Context-dependent relationships between the BMPs gbb and dpp during development of the Drosophila wing imaginal disk

In the course of development, cells within a developing tissue receive many different kinds of signals, mediated by a variety of signaling systems, and the cells are able to integrate and coordinate these signals into actions appropriate to their role in the developmental program. Studies on most of these systems have shown that proper signaling involves multiple ligands and multiple receptors. For example, at least three types of ligands have been described for the epidermal growth factor (EGF) receptor: transforming growth factor (TGF)α proteins, Neuregulins and Amphiregulins, all of which contain the canonical cysteine-rich EGF-repeat, yet each has different effects on receptor function (Moghal and Sternberg, 1999; Wells, 1999). Perhaps most diverse in this regard is the TGFβ signaling system. In vertebrates, TGFβ superfamily members include at least 16 ligands, many combinations of which are expressed in overlapping patterns within the same tissue. These ligands signal through heteromeric receptor complexes consisting of Type I and Type II receptors that are also diverse (Hogan, 1996; Massagué, 1998). This kind of complexity within a single signaling system raises the question of how cells distinguish between specific ligands or ligand-receptor pairs, and how these different combinations influence the development of the organism.

The TGFβ superfamily is comprised of more than 25 structurally related members that have been grouped into four families, TGFβs, bone morphogenetic proteins (BMPs), Activins and MIS. All of the members are produced as pro-protein dimers consisting of an N-terminal pro domain and a C-terminal ligand domain, the latter of which is cleaved from the pro region during secretion to release the biologically active ligand dimer. Studies in a number of systems have shown that the signaling potential of these ligands may be regulated in a number of different ways. For example, the ligands are subject to regulation by extracellular antagonists such as Chordin, Noggin and Follistatin, which act directly on the ligands thereby preventing their interaction with receptors (Piek et al., 1999). Ligand function can also be affected by the formation of heterodimers that may have properties distinct from their corresponding homodimers (Yu et al., 1987; Petraglia, 1989).

In Drosophila, as in vertebrates, there are multiple TGFβ superfamily ligands and multiple receptors. The three characterized ligands, all members of the BMP family, are the BMP2/4 homolog dpp, the BMP5/6/7/8 homolog gbb, and the more distantly related screw (scw) (Padgett et al., 1987; Wharton et al., 1991; Arora et al., 1994). The BMP receptors include two Type I receptors, thick veins (tkv) and saxophone (sax), and a single Type II receptor, punt (put) (Nellen et al., 1994; Brummel et al., 1994; Letsou et al., 1995). dpp is a central figure in all characterized BMP signaling events in Drosophila, and has been implicated in numerous functions throughout the life cycle of the fly. In two of these functions, specifically, dorsal-ventral patterning in the embryo and anteroposterior (A/P) patterning in the wing disk, it has been proposed that dpp has morphogenetic properties in that multiple cell fates are specified as a function of different levels of dpp activity (Podos and Ferguson, 1999).

In the embryo, dpp acts in combination with scw to specify pattern elements in the dorsal epidermis through a gradient of BMP signaling whose high point lies along the dorsal midline (Ferguson and Anderson, 1992; Wharton et al., 1993; Neul and Ferguson, 1998; Nguyen et al., 1998). According to the current model, formation of this activity gradient depends on three features of the system: specificity of each ligand for a different receptor complex, strict dependency of Scw signaling on dpp, and antagonism of Scw activity by short gastrulation (sog), the Drosophila ortholog of Chordin. In brief, it is thought that Dpp signals uniformly throughout the dorsal 40% embryo (where the dpp RNA is expressed) through a receptor complex composed of Tkv and Put. Scw is thought to signal through a receptor complex composed of Sax and Put, and this signaling is limited to the dorsal regions of the embryo in two ways. First, Scw signaling is strictly dependent on Dpp signaling, and thus, while scw is expressed throughout the embryo, Scw signaling only occurs in the dorsal 40% of the embryo where dpp is expressed. Notably, this dependency does not require the formation of Scw:Dpp heterodimers, as restriction of scw expression to ventral cells does not compromise its ability to act in conjunction with Dpp signaling to generate a normal dorsal-to-ventral gradient. Second, Scw activity is negatively regulated by a gradient of Sog diffusing dorsally from its site of expression in the ventral ectoderm, such that Scw activity is highest along the dorsal midline and grades off ventrally. As Scw signaling acts to augment Dpp signaling, the highest levels of BMP signaling in the embryo lie along the dorsal midline, where Scw activity is highest, and the levels grade off ventrally (Podos and Ferguson, 1999).

In the wing disk, dpp has a number of developmentally and genetically separable functions. Throughout larval development, dpp is expressed in a narrow band of cells that lie just anterior to the A/P compartment boundary (Masucci, et al., 1990; Posakony et al., 1991). From this localized site of expression, dpp acts long range across the disk to promote disk proliferation, predominantly during early larval development (Spencer et al., 1982; Burke and Basler, 1996), and specification of vein territories during later larval development (deCelis et al., 1996; Sturtevant et al., 1997). It has been proposed that this ‘stripe’ of expression serves as a localized source for a gradient of dpp activity that activates the expression of target genes spalt (sal) and optomotor blind (omb) with respect to different activity thresholds (Lecuit et al., 1996; Nellen et al., 1996). Mutations in the disk region of dpp (i.e. dpp^d alleles) affect these functions, and give rise to small disks in mutant larvae (Spencer et al., 1982), or, in adult viable combinations, to small wings that show loss of vein and intervein territories (Segal and Gelbart, 1985; deCelis, et al., 1996). Similar phenotypes have been observed by clonal analysis with null and dpp^d alleles (Posakony, et al., 1991), and, as mutant wings can be recovered showing phenotypes far from the site of the clone, these studies confirm that dpp acts non-autonomously (i.e. at a long range) from this focus along the A/P compartment boundary.

