A positive role for Short gastrulation in modulating BMP signaling during dorsoventral patterning in the Drosophila embryo

Patterning along the dorsoventral (DV) axis in both arthropod and vertebrate embryos is established by activity gradients of Bone Morphogenetic Proteins (BMPs) (Dale and Wardle, 1999; Holley and Ferguson, 1997). In Drosophila, high levels of BMP activity in the dorsal 10% of embryonic cells promote differentiation as extraembryonic amnioserosa, lower BMP levels cause dorsolateral cells to become dorsal epidermis, and the absence of BMP signaling allows ventrolateral cells to differentiate as neurogenic ectoderm (Podos and Ferguson, 1999). While the biological output of the activity gradient is straightforward with respect to the tissue types it generates, the molecular complexities underlying the establishment and interpretation of the activity gradient are just beginning to be understood.

One of the unusual features of the BMP activity gradient is that it is generated by the combined action of two BMP ligands, Decapentaplegic (DPP) and Screw (SCW). DPP, which is expressed at uniform intensity throughout the dorsal 40% of the embryonic circumference, is absolutely necessary and under experimental conditions can be sufficient to generate all positional values within the ectoderm. However, in vivo SCW, which is uniformly expressed at the blastoderm stage, is necessary for production of the most dorsal cell fates (Arora et al., 1994). Moreover, functional data suggests that SCW acts through the type I receptor Saxophone (SAX), and that SAX signaling synergizes with signaling by the DPP type I receptor Thickveins (TKV) to promote dorsal cell fates (Neul and Ferguson, 1998; Nguyen et al., 1998).

Because both BMP ligands are expressed at uniform intensities over the regions in which they are required, the BMP activity gradient must be generated by post-transcriptional modification of ligand activity. The short gastrulation (sog) gene, which is expressed in the ventrolateral regions of the embryo, encodes a large cysteine-rich extracellular protein (François et al., 1994; Holley et al., 1995) that antagonizes BMP activity ventrally. There is a DPP-dependent expansion of dorsal cell types in sog mutants, and ectopic dorsal expression of SOG can lead to a partial ventralization of the embryo (Biehs et al., 1996; Ferguson and Anderson, 1992b; Holley et al., 1995). In Xenopus, the functional homologue of SOG, Chordin (Chd), inhibits BMP4 by binding to it and preventing BMP4 from activating its receptor (Piccolo et al., 1996). Overexpression of SOG can block SCW, but not DPP activity, suggesting that the inhibitory action of SOG is likely to be directed against the SCW protein (Neul and Ferguson, 1998; Nguyen et al., 1998).

The existence of a diffusible BMP inhibitor raised the possibility that a BMP activity gradient could be formed through a gradient of the SOG inhibitor. Support for this hypothesis came from the characterization of a secreted zinc metalloprotease, tolloid (tld) in flies and Xolloid (Xld) in Xenopus (Piccolo et al., 1997; Shimell et al., 1991). tolloid is expressed dorsally in the same domain as dpp and is required for production of dorsal structures. In vitro, TLD cleaves SOG, and this cleavage is stimulated by the presence of DPP (Marqués et al., 1997). Similarly, Xld cleaves Chd and Chd/BMP4 complexes, releasing biologically active BMP (Piccolo et al., 1997). These results suggested that a ventral source of the SOG inhibitor coupled with a dorsal sink for SOG (the cleavage of SOG by TLD) could result in a ventral-to-dorsal gradient of SOG, causing a reciprocal dorsal-to-ventral gradient of DPP activity (Marqués et al., 1997).

The inhibitory function of SOG, however, is not sufficient to fully explain its mutant phenotype. Specifically, in addition to their ventral defects, sog mutants also lack amnioserosa, the most-dorsal tissue type (Zusman et al., 1988). Thus, SOG must function outside of its domain of expression to promote amnioserosa formation. All other mutants that lack amnioserosa have defects that reduce BMP signaling, and it is not immediately obvious how loss of a negative regulator of BMP signaling would disrupt amnioserosa formation. We had previously proposed a model to explain this phenotype in which binding of BMP ligand to SOG allowed for diffusion of the ligand-SOG complex to the dorsal region of the embryo where release of the ligand from the inhibitory complex, possibly by TLD cleavage of SOG, would lead to elevated ligand signaling (Holley et al., 1996). However, at that time this model lacked experimental proof.

Specification of amnioserosa is well correlated with the expression pattern of the direct DPP target gene zerknüllt (zen) at the cellular blastoderm stage (Ray et al., 1991; Rushlow et al., 2001). At this stage in wild-type embryos, the expression of zen is restricted to the dorsal 10% of embryonic nuclei that will become amnioserosa. In ventralized mutants that lack amnioserosa, such as dpp, tld and scw, transcription of zen disappears at this stage (Ray et al., 1991). In contrast, in sog null mutants, zen transcription encompasses the dorsal 40% of embryonic nuclei (Ray et al., 1991; Rushlow and Levine, 1990). This expanded pattern has been interpreted to suggest that SOG acts solely as a negative regulator of zen transcription. Recently, however, Rushlow et al. (Rushlow et al., 2001) demonstrated that loss of various subsets of MAD binding sites in the zen promoter also resulted in zen transcription in the dorsal 40% of the embryo, suggesting that the spatial extent of zen transcription may not be a linear readout of the strength of BMP signaling at the cellular blastoderm stage.

