An activity gradient of decapentaplegic is necessary for the specification of dorsal pattern elements in the Drosophila embryo

Within the past two centuries, two related models have been proposed as mechanistic solutions to the problem of organization and specification of cells in embryonic fields (Slack, 1991). According to the localized determinant model, embryonic pattern is established through the action of discrete morphogenetic substances that are localized to different regions of the egg cortex. During cleavage, these determinants are partitioned into different blastomeres, which consequently assume different states of differentiation. According to the gradient model, pattern is established by one or a few multifunctional regulatory substances, called morphogens, that are distributed in monotonic gradients. These morphogens specify cell fate in a concentration-dependent fashion such that cells acquire different fates based on the concentration of morphogen that they contain (Crick, 1970).

Embryological, genetic and molecular studies on Drosophila have provided the first concrete evidence for the gradient model. Two maternal morphogen gradients have been identified that establish aspects of the anterior-posterior and dorsal-ventral axes (reviewed by Nüsslein-Volhard, 1991). Specifically, along the anterior-posterior axis, the bicoid (bcd) gene product is distributed in a monotonic gradient with a high point at the anterior end of the embryo (Driever and Nüsslein-Volhard, 1988a,b). Along the dorsal-ventral axis, the dorsal (dl) gene product is dif-ferentially compartmentalized into a steep gradient with a high point ventrally (Steward et al., 1988; Roth et al., 1989; Rushlow et al., 1989; Steward, 1989).

The dl gradient specifies positional information along the dorsal-ventral axis by the differential activation and repression of downstream zygotic genes (Roth et al., 1989; Ray et al., 1991). dl directs the differentiation of ventral tissues by the activation of the zygotic genes twist (twi) and snail (sna), which are required for specification of the ventrally derived mesoderm (Simpson, 1983; Thisse et al., 1987). Conversely, dl directs the differentiation of dorsal tissues by repressing the expression of dpp and zerknüllt (zen) ventrally, thereby restricting the activities of these genes to the dorsal regions of the embryo where they are required for the specification of the dorsal ectoderm and amnioserosa (Irish and Gelbart, 1987; Wakimoto et al., 1984).

Significantly, the dl-directed activation and repression of these zygotic genes is restricted to the ventral regions of the embryo. Thus, while the dl morphogen may directly specify a number of different cell states in the ventral part of the embryo, it is apparently not involved in the specification of cell states within the dorsal part of the embryo. This is plainly evidenced by the fact that dorsal pattern elements, specifically dorsal epidermis and amnioserosa, are observed in embryos that completely lack dl function (Ray et al., 1991). Therefore, since the dl gradient appears to be the only maternal system defining dorsal-ventral polarity (Nüsslein-Volhard, 1991), it follows that the specification of these dorsal pattern elements must be carried out by downstream zygotic genes.

Based on mutant phenotype, seven zygotic loci have been identified that are likely to be involved in the specification of dorsal pattern elements: dpp, screw (scw), short gastru - lation (sog), shrew (srw), tolloid (tld), twisted gastrulation (tsg), and zen (reviewed by Rushlow and Arora, 1990). Loss-of-function alleles of these genes produce a variety of related mutant phenotypes, consisting, to a greater or lesser degree, of an expansion of lateral or ventral structures at the expense of more dorsal ones. For example, embryos bearing loss-of-function alleles of sog, srw and tsg show an expansion of the dorsal-most part of the dorsal epidermis at the expense of the amnioserosa. Similarly, embryos bearing loss-of-function alleles of scw and tld show an expansion of lateral aspects of the dorsal epidermis at the expense of the rest of the dorsal epidermis and amnioserosa. Null alleles of dpp produce the most severe mutant phenotype, in which ventral pattern elements are expanded at the expense of all dorsal pattern elements (Irish and Gelbart, 1987; Ray et al., 1991; Ferguson and Anderson, 1992; Arora and Nüsslein-Volhard, 1992).

Of these seven zygotic genes, dpp is unique, not only because of the severity of its mutant phenotype, which is reminiscent of the ventralized phenotypes produced by the maternal genes, but also because of the nature of its interactions with the other zygotic genes. For instance, the initial expression pattern of dpp in the dorsal-most 40% of the blastoderm embryo is autonomous with respect to the other genes. That is, scw, srw, sog, tld, tsg, and zen mutations do not affect the initiation or maintenance of dpp gene expression. By contrast, the expression of at least two of these genes, zen and tld, depends on dpp activity (Ray et al., 1991; R. Ray, unpublished data). These findings suggest that the genes of this group participate in a single pattern forming system, in which dpp is a central element.

In this report, we provide support for this proposal through a genetic and molecular analysis of dpp function in early embryonic development. Our analysis indicates that dpp is an integral part of a zygotic activity gradient that differentially specifies cell fate within the dorsal ectoderm. We show that the dorsal-ventral requirement for dpp is highly dosage sensitive, and that there is a graded requirement for the dpp product in the early embryo. Cells fated to the amnioserosa are most sensitive to reduction in dpp activity; cells fated to the dorsal epidermis are less sensitive; and cells fated to the ventral epidermis are the least sensitive. Furthermore, we show that the fate map of the embryo along the dorsal-ventral axis responds to decreases and increases in the level of dpp activity. Finally, we present evidence that the dl and dpp gradients control different aspects of the dorsal-ventral pattern, and that they elaborate dorsal and ventral pattern elements from a common, central ground state.

