS.SMAD4^100T/MS1096.Gal4 wings is a stout mechanosensory bristle in place of Twin Sensillum One(Fig. 4C). A sensillum on the ventral surface of L3 has only recently been reported(Aigouy et al., 2004)(Fig. 4D). A stout mechanosensory bristle is occasionally seen in the place of this sensillum on a UAS.SMAD4^100T/MS1096.Gal4 wing(Fig. 4E). Three sensilla(Sensilla Campaniformium of Dorsal Radius) are located on the dorsal surface of L3 (Fig. 4F). The second most frequent transformation seen in UAS.SMAD4^100T/MS1096.Gal4 wings is a thin mechanosensory bristle replacing either the proximal (Fig. 4G), middle (Fig. 4H) or distal (not shown) sensillum at roughly equal frequencies. The most posterior ectopic bristle on a UAS.SMAD4^100T/MS1096.Gal4 wing is rare, and is located on the ventral surface of L5 (Fig. 4A). We could not detect a campaniform sensillum in this location and none has been reported. Thus, five out of the six ectopic bristles on UAS.SMAD4^100T/MS1096.Gal4 wings are derived from the transformation of sensilla.

Though never previously associated with TGFβ signaling, the presence of ectopic anterior margin bristles on the wing blade is not unprecedented. Wingless (WG) is a secreted signaling molecule expressed along the presumptive wing margin in imaginal discs. Numerous studies have shown that mechanosensory bristle development along the margin requires WG as well as components of the canonical WG signaling pathway (Couso et al., 1994). Alternatively, when ectopic WG signaling is activated in the presumptive wing blade, for example in zeste white 3(zw3; shaggy, sgg – FlyBase) null clones, ectopic mechanosensory bristles result (Blair,1992). Several thin mechanosensory bristles on the wing blade emanating from unmarked clones of zw3^M11 cells are shown(Fig. 4I). Ectopic bristles can be generated anywhere on the wing blade by zw3^M11clones.

The unexpected transformation of sensilla to bristles on wings expressing UAS.SMAD4^100T suggests the hypothesis that the R100T mutation in SMAD4^100T conveys a novel activity upon the encoded protein. Thus, SMAD4^100T is the second gain-of-function allele associated with a human tumor. Furthermore, the similarity between the UAS.SMAD4^100T wing phenotype and that of zw3^M11 clones suggests a second hypothesis –that SMAD4^100T has the ability to activate WG target genes.

To test our hypothesis that the R100T mutation conveys a novel activity upon SMAD4, we examined wings expressing Drosophila TGFβ family members with known roles in wing development (DPP and GBB)(Ray and Wharton, 2001). We wondered whether these genes could generate ectopic bristles on the wing blade. In these experiments, UAS.Dpp expression was absolutely lethal, and significant lethality was also associated with UAS.Gbbexpression. In the two experiments where adult flies were obtained, we found that UAS.Gbb inhibits triple row bristle formation. ap.Gal4is expressed in all dorsal cells of the wing disc(Diaz-Benjumea and Cohen,1993). We observed that the proximal 25% of the anterior margin of a UAS.Gbb/ap.Gal4 wing has no bristles(Fig. 5A). More distally, the triple row reappears but is quite irregular(Fig. 5B,C). When UAS.Gbb was driven in all cells of the wing blade with C765.Gal4 (de Celis et al.,1996), the phenotype was similar but less severe (data not shown). In summary, none of our studies of UAS.Dpp or UAS.Gbbgenerated sensilla to bristle transformation, supporting our hypothesis that UAS.SMAD4^100T is a gain-of-function allele with novel activity.

To test the second hypothesis (that the novel activity of the SMAD4^100T protein is the ability to activate WG target genes), we conducted three experiments. In the first, we tested the ability of SMAD4^100T to suppress phenotypes generated by DN-TCF, a dominant-negative form of the WG pathway transcription factor Drosophila TCF (Pangolin, PAN – FlyBase). For this test, we used MS1096.Gal4, a driver strongly expressed throughout the wing blade. The wings of UAS.DN-TCF/MS1096.Gal4 flies(Fig. 6A) are very similar to those with impaired WG signaling (e.g. Couso et al., 1994). These wings are very small and lack a margin. However, if two(Fig. 6B) or three(Fig. 6C) copies of UAS.SMAD4^100T are also present, then the UAS.DN-TCF/MS1096.Gal4 wing phenotype is visibly suppressed in a dosage-dependent fashion. If instead we add copies of UAS.lacZ, then the UAS.DN-TCF/MS1096.Gal4 wing phenotype is unaffected (data not shown). This result suggests that, even in the presence of a strong antagonist of WG signaling, SMAD4^100T is capable of activating WG target genes involved in wing growth and margin formation.