Another extracellular modulator of BMP signaling, twisted gastrulation (tsg), is expressed in the dorsal 40% of the embryo and encodes a secreted cysteine rich protein that has weak similarity to Connective Tissue Growth Factor (CTGF) (Mason et al., 1994). tsg mutants also display a weakly ventralized phenotype, with loss of amnioserosa (Mason et al., 1994; Zusman and Wieschaus, 1985), suggesting that TSG is also required for maximal BMP signaling. However, zen transcription is also expanded in tsg mutants, in a pattern similar to that observed in sog mutants, although the intensity of zen protein level is decreased relative to that observed in sog mutants (Rushlow and Levine, 1990). A recent series of papers concerning the action of TSG and its vertebrate homologues have presented conflicting interpretations as to its function. These papers have shown that TSG physically interacts with SOG; however, different conclusions were reached as to the biological function of the complex. Oelgeschlager et al. (Oelgeschlager et al., 2000) suggested that TSG acts to potentiate BMP signaling, while others conclude that a TSG/SOG complex acts to inhibit BMP signaling (Chang et al., 2001; Ross et al., 2001; Scott et al., 2001; Yu et al., 2000). Although these different interpretations have not yet been resolved, the data are suggestive that both SOG and TSG may have multiple functions in vivo.

In this paper we provide direct genetic evidence that SOG has a dual function in modulating the BMP activity gradient. In addition to its previously described ability to locally inhibit BMP signaling ventrally, we unambiguously establish that SOG is required for maximal BMP activity dorsally. We show that a weak homozygous viable allele of sog is enhanced to lethality by reduction in the activity of the essential DPP signal transduction components, Mad or Medea, and that this enhancement can be observed as a reduction in the level of transcription of the direct DPP target zen. During the course of this work Ashe and Levine (Ashe and Levine, 1999) showed by gain-of-function experiments that SOG can inhibit the transcription of a dorsal-specific gene at the site of expression and can activate the transcription of the same gene in a long range fashion. We confirm their findings, and in addition, we present results that suggest that the positive activity of SOG is directed toward DPP rather than SCW. We demonstrate, in Drosophila, that Xenopus Chd has the same ligand specificity as SOG, but that it does not have the same ability to potentiate BMP activity. We discuss our data in the context of the recent findings concerning the activities of SOG and TSG, and we present a model that may help reconcile their mutant phenotypes and their demonstrated biological activities.

Wild-type embryos were from a stock containing a P[Kr-lacZ] insertion on the third chromosome. sog^P129D, listed in FlyBase as sog^supp, was recovered as a dominant extragenic suppressor of tld^10E in a P-element mutagenesis screen (Ferguson and Anderson, 1992b). Mad¹² (Sekelsky et al., 1995), Med¹³ and Med¹⁵ (Hudson et al., 1998), zen¹, snk¹ and snk²²⁹ (FlyBase, 1999) have been described previously.

zen¹ is a weakly temperature-sensitive mutation. To determine the phenotype of sog^P129D; zen¹ embryos at 25°C, a sog^P129D/FM7; zen¹/TM3 doubly balanced stock was grown at 18°C. Rare double mutant escapers of genotype sog^P129D; zen¹ were transferred to 25°C, and the phenotypes of the progeny embryos were examined.

To obtain sog null embryos with three copies of dpp⁺, sog^YSO6/FM7 P[ftz-lacZ] females were crossed to Dp(2;2)DTD48/CyO; P[Kr-lacZ] males. Stage 13 sog embryos were identified as those that did not stain for the P[ftz-lacZ] insert. Stage 6a sog embryos were identified as those that did not express the sog transcript (sog^YSO6 is a transcript null allele (Francois et al., 1994)). In both cases, 1/2 of the sog embryos carried an extra copy of dpp⁺.

Dorsalized embryos were obtained by mating snk²²⁹/snk¹ females with wild-type males. Embryos lacking scw were obtained as the progeny of a Df(2L)OD16 P[Kr-lacZ]/CyO P[ftz-lacZ] stock that did not stain for the P[ftz-lacZ] insert.