Drosophila were cultured on standard cornmeal yeast extract dextrose medium at 25°C. In all cases, the strain referred to in the text as ‘wild type’ is Oregon-R. In the experiments reported here, ten ethyl methyl sulfonate (EMS) induced, and two γ-ray induced dpp alleles were used. The ten EMS induced alleles are: dpp^hr4, and dpp^hr27, which were generated in an F₂ lethal screen based on failure to complement the dpp disk alleles dpp^d2 and dpp^d14, respectively (Spencer et al., 1982); dpp^e87, which was isolated in an F₂ lethal screen based on failure to complement the embryonic lethal dpp allele dpp^hr4 (Irish and Gelbart, 1987); dpp^e90, formerly dpp^IC, which was isolated in an F₂ lethal screen as a mutation that affected the pattern of the larval cuticle (Nüsslein-Volhard et al., 1984); and dpp^hr92, dpp^hr93, dpp^H91, dpp^H94, dpp^H95, and dpp^H96, which were isolated as dpp alleles that failed to complement a transvection-sensitive chromosomal rearrangement of the disk mutation dpp^d-ho (R. Blackman and W. Gelbart, unpublished data). The two γ-ray induced alleles, dpp^hr56 and dpp^H46, were isolated in an F₁ screen based on failure to complement a rearrangement of dpp^d-ho (Irish and Gelbart, 1987). A synthetic duplication of dpp, CyO23, was constructed by jumping a P-element rescue construct, P20 (R. Padgett, unpublished data), onto the balancer chromosome CyO. This new balancer chromosome completely rescues the haplolethality associated with null alleles of the dpp locus. The dpp stocks used in the experiments described in this report are as follows: dpp^e87 (V. Twombly); S dpp^e90cn/CyO (C. Nüsslein-Volhard), net dpp^hr4/CyO, dpp^hr4cn bw/CyO; net dpp^hr27/CyO; dpp^hr27cn bw/CyO; dpp^hr56Sp cn bw/CyO; dpp^hr56cn bw/CyO; dpp^hr92L/CyO; dpp^hr92cn bw/CyO; dpp^hr93L/CyO; dpp^hr93cn bw/CyO; dpp^H94L/CyO23; dpp^H94cn bw/CyO23; dpp^H95L/CyO23; dpp^H95cn bw/CyO23; dpp^H96L/CyO23; dpp^H96cn bw/CyO23; dpp^H91L/CyO23; dpp^H91cn bw/CyO23; dpp^H37cn bw/CyO23; dpp^H46Sp cn bw/CyO23; Dp(2;2)DTD48 (4 copies of dpp), Dp(2;2)B16 dp cl cn bw (4 copies of dpp, V. Twombly); Dp(2;2)B16 dp cl dl¹cn sca/CyO; dpp^H46Sp dl¹cn sca/CyO23.

To test for the dominant lethality of dpp mutations, males bearing the mutant alleles were crossed to virgin Oregon-R females. The genotype of the males was either Dp(2;1)G146, dpp⁺/Y; dpp*/CyO, or dpp*/CyO, or CyO23. Flies were reared in small cages constructed by inverting a 150 ml tripour beaker (VWR) onto a 60 mm Petri dish (Falcon) filled with grape juice agar. Embryos were collected every 12 hours and individually laid in a grid pattern on grape juice plates. The grids were scored after 24 hours and the number of hatched and phenotypically mutant embryos were recorded. More than 300 embryos (n) were scored for every genotype in multiple, independent experiments. The numbers from all experiments were pooled, and the percentage lethality was determined. In order to make the embryonic data comparable with the adult data (described below) we devised the statistic ‘percentage lethality of expected’, that was calculated by dividing the number of phenotypically mutant embryos by half of the total number of embryos (n/2).

To test for the dominant lethality of the alleles as determined by adult survivorship, crosses identical to those performed for the embryonic viability studies were reared in bottles. Following eclosion, the adults were scored and the number of Cy and Cy⁺ flies were recorded. To determine the statistic ‘percentage lethality’, the number of Cy⁺ flies was subtracted from the number of Cy flies, and this difference, representing the predicted number of individuals that did not survive to adulthood, was divided by the number of Cy progeny. This approach necessarily introduces a bias if the number of survivors is very small, thus as the severity of the genotype approaches that of dpp^null/+ (haploinsufficient), the statistic often reaches or exceeds the theoretical maximum of 100%. In such cases, the percentage survivorship is indicated in the text to illustrate this fact.

Cuticle preparations were done as follows: mutant embryos were picked from a grape plate after having aged for 36 hours at 25°C. They were dechorionated in 50% Chlorox (sodium hypochlorite) for 3 minutes in wire mesh baskets and then transferred to eppendorf tubes containing 1 ml of 0.1% Triton X-100. Dechorionated embryos were washed twice with Triton X-100 and then twice with 1:9, methanol: 50 mM EGTA (ME). After ME washes, 1 ml of fresh ME was added, followed by 0.5 ml n-heptane. The tubes were inverted until the devitellinized embryos sank to the bottom, after which the methanol and heptane were removed and replaced with 200 μl 1:4, glycerol: acetic acid. The tubes were then incubated at 60 °C for 1 hour. The cuticles were mounted under a coverslip in a small drop of Hoyer’s medium (Wieschaus and Nüsslein-Volhard, 1986). The slides were returned to 60°C for 24 hours, and a 100 g weight was placed on the coverslip. Cuticles were analyzed and photographed under phase contrast optics using an Olympus stereomicroscope and Kodak Tech Pan Film.

Embryos were removed from the grape juice agar plates, and then dechorionated in 50% Chlorox in wire mesh baskets and washed extensively with water. The dechorionated embryos were transferred to a glass scintillation vial containing 2 ml of 3.7% formaldehyde (Mallinckrodt) and 2 ml of n-heptane, and fixed for 15 minutes with constant agitation. After fixation, the lower phase (formaldehyde) was removed with a pasteur pipette and the embryos were washed twice with fresh heptane. Then, 3 ml of ME (see above) were added, and the vials were shaken until the devitellinized embryos sank to the bottom. The embryos were collected and washed in ME, 1:1 ME: × PBS (130 mM NaCl, 7 mM Na ₂HPO₄ 7H₂O, 3 mM NaH₂PO₄ H₂O), and × PBS for 10 minutes each. The tissue was blocked for 1 hour in PBT (× PBS; 1% bovine serum albumin fraction V, Boehringer Mannheim; 0.1% Triton X-100, Boehringer Mannheim), and exposed to the primary antibody (anti-Kr, kindly provided by C. Rushlow) for 1 hour at room temperature or overnight at 4°C. The embryos were washed in PBT for 1.5 hours with four changes of buffer, and then exposed to the secondary antibody (peroxidase-conjugated goat anti-rabbit, Vector) using the same conditions as described above. After the secondary labelling, the embryos were washed for 1.5 hours with four changes of buffer. DAB staining solution (500 μg/ml DAB in PBT, 0.2% H₂O₂) was added, and the reaction was allowed to proceed for 10 minutes. Embryos were washed 3 times with PBT to stop the reaction, and then dehydrated with ethanol. They were mounted in methyl salicylate and photographed using a Olympus stereomicroscope and Kodak Tech Pan film.