During pupal development, dpp ceases to be expressed along the A/P compartment boundary, and novel transcription of dpp is detected along the lengths of the presumptive veins (Yu, et al., 1996; deCelis, 1997). At this stage of development, dpp is thought to contribute to the process of vein promotion whereby vein and intervein tissues in the wing are defined and refined. Mutations in the shortvein region of dpp (i.e. dpp^s alleles) affect this function, and, in adult viable combinations show truncation of the distal tips of the longitudinal veins and loss of the crossveins (Segal and Gelbart, 1985). Based on clonal analyses with null and dpp^s alleles, it has been shown that the vein loss associated with the mutant clones respects the clone boundaries, indicating that, for this function, dpp acts more or less autonomously (Posakony et al., 1991; deCelis, 1997). Thus, in contrast to its long-range functions during larval development, dpp appears to act locally during pupal development to promote the vein fate.

The developmental events that require dpp during wing development do not involve scw, which is not expressed after the embryonic stages (Arora et al., 1994), but may involve gbb. gbb is broadly expressed in the wing disk (Khalsa et al., 1998), and gbb mutants show phenotypes that are to some extent similar to the wing phenotypes of dpp (Khalsa et al., 1998; Wharton et al., 1999). Despite these similarities, the nature of the relationship between dpp and gbb, and how these two BMPs interact to pattern the wing properly is not clearly understood. Indeed, while a previous study based on the overexpression of gbb and dpp in combination with dominant-negative receptor constructs suggested that the relationship between gbb and dpp in the wing was similar to that of scw and dpp in the embryo (Haerry et al., 1998), these results were not entirely consistent with corresponding loss-of-function data (Khalsa et al., 1998), and prompted a more detailed analysis of gbb function in wing development.

We present a detailed clonal analysis of gbb in the wing. Our results show that, like dpp, gbb has two different types of foci in the disk, local and long range, and these foci correlate both spatially and functionally with the local and long-range functions of dpp. This coincidence of the foci of gbb and dpp in the disk indicates that the two BMPs act from the same sites to regulate disk proliferation and vein specification in the larval imaginal disk, and vein promotion in the pupal wing. Function-by-function comparisons of the phenotypes of gbb mutants and null clones with the phenotypes of dpp and sax clones clearly demonstrates that the relationship between gbb and dpp in the wing is not only different from that proposed for dpp and scw in the embryo, but that the relationship between gbb and dpp depends on the developmental process they affect. These results provide evidence that there is not a single type of relationship between different BMPs that is co-opted into a variety of different developmental contexts, but rather that the BMPs have evolved relationships that are specific to the developmental context in which they act.

Flies were raised on standard Drosophila cornmeal-yeast medium at 25°C unless otherwise indicated. The amorphic gbb alleles, gbb¹ and gbb², and hypomorphs, gbb³ and gbb⁴, have been described elsewhere (Wharton et al., 1999). The gbb mutant chromosomes used in this report are derivatives of the original mutagenized chromosomes. The dp cn bw gbb¹ chromosome was recombined with an isogenic pk cn stock to generate pk cn gbb¹ which was used to generate the FRT-G13 sha^IN gbb¹ lines used for the clonal analysis. The pk cn gbb¹ chromosome was in turn recombined with an isogenic b pr cn bw stock to generate the b pr cn bw gbb¹, which was then recombined with Oregon R to generate the bw gbb¹ chromosome. We have shown that the lethality of this chromosome is specific to gbb, as we can rescue it to viability with a single copy of a gbb⁺ transgene. For gbb², the mutagenized dp b cn gbb² chromosome was recombined with b pr cn bw to generate b pr cn bw gbb² chromosome that was then used to generate the FRT-G13 sha^IN bw gbb² chromosome. The b pr cn bw gbb⁴ chromosome was recombined with Oregon R to generate bw gbb⁴ and cn bw gbb⁴ chromosomes.

For the rescue experiments with additional copies of dpp, recombinants were made directly between two dpp duplications, Dp(2;2)B16 and Dp(2;2)DTD48 (Wharton et al., 1993), and the pk cn gbb¹ and b pr cn bw gbb⁴ chromosomes. For gbb mutant clones bearing additional copies of dpp, double recombinants were isolated directly from a cross of FRT-G13 sha^IN bw gbb² × FRT-G13 Dp(2;2)DTD48 to generate the stock FRT-G13 sha^IN Dp(2;2)DTD48 bw gbb². For gbb-null clones produced in a background carrying three copies of the dpp locus, Dp(2;2)DTD48 was recombined onto a chromosome carrying FRT-G13 sha^IN to generate the FRT-G13 Dp(2;2)DTD48 which was, in turn, recombined with FRT-G13 M(2)53 to generate FRT-G13 Dp(2;2)DTD48 M(2)53.

For dpp clones, the dpp^H46 ck¹³ FRT-40A bw and Dp(2;2)B16 dp πM-36F FRT-40A stocks were generated by recombination from the crosses dpp^H46 Sp cn bw X P[y⁺]25D ck¹³ FRT-40A (Bloomington) and Dp(2;2)B16 dp cl cn bw × 2πM FRT-40A (T. Xu), respectively.