We performed a screen to identify maternal and X-linked zygotic mutations that dominantly enhanced the phenotype of sog^P129D to lethality. sog^P129D, y cv v f car males were starved for 8 hours, fed 25 mM EMS in 1% sucrose for 24 hours, and mated to sogYL26, wa/FM7b, y wa females. F₁sog^P129D, y cv v f car (*)/FM7b, y wa females were individually mated both to sog^P129D, f car males (to test for the presence of an enhancer mutation), and to FM7b males (to recover the mutant chromosome). In the F₂ generation, we scored the ratio of sog^P129D, y cv v f car (*)/sog^P129D, f car females to sog^P129D, f car/FM7b, y wa females to minimize the false positives that could arise from low progeny numbers. To distinguish X-linked zygotic from autosomal maternal enhancers, individual F₃ candidate females carrying the mutagenized chromosome in trans to FM7b were crossed to sog^P129D, f car males. An X-linked zygotic enhancer should be evident in the progeny of every female analyzed. However, only 50% of the F₃ females should carry a dominant maternal enhancer on either the second or third chromosome.

The maternal enhancer mutation on chromosome 2, Mad^ES1, was recovered through iterated matings of individual candidate females to sog^P129D, f car; S cn bw/CyO, cn pr males. Each female progeny that carried CyO was assessed for dominant enhancement of the sog^P129D mutation. Those that carried the enhancer mutation were mated again to the same males, and the enhancer mutation was observed to segregate away from the balancer chromosome. The two maternal enhancer mutations on chromosome 3, MedES2 and MedES3, were recovered in trans to the balancer chromosome TM3, Sb e through a similar strategy of iterated matings of individual candidate females to sog^P129D, f car; srp e/TM3, Sb e males.

The genetic identities of the enhancer mutations 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 and was shown to be an allele of Mad by non-complementation with the phenotypically null mutation Mad¹². The two enhancer mutations on chromosome 3 were mapped distal to the visible marker claret and were identified as alleles of Med by non-complementation with the null mutation Med13.

The degree of enhancement of sog^P129D by various maternal-effect mutations was obtained by pooling viability data from at least 10 crosses of sog^P129D males to individual females doubly balanced for both mutations.

Embryonic cuticle preparations were performed as described previously (Wieschaus and Nüsslein-Volhard, 1980). β-galactosidase expression was assayed in stage 13 embryos as described by Ferguson and Anderson (Ferguson and Anderson, 1992a).

Capped mRNAs were generated by in vitro transcription with SP6 mRNA polymerase from the Message Machine kit (Ambion). pSP35T plasmids containing sog or chd (Holley et al., 1995) were linearized using SacI and XbaI respectively. chd mRNA encodes a sog/chd chimera that contains the N-terminal 89 amino acids of SOG, followed by two serine residues, followed by Chd starting at amino acid 35 (Holley et al., 1995). pSP35T-scw was linearized with PstI. dpp mRNA was expressed from the pGEM-7zf(+)-dpp vector (Ferguson and Anderson, 1992a), which was linearized using XhoI.

mRNA was injected into preblastoderm embryos as described previously (Ferguson and Anderson, 1992a). Microinjection needles were pulled using standard microinjection parameters, broken to produce an opening of diameter less than or equal to 8.7 μm, and subsequently ground to a uniform bevel using a Narishige microgrinder. A Narishige picospritzer was used to deliver a constant bolus of approximately 20 pl for embryos in Fig. 5 and Fig. 6, or of approximately 100 pl for embryos in Fig. 8 and Fig. 9. The diameter of the injected bolus was measured by an eyepiece reticle.

In situ hybridization for the detection of zen, sog, Race (also known as Ance), and chd transcripts was performed as described previously (Patel, 1996). zen, sog and chd riboprobes were labeled with digoxigenin (DIG RNA Labeling Mix, Roche), and Race with fluorescein (Fluorescein RNA Labeling Mix, Roche). All embryos were histochemically stained with the appropriate antibody coupled to alkaline phosphatase (Roche). NBT/BCIP (Promega) was used as substrate for the AP reaction to give the ‘blue’ color. T-NBT/BCIP (Agarwala et al., 2001) was used as an AP substrate to give the ‘brown’ color.

Embryos injected with sog or chd mRNA were allowed to develop under oil, and before fixation eggs were gently punctured on the ventral side with an empty injection needle to allow the even penetration of fix into the embryo. Fixation was carried out on the coverslip after the oil had been removed by repeated rinsing with heptane. Embryos were fixed for 18 minutes in a 35 mm Petri dish containing 3.7% formaldehyde in fixation buffer (100 mM Hepes pH 6.9, 2 mM MgSO₄, 1 mM EGTA). The coverslip containing the fixed embryos was placed in methanol for 5 minutes and then rinsed with PBS pH 7.4 containing 0.1% Triton X-100. Embryos were hand-dissected in this solution and then subjected to hybridization as described above.

Photomicroscopy was performed with an Axiophot microscope (Zeiss) with DIC Nomarski or darkfield optics. Darkfield cuticle images were taken with TECHPAN film and digitized using a Polaroid Sprint Scan 35 slide scanner. DIC images were obtained with a Progres 3012 digital camera (Kontron Electroniks). Figures were assembled using Adobe PhotoShop.