The dorsal-ventral patterning defect observed in dpp mutant embryos represents the first requirement for this gene during the development of the fly (St. Johnston et al., 1990). All embryos homozygous for a null dpp allele are 100% lethal, while embryos heterozygous for this same allele are approx. 95% lethal. This haploinsufficiency, or inability to survive as a heterozygote, demonstrates the sensitivity of the animal to the level of the dpp product. More than 30 embryonic lethal dpp mutations have been isolated that affect the dorsal-ventral pattern of the embryo (Spencer et al., 1982; Irish and Gelbart, 1987). The majority of these alleles behave as nulls, as they show a high percentage of dominant lethality when heterozygous (dpp*/+). However, some of these alleles display a lower percentage of dominant lethality and represent a class of hypomorphic, or partial loss-of-function, alleles (Irish and Gelbart, 1987). We have used these hypomorphic alleles to determine the nature of the requirement for dpp in the early embryo.

To determine the relative severity of these hypomorphic dpp alleles, we have assayed for the level of dominant lethality specific to each allele and compared this to the level characteristic of null dpp alleles. In these experiments, the viability of each allele as a heterozygote in trans to wild type (dpp*/+) was determined both in terms of the number of progeny that died as embryos, and the number of progeny that survived to adulthood (Fig. 1A,B). We have used the dpp^H46 allele as a standard null, since it is known to be deficient for a significant portion of the dpp coding region (St. Johnston et al., 1990; Table 1). As a heterozygote, dpp^H46 shows 98% dominant lethality in embryos, and 99.8% lethality in adults, consistent with the level expected for a null allele. Two of the other alleles we have tested, dpp^H91 and dpp^H96, exhibit similar levels of dominant lethality in both embryos and adults, and thus are not distinguishable from nulls. The weakest allele we have identified is dpp^e87, which exhibits no dominant lethality (Fig. 1A,B) and is, in fact, 60% viable as a homozygote (Fig. 1C, Table 1). Two other alleles, dpp^e90 and dpp^hr56, while 100% lethal as homozygotes, are nevertheless about 30% viable in trans to dpp^e87 (Fig. 1C), and thus are also relatively weak alleles. Each of the remaining alleles tested exhibit a distinct level of dominant lethality which lies in between the level observed for dpp^e87 and that observed for dpp^H46. Thus, these dpp alleles describe a simple allelic series. Based on these data, the order we have observed for this allelic series from weakest to most severe is: dpp^e87 <dpp^e90 <dpp^hr56 <dpp^hr4 <dpp^hr92 <dpp^hr27 <dpp^hr93 <dpp^H94 <dpp^H95 <[dpp^H96=dpp^H91 =dpp^H46].

Based on these viability studies, we have arbitrarily designated those alleles that exhibit >50% dominant lethality dpp^H, and those that exhibit <50% lethality as dpp^hr alle-les (Table 1). For simplicity, in this report we consider the designation dpp^H46 to be equivalent to dpp^null. The two weak alleles, dpp^e87 and dpp^e90, have been given a special designation based on their complementation behavior with other dpp alleles, as described elsewhere (Spencer et al., 1982).

The haploinsufficient nature of the dpp locus indicates that in a large percentage of dpp^H46/+ embryos, the single remaining copy of the dpp gene does not produce enough functional dpp product to meet the early dorsal-ventral requirement. Similarly, the lower levels of dominant lethality we observe for the hypomorphic alleles suggest that some of the embryos heterozygous for these alleles do not produce enough functional dpp product but that a significant portion of embryos must produce just enough product to survive. However, in the case of the hypomorphic alle-les, the loss of dpp function does not appear to be due to a difference in the absolute level of protein (as was true for the null) or in the spatial regulation of dpp transcription. Our molecular studies indicate that at least six of these hypomorphic alleles result in single amino acid changes in the dpp protein sequence (Wharton et al., unpublished data), and thus they are not likely to affect the levels of dpp transcription. Furthermore, in whole-mount in situ hybridizations of mutant embryos, we have not observed any changes in the initial expression pattern, even in the null. Therefore, it is very likely that such alleles result in mutant dpp proteins of differing activities and not simply in different concentrations of the dpp product. Thus, we must conclude that the different levels of dominant lethality reflect different levels of dpp protein activity.

To address the role of dpp in the patterning of the dorsal ectoderm, we have examined the cuticular phenotypes of embryos mutant for the alleles discussed above as homozygotes and as transheterozyogotes with other dpp alleles. Cuticle preparations representative of these data are shown in Figs 2, 3 and 4, and an analysis of the data is presented in Table 2.

The weakest dpp mutant phenotypes that we observe are produced by embryos bearing homo- and heteroallelic combinations of those alleles that show the lowest percentages of dominant lethality, specifically, dpp^e87, dpp^e90, and dpp^hr56 (see above). In addition, these same phenotypes are produced by embryos that die due to the dominant lethality associated with the stronger alleles. Thus, the mutant phenotypes resulting from dpp genotypes that exhibit dominant lethality (i.e. heterozygotes of the form dpp*/+) are similar to those that are produced by embryos homozygous for the weak alleles.