All other strains are listed in FlyBase (http://fly.ebi.ac.uk:7081).

Clones were generated using the FLP/FRT technique (Xu and Rubin, 1993) in the progeny of the following crosses:

y w¹¹¹⁸; FRT-G13 sha^IN gbb¹/SM6a X y w* hs-FLP1; FRT-G13 2πM

w hs-FLP1; FRT-G13 M(2)53/CyO X y w¹¹¹⁸; FRT-G13 sha^IN gbb¹/SM6a

w hs-FLP1; Dp(2;2)B16 dp πM FRT-40A/CyO X dpp^H46 ck¹³ FRT-40A/CyO23,dpp⁺

w hsFLP1; FRT-G13 M(2)53/CyO X FRT-G13 sha^IN Dp(2;2)DTD48 bw gbb²/SM6a

w hsFLP1; FRT-G13 Dp(2;2)DTD48/CyO X FRT-G13 sha^IN gbb¹/SM6a

w hsFLP1; FRT-G13 Dp(2;2)DTD48 M(2)53/CyO X FRT-G13 sha^IN gbb¹/SM6a

w hs-FLP1; FRT-G13 M(2)53/CyO X y w¹¹¹⁸; FRT-G13 sax⁴ sha^IN/SM6a

Crosses were made in bottles, brooded every 12 or 24 hours, aged for 24 hours, and then heat-shocked for 2 hours at 37°C. Wings of flies of appropriate genotype were mounted in DPX mountant (EM Sciences) and analyzed on a Nikon Microphot-FXA photomicroscope. Images were collected with a SPOT-RT color digital camera (Diagnostic Instruments). For each experiment, at least 500 clones were mapped and analyzed, and in some cases, for example, for the dpp clones affecting the posterior cross vein and the large sax clones occupying the anterior or posterior compartments, many more were scored for these particular characteristics.

For rescue of gbb¹/gbb⁴ transheterozygotes, crosses of pk cn gbb¹/SM6a, Dp(2;2)B16 dp cn gbb¹/SM6a, Dp(2;2)DTD48 gbb¹/SM6a to b pr cn bw gbb⁴/SM6a and a cross of Dp(2;2)B16 dp cn gbb¹/SM6a to Dp(2;2)DTD48 bw gbb⁴/SM6a were scored for the presence of Cy⁺ progeny. The statistic ‘percent of expected’ was calculated by dividing the total number of Cy⁺ progeny by half the total number of Cy progeny. Pharate and pupal lethals were scored 2 days after the last eclosed progeny were collected. For each cross more than 1000 progeny were scored from multiple broods. For rescue of gbb¹ homozygotes crosses of pk cn gbb¹/SM6a, Dp(2;2)B16 dp cn gbb¹/SM6a, Dp(2;2)DTD48 gbb¹/SM6a to bw gbb¹/SM6a and a cross of Dp(2;2)B16 dp cn gbb¹/SM6a to Dp(2;2)DTD48 gbb¹/SM6a were scored.

We have previously provided evidence that gbb plays a role in the development of the wing imaginal disk (Khalsa et al., 1998; Wharton et al., 1999). Amorphic alleles of gbb, e.g. gbb¹ and gbb², are larval/pupal lethals, and the mutant larvae have variably reduced wing imaginal disks. Hypomorphic alleles of gbb, e.g. gbb⁴, are semi-viable as homozygotes and in trans to null alleles (see Fig. 6) (Wharton et al., 1999). The wings of the mutant flies are reduced in size compared with wild type and show defects in wing vein patterning (Fig. 1). The weakest vein defect observed is the specific loss of the posterior cross vein (PCV). Moderate phenotypes, which are most commonly observed, show loss of the PCV as well as truncations of the distal tips of longitudinal veins L4 and L5 (Fig. 1B, compare with Fig. 1A). The most severe wing phenotype shows complete loss of L5 except for the most proximal portion, and loss of the PCV and distal tip of L4 common to the weaker phenotypes (Fig. 1C). These phenotypes suggest a number of possible roles for gbb in wing development: the small disk and wing phenotypes suggest a role in disk proliferation, while the venation defects suggest a role in either the establishment of vein territories or wing vein promotion.

To better understand the role that gbb plays in these developmental functions, we have used clonal analysis to look for functional foci of gbb in the wing disk. Clones were generated in both wild type and Minute backgrounds, and the results from both analyses reveal a number of general features of gbb mutant clones. First, when compared with marked, wild-type clones induced in parallel, the presence of a null allele of gbb has no effect on clone frequency, clone distribution or clone size (data not shown). Thus, gbb is neither cell lethal, either generally or within specific regions of the disk, nor does it have a significant autonomous effect on cell growth or proliferation. Second, the regions of the wing where gbb clones are associated with mutant phenotypes are fairly restricted. Thus, despite the broad expression of gbb in the wing disk (Khalsa et al., 1998), its function is limited to specific areas. Third, mutant phenotypes are observed only in clones or regions of clones where there is a dorsoventral overlap of mutant tissue. Mutant clones entirely confined to either the dorsal or the ventral surface of the wing – even if the clone occupies the entire anterior or posterior compartment – do not show mutant phenotypes. In the following discussion, we refer to clones with a dorsoventral overlap as ‘double-sided’ clones.