The homozygous viable sog mutation, sog^P129D, which is caused by insertion of a P element 106 bp 5′ to the beginning of the sog transcript (Holley, 1997), reduces the level of sog transcript relative to wild type (Fig. 1). The sog^P129D phenotype is very sensitive to changes in levels of BMP signaling: while 80% of sog^P129D embryos survive to adulthood, only 10% survive either in trans to a null allele of sog or with two extra copies of Mad¹². We thus reasoned that loss of one copy of a gene that cooperates with sog could enhance the sog^P129D mutant to lethality. Specifically, although loss of sog activity does not completely eliminate neurogenic ectoderm (Ferguson and Anderson, 1992b), ectopic dorsal localization of sog mRNA is sufficient to inhibit formation of dorsal tissues (Holley et al., 1995), suggesting that other genes act in concert with SOG to repress BMP signaling ventrally. We had originally planned to screen for such genes; however, the results of our experiments, as described below, instead gave us insight into a second function of SOG.

We conducted a screen of 3000 chromosomes from which we recovered three independent mutations that acted as dominant maternal-effect enhancers of sog^P129D (Materials and Methods). Females carrying one copy of any of the three mutations enhanced the sog^P129D phenotype such that only 2 to 4% of sog^P129D embryos survived to adulthood (Table 1). Genetic mapping and complementation analysis (Materials and Methods) indicated that one of these mutations (ES1) is an allele of Mad and that the other two (ES2 and ES3) are alleles of Medea. Mad and Medea are essential components of the BMP signaling pathway (Das et al., 1998; Hudson et al., 1998; Newfeld et al., 1996; Wisotzkey et al., 1998), coupling DPP receptor activation to the control of target gene expression (Massagué and Chen, 2000).

If SOG acts only to inhibit BMP signaling, loss-of-function mutations in essential components of the BMP signal transduction pathway should suppress rather than enhance the sog mutant phenotype. We therefore tested whether known loss-of-function alleles of Mad and Medea behaved in the same fashion. We observed that the null alleles Mad¹² and Med13 enhanced sog^P129D to an extent similar to that caused by the alleles obtained in our screen, while a hypomorphic allele of Medea, Med15, acted as a less severe enhancer of the sog mutant (Table 1). Furthermore, this enhancement appears to be extremely sensitive to the level of sog activity, since loss of one copy of maternal Mad also caused some sog^P129D heterozygotes to die as embryos (Table 1). These findings thus provided strong support for the previous suggestions that SOG plays a positive role in BMP signaling.

In order to understand the biological basis underlying this genetic interaction, we characterized the lethal phenotype of the enhanced sog^P129D embryos. The mutant embryos die with cuticular defects similar to those of weakly ventralized embryos, possibly indicative of a defect in amnioserosa specification (Fig. 2A,C). To directly assay the amount of differentiated amnioserosa, we analyzed the expression of β-galactosidase from a P[Kr-lacZ] construct, which is expressed in this tissue type (Fig. 2B). While embryos from Mad^ES1/+ females have a grossly normal amount of amnioserosa (not shown), sog^P129D embryos laid by Mad^ES1/+ females have a reduced amount of amnioserosa (Fig. 2D).

These data focused our attention on the role of SOG in specifying amnioserosa. sog was originally isolated as a result of a failure in germ band elongation, indicative of the lack of amnioserosa (Zusman et al., 1988). Although in sog null mutants Kr-lacZ expression is greatly reduced or absent (Fig. 7E), the genetic basis for this defect has remained unclear. The fact that amnioserosa specification in a weak sog mutant is enhanced by reduction in the activity of Mad and Medea strongly suggests that the defect in amnioserosa formation in sog null embryos results from dorsal reduction of BMP signaling.

In the wild-type embryo, the pattern of zen transcription correlates with the future differentiation of amnioserosa. In sog mutant embryos, however, this correlation does not hold: the pattern of zen transcription is expanded to encompass the dorsal 40% of the embryo, but the amnioserosa does not differentiate. The pattern of zen transcription in sog null embryos therefore could reflect not only the loss of negative regulation of BMP signaling, possibly manifest as an expansion in the spatial extent of zen transcription, but also the elimination of an activity that promotes BMP signaling, possibly manifest as a decrease in the level of zen transcription. If this hypothesis is correct, a reduction in the level of zen transcription due to loss of the positive function of SOG might allow us to observe a synthetic interaction between weak alleles of the two genes. While sog^P129D and zen1 are each homozygous viable at 25°C (Fig. 3A), double mutant embryos die with a partially ventralized cuticular phenotype (Fig. 3A,B). Thus, reduction of SOG function can enhance a weak zen mutation, indicating that SOG is needed for maximal zen activity.