The cuticular phenotypes characteristic of these allelic combinations consist of progressive deletion of the anterior-dorsal cuticular structures of the head, and invagination of the telson and posterior abdominal segments. In the head region, the labral sense organ, dorsal bridge and dorsal arms, which map to the dorsal anterior tip of the blastoderm embryo (Jürgens et al., 1986), are the first structures to be affected, and the only structures that are absent in embryos homozygous for dpp^e87 and dpp^e90 (Fig. 4B). No alterations in tail morphology are observed in these cuticles (Fig. 2B,C). In dpp^e87/dpp^e90transheterozygotes and dpp^hr56 homozygotes (Fig. 2D,E), deletions in the head region include structures that map to more ventral positions on the blastoderm fate map. Specifically, in addition to those structures absent from dpp^e87 and dpp^e90 homozygotes, these embryos also show defects in the vertical plates and the anterior portions of the ventral arms (Fig. 4C). Furthermore, these embryos show progressive changes in the morphology of the tail region. In dpp^e87/dpp^e90 embryos, the telson and posterior abdominal segments are often curled onto the dorsal side of the embryo as if the germ band had failed to retract completely (Fig. 2D). In dpp^hr56 embryos the defect is more severe, and the entire telson and eighth abdominal segment are found inside the embryo beneath the fifth and sixth abdominal segments (Fig. 4G).

Many of these weaker genotypes produce a continuum of mutant phenotypes. This is particularly true for those genotypes that produce both wild-type and mutant progeny. For example, 60% of the embryos that are homozygous for dpp^e87 survive to adulthood, and 40% die as embryos (Fig. 1C). Of the 40% that die as embryos, some produce a cuticle like the one shown in Fig. 2C, that lacks most of the dorsal elements of the cephalopharyngeal skeleton, while others produce a cuticle that lacks only the labral sense organ and a small part of the dorsal bridge, and others still that lack only the labral sense organ (cf. Fig. 2C, 4B). Similarly, 80% of embryos of the genotype dpp^hr92/+ survive to adulthood, and 20% die as embryos (Fig. 1A,B). Of the 20% that die, some produce a cuticle like that shown in Fig. 2D (the most severe mutant phenotype observed for dpp^hr92/+), while others produce cuticles like those shown in Fig. 2B,C (data not shown). Finally, only 5% of the embryos of the genotype dpp^H46/+ (haploinsufficient) survive to adulthood, and 95% die as embryos (Fig. 1A,B). Of the 95% that die as embryos, some produce a cuticle similar to that produced by dpp^hr56 homozygotes (Fig. 2F, cf. Fig. 2E), while others produce cuticles like those shown in Fig. 2C,D (for example, Fig. 2B). Thus, each of these genotypes produces a continuum of phenotypes that are representative of all or a portion of the range of phenotypes characteristic of homo- and heteroallelic combinations of the weak alleles.

A careful analysis of the phenotypes produced by embryos mutant for weak alleles, or those that are hap-loinsufficient, revealed that the progressive change in tail region morphology that we observed in the allelic series was not accompanied by the reduction or loss of any posterior cuticular structures. In light of this, we suspected that this aspect of the phenotype might be a consequence of defects elsewhere in the embryo. In particular, the pheno-type of dpp^e87/dpp^e90 embryos (Fig. 2D), which is indicative of a defect in germ band extension or retraction, suggested the possibility that the tail defect could be due to a progressive deletion of the dorsally derived amnioserosa. To test this hypothesis, we immunostained mutant embryos using an antibody directed against the Krüppel protein (Kr), which accumulates in amnioserosa cells shortly after gastrulation (Gaul et al., 1987). We find that the internalization of the posterior structures can be directly correlated with the progressive loss of amnioserosa cells. This result is best illustrated in haploinsufficient embryos of the genotype dpp^H46/+. Consistent with the variability characteristic of this genotype, we find embryos that have anything from 0 to 150 amnioserosa cells (Fig. 5). Furthermore, based on the morphology of these embryos, it appears that this reduction in the number of amnioserosa cells specifically results in the invagination of the posterior segments and telson. We have done similar immunostainings on embryos mutant for the weak hypomorphic alleles and we find that the more severe the allele or allelic combination, the fewer amnioserosa cells are found. Thus, in embryos mutant for dpp^e87 and dpp^e90, the number of amnioserosa cells is essentially the same as that observed in wild-type controls (about 200). In the heteroallelic combinations between dpp^e87, dpp^e90, and dpp^hr56, the number of amnioserosa cells ranges from 20 to 100 cells. In dpp^hr56 homozygotes few if any amnioserosa cells are observed (data not shown).

Taken together, the above results indicate that the weaker hypomorphic dpp alleles manifest progressive deletions of dorsally derived structures: weak alleles, like dpp^e87, only show deletions of anterior dorsal regions of the head; more severe allelic combinations like dpp^e87/dpp^hr90, and dpp^hr56, result in further deletions of head structures and progressive deletion of the dorsally derived amnioserosa. Thus, the weakest dpp alleles affect only the dorsal-most pattern elements on the blastoderm fate map.

The moderate and severe hypomorphic dpp alleles show further deletions in the head, as well as a progressive expansion of the ventral denticle belts which was not observed in the cuticles of the weak alleles. In addition, deletions of the structures in the tail region are observed (Table 2, Figs 2,3). There are several distinct transition points within this part of the allelic series that are evident due to the loss of prominent cuticular structures. The first is evidenced by a radical change in the morphology of the head and thorax. This is illustrated by the change in morphology between the cuticles of dpp^H46/+ and those of dpp^H46/dpp^e87 and dpp^hr4 (Fig. 2G,H). In this transition, the head loses most of the remaining cephalopharyngeal skeleton, in particular the ventral arms and pharnyx walls, and becomes globular. In fact, the only scorable head structures present in these embryos are the mouth hooks and antennal sense organs (Fig 4D). A characteristic defect associated with these changes is a prominent constriction in the first thoracic seg-ment which essentially obscures all structures in this region of the cuticle (see Fig. 2G,H; Table 2).