We made two additional observations in these experiments. First, siblings to our experimental flies (those without UAS.DN-TCF) showed dosage-dependant dominant-negative phenotypes for DPP-dependent processes such as vein formation (Fig. 6D-F). Thus, above a certain threshold, UAS.SMAD4^100T resembles the other MH1 mutations in our study – it has dominant-negative effects on DPP signaling. Second, we found that the UAS.DN-TCF/MS1096.Gal4genotype is male-specific lethal. In a cross of MS1096.Gal4homozygous females to UAS.DN-TCF homozygous males, we obtained 199 female progeny and no male progeny. By using the F1 females in crosses to males containing various numbers of UAS.SMAD4^100Tinsertions, we obtained male progeny bearing MS1096.Gal4 and UAS.DN-TCF. The suppression of male-specific lethality suggests that SMAD4^100T is able to activate WG target genes outside of wing discs.

To further support this observation, we expressed SMAD4^100T in the embryonic ventral epidermis. In most abdominal segments, the ventral epidermis is composed of twelve rows of cells: six rows that secrete smooth cuticle followed by six rows that secrete protrusions called denticles(Fig. 7A). Cells choose to secrete naked cuticle or denticles according to positional information supplied, in part, by WG and Engrailed (EN)(Gritzan et al., 1999; Alexandre et al., 1999). WG signals instruct cells to secrete naked cuticle. EN is expressed just posterior to WG and an EN-expressing cell secretes the first denticle row(Fig. 7B). When a constitutively active form of the WG pathway signal transducer Armadillo(CA-ARM) is expressed via en.Gal4, the first row of denticles in all segments is transformed into naked cuticle(Fig. 7C). When SMAD4^100T is expressed with en.Gal4, patches of cells within the first denticle row are transformed into naked cuticle on one or more segments of most embryos (Fig. 7D). This result is consistent with a recent report that weak global expression of CA-ARM results in a patchy loss of denticles(Hayward et al., 2005). We conclude that SMAD4^100T is capable of phenocopying activated WG signaling in the ventral epidermis. These data support the idea that the ability of SMAD4^100T to activate WG target genes is not context dependent.

Moving from the phenotypic to the molecular level, we looked directly at the expression of Achaete (AC) in wing discs expressing SMAD4^100T. AC is expressed in the presumptive anterior wing margin(Fig. 8A) coincident with WG. Genetic analyses have shown that ac is required for the development of the margin bristles and that ac is a direct target of WG signaling(Skeath and Carroll, 1991; Blair, 1992). As a result, wing discs bearing clones of cells homozygous for a mutation in zw3 have numerous regions of ectopic AC expression in the presumptive wing blade(Fig. 8B). The same result is seen when examining AC expression in wing discs from individuals expressing UAS.SMAD4^100T (Fig. 8C). In a UAS.DPP-expressing disc, the pattern of AC expression along the presumptive margin is normal, although the disc is overgrown,reflecting the ability of DPP to influence wing growth(Fig. 8D). In a UAS.GBB-expressing disc, AC expression is absent in the most proximal region of the anterior margin, reflecting the ability of GBB to inhibit bristle formation in this region (Fig. 8E). Overall, our studies of DN-TCF, CA-ARM and AC strongly support the hypothesis that the SMAD4^100T protein is capable of activating the expression of WG pathway target genes.

From a larger perspective, our study establishes guidelines for interpreting data from transgenic analyses of human tumor alleles. In principle, studies of this type can be applied to mutant alleles of any well-conserved tumor suppressor gene or oncogene. Furthermore, our unexpected finding that all of the tested mutations in the DNA-binding domain of SMAD genes are gain of function, whereas all of the tested mutations in the multimerization domain are loss of function, indicates that missense mutations in modular proteins should be experimentally characterized rather than defaulted to the loss-of-function category.