To determine whether this genetic enhancement could be observed at the level of zen transcription, we examined zen expression in sog^P129D embryos laid by Mad^ES1/+ females. Because zen transcription is very dynamic, we confined our observations to a 5-minute period (stage 6a, according to Campos-Ortega and Hartenstein (Campos-Ortega and Hartenstein, 1985)) immediately after the onset of gastrulation. Compared to wild-type embryos (Fig. 4A), sog^P129D embryos showed both a spatial expansion of zen expression and a slight reduction in intensity of expression (Fig. 4B). Embryos laid by Mad^ES1/+ females expressed zen in the dorsal 10% of nuclei but with a reduced intensity (Fig. 4C). In contrast, zen transcription in both heterozygous and homozygous sog embryos laid by Mad^ES1/+ females was greatly reduced compared to either parental strain and in some embryos was almost completely eliminated (Fig. 4D-F). These data directly demonstrate that SOG is required for maximal expression of a DPP target gene.

Interestingly, in the enhanced sog^P129D embryos that retained some zen transcription, the spatial extent of zen transcription appeared to be broader than in the sog^P129D embryos alone. This result, together with data from Rushlow et al. (Rushlow et al., 2001), suggests that refinement of zen transcription to the dorsal 10% of the embryonic nuclei may require maximal BMP signaling, and that lower levels of BMP signaling at the cellular blastoderm stage could permit zen to continue to be transcribed in more lateral regions of the embryo. Such a hypothesis could possibly explain the expanded pattern of zen transcription in sog null mutants without the necessity of invoking loss of negative regulation of BMP signaling as the sole cause for this defect.

We then wondered whether, in the background of a sog null mutation, reduction in Mad activity would result in a similar decrease in zen expression. We thus assayed zen transcription in both sogYS06 embryos and in sogYS06 embryos laid by Mad¹²/+ females. We found that the level of zen transcription is variable in sogYS06 embryos (Fig. 4G), and we observed approximately the same range of variability in sogYS06 embryos from a Mad¹²/+ female (Fig. 4H). Thus, in the absence of sog activity, Mad no longer displays a dosage-sensitive phenotype in this assay. Possibly, an overall elevation in the level or activity of MAD caused by loss of SOG eliminates its dosage dependence in this genetic background. We hypothesize that the complete lack of SOG in sog^YSO6 mutants may mask the decrease in zen transcription clearly seen in sog^P129D embryos laid by Mad heterozygote mothers.

While SOG’s negative activity has been shown to be exerted at the site of its transcription in the ventral regions of the embryo, its positive activity is required for amnioserosa formation in the dorsal-most cells, and thus must occur at a distance from its site of expression. To examine both activities simultaneously it is necessary to assess the effects of localized sog expression over a large field of equipotent cells such as is found in a dorsalized embryo. Embryos laid by females mutant for the dorsal group of genes (Anderson and Nüsslein-Volhard, 1984) express dorsal-specific genes, such as dpp, zen and tolloid, around the embryonic circumference and do not express ventral specific genes, such as sog (Ray et al., 1991; Rushlow et al., 1987). Although all cells in dorsalized embryos therefore have a pattern of gene expression similar to that of the most dorsal cells in the wild-type embryo, the great majority of embryonic cells in dorsalized embryos differentiate as dorsal ectoderm and not as amnioserosa (Wharton et al., 1993). We and others had previously speculated that the absence of amnioserosa in dorsalized embryos could be due to the lack of the ‘potentiating’ ability of SOG upon BMP signaling (Neul and Ferguson, 1998; Nguyen et al., 1998).

During the course of this work, Ashe and Levine (Ashe and Levine, 1999) presented data that directly verified this hypothesis. They showed that embryos laid by dorsalized mutant mothers express Race, a dorsal marker, in a circumferential domain that is restricted to the anterior region of the embryo (Fig. 5A,B). They misexpressed SOG in this dorsalized background from the eve-stripe2 enhancer in a narrow stripe around the circumference of the embryo and observed both local inhibition of the Race transcript in the eve-stripe2 domain and ectopic expression of Race at a distance from the site of SOG expression (Ashe and Levine, 1999).

We confirmed these results by injecting dorsalized snk mutant embryos with a small bolus of sog mRNA (4 μg/μl) in a dorsoanterior position at a site that overlapped the domain of Race expression. After doubly staining these embryos for sog and Race mRNA expression, we found that sog both inhibited the expression of Race locally at the site of dorsoanterior injection (98%; n=102) and ectopically activated Race expression at a distance from the site of injection (80%; n=102: Fig. 5C,D).