The next transition occurs in the range of phenotypes produced by embryos homozygous for dpp^hr4, dpp^hr27, dpp^hr92, and heteroallelic combinations thereof. In this transition progressive loss of the posteriorly located Filzkörper is observed. In homozygous dpp^hr4 embryos, at least one normally differentiated Filzkörper is observed in every case (Fig. 2H); in homozygous dpp^hr92 embryos, the Filzkörper are either normally differentiated as in dpp^hr4 (rare), irregular or partially deleted (Fig 4H), or entirely absent (Fig. 3A); in homozygous dpp^hr27 embryos, small remnants of the Filzkörper are occasionally observed (Fig. 4I), but in general they are entirely absent (Fig. 3B). In addition to this defect, further degeneration of the mouth hooks and antennal sense organs is observed. The change here is particularly interesting as the pairs of mouth hooks and antennal sense organs are not lost, but rather fuse to form a single irregular mouth hook and antennal sense organ (Fig. 4E). These fused structures are present as such in all of the more severe genotypes discussed below.

Subsequent to the loss of the Filzkörper and the fusion of the mouth hooks and antennal sense organs, the most obvious transition that is observed in the cuticles is the progressive expansion of the ventral denticle belts. In embryos homozygous for dpp^hr93 and dpp^H94, and in transheterozygous combinations with dpp^H46, the tips of the denticle belts in the most anterior and most posterior segments are observed to meet at the dorsal side of the embryo (Fig. 3C,D,E,F) indicating a complete loss of dorsal structures in these regions. The tips of the denticle belts in the midabdominal segments are expanded, but in general, they do not completely encircle the embryo and a small patch of dorsal tissue remains on the dorsal side (Fig. 3D,E,F). In the more severe genotypes like dpp^H95/dpp^hr93, dpp^H95/dpp^H94 and dpp^H95/dpp^H91 further expansion of the ventral denticle belts is evidenced by an increase in the width of the denticles that reach the dorsal side of the embryo (Fig. 3G). Finally, in dpp^H46 homozygotes (complete null) no dorsal structures are observed, and the ventral denticle belts encircle the entire embryo (Fig. 3H, Irish and Gelbart, 1987). The most ventral pattern elements in the tail region of the embryo, the anal pads (Jürgens, 1987), are observed in all mutant embryos except the null (Fig. 3H; Table 2).

From our analysis of the phenotypic manifestations of these 12 hypomorphic dpp alleles, we draw the following con-clusions. First, the different levels of dpp activity that we defined based on the relative percentage of dominant lethality give rise to distinct mutant phenotypes. Second, the severity of the mutant phenotype characteristic of a particular allele is inversely proportional to the level of residual dpp activity it is thought to produce. Thus, dpp alleles that exhibit little or no dominant lethality (i.e. that have the highest level of residual activity) give rise to the mildest mutant phenotypes consisting of progressive deletions of the dorsally derived amnioserosa. Furthermore, those dpp alleles that exhibit intermediate percentages of dominant lethality, i.e. dpp^hr4, dpp^hr27, dpp^hr92, and dpp^hr93, give rise to more severe mutant phenotypes which consist of a complete loss of the amnioserosa and the progressive loss of the dorsal ectoderm. Finally, those alleles that exhibit high percentages of dominant lethality similar to the null alleles, give rise to mutant phenotypes which consist of near complete or complete loss of all dorsal structures (Fig. 6). Similar progressive dorsal-to-ventral deletions of cuticle structures are observed in the head and tail regions of the embryo as discussed above.

In light of this relationship between the activity levels, as defined by the dominant lethality assay, and the cuticular phenotypes, it is evident that there is a graded requirement for dpp activity in the early embryo. Peak levels of dpp activity are required for initial specification of the dorsally derived amnioserosa; intermediate levels are required for specification of the dorsal epidermis, and low levels are required for the specification of the ventral epidermis. Furthermore, the progressive nature of the phenotypes we observe, i.e. the progressive loss of the amnioserosa, and the progressive loss of the head structures and Filzkörper, as well as the fact that the range of phenotypes characterisitc of the different mutant genotypes overlap to such a great extent, imply that the graded requirement for dpp activity is continuous or nearly so. Thus, from these data, it is apparent that, like the ventral pattern elements, dorsal pattern elements in the Drosophila embryo are specified by a gradient. However, unlike the ventral pattern elements, which are specified by the maternal dl gradient, the dorsal pattern elements appear to be specified by a zygotic gradient system involving the dpp gene product.

Significantly, the range of mutant phenotypes that we have observed for dpp mutations alone encompasses the entire repertoire of phenotypes characteristic of all of the other zygotic genes that affect the dorsal-ventral pattern. The mildest phenotypes we observed are similar to those produced by embryos bearing loss-of-function alleles of sog, srw, tsg, and zen, while the more moderate dpp mutant phenotypes are representative of those produced by tld and scw alleles. In particular, sog, srw, and tsg embryos produce mutant phenotypes which are similar to those produced by dpp^e87 and dpp^e90 homozygotes and trans - heterozygotes (Fig. 2C,D). Null alleles of zen (e.g. zen^w36) produce mutant phenotypes virtually identical to that produced by dpp^hr56 homozygotes (Fig. 2E). Finally, tld and scw embryos exhibit mutant phenotypes similar to those produced by embryos homozygous for dpp^hr4, dpp^hr27, and dpp^hr92(Fig. 3A,B). This similarity of the hypomorphic dpp mutant phenotypes and those of the other zygotic genes, as well as the demonstration that tld, srw, scw and sog show genetic interactions with the dpp gene (Ferguson and Anderson, 1992; V. Twombly, personal communication), strongly suggest that the gradient system that we have defined based on the loss-of-function mutant phenotypes of dpp must involve these other zygotic genes as well.

In light of the fact that these other gene products are likely to be involved in the gradient system defining dorsal pattern elements in the embryo, the nature of loss-of-function mutations, while clearly indicating the presence of a gradient, does not distinguish the central (i.e. graded) element in the system. That is, in a multicomponent gradient system like the one considered here, any element that affects the efficacy of the system as a whole will produce a graded allelic series like that we have described for dpp (and, in fact, such a series has been described for tld; Ferguson and Anderson, 1992). Thus, a graded series of mutant phenotypes only suggests that the particular gene product contributes to the gradient system as a whole; it does not imply that it is the graded element. In order to define the key element in a gradient system, it is necessary to define which element in the system is limiting, since this must, by definition, be the graded element. Given that any product already in excess is unlikely to have an effect on the system when present in even greater excess, the simplest demonstration that a particular element is limiting is to determine whether or not gain-of-function phenotypes are consistent with a general increase in the activity of the gradient. In the next section we address this issue.

As we have previously stated, dpp is a dosage sensitive gene. A deviation from the normal diploid dose of dpp⁺ is sufficient to induce a shift in the fate map of the early embryo along the dorsal-ventral axis, such that cell fates that map to the dorsal regions of the embryo are lost and are replaced by more ventral fates. In order to characterize the effects of increasing dpp dosage in the wild-type embryo, we have employed two insertional duplications of dpp, Dp(2;2)DTD48 and Dp(2;2)B16 (see Materials and methods). Since reducing dpp gene dosage affected the dor-sally derived amnioserosa, we expected that increasing the gene dosage would also affect this tissue. So, we have used the Kr antibody to detect the domain of amnioserosa cells.

We have found that increasing the dpp gene dosage can cause shifts in the fate map along the dorsal-ventral axis. In wild-type embryos bearing two copies of the dpp gene (i.e. disomic) we have observed an average of 130 (n=36) amnioserosa cells (Fig. 7A). When the gene dosage is increased from two to four doses (i.e. tetrasomic), we find that the average number of amnioserosa cells increases from 130 to 325 (n=36), and in exceptional embryos the number can reach 400 or more (Fig. 7B). Thus, by doubling the dpp gene dosage, we observe 2.5× the normal number of amnioserosa cells (on average). Notably, despite this dramatic change in the dorsal-ventral fate map, embryos tetrasomic for dpp are adult viable and exhibit no mutant phenotype. This fact implies a certain degree of autonomy of the more ventral fates (which are presumably specified by the dl gradient) as they are not simply overrun by the expansion of the dorsal pattern elements. We suspected that this ventral patterning system might be limiting the expansion of the dorsal fates, so we repeated the above experiment in embryos derived from dl mutant mothers. These embryos lack the ventral structures of the wild-type embryo. dl^- embryos that are monosomic for dpp (i.e. dpp^H46/+) completely lack amnioserosa cells (Fig. 8A), consistent with the phenotype we have observed in dl⁺ embryos that are monosomic for dpp (Fig. 5E). In dl embryos that are disomic for dpp (+/+) we observed 20-200 amnioserosa cells, as has been previously described (Fig. 8B; Ray et al., 1991). If the dpp^- gene dosage is increased further, progressively more amnioserosa cells are observed. In dl embryos trisomic for dpp, we observed 500-1000 amnioserosa cells (Fig. 8C), and in those tetrasomic for dpp, 2000 or more (Fig. 8D). Thus, increasing the dosage of dpp in the context of an apolar dl^- embryo completely transforms the segmented region of the embryo into amnioserosa.

From these results, it is clear that increasing the dosage of the dpp gene, and thus presumably the level of dpp activity in the embryo, alters the distribution of fates along the dorsal-ventral axis. Increasing the dpp gene dosage in the context of the wild-type embryo produces a corresponding expansion of the dorsally derived amnioserosa, and thus must result in a ventral shift in the fate map of the embryo. Increasing the dpp gene dosage in the context of an apolar dl^- embryo results in the transformation of a uniform field of dorsal epidermal cells into amnioserosa cells. Both of these experiments imply that cell fate in the embryo is very sensitive to the level of dpp activity, and that the dpp gene product has the potential to specify at least two different cell fates along the dorsal-ventral axis. In similar experiments, in which the dosage of dpp is reduced in the context of an apolar dl^- embryo by the introduction of loss-of-function dpp alleles, we find that a number of other cell types characteristic of dorsal and lateral ectoderm are produced (Fig. 9, data not shown). Thus, just as the hypomorphic series in the context of the wild-type embryo shows progressive, nested deletions of dorsal and lateral structures (Fig. 6), a hypomorphic series can be generated in the context of an apolar embryo that shows an ordered series of different cell states from dorsal to ventral as the level of dpp activity is reduced. We conclude that it is the activity of the dpp gene product that is graded along the dorsal-ventral axis, and that the gradient that defines pattern within the dorsal regions of the embryo is, in fact, a dpp gradient.

Along the anterior-posterior axis, distinct regions of the body plan are determined largely independently by three maternal patterning systems. The anterior system is required for specification of the head and thorax, the posterior system for the abdomen and the terminal system for the unsegmented acron and telson. The independence of these systems is illustrated by the general additivity of the mutant phenotypes. If the activity of any two of the systems is disrupted, the resulting embryo shows a partial pattern that reflects the function of the remaining system. In the absence of all three systems no apparent pattern is formed (Nüsslein-Volhard, 1991). On the molecular level, the lack of all pattern in the cuticle of the triple mutant is a reflection of the uniform expression of the zygotic gap gene Krüppel along the entire anterior-posterior axis (Lehmann and Frohnhöfer, 1989). In light of this regulation, it has been suggested that the expression of Kr defines a central ground state from which the three maternal systems elaborate the anterior-posterior pattern.