Double-sided gbb clones that encompass the entire posterior compartment show loss of the PCV as well as loss of the distal quarter of L5 (Fig. 2A), consistent with the phenotypes of gbb hypomorphs (compare with Fig. 1B). Smaller clones covering either just the PCV or just the distal tip of L5 also show loss of the corresponding vein (data not shown, Fig. 2E), indicating that these two foci are independent of one another.

Both the PCV and L5 foci have only short range effects on the veins, and thus constitute ‘local’ requirements for gbb. For the PCV, in double-sided clones that cover only the anterior or posterior half of the PCV, the vein is absent in the mutant cells and stops either precisely at the boundary of the dorsoventral overlap or within two to three cells of it (Fig. 2B,C,G). Notably, the PCV can stop on either the wild-type or the mutant side of the clone boundary, indicating that this effect is not simply a consequence of gbb product diffusing from wild-type cells into mutant tissue. Interestingly, this partial loss of the PCV requires that the double-sided clone include at least one of its two junctions with the longitudinal veins. Double-sided clones that occupy half of the L4/L5 intervein but stop short of the L4 or L5 vein do not affect the PCV (Fig. 2D). These latter clones suggest that gbb may not be required uniformly along the length of the PCV, but rather more strongly near the junctions with the longitudinal veins and less so in the intervein between them.

For the distal tip of L5, double-sided clones that fall within the distal quarter of L5 show loss of the vein only within the mutant tissue with the vein stopping within two to three cells of the dorsoventral overlap of the clone (Fig. 2F). Paradoxically, the distal tip of L5 can also be lost in association with clones covering proximal L5, even if the distal quarter of the vein is wild type for gbb function on the ventral, dorsal or even both surfaces (Fig. 2G). Such clones imply a degree of long-range non-autonomy for the L5 focus. However, as clones of this type invariably cover part of the vein that is normally absent in larger clones covering all of L5, we would predict that such clones should be associated with a gap between the proximal boundary of the focus and the boundary of the dorsal-ventral overlap of the clone. It is possible that such short gaps in the middle of the vein interfere with the differentiation of vein tissue more distally, and consequently result in the deletion of the entire distal tip.

In addition to the loss of the PCV and the distal tip of L5, we have noted that hypomorphic gbb alleles also affect the distal tip of L4 (Fig. 1B), yet, gbb clones that occupy the entire posterior compartment have little or no effect on this vein (Fig. 2A). This difference is also a reflection of the local non-autonomy of gbb in the disk. While posterior clones show little or no effect on L4, we do observe loss of L4 when double-sided mutant tissue covers the regions both anterior and posterior to the vein (data not shown). As L4 lies just posterior to the A/P compartment boundary (Fig. 1A), L4 loss is only observed in the statistically rare instances when clones are induced in both the anterior and posterior compartments that happen to fall next to one another along the distal tip of L4. As such, the failure to lose L4 in posterior clones, though the vein is entirely within mutant tissue, is presumably due to wild-type gbb product diffusing locally from the anterior compartment to compensate for its loss in the posterior.

From these data, we conclude that there are three independent foci of gbb function in the posterior compartment, along the length of the PCV and the distal quarters of L4 and L5. The phenotypes associated with clones overlying these foci are confined to them, and have only short range effects on their respective veins. As such, they reflect local requirements for gbb, and most probably reflect a role for gbb in vein promotion.

The phenotypes of gbb hypomorphic mutations have little or no effect on the patterning of veins in the anterior compartment (Fig. 1B,C) (Khalsa et al., 1998; Wharton et al., 1999), and this is also true for null gbb clones. Double-sided clones covering the entire anterior compartment exhibit no defects in the costal vein or longitudinal veins L1, L2 or L3 (Fig. 3A). Thus, gbb is not required locally for promotion of veins in the anterior compartment. However, such clones are associated with a reduction in the overall size of the wing blade and loss of all but the most proximal region of L5 (Fig. 3A,B).

The difference in wing size associated with these clones can be seen clearly in comparisons of wings of the same fly, one of which has a large anterior clone, and the other not. In such situations, the wing lacking the clone serves as a size control for the wing with the clone. Wild-type left and right wings differ in area by no more than 2%, and we observe the same range of variability in flies with a wild type double-sided anterior clone on one of its two wings (Fig. 3C). By contrast, wings with a gbb mutant double-sided anterior clone show approximately a 30% reduction in overall wing size (Fig. 3D). Notably, both the anterior and posterior compartments show a reduction in size in these wings. For the pair in Fig. 3D, the overall left:right (L:R) wing ratio is 0.70, for the anterior compartments, the ratio is 0.74, and for the posterior compartments 0.69. Thus, gbb clones in the anterior compartment affect wing size over the entire wing blade.

Large anterior gbb clones are also associated with a loss of most of L5. The truncation of L5 is mildly variable, and may or may not be accompanied by the loss of the PCV. In the most severe cases, as in Fig. 3A, L5 is truncated back beyond the point of its junction with the PCV, and in these instances the PCV is always absent. In weaker examples, L5 extends further distal and abruptly turns up to make a right angle junction with L4 (Fig. 3B). Such veins are presumably chimeras of L5 and the PCV. Notably, although most of L5 is absent in these wings, the intervein between L4 and L5 is still present.

Fine mapping of this gbb focus indicates that the requirements for proliferation and specification of L5 map to the same region of the anterior compartment: just anterior to the A/P compartment boundary. Double-sided clones covering the region between longitudinal veins L1 and L3, show no effect on overall wing size, or on patterning of L5 (Fig. 4A,B). Similarly, double-sided clones that occupy only the region between L3 (but not including L3) and L4 are wild-type in size and pattern (Fig. 4E). By contrast, double-sided clones with an anterior border in the intervein between L2 and L3 and a posterior border running the length of the A/P compartment boundary show the mutant phenotypes associated with clones covering the entire anterior compartment (Fig. 4C,D). Thus, the anterior focus for gbb falls in a broad band of cells that lie just anterior to the A/P compartment boundary.