Given that our screen identified second site mutations that affected SOG’s positive function, but none that affected its negative activity, we wondered whether the positive function of SOG was more sensitive to a reduction in the level of SOG protein. We tested this hypothesis by injecting a range of concentrations of sog mRNA into snk embryos (Table 2). At concentrations higher than in our original experiment (40 μg/μl), injected snk embryos did not express Race anywhere along the AP axis (Fig. 6A), suggesting that local inhibition of Race activity is dominant over long-range activation. However, at concentrations (0.4-0.2 μg/μl) lower than those that produced long-range activation, local inhibition of anterior Race transcription was still present (Fig. 6D). At even lower concentrations no inhibition was observed (Fig. 6E). These experiments indicate that the long-range positive activity of SOG requires a higher level of protein than does its local inhibition of BMP signaling.

Because the negative activity of SOG has been postulated to be directed toward the inhibition of SCW rather than DPP (Neul and Ferguson, 1998; Nguyen et al., 1998), we wished to determine whether the positive activity of SOG could be similarly ascribed to action on either ligand. Previous results had demonstrated that injection of high concentrations of dpp mRNA is sufficient to restore amnioserosa in all zygotic ventralizing mutants (Neul, 1998). The phenotype of most of these mutants can also be weakly suppressed by a doubling of the Mad¹² gene dosage (Ferguson and Anderson, 1992b) E. Ferguson unpublished results). However, unlike all other mutants tested, we found that the addition of only one extra copy of the Mad¹² gene restored large amounts of amnioserosa to a sog null embryo (Fig. 7D-F). We also found that an extra copy of Mad¹² partially restored the expression of Race in a sog mutant (Fig. 7A-C). Thus, sog mutants appear uniquely sensitive to small elevations in Mad¹² gene dosage.

We then tested whether elevation of SCW activity was sufficient to rescue a sog phenotype. While a low concentration of scw mRNA (27ng/μl) is sufficient to rescue the amnioserosa defect of a scw mutant embryo, injection of scw mRNA at a hundred-fold greater concentration was not sufficient to restore amnioserosa in a sog mutant (Fig. 8A,B). The lack of response of sog embryos to elevation of SCW activity, coupled with the extreme sensitivity of the sog mutant to elevation of Mad¹² gene dosage, suggests that the positive activity of SOG is exerted mainly, if not exclusively, upon DPP.

Chd and SOG are components of a conserved system for dorsoventral patterning in arthropods and vertebrates (Holley et al., 1995; Schmidt et al., 1995). Although the sequence similarity between SOG and Xenopus Chd is only 27% (François and Bier, 1995), the overall architecture of the two proteins is conserved. Both proteins contain four cysteine-rich domains (CR) that contain most of the sequence identity (e.g., 47% amino-acid identity in the first CR repeat). Because the CR repeats have been shown to bind BMPs (Larraín et al., 2000), we tested whether Chd showed a ligand specificity similar to that described for SOG (Neul and Ferguson, 1998; Nguyen et al., 1998). Using the assay developed by Neul and Ferguson (Neul and Ferguson, 1998), we co-injected chd mRNA with biologically equivalent concentrations of either scw or dpp mRNAs into scw mutant embryos and observed whether the co-injection of chd mRNA could block the activity of either ligand (i.e., restoration of amnioserosa in the scw mutant). We observed that injection of 4.5 μg/μl of chd mRNA completely abolished the activity of injected scw mRNA (Fig. 9A,C), but it did not block the activity of injected dpp mRNA (Fig. 9B,D). This result suggests that in Drosophila, Chd, like SOG, locally inhibits BMP signaling by abrogating the effects of the SCW ligand.

We then assayed whether Chd had a positive effect upon BMP signaling. We injected increasing amounts of chd mRNA (Table 3) into snk mutant embryos and stained for both chd and Race mRNAs (data not shown) or just for Race mRNA (Table 3). High amounts of chd mRNA (4.5 μg/μl) caused a strong inhibition of Race (Fig. 6F), abolishing expression throughout the dorsal side of the embryo; lower quantities (2.25 μg/μl) resulted in partial inhibition of dorsal Race staining (Fig. 6G), and concentrations below 1.5 μg/μl had no inhibitory effect (Fig. 6H). While the range of chd mRNA concentrations tested caused the same classes of inhibition of Race expression as did sog mRNA, we never observed ectopic Race at a distance from the site of injection in any chd-injected embryo. These results indicate that in this Drosophila assay Chd lacks the ability to act as a long-range enhancer of BMP signaling.

Morphogen gradients, once a purely theoretical concept, are now viewed as central players in the establishment of cell identity in a broad range of developmental processes. However, the exact biological mechanisms used to establish and maintain a morphogen gradient vary depending on the biological context. In the Drosophila embryo, while DPP can act in a dose-dependent fashion to specify different cell fates along the DV axis, in vivo its activity is modulated spatially by other components of the patterning system. In particular, SOG, a diffusible BMP-binding protein, has been shown to inhibit BMP signaling ventrally by preventing ligand access to the BMP receptors. In this paper we characterize a novel aspect of SOG’s function. Specifically, we demonstrate that SOG functions cell non-autonomously to elevate BMP signaling on the dorsal side of the embryo. Thus, the interpretation of any experiment to elucidate the role of SOG in the control of dorsoventral patterning must take into account the two apparently opposing functions of the protein.