In light of the results we have presented in this report, it would appear that the dorsal-ventral axis is specified by at least two gradients, one maternal and one zygotic, which are quite dependent, as discussed below. Nevertheless, like the anterior, posterior and terminal systems, the dl and dpp systems appear to control different aspects of the dorsal-ventral pattern. The dl gene product defines the ventral pattern elements by the activation of zygotic genes that are expressed ventrally and the repression of genes that are expressed dorsally, and the dpp system defines dorsal pattern elements via an intercellular signalling mechanism that has yet to be defined (see Discussion). However, do these two systems define the entire dorsal-ventral pattern? By analogy with the anterior-posterior system, this question should be answered by the phenotype of embryos that lack the function of both dpp and dl. Such embryos can be generated, and have been generated (Irish and Gelbart, 1987), and an example of such an embryo is shown in Fig. 9A. Embryos lacking the function of both dpp and dl are lateralized. Thus, as regards the cuticular phenotype, the dpp and dl systems do not account for all of the dorsal-ventral pattern elements. However, if we consider the molecular basis for the dpp, dl and dpp,dl double-mutant phenotypes, this result is, in fact, in keeping with that of the triple mutant in the anterior-posterior system. First, the dl mutant pheno-type, as noted above, results from the failure to activate the expression of genes, like twi and sna, that direct the development of ventral regions, and a failure to repress the expression of genes, like dpp and zen, that direct the development of dorsal regions of the embryo (in this way, the dpp system is dependent on the dl system). Based on the results presented in this report, it is this uniform expression of dpp in dl embryos that leads to the ‘dorsalization’ of the embryo. Second, in a dpp mutant embryo the dorsal fates are lost and they are supplanted by more lateral fates (see Fig. 3H), but the ventral fates are not affected since these are specified by the dl system. Finally, in the dpp, dl double mutant, both dorsal and ventral fates are absent, and both are supplanted by the lateral fates (Fig. 9B).

Based on this interpretation of the mutant phenotypes, the following conclusions can be made. The dorsal expansion of lateral pattern elements in dpp mutant embryos suggests that in the wild-type embryo, the activity of the dpp system serves to confine these elements to more lateral regions. In other words, the dpp system elaborates the pattern in the dorsal ectoderm with respect to the lateral fates. Further evidence for the negative effect that dpp function has on the lateral fates is revealed by the dl mutant phenotype. As at least some lateral fates can be specified in dl embryos (as evidenced from the phenotype of the dpp, dl double mutant), the failure to produce lateral structures in a dl mutant embryo must be the result of a negative effect from dpp, which is expressed throughout the embryo. Finally, the ventral expansion of the lateral pattern elements in the dpp,dl double mutant embryo implies that the ventral gene expression induced by the dl gradient also acts to confine lateral pattern elements to the lateral regions of the embryo. In other words, the dl system elaborates the pattern in the ventral ectoderm with respect to the lateral fates. Thus, like the Kr expression domain, which is defined entirely by negative regulatory cues, the lateral domain of the dorsal-ventral pattern is also defined entirely by negative regulatory cues. Thus, this pattern element is a ground state from which the dl and dpp gradients define ventral and dorsal pattern elements, respectively.

We have presented evidence supporting the hypothesis that the dpp gene product is an integral part of a dorsal-to-ventral activity gradient in the early embryo that specifies fates within the ectodermal anlagen. We have demonstrated that there is a graded requirement for the dpp gene product in the early embryo. A high level of dpp activity is necessary to specify the amnioserosa, and progressively lower levels of dpp activity are required to specify dorsal and lateral ectoderm. Furthermore, we have provided evidence that the dpp gene product is the central element of this system.

Significantly, the dpp system differs from the previously described maternal gradient systems in two ways. First, the dpp gene is strictly zygotic and has no germ-line dependent maternal activity (Irish and Gelbart, 1987). Thus, the dpp system is organized around a zygotic gene rather than a maternal one. Second, the dpp gene product is homologous to members of the TGF-β superfamily of secreted growth factors (Padgett et al., 1987). Thus, dpp does not instruct cell fate based on direct transcriptional activation or repression of downstream genes. Rather the mechanism by which dpp specifies cell fate must involve intercellular signalling. Given that there is no paradigm for a gradient system organized around a growth factor molecule, it is of interest to consider how such a gradient would be established and how it would function.

For instance, it is possible that the dpp protein is distributed in a concentration gradient, analogous to the bicoid system. In such a scenario, the fate of a particular cell could be specified based on the number of receptors that are activated by the dpp ligand. As such, a cellular milieu containing high concentrations of dpp product would promote one cell fate, while a milieu with a lower concentration would promote a different cell fate. While such a model is, of course, hypothetical, concentration-dependent specification of cell fate by a TGF-β-related molecule is not without precedent. Studies on Xenopus XTC-MIF (activin A) have indicated that this TGF-β member has the potential to induce different cell types in a dose-dependent fashion (Green and Smith, 1990). However, if dpp protein was in fact distributed in such a concentration gradient, then this graded distribution must be established by transcriptional or translational regulation. Two possible types of transcriptional regulation can be eliminated. First, dpp transcripts do not accumulate in a concentration gradient. Based on in situ localization of dpp transcripts in wild-type embryos, the expression of dpp is uniform throughout the dorsal ectoderm (St. Johnston and Gelbart, 1987). Second, it is unlikely that the initial expression of dpp could serve as a localized source from which the protein might diffuse, since dpp is expressed in nearly all the cells that are affected in dpp mutant embryos. The possibility that the dpp gradient is regulated at the level of differential translation of message cannot be eliminated at the present time, and is a reasonable possibility.