For SMAD tumor suppressor genes, our identification of two gain-of-function alleles of SMAD4 (dominant negative and neomorphic) falsifies the prevailing hypothesis that all SMAD tumor mutations are loss-of-function mutations. Instead, our data support an alternative hypothesis: that there are multiple classes of SMAD mutation and that each class is associated with a different mechanism of tumorigenesis.

This alternative hypothesis may also apply to other TGFβ signaling pathway components with tumor-associated mutations. For example, mutations in TGFβ receptors are found in tumors from the same tissues that exhibit SMAD mutations [e.g. pancreas (Hempen et al., 2003), colon (Peltomaki,2001), breast (Pouliot and LaBrie, 1999), lung (Zhang et al., 2004)]. However, missense mutations in TGFβ receptors conferring gain-of-function activity have not been identified in tumors,because the most common mutational assays are loss of expression and polyA tract sequencing.

Our data for SMAD4^100T are unprecedented in studies of TGFβsignaling in flies. This suggests that SMAD4^100T may induce tumors in humans by an unexpected method. The fact that SMAD4^100Texpression mimics activated WG signaling and suppresses an antagonist of WG signaling further suggests that SMAD4^100T utilizes a mechanism of tumorigenesis associated with loss-of-function mutations in Adenomatous Polyposis Coli (APC). A model based on this interpretation is shown in Fig. 9.

In vertebrates and flies, APC serves as a component of the highly conserved WG/int-1 (WNT) signal transduction pathway. Like ZW3 and its homolog Glycogen Synthase Kinase-3β (GSK3β), APC functions as a WNT antagonist via participation in a cytoplasmic retention complex that prevents Armadillo (or its vertebrate homolog β-catenin) from accumulating in the nucleus in the absence of WNT signals. Studies in flies have shown that homozygous null clones bearing mutations in any member of the retention complex (ZW3, Drosophila APC and Drosophila Axin) lead to the same phenotype: ectopic anterior margin bristles on the wing blade as a result of the unregulated expression of WG target genes such as AC(Blair, 1992; Akong et al., 2002; Hamada et al., 1999). First identified in the rare inherited cancer Familial Adenomatous Polyposis,homozygous mutations in APC are now found in roughly 85% of all colon tumors(Kinzler and Vogelstein,1996). In studies of mice engineered to homozygose APC null mutations only in cells of their intestinal epithelium, the immediate consequence of APC loss was ectopic expression of WNT target genes via constitutively nuclear β-catenin(Sansom et al., 2004).

Given their roles in their respective signal transduction pathways,loss-of-function mutations in APC cause tumors by a fundamentally different mechanism than loss-of-function SMAD mutations. Specifically, inactive APC proteins cannot block a mitogenic WNT signal (an oncogenic mechanism), whereas inactive SMAD proteins cannot transduce an anti-mitotic TGFβ signal (a tumor suppressor mechanism). Interestingly, one study of SMAD4^100Tin mammalian cells suggested that this allele could employ an oncogenic mechanism of tumorigenesis (Dai et al.,1999), a proposal consistent with our data.

An examination of the primary difference between the phenotypes of SMAD4^100T and zw3, Apc and Axin mutant clones may shed light on SMAD4^100T-associated tumor formation. The primary difference is the location of ectopic bristles on the wing blade. In wings expressing SMAD4^100T, ectopic mechanosensory bristles are derived from a cell fate transformation of mechanosensory receptors(campaniform sensilla). Cells extruding sensilla and those extruding bristles are derived from Sensory Organ Precursor cells(Aigouy et al., 2004). Alternatively, zw3, Apc and Axin mutant clones generate cell fate transformations anywhere on the wing blade – regardless of cell lineage. This discrepancy suggests that SMAD4^100T is not as potent an activator of WNT target genes as zw3, Apc or Axinmutations. SMAD4^100T may only activate WNT target genes in cells predisposed to tumorigenesis, perhaps by pre-existing mutations.

This possibility is supported by our studies of wings with an increasing dosage of UAS.SMAD4^100T. When wild-type WG signaling is present,increasing the UAS.SMAD4^100T copy number did not increase the frequency of sensillum to bristle transformation. However, in wings with reduced WG signaling due to the activity of UAS.DN-TCF, increasing the copy number of UAS.SMAD4^100T quantitatively increased the rescue of WG-dependent functions outside of sensilla (e.g. in wing outgrowth and anterior margin formation).