From these data we conclude that there is a single focus for gbb in the anterior compartment, lying just anterior to the A/P compartment boundary, from which gbb affects wing size and specification of the L5 vein territory. The phenotypes associated with clones overlying this focus are not confined to it, but affect either the whole wing blade (wing size) or vein structures far from the site of the clone (L5). As such, this focus constitutes a long-range requirement for gbb in wing disk patterning. The wing size phenotype reflects a role in disk proliferation and the loss of L5 a role in specification of vein territories. Significantly, the sum of the phenotypes associated with anterior and posterior gbb clones is essentially the same as the most severe wing phenotypes associated with the gbb hypomorphs (Fig. 1C), indicating that all of the gbb wing disk functions are recapitulated in the phenotypes of the hypomorphic alleles.

The four gbb foci we have mapped correspond in location and function with the established foci for dpp in the disk. We have shown that gbb is required locally for vein promotion at the distal tips of L4 and L5. Mutations in the shortvein region of dpp (i.e. dpp^s alleles) also result in loss of the distal tips of the longitudinal veins, though the phenotypes are more severe than what is observed for gbb. Weak dpp^s alleles show truncations of distal L4 and L5 to half their normal length, complete loss of the PCV, and loss of distal L2 in the anterior compartment (Fig. 5A) (Segal and Gelbart, 1985). Consistent with a previous clonal analysis, which showed that dpp was required for the distal half of L5 (Posakony, et al., 1991), we find that in dpp null clones, dpp is required locally along at least half the length of L4 and L5 (Fig. 5C,D). Thus, the gbb foci represent a subset of the regions of the longitudinal veins that require dpp. Moreover, as the sites of the gbb foci correspond to those regions of L4 and L5 most sensitive to the loss of dpp, we conclude that gbb is required to achieve maximal levels of BMP signaling at the distal tips of the veins.

The local focus for gbb along the PCV is also a focus for dpp function. As no allele of dpp results in the specific loss of the PCV, and as previous clonal analyses (Posakony et al., 1991; deCelis, 1997) did not show a specific effect of dpp on the PCV, we have generated null dpp clones specifically over the PCV to determine if such clones show the same behaviors as the gbb clones described above. We find that, indeed, null dpp clones covering all or part of the PCV show the same behavior as gbb clones: the dpp clones have no effect on the PCV unless there is a dorsoventral overlap of mutant tissue, and double-sided clones that cover half of the PCV show loss of the vein in the mutant tissue up to or within two to three cells of the clone boundary (Fig. 5C,D). Thus, gbb and dpp are both required for promotion of the PCV. As such, we conclude that gbb and dpp act together to achieve the levels of BMP signaling required for this vein.

The anterior focus of gbb is associated with two different functions, disk proliferation and L5 specification, and this focus coincides precisely with the major focus for dpp in the disk that has been implicated both in proliferation (Burke and Basler, 1996) and in specification of wing vein territories (deCelis et al., 1996; Sturtevant et al., 1997). Expression of dpp in this focus is regulated by cis-acting sequences in the disk region of the dpp gene (Blackman et al., 1991), and dpp alleles that affect this region (i.e. dpp^d alleles) exhibit phenotypes that can be related to those of our gbb clones. The most severe dpp^d alleles are pupal lethals (Spencer et al., 1982), and the mutant larvae have small disks very similar to those of gbb null larvae (Khalsa et al., 1998). Weaker dpp^d allelic combinations are adult viable, and the wings of these flies are also reduced in size. However, the dpp phenotypes are more severe than those associated with gbb clones. The weakest dpp^d heteroallelic combinations give rise to ‘winglets’ that may be no more than one tenth the size of a normal wing (Spencer et al., 1982), and it is only in heteroallelic combinations of dpp^d and dpp^hr alleles that wings are produced of comparable size with those with null gbb clones (Fig. 5B) (Segal and Gelbart, 1985; deCelis, 1997). Thus, while it is clear that gbb and dpp both contribute to proliferation in the wing disk, dpp has a much more profound effect, suggesting that the role of gbb in this process may be facilitatory.

The specific loss of L5 associated with the anterior focus is the only gbb wing phenotype which cannot be correlated with a phenotype of dpp. That dpp is involved in the specification of vein territories has been clearly established (deCelis et al., 1996; Sturtevant et al., 1997). dpp function along the A/P compartment boundary is required for expression of the transcription factor sal (Lecuit et al., 1996; Nellen et al., 1996), and this gene is required for specification of L2 and the intervein between L4 and L5 (deCelis et al., 1996). Indeed, the phenotypes of weak dpp^d/dpp^hr combinations as well as clones of sal consist of a loss of L2 in the anterior compartment and loss of the intervein between L4 and L5 in the posterior compartment (Fig. 5B) (deCelis, 1997). However, as anterior clones of gbb affect neither of these structures, our data are not consistent with a role for gbb in contributing to the maximal levels of BMP signaling along the A/P compartment boundary. Rather, as the structures affected in gbb clones lie further away from the source than the domain of sal expression, it appears that gbb is required for the low points of BMP gradient at sites far from the A/P compartment boundary. As such, the gbb and dpp phenotypes are not the same because the dpp hypomorphs affect the high point of the gradient, and less so the low points, while gbb mutations affect the low points of the gradient and not the high point.