As an alternative to a bcd-like concentration gradient, it is possible that the dpp protein is part of an activity gradient, such that the activity of the dpp protein is modulated along the dorsal-ventral axis. As such the dpp signal is positively or negatively modulated by interactions with extracytoplasmic factors, and that the modified dpp ligand or ligand-factor complex can elicit different cellular responses upon receptor binding. Evidence in favor of this type of model is derived from genetic studies on the interactions between dpp and the other zygotic genes which affect the patterning the dorsal ectoderm. Three of these genes, tld, srw, and scw, have been shown to interact genetically with dpp, and the nature of these interactions strongly suggests that the wild-type function of these gene products is to potentiate the dpp signal (Ferguson and Anderson, 1992; V. Twombly, personal communication). Based on molecular analysis of tld, it is conceivable that this gene could potentiate the dpp signal via post-translational modulation. It has been shown that tld encodes a secreted protein with sequence similarity to the vertebrate bone morphogenetic protein BMP-1. The regions of homology include a metalloprotease domain and several motifs implicated in protein-protein interaction (Shimell et al., 1991). Based on potential interactions of BMP-1 and TGF-β-related ligands (Wozney et al., 1988), Shimell et al. (1991) have suggested that the tld gene product may be involved in the post-translation modification of the dpp protein. Thus, it is compelling to hypothesize that this post-translational modification may be partially responsible for establishing the dpp gradient.

In light of the similarity between the phenotypes of dpp alleles and the other zygotic dorsal-ventral patterning genes, it is possible to construct a model in which these various factors interact in a combinatorial manner so as to generate such an activity gradient of dpp. For instance, if these genes were expressed in overlapping, nested domains centered on the dorsal midline, the combinatorial interactions between these potentiating gene products and the dpp ligand could serve to create an activity gradient of dpp. However, as dl is the only source of polarity in the dorsal-ventral system this type of model requires that these different over-lapping expression domains be determined by the dorsal gradient, and at present there are no data, either genetic or molecular, to support this differential transcriptional regulation of tolloid, shrew, and screw. Furthermore, in situ localization of tld transcripts has revealed that this gene is expressed in the dorsal-most 40% of the blastoderm embryo, similar to dpp (Shimell et al., 1991). Therefore, with the present data it is not possible to pursue this type of model further.

Nevertheless, the models we have suggested above clearly illustrate a fundamental constraint on the dpp system. Inevitably the dpp gradient must be tied to the dl gradient since it is the only source of polarity in the dorsalventral system. Thus, regardless of whether a model is proposed in which dpp transcripts are translationally regulated, or one invoking the post-translational modulation of dpp activity by several potentiating gene products, the polarity of the dpp system must be derived from dl. In the first case, it is necessary to propose a translational regulator, perhaps an RNA binding protein, that is specified by dl and mediates the graded translation of the dpp message. In the second case, as we have discussed above, it is necessary to propose that the different expression domains of the post-translational modifiers are established by the dl gradient.

Returning to the post-translational modification model, if dl does not affect the differential expression of proteins that interact positively with dpp, the dl gradient might establish a new zygotic ventral-to-dorsal gradient that interacts antagonistically with the dpp system. Thus, dl may indirectly generate the dpp gradient. This is analogous to proposing a graded regulatory molecule which mediates the differential translation of the dpp message, but in this case, the regulatory molecule is a secreted protein that modulates dpp activity post-translationally.

This bipolar model would explain several puzzling results that have been obtained in studies involving the zygotic dorsal-ventral patterning genes. For instance, consider the two mesoderm specific zygotic genes twist and snail. A double mutant for twi and sna completely deletes the mesoderm and adjacent mesectoderm, yet this is not accompanied by a corresponding expansion of the neurogenic ectoderm. Rather, the neurogenic region is simply shifted ventrally (Rao et al., 1991). This could be explained in light of a bipolar gradient system in that rather than expanding the ventral fates, the entire axis is redistributed over a greater number of cells. Similarly, we have shown here that an intercalation of fates is observed in embryos bearing duplications of dpp. In this case, the expanded amnioserosa reduces the number of cells that can contribute to the epidermal fates, yet all of the fates are established.

The idea of a bipolar system is also made compelling by a genetic analysis of sog. The sog gene has been placed among the zygotic ventralizing genes, based on the effect it has on the specification of the amnioserosa and on the expression pattern of zen (Ray et al., 1991). However, the focus of sog activity has been fate mapped ventrally (Zusman and Wieschaus, 1985) suggesting that it is active in ventral and not dorsal regions of the embryo. Based on these two characters alone, it is tempting to propose that sog may be a component of a ventral gradient like the one proposed above. This proposal is supported by the fact that sog interacts antagonistically with dpp in a dosage sensitive fashion. It has been shown that increasing the dosage of the dpp gene in a sog mutant embryo effectively dorsalizes the embryo (Ferguson and Anderson, 1992). Furthermore, we have found that sog/+;dpp^H46/+ animals survive more frequently than sog⁺/sog⁺; dpp^H46/+ animals indicating that sog is a dominant suppressor of dpp haplolethality. Thus, sog is a gene that is presumably expressed ventrally and that acts antagonistically to the dpp gradient. The possibility that sog is a component in the proposed ventral gradient that establishes the dpp system must await further data.

It is important to note that the proposal that sog is part of a ventral-to-dorsal gradient is consistent with either a model invoking translational regulation of dpp or one invoking post-translational modification of the dpp protein. In the first case, it might be suggested that the sog gene encodes an RNA binding protein that sequesters dpp transcripts in more ventral regions. In the second case, a wide variety of protein types might be suggested. For instance, the ventral gradient could be composed of a specific protease or ligand binding protein, either of which could differentially inactivate the dpp protein along the dorsal-ventral axis. Alternatively, analogous to the activin/inhibin relationship, where the function of hetero- and homodimers differs dramatically (Vale et al., 1986; Ling et al., 1986), the ventral gradient could be composed of another TGF-β-like molecule. In such a system, the position specific ratio of homo- to heterodimers could be involved in the specification of discrete cell fates.

We would like to thank H. Duncan for technical contributions. We are indebted to C. Rushlow for providing the Krüppel anti-body and for helpful discussions. We would like to thank the members of the Schüpbach laboratory for help in preparation of the final figures, and Trudi Schüpbach for critical reading of the manuscript. During the course of this work, K. A. W. was supported by a National Institutes of Health postdoctoral fellowship, and R. P. R. by Office of Naval Research and National Institutes of Health predoctoral training fellowships. This work was funded by a National Institutes of Health Grant to W. M. G.