We identified loss-of-function mutations in Mad or Medea as dominant enhancers of a weak homozygous-viable sog mutation, and showed that the enhanced embryos have defects in amnioserosa specification. Furthermore, we demonstrated synthetic lethality between weak homozygous-viable alleles of sog and zen, indicating that both are required for maximal production of amnioserosa. Lastly, we showed that there was a dramatic decrease in the level of zen transcription in sog^P129D embryos that were derived from Mad/+ females, compared to the level of zen transcription in either genotype alone. Taken together, these results unambiguously demonstrate that the positive action of SOG is exerted before gastrulation to attain the maximal expression of a direct BMP target gene.

In a series of elegant experiments, Ashe and Levine (Ashe and Levine, 1999) showed that expression of SOG in a specific anterior-posterior position in dorsalized embryos causes the inhibition of the dorsal-specific gene Race locally, and the upregulation of Race transcription at a distance from the site of expression, formally demonstrating the dual action of SOG upon a field of equipotent cells. However, Race is not necessary for amnioserosa specification, since sog^P129D embryos, which have functional amnioserosa, lack Race expression (Fig. 1C). Our results demonstrate that the positive activity of SOG is also exerted upon expression of zen, which is known to be required for amnioserosa specification. These results provide a direct link between SOG’s effect on zen transcription and loss of amnioserosa in sog mutants.

We confirmed the results of Ashe and Levine (1999) by mRNA injections, and used our ability to vary the amount of mRNA to demonstrate that the minimal amount of sog mRNA that must be injected to observe the ectopic transcription of Race is four-fold higher than the minimal amount necessary to locally inhibit Race transcription. Thus, a small decrease in the concentration of SOG affects the positive activity of SOG to a greater extent than it affects the negative function. This could be for any of a number of reasons, including a marked decrease in the concentration of SOG (or one of its proteolytic fragments) as it diffuses away from its site of synthesis.

These results also correlate well with our phenotypic and genetic analysis of sog^P129D, which causes a reduction in the level of sog transcription. Although this allele is homozygous viable, it appears to cause a preferential reduction in the positive activity of SOG, as evidenced by the loss of Race transcription in the amnioserosa. The preferential loss of positive activity in the sog^P129D mutant could also explain why we isolated second site mutations that decreased the positive function of SOG, but did not recover mutations in genes such as brinker (Jazwinska et al., 1999) that cooperate with SOG to repress BMP signaling ventrally.

We and others had previously suggested that the inhibitory function of SOG is primarily directed against the SCW ligand. We now present data that suggest that the positive function of SOG may be directed towards DPP. In particular, we show that a 50% increase in dpp copy number is sufficient to restore amnioserosa to sog mutant embryos, indicating that sog is more sensitive than any other known ventralizing mutation to an increase in Mad¹² gene dosage. We also demonstrated that the lack of amnioserosa in sog embryos was not rescued by injection of an amount of scw mRNA far in excess of that required to rescue a scw mutant. These results are strongly suggestive that SOG’s positive function may be directed against DPP, not against SCW.

These findings may allow us to clarify a series of recent results concerning the action of a second extracellular factor that modulates DPP activity, the product of the twisted gastrulation (tsg) gene. While the phenotype of tsg embryos, a partial ventralization caused by lack of amnioserosa, is suggestive of a positive activity of TSG upon BMP signaling, TSG has been shown to form a complex with SOG, and coexpression of TSG with SOG is sufficient to block DPP activity (Yu et al; 2000; Ross et al., 2001). Similar results have been demonstrated for the vertebrate homologs of the TSG and SOG proteins (Chang et al., 2001; Scott et al., 2001). These results have been primarily interpreted to suggest that, in vivo, SOG and TSG cooperate to block DPP signaling. In contrast, Oelgeschlager et al. (Oelgeschlager et al. 2000) suggest that the Xenopus homologue of TSG promotes BMP signaling primarily by antagonizing BMP binding to one of the cysteine rich domains (CR1) of Chd.

We would like to combine our data with the published results to reconcile the different interpretations of the biological functions of SOG and TSG. Specifically, we propose that the positive activity we have shown for SOG, which we have postulated is directed toward DPP, is in fact mediated by a tripartite complex composed of SOG, TSG and DPP. We suggest that this complex promotes BMP signaling by sequestering DPP from its receptor, thus allowing the dorsally directed diffusion of laterally produced DPP ligand. We suggest that the cleavage of SOG by TLD in the dorsal-most cells releases the DPP ligand from the tripartite complex, allowing it to signal. After cleavage of SOG by TLD, TSG could function in a manner similar to that described by Oelgeschlager et al. (Oelgeschlager et al., 2000), antagonizing the further binding of DPP by the proteolytic fragments of SOG. This model, and the one proposed by Ross et al. (Ross et al., 2001), are conceptually based on the model presented by Holley et al. (Holley et al., 1996), except that these models incorporate the recently described functions of TSG.