To better understand the relationship between gbb and dpp in the disk, we tested for suppression of the gbb mutant phenotypes by additional doses of the dpp locus. As suggested above, one possible function for gbb may be to augment the levels of BMP signaling provided by Dpp. As such, we would expect that raising the level of dpp expression in the disk would compensate for the loss of gbb. That is, if we increase the levels of dpp activity at those sites where our clonal analysis indicates that gbb is active, we should be able to suppress the corresponding gbb mutant phenotypes. To do this, we took advantage of two duplications of the dpp locus, Dp(2;2)B16 and Dp(2;2)DTD48, to generate gbb mutant flies bearing three or four copies of the dpp locus (see Materials and Methods).

As the hypomorphic gbb mutations are to some degree sensitive to genetic background (Khalsa et al., 1998), it was necessary to demonstrate that any rescue associated with the Dp(dpp) gbb recombinants was due to the additional copies of dpp rather than other modifying factors on the chromosome. To show this, we tested for the ability of extra copies of the dpp gene to rescue the lethality associated with hypomorphic and amorphic gbb alleles and found that additional doses of dpp do show a dose-dependent rescue of gbb lethality (Fig. 6A,B). For gbb¹/gbb⁴ transheterozygotes, 2% of expected are viable to adulthood, with three copies of dpp, 25% or 30% are viable, depending on which of the two duplications was used, and with four copies, 75%. Moreover, although four copies of dpp cannot rescue the lethality of gbb¹ homozygotes, we did observe a dose-dependent rescue of the lethal phase from larval to pupal lethal. For gbb¹ homozygotes, 10% of the expected class form pupae, with three copies of dpp, 30% or 60% depending on the duplication, and with four copies, 80% of expected form pupae. Thus, in both these assays we see a dose-dependent rescue of gbb phenotypes with additional copies of the dpp locus.

To examine the effects of additional doses of dpp on the phenotypes associated with specific gbb foci, we have generated clones that are both null for gbb and carry four copies of the dpp gene (see Materials and Methods). Duplications of dpp are able to rescue the distal tip of L5 in posterior clones (Fig. 6C,D), and while we cannot use clonal analysis to assay the effects of additional doses of dpp on the distal tip of L4, we do see rescue of this phenotype in gbb¹/gbb⁴ transheterozygotes bearing four copies of dpp (data not shown). Thus, for the L4 and L5 vein promotion foci, additional doses of dpp are able to rescue the gbb mutant phenotypes. By contrast, we never observe rescue of the PCV by dpp duplications in either gbb mutants or clones (Fig. 6C). This result suggests that either gbb and dpp act independently at this focus, or that the four doses of dpp are not sufficient to compensate for the loss of gbb. In favor of the latter hypothesis, it is worth noting that the PCV is the wing structure that is most sensitive to loss of gbb, and it is not clear that this is the case for dpp. As such, it is possible that the relationship between gbb and dpp is reversed in this case, and gbb is the central figure in PCV promotion while the role of dpp is secondary.

The truncation of L5 associated with the anterior focus can also be rescued with additional doses of dpp (Fig. 6E). Given that this function of gbb seems to reflect a requirement in extending the range of BMP signaling in the disk, we can account for this result in two possible ways. On the one hand, gbb may simply act to augment the levels of dpp signaling at the low points of the dpp gradient. As such, the additional doses of dpp increase these levels and compensate for the loss of gbb. Alternatively, gbb may be required for signaling in regions beyond the normal limit of the spread of dpp across the disk. In this case, the rescue by dpp would reflect an increase in the spread of dpp, owing to the higher levels of dpp at the source along the A/P compartment boundary.

In the above clonal experiments, while it is possible to confine the clones carrying four copies of dpp to the known sites of gbb foci in the wing, because of the method employed to make the clones, the wing cells outside the clone carry three copies of dpp rather than the wild-type two copies. As such, it is possible that the rescue of gbb phenotypes we observe is due to this additional copy of dpp in the background and not the four copies within the clones. This is a particularly relevant issue with regard to the long-range anterior focus for which the responding cells in the posterior compartment all carry three copies of dpp. To control for this, we performed the ‘reciprocal’ experiment to the one above and generated gbb null clones that carry the wild-type two copies of dpp in a background that carries three copies. We found that, in all cases, three copies of dpp outside of the clones could not rescue either local or long-range phenotypes associated with gbb clones (data not shown). For the anterior focus, this result demonstrates that the rescue of L5 that we observed in the clones carrying four copies of dpp is strictly due to the additional copies of dpp within the focus along the A/P compartment boundary. Moreover, as this focus corresponds to the early stripe expression of dpp in the disk, the result provides further evidence that the loss of L5 in gbb mutants and clones identifies a vein specification function associated with the global patterning functions of dpp, and not a vein promotion function.