The transformation of campaniform sensilla to mechanosensory bristles was reported once previously. Several transheterozygous mutant genotypes of ash2, a member of the Trithorax group of transcriptional regulators,generate this phenotype (Adamson and Shearn, 1996). Although the ASH2 protein contains a zinc finger motif, its function has not yet been demonstrated biochemically. Studies of its yeast homolog suggest that ASH2 may function in chromatin remodeling and transcriptional activation as part of a complex containing histone methyltransferases (Janody et al.,2004). The similarity of SMAD4^100T and ash2 mutant phenotypes, and the ability of SMAD4^100T to suppress DN-TCF phenotypes, suggest that SMAD4^100T contributes to the activation of WNT target genes downstream of APC, perhaps by participating in transcription factor complexes.

Three previous reports have shown physical interactions between the wild-type SMAD proteins and transcriptional effectors of WNT signaling(β-catenin and TCF). One study used Xenopus embryos to demonstrate that SMAD4/β-catenin/TCF complexes activate the transcription of the WNT target gene twin(Nishita et al., 2000). Recently, SMAD1/β-catenin/TCF complexes were detected in renal medullary dysplasia in ALK3 transgenic mice (Hu et al., 2003), and in human dysplastic renal tissue(Hu and Rosenblum, 2004). We are currently testing the possibility that SMAD4^100T cooperates with ARM and/or TCF to activate the transcription of AC.

At this time, therapeutic research on SMAD-associated tumors is guided by the current hypothesis that all SMAD mutations lead to tumors via a loss of a TGFβ-encoded anti-mitogenic signal. As a result, effort is focused on restoring the wild-type function in tumors by gene replacement. However, we have shown that two SMAD4 tumor alleles are gain-of-function mutations. One important feature of gain-of-function mutations is that they exert their effect even in the presence of a wild-type allele on the homologous chromosome. Thus, it seems unlikely that gene replacement will be successful in inhibiting tumorigenesis in cells with SMAD4 gain-of-function mutations

In individuals with APC mutant colon tumors (those with unregulated WG target gene expression such as Familial Adenomatous Polyposis), the transition from adenomatous polyps to carcinoma will take roughly ten years. Alternatively, for TGFβ receptor mutant colon tumors [those unable to respond to a TGFβ-encoded anti-mitogenic signal such as Hereditary Non-Polyposis Colorectal Cancer (HNPCC)], progress from adenomatous polyps to carcinoma takes less than three years(Souza, 2001). Given these data, if SMAD4^100T induces tumorigenesis by an `APC-like'mechanism while SMAD4 dominant-negative and loss-of-function alleles stimulate tumorigenesis by an `HNPCC-like' mechanism, then the prognosis for cancer patients with a SMAD4^100T mutation is distinctly different from that of patients with other SMAD mutations.

This raises several issues for future investigation. First, how many different mutations in SMAD4 can generate an `APC-like' gain-of-function allele? To date, three SMAD4 missense mutations near codon 100 have been identified in colon tumors [Y95N, C115R and N118K(Iacobuzio-Donahue et al.,2004)]. All of these mutations (including R100T) occur in the L2/L4 double-loop region identified in the crystal structure of the SMAD3 MH1 bound to DNA. This loop occurs at the surface of the molecule and is important for macromolecular interactions (Shi et al., 1998). Are these also gain-of-function mutations? Second,what is the relative frequency of `APC-like' gain-of-function SMAD4 alleles versus `HNPCC-like' loss-of-function alleles in tumors from various tissues?Third, can an accurate and efficient diagnostic test be developed to distinguish between `APC-like' and `HNPCC-like' alleles in tumors with a SMAD4 mutation? Answering these questions will require a continued collaboration between model organism geneticists and oncologists.

We thank G. Riggins, M. Schutte and R. Ray for cDNAs, and the Developmental Studies Hybridoma Bank for antibodies. T. Haerry, S. Hayashi, M. Hoffmann, A. Manoukian, L. Marsh, L. Raftery, E. Siegfried, K. Wharton and the Bloomington Stock Center contributed flies. A. Schmid and C. Lorson provided technical advice. A. Johnson generated the UAS.Mad¹ plasmid and assisted with image analysis. B. Johnson, M. Stinchfield, B. Celaya and N. Emmert helped with fly pushing. This study was supported by the NIH (CA095875 to S.J.N.). M.B.O. is an Associate Investigator of the Howard Hughes Medical Institute.