We propose that the antagonistic aspects of the complex toward DPP signaling that were observed through overexpression studies in various developmental contexts are due to the lack of, or improper stoichiometry of, a component present in the embryo that leads to disassociation of the tripartite complex. For example, if in these developmental contexts the TLD protease was not present in sufficient quantities to release the ligand from the tripartite complex, there would be continued DPP sequestration by the TSG/SOG complex leading to antagonism of BMP signaling. Alternatively, there may be one or more components present in the embryo whose function is to antagonize the stability of the tripartite complex, also potentiating ligand release. In that regard, it is interesting that mutations in the shrew gene cause a phenotype (lack of amnioserosa) that is similar to that caused by sog or tsg mutations. Possibly, shrew, the sequence of which has not been reported, could encode a component that aids in the formation or dissociation of the tripartite complex.

The conservation of the molecular mechanisms underlying the process of dorsoventral pattern formation between arthropods and chordates has allowed functional studies of vertebrate proteins to be carried out in Drosophila. SOG and Chd have been shown to be functionally interchangeable in their BMP inhibitory function (Holley et al., 1995). In this paper, we demonstrate that Chd, like SOG, preferentially inhibits SCW signaling, and under the assay conditions used is not capable of blocking DPP activity. This finding suggests that in vertebrates, as in flies, the activity of different ligands can have differential responsiveness to particular inhibitors. Although SCW does not have a known vertebrate ortholog, phylogenetic analyses have placed it in the 60A/BMP7 subgroup (Miya et al., 1997) or in a separate clade with mouse GDF3 (Newfeld et al., 1999). Furthermore, the likely receptor for SCW, SAX, has been placed in the same subfamily as the BMP7 receptor, ALK2, based in part on homology in the L45 loop that is critical in determining signaling specificity (Macias-Silva et al., 1998; Newfeld et al., 1999; Persson et al., 1998). Thus, we would propose that in vivo, Chd alone preferentially inhibits BMP7, not BMP2/4.

We wondered whether Chd, like SOG, would display a long-range positive activity. We found that although injection of Chd mRNA inhibits Race expression in the anterior domain of dorsalized embryos, it does not cause activation of Race transcription in the posterior domain. Thus, in Drosophila, Chd can inhibit BMP signals locally, but does not display long-range activation of BMP signaling.

If our model concerning the positive action of SOG is correct, Chd’s inability to promote a long-range elevation of BMP signaling could be for any of a variety of reasons. For example, data from Ashe and Levine (Ashe and Levine, 1999) indicated that wild-type levels of the metalloprotease TLD are required for the positive activity of SOG. If the positive activity of SOG is mediated by a proteolytic fragment of SOG, the differences in the in vitro patterns of cleavage of SOG and Chd (Marqués et al., 1997; Piccolo et al., 1997) could be critical for determining biological function. Another possibility is that Chd is unable to form sufficiently stable tripartite complexes with DPP and TSG to permit long-range diffusion. Alternatively, if Chd displays a higher affinity for DPP than does SOG, possibly sufficient DPP remains associated with the proteolytic fragments of Chd to prohibit long-range signaling. In support of this, Oelgeschlager et al. (Oelgeschlager et al., 2000) have proposed that one function of TSG is to remove BMP ligands from the proteolytic fragments of Chd. If either of the last two explanations were correct, elevation of TSG levels in the Drosophila embryo might permit Chd to display a positive activity. We therefore tested whether coinjection of tsg and chd mRNAs would result in positive long-range signaling. Although we injected a ten-fold concentration range of tsg mRNA with the chd mRNA, none of the embryos in our injection experiments displayed long-range activation of Race transcription (data not shown).

Although we cannot yet ascertain the mechanistic difference between SOG and Chd function, our results do provide an opportunity to determine the domains of SOG that are important for its positive activity. Interestingly, all known zygotic-lethal sog mutations are defective in both activities of the gene (E. Ferguson, unpublished), even though absence of its positive activity alone would have been sufficient to confer embryonic lethality. More generally, up to now there has been striking conservation of function of each individual component of the dorsoventral patterning system in arthropods and chordates. The identification of the basis of functional differences in apparently homologous proteins could provide insights into the degree of evolutionary divergence that can exist within the constraints of a conserved signaling system.

We thank Jannette Rusch for Race and zen cDNAs and Kavita Arora for scwpBS; Nipam Patel for advice and use of equipment for photomicroscopy; and the Patel laboratory and Timothy Sanders for help with in situ hybridizations. We also thank Michele Markstein, Jeff Neul and Steve Podos for discussions, and the members of the Ferguson laboratory for critical reading of the manuscript. This work was supported by a grant from the National Institutes of Health (GM50838). E. D. was a William B. Graham Fellow (Baxter Foundation).