In the embryo, cooperative signaling by dpp and scw are required for elaboration of the pattern in the dorsal ectoderm (Neul and Ferguson, 1998; Nguyen et al., 1998). According to the current model, this cooperation is achieved by dpp and scw signaling through different receptor complexes composed of tkv and put, and sax and put, respectively. Given that many of the relationships we observe between dpp and gbb are similar to the relationships between dpp and scw in the embryo, we were interested in determining if the gbb-dependent signals we had characterized in the wing were also transduced by sax. In a previous clonal analysis, it has been shown that large sax null clones resulted in reduced wing size, blunting of the wing tip, ectopic venation and mis-patterning of the anterior wing margin (Singer et al., 1997). However, as many of these phenotypes appeared to be associated with the creation of discontinuities in the BMP gradient, we were interested in establishing the phenotype of null sax clones occupying the entire anterior or posterior compartments, and comparing these phenotypes with those of our gbb clones. We have found that large sax null clones that occupy the entire anterior or posterior compartment give rise to essentially normal wings (Fig. 7A,C), though anterior clones are associated with a slight reduction in wing size. Notably, the ectopic vein and margin bristle phenotypes that have been described previously (Singer et al., 1997) are not observed in clones that encompass an entire compartment, but only when the clone boundary subdivides a compartment (Fig. 7B), indicating that it is not the loss of sax, per se, but the discontinuity between sax+ and sax^– cells that results in this phenotype. Given these results, we conclude that the gbb signals that we have characterized in the wing are not transduced by sax.

We have used clonal analysis to map foci for gbb in the developing wing imaginal disk. Our results show that gbb has two distinct types of functions: local and long range. The local foci are confined to the posterior compartment, and affect the promotion of the PCV and the distal tips of the longitudinal veins L4 and L5. The long-range focus lies in the anterior compartment comprising a broad band of cells along the A/P compartment boundary and affects disk proliferation and the specification of L5. These gbb foci are coincident with the foci for dpp in the disk, and many of the phenotypes associated with the gbb clones are rescued by additional copies of the dpp locus. Thus, gbb and dpp contribute to the same functions in the disk and gbb functions are to some extent redundant with those of dpp. Comparison of the foci and phenotypes of gbb and dpp mutants and clones indicates that the relationship between gbb and dpp is different for different functions. For promotion of distal tips of L4 and L5, gbb function is restricted to those areas that require the highest levels of dpp signaling, and as these phenotypes can be rescued with additional copies of dpp, we conclude that gbb is required to augment the levels of dpp signaling. For promotion of the PCV, the case is not so clear. We have shown that both gbb and dpp are required for PCV promotion. However, as dpp duplications do not rescue this phenotype, it is possible that gbb and dpp act independently or that the contribution of gbb to this process is sufficiently great that it cannot be compensated for by the additional doses of dpp. The requirement for gbb in the specification of L5 is not consistent with an augmentation of dpp signaling, as gbb mutants and clones do not affect structures specified by the high point of the dpp gradient. Rather, gbb clones affect structures far from the source along the A/P compartment boundary, suggesting that gbb signaling contributes to the low levels of BMP signaling at the extremes of the gradient.

We have noted that mutant phenotypes are observed only in gbb clones when the mutant tissue encompasses the entirety of the focus on both the dorsal and ventral surfaces of the wing. For example, clones that occupy the dorsal-anterior quadrant of the wing exhibit no defects in the patterning or size of the wing, while clones that occupy both the dorsal-anterior and ventral-anterior quadrants affect both these aspects of wing development (see Fig. 3). One explanation for this phenomenon is that Gbb exhibits long-range non-autonomy in the disk, and, in fact, there is some evidence for this, as we have found that small patches of wild-type cells along the A/P compartment boundary in the context of a large mutant clone are able to rescue loss of L5 completely in the posterior compartment (data not shown). However, gbb clearly does not act in a broadly non-autonomous fashion in all of its functions: gbb clones that cover the PCV or distal L5 exhibit vein defects that respect the clone boundaries indicating that the presumptive vein cells within the clone cannot be rescued by the wild-type Gbb present in the adjacent cells (see Fig. 2). For these functions, the ‘rescue’ observed in single-sided clones implies pattern regulation occurring between the two wing surfaces. Indeed, it has long been asserted that there are signaling events between the dorsal and ventral surfaces of the wing as it has been shown for several genes that loss of veins on one surface can be compensated for by the wild-type pattern in the opposing surface (Garcia-Bellido and de Celis, 1992). The requirement for dorsal-ventral overlap that we have observed with gbb mutant clones is indicative of such a signaling mechanism, and given these results, as well as those from previous studies that have shown a requirement for dorsal-ventral overlap in clones of dpp and sog (Posakony et al., 1991; Yu et al., 1996; deCelis, 1997), it is plausible that the BMPs themselves might be responsible for mediating these signaling processes.

Perhaps the most striking result from our clonal analysis is that the requirements for gbb in the wing disk are localized even though the gene is widely expressed. This result implies that Gbb activity is in some way restricted post-transcriptionally. Two models seem the most likely to account for this effect. First, as we have shown that all gbb foci are coincident with sites of dpp expression in the disk, it is possible that Gbb and Dpp form heterodimers, and that Gbb is only active in this form. Heterodimer formation has been documented for a number of different TGFβ superfamily members, and in some cases heterodimers and homodimers have been shown to have distinct properties. For example, heterodimers of BMP2 or BMP4 and BMP7 are much more potent in the induction of ventral mesoderm and bone induction than their respective homodimers (Isreal et al., 1996; Nishimatsu and Thompsen, 1998). Activins and Inhibins illustrate a different relationship: the homodimeric Activins having the opposite biological effects of the heteromeric Inhibins (Yu et al., 1987; Petraglia, 1989).

An alternative model is that the restriction of gbb function in the disk is achieved through local activation of Gbb homodimers, which may be achieved by specific agonists expressed within the foci or antagonists expressed everywhere else. Possible agonists include the Drosophila BMP-1 homologs tolloid and tolkin (Shimell et al., 1991; Nguyen et al., 1994; Finelli et al., 1995), or Drosophila homologs of the subtilisin-like proprotein convertases or furins, that are thought to be involved in the cleavage of BMP pro-proteins into the active ligand (Cui et al., 1998; Constam and Robertson, 1999). In addition, the recently characterized secreted protein crossveinless 2 (cv-2) may act as an agonist of BMP signaling specifically in the presumptive crossveins (Conley et al., 2000) (see below). The antagonist sog is a likely candidate for restricting BMP activity during pupal development (i.e. for vein promotion functions) as it has been shown to be expressed in all intervein cells at this time (Yu et al., 1996). Moreover, there is some evidence that sog function in the wing may specifically antagonize gbb (Yu et al., 2000), and thus may very well account for the restriction of gbb function to the presumptive veins.

Our clonal analysis has identified four processes that require gbb during wing development, disk proliferation, specification of the L5 vein territory, promotion of the PCV and promotion of the longitudinal veins L4 and L5. Based on the criteria of comparisons of gbb clone phenotypes with dpp and sax phenotypes, the ability for the gbb mutant phenotypes to be rescued by additional copies of dpp, and the spatial requirements for gbb during wing development, it is clear that each of these functions employs a different relationship between dpp and gbb, and each of these relationships is distinct from that which has been established for dpp and scw in the embryo (Table 1).

Some of the features of dpp and gbb function indicate consistent distinctions between gbb-dependent functions in the wing and scw-dependent functions in the embryo. For example, as discussed above, the explicit coincidence of dpp expression and gbb functions in the wing disk raises the possibility that heterodimer formation may play a role in some or all of the wing functions of dpp and gbb, while this does not appear to be the case for dpp and scw in the embryo (Nguyen et al., 1998). In particular, the absolute requirement for both gbb and dpp in PCV promotion, as evidenced by the failure of dpp duplications to rescue PCV loss in gbb mutants and clones, is entirely consistent with a mechanism requiring Gbb:Dpp heterodimers. Similarly, comparison of sax clone phenotypes with those of dpp and gbb indicate that sax is not required for the transduction of gbb signals in the wing, with the possible exception of the disk proliferation function. This is in contrast to the situation in the embryo, where sax is proposed to be a dedicated receptor for scw.

We have also observed distinctions in the relationships between dpp and gbb in different wing functions. Comparing the specification of the L5 territory and promotion of the distal tips of L4 and L5, gbb acts differently to modulate the activity of dpp. In the case of the vein promotion functions, gbb is required for maximal levels of BMP signaling at the distal tips of L4 and L5 – which is similar to what has been described for scw and dpp in the embryo. By contrast, for the specification of L5 during larval development, gbb is required for the specification of fates at the low points of the BMP gradient. Indeed, as the gbb clone phenotypes do not reflect the phenotypes of the dpp target gene sal (which requires maximal levels of BMP signaling for its expression), it follows that the expression of this gene, and thus the high point of the BMP gradient, is normal in wings bearing gbb null clones in the anterior compartment. This relationship is quite distinct from that of dpp and scw in the embryo.

The promotion of the PCV, while similar in many ways to the longitudinal vein promotion functions, is distinct in that it is one gbb function in the wing that cannot be rescued by additional copies of the dpp gene. It is relevant here that hypomorphic mutations of gbb and dpp show distinct phenotypes with regard to PCV promotion. For gbb, it is clear that the PCV is the structure most sensitive to a reduction in gbb activity as the weakest alleles show specific loss of it. By contrast, weak shortvein alleles show truncations of the distal tips of the L2, L4 and L5, but the PCV is intact (Segal and Gelbart, 1985). This suggests that for PCV promotion the relationship between dpp and gbb may be reversed, and gbb may play the more central role. This notion is supported by analysis of the distribution of the phosphorylated from of the Smad protein Mad (pMad), which can be detected in the presumptive PCV before the localized expression of dpp is detected at this site by in situ hybridization (Conley et al., 2000). Conley et al. have suggested that the localized expression of cv-2 in the presumptive PCV cells may account for the early appearance of pMad in this vein. It is tempting to speculate that cv-2 may localize the activity of gbb to the presumptive PCV, which results in the subsequent activation of dpp expression.

Given these different functions and the different relationships between BMP ligands specific to each, it is evident that there is not a ‘canonical’ relationship between BMP2/4- and BMP5/6/7/8-like molecules that is co-opted like a cassette into different developmental contexts. Rather, it seems that specific relationships have evolved between the two types of ligands that fulfill particular functional requirements during development. Moreover, as many of the distinctions appear to be occurring at the level of ligand activation, distribution, and ligand-receptor interactions, it follows that extracellular modulation of BMP ligands plays a major role in the establishment of these particular relationships. Identifying and understanding the roles of such extracellular factors will be key to understanding the molecular mechanisms underlying these different signaling events.

We thank W. M. Gelbart, F. M. Hoffmann and T. Xu for providing fly stocks and reagents. We also thank Erdem Bangi, Caroline Savery and Lorena Soares for critical reading of the manuscript, and all of our colleagues at Brown University for stimulating discussions. We are indebted to our two anonymous reviewers for many helpful suggestions and comments that greatly improved the manuscript. During the course of this work, K. A. W. was supported by an Established Investigatorship from the American Heart Association and with funds contributed in part by the AHA, Maine Affiliate. This work was supported by a grant from the American Cancer Society, DDC-98790 to K. A. W.