The role of brinker in mediating the graded response to Dpp in early Drosophila embryos

In both early insect and early vertebrate embryos, gradients of BMP activity are established that are responsible for most aspects of patterning along the embryonic dorsoventral axis (Ferguson, 1996). In Drosophila, the BMP2/4 homolog Decapentaplegic (Dpp) is an important component of a BMP activity gradient, which has its highest levels along the dorsal midline of cellular blastoderm embryos and declines towards more lateral regions. The high levels determine the cell fate of the extraembryonic amnioserosa while lower levels specify the dorsal epidermis. The absence of Dpp activity in ventral regions is required for the formation of the neurogenic ectoderm, which gives rise to both the ventral epidermis and the central nervous system. Thus, as in vertebrates, BMP in flies promotes the formation of epidermis and suppresses that of neural tissues (for reviews see Rushlow and Roth, 1996; Biehs et al., 1996; Wilson et al., 1997).

The importance of BMP activity gradients in DV patterning poses two questions. How are these gradients formed in early embryos and how is the graded distribution transformed into distinct cellular responses? Both in vertebrates and in insects gradient formation depends in part on secreted inhibitors that are expressed complementary to the BMPs (for a review see Cho and Blitz, 1998). One of these inhibitors is chordin in vertebrates and its homolog short gastrulation (sog) in Drosophila (Sasai et al., 1994; François et al., 1994; Holley et al., 1995; Schmidt et al., 1995; Biehs et al., 1996). In vertebrates the regulatory mechanisms that lead to the expression of chordin and BMP in complementary domains are not well understood. However, in Drosophila, sog and dpp are target genes of the maternal Dorsal (Dl) protein gradient, which has peak levels in ventral nuclei and declining levels laterally. Dl activates sog and represses dpp and thus confines dpp to the dorsal side of the embryo and sog to a ventrolateral stripe abutting the dpp domain. In ventralmost regions Dl activates the expression of the genes twist and snail, which repress sog and specify the mesodermal anlagen, but themselves are not affected by Dpp signaling (for a review see Rusch and Levine, 1996).

Chordin and sog are BMP-binding proteins which compete with ligand binding to the receptor and thus prevent signaling. They are cleaved by metalloproteases (tolloid (tld) in Drosophila and its homologs in vertebrates) expressed in the same domain as the BMPs (Piccolo et al., 1997; Marqués et al., 1997). Cleavage of the ligand/inhibitor complex results in the release of the active ligand. Thus, the activity gradient is not formed by diffusion of the ligand itself, but rather by diffusion of its inhibitor and of ligand/inhibitor complexes. Moreover, the actual gradients might result from the combined activity of several ligands interacting with different receptors and being differentially susceptible to inhibition, as has been recently shown for Drosophila. Here, the Dpp activity gradient results from an intricate interplay of Dpp with a second BMP- like ligand, Screw (Scw; Neul and Ferguson, 1998; Nguyen et al., 1998). Dpp and Scw each preferentially signal through different type I receptors. While Dpp signals through Thick veins (Tkv) and is required for all aspects of the DV pattern, Scw signals through Saxaphone (Sax) and is required only to induce dorsalmost cell fates. To explain this observation a synergistic interaction between Tkv activated by Dpp and Sax activated by Scw has been proposed, which potentiates Dpp signaling in dorsalmost positions.

How BMP gradients are transformed into a series of different cellular responses depends on the way target genes are regulated in an activity-dependent manner. Here, the SMAD proteins, the cytoplasmic signal transducers of all TGF- β family members, play a crucial role since they directly mediate between receptor activation and transcriptional response (for reviews see Heldin et al., 1997; Whitman, 1998). Receptor activation leads to phosphorylation of Class I SMADs, which then associate with Class II SMADs and translocate into the nucleus to directly bind to DNA recognition sites in target gene promoters (for a review see Derynck et al., 1998). In vertebrates dose-dependent responses to BMP can be mimicked by injection of different amounts of Class I SMAD proteins (Suzuki et al., 1997; Wilson et al., 1997). Therefore, it was suggested that graded BMP signaling leads to a gradient of nuclear concentrations of kinase-activated SMAD proteins, which in turn bind to target promoter sites with different affinities, thereby eliciting concentration-dependent threshold responses (Wilson et al., 1997). However, the situation now appears to be more complex. Firstly, BMP signaling also activates negatively acting SMADs, which do not enter the nucleus but compete with Class I and Class II SMADs for receptor binding or complex formation, respectively. This results in a non-linear relationship between receptor activation and nuclear SMAD concentrations (for a review see Whitman, 1998). Secondly, the actual transcriptional response to SMAD activation appears to be strongly dependent on the simultaneous binding and probable complex formation with other transcriptional regulators such as the transcription factors FAST-1, FAST-2, or Jun/Fos in vertebrates (for a review see Derynck et al., 1998). The spatial distribution of such regulators themselves could be important for gradient interpretation.

So far attention has been paid mainly to proteins which, together with the SMADs, act as activators of transcription. However, one case is known from C. elegans where activated SMADs counteract the specific repression exerted by a DNA- binding inhibitory SMAD (Patterson et al., 1997; Thatcher et al., 1999). In addition, recent evidence was provided from Drosophila that one purpose of SMAD signaling is to antagonize transcriptional repression of Dpp target genes (Campbell and Tomlinson, 1999; Jaźwińska et al., 1999; Minami et al., 1999). In the wing, the Brinker (Brk) protein acts as a repressor of low and intermediate level Dpp target genes. These genes are expressed upon loss of brk even in the absence of SMAD signaling, suggesting that Brk acts as a direct transcriptional repressor. Since brk expression itself is negatively regulated by Dpp, target gene activation at least in part results from downregulation of brk.

Here, we show that in the early embryo brk plays a role comparable to that in the wing disc, with some differences. In contrast to the wing disc, loss of brk in the embryo leads to ectopic expression of dpp. However, by injection of activated Dpp receptor RNA we confirm that in the embryo dpp is its own target, and thus dpp repression by brk is a special case of target gene regulation. Moreover, most genes that are repressed by brk in the embryo are also targets of repression by maternal Dl, suggesting that brk could act as a corepressor of Dl. However, using double mutant analysis and ectopic expression of brk we show that brk represses Dpp target genes independently of Dl. Consistent with brk’s role in the wing disc, embryonic target gene activation in the absence of brk is independent of SMAD activity. Thus, in both contexts, brk acts parallel or downstream to SMADs as a specific repressor of low and intermediate level Dpp target genes. brk is expressed like short gastrulation (sog) in ventrolateral regions of the embryo abutting the dorsal dpp domain, and in brk mutants dpp expression expands to cover the entire ectoderm. In this situation sog is largely responsible for Dpp gradient formation since brk sog double mutant embryos have almost no polarity information in the ectoderm. The double mutants consist mainly of mesoderm and unstructured dorsal epidermis. Thus, brk and sog together specify the neuroectoderm of Drosophila embryos.

To distinguish hemizygous or homozygous mutant embryos from their heterozygous siblings, balancers were used which carried early expressing lacZ transgenes (FM7c ftz-lacZ; CyO hb-lacZ; TM3 hb- lacZ) whenever possible. In all other cases the genetically expected number of embryos with a given phenotype was confirmed.

brk deficiencies:

Df(1) ct^J4, In (1)dl⁴⁹, f / FM7c ftz-lacZ

Df(1) Sxl-ra/Y y⁺ct⁺Sxl⁺

Df(1) ct^4b1/ FM7

Df(1) RF19 / FM7

brk: yw brk^M68/ FM7c ftz-lacZ

brk germline clones: germline clones were induced in brk^M68FRT 101 / ovo^DIFRT101 second or third instar larvae by the Flp-FRT technique (Chou and Perrimon, 1996).

sog: sog_YS06/ FM7c ftz-lacZ

brk sog: yw brk^M68sog^YS06/ FM7c ftz-lacZ

sog plus four copies of dpp⁺: sog^YS06/ FM7c ftz-lacZ; Dp (2; 2) DTD48 /CyO23, P[dpp⁺]

brk dpp: yw brk^M68/ FM7c ftz-lacZ; dpp^H46sna^IIGtwi^S60/ CyO23, P[dpp⁺]. dpp^H46 is a null allele (Wharton et al., 1993). dpp mutant embryos were identified by absence of ventral furrow formation, which is eliminated by loss of sna and twi.

brk tkv: yw brk^M68/ FM7c ftz-lacZ; tkv^O1/tkv^Sz3 crossed to FM7c ftz-lacZ; tkv^O1/ CyO hb-lacZ. tkv^O1 is an antimorphic allele, tkv^Sz3 is a strong hypomorphic allele (Terracol and Lengyel, 1994; Nellen et al., 1994). Eggs from transheterozygous tkv^O1/tkv^Sz3 females fertilized with tkv^O1 sperm develop ventralized embryos that are comparable to those of dpp null.

brk Medea: yw brk^M68/ FM7c ftz-lacZ; hs-flp 38/+; Med¹e FRT82B / ovo^DIFRT82B females crossed to FM7c ftz-lacZ;; Med¹e FRT82B / TM3 hb-lacZ. Med¹ is a null allele (Das et al., 1998). Med¹ germline clones were induced in the second or third instar larvae by the Flp-FRT technique (Chou and Perrimon, 1996).

brk Tl^9Q: yw brk^M68/ +;; Tl^9Q/+ were crossed to wild-type males. Tl^9Q causes a dominant maternal ventralization (Anderson et al., 1985).

dpp dl: dpp^H46dl¹/ CyO23, P[dpp⁺] were crossed to Dp (2; 2) DTD48 dl¹/ CyO to generate dpp^H46dl¹/Dp (2; 2) DTD48 dl¹ females, which were crossed to dpp^H46/ CyO23, P[dpp⁺] males (Wharton et al., 1993).

UAS brk: Jaźwińska et al. (1999).

tub GAL4: this stock carries a transgene in which the DNA binding domain of GAL4 fused to the VP16 transcriptional activation domain is expressed from the α4-tubulin promoter (gift of D. St Johnston).

For the analysis of embryonic cuticle, non-hatched larvae were dechorionated and mounted in a mixture of Hoyer’s medium and lactic acid 2:1 (Roberts, 1998).

Immunostaining of embryos with rabbit anti-Krüppel was done as described previously (Roth et al., 1989). Detection of transcripts in situ was performed as outlined in Tautz and Pfeifle (1989) and double- label in situ hybridization according to O’Neill and Bier (1994) and Hauptman and Gerster (1996). The embryos were mounted in Durcupan-ACM and sectioned as described in Roth et al. (1989). LacZ expression was detected using rabbit anti-βGal (Cappel) antibodies and by in situ hybridization with an antisense LacZ probe.

Capped Tkv* mRNA (Hoodless et al., 1996) and brk mRNA were made from pCS2-Tkv* or pCS2-brk (Rupp et al., 1994), respectively, linearized with NotI and transcribed using SP6 mRNA polymerase using the Message Machine kit (Ambion). mRNA was injected into preblastoderm embryos as previously described (Ferguson and Anderson, 1992a). For both Tkv* and brk, the injected mRNAs had concentrations of approximately 100 ng/ μl.

The Dpp activity gradient that is established in the dorsal half of cellular blastoderm embryos subdivides the ectoderm into three major territories that can be identified by the cell types they will comprise and/or by the cuticular structures they secrete (Fig. 1A-C; Ferguson and Anderson, 1992a,b; Wharton et al., 1993; Biehs et al., 1996). High levels of Dpp activity specify the dorsalmost region, which differentiates into large amnioserosa cells that express Krüppel (Kr) protein (Fig. 1C; Gaul et al., 1987). These cells form extraembryonic tissue that does not produce cuticle. Lower Dpp levels in the dorsolateral region specify the non-neurogenic ectoderm harboring peripheral nervous system precursors that can be identified as small segmentally repeated snail (sna)-expressing cells (Alberga et al., 1991; PNS in Fig. 1B). This region gives rise to cuticle with weakly pigmented dorsal hairs (DE, dorsal epidermis in Fig. 1A). Finally, the ventral (neurogenic) ectoderm requires very low levels or absence of Dpp to form the neuroblasts of the central nervous system, which also express sna (Alberga et al., 1991; CNS in Fig. 1B). This region secretes cuticle with heavily pigmented ventral denticles (VE, ventral epidermis in Fig. 1A).

The cuticle of brk mutant embryos has an enlarged region carrying dorsal hairs and a smaller region carrying ventral denticles (Fig. 1D). The number of sna-expressing neuroblasts in the ventral neurogenic region is reduced (Fig. 1E). This indicates that brk mutations lead to an expansion of dorsolateral fates and a reduction of ventrolateral fates. However, despite these lateral fate shifts, the number of Kr- expressing amnioserosa cells is not different from wild type (Fig. 1F). Thus, brk specifically affects cell fates depending on low or intermediate levels of Dpp signaling, while those that require peak levels are not altered.

To identify the underlying causes of the visible changes in cell fate, we examined the effect of brk on the expression of two groups of DV patterning genes. The first group consists of dpp, zen and tld, whose expression is initiated very early in syncytial blastoderm stages. Since they are ventrally repressed by Dl protein their expression domains are confined to the dorsal 40% of the egg circumference (Figs 2A, 3A; Rushlow et al., 1987; Ray et al., 1991; Shimell et al., 1991; data not shown for tld). In brk mutant embryos dpp, zen and tld expression is initiated normally as shown for zen in Fig. 3E. However, in contrast to wild type their expression domains expand ventrally during mid-cellularization (Figs 2A,D, 3B,F). Double in situ hybridization shows that dpp expression abuts, and sometimes overlaps, the first cell row of the sna domain whose blastoderm expression marks the mesodermal anlagen (Fig. 2D). These data demonstrate that brk is not required for the early ventrolateral repression of these genes, but is essential to prevent their lateral expansion during cellularization. Since at these stages Dl protein is still present in the lateral nuclei, brk might function to augment the Dl-mediated repression of dpp, zen and tld.

The second group of DV patterning genes includes rhomboid (rho, dorsal expression domain; Bier et al., 1990), u- shaped (ush; Haenlin et al., 1995; Cubadda et al., 1997) and pannier (pnr; Winick et al., 1993; Heitzler et al., 1996), which are not direct targets of repression by maternal Dl. The initiation of their expression during cellularization requires prior formation of the Dpp activity gradient (rho, Bier et al., 1990; ush, C. R., unpublished results; pnr, Winick et al., 1993). Therefore, they are candidates for being direct targets of Dpp signaling in the embryo. They are expressed in domains straddling the dorsal midline that are 12 (rho), 14 (ush) and 32 (pnr) cells wide at cellular blastoderm (cell counts at approx. 50% egg length; Fig. 3I-K). The two narrowly expressed genes rho and ush are not changed in brk mutant embryos (Fig. 3M,N). This is also true for late zen expression, which in brk mutant embryos, as in wild type, refines to a narrow 5- to 6- cell-wide stripe along the dorsal midline despite the prior expansion (Fig. 3C,D,G,H). However, pnr expression expands in brk mutant embryos and low ectopic pnr levels can be seen in a broad lateral domain that stops about five cells short of mesodermal sna expression (Fig. 3O). Thus, brk does not affect Dpp target genes that are expressed in dorsalmost regions and supposedly depend on highest Dpp levels. However, a target gene that is expressed in a wider domain, and is therefore presumably activated by intermediate levels of Dpp, is expanded.

The described alterations in DV patterning appear to result from complete loss of brk function, and deficiencies uncovering brk produce phenotypes identical to the original point mutation brk^M68 (see Materials and methods). This mutation contains a stop codon causing an early truncation of the protein (Jaźwińska et al., 1999). The phenotypes are also not enhanced in embryos derived from females in which germline clones homozygous for brk^M68 had been induced (see Materials and methods), indicating that the zygotic phenotype is not ameliorated by maternal expression.

In summary, brk mutations affect the Dpp activity gradient in the embryo by expanding the domains of expression of dpp and one of its activators (tld) into ventrolateral regions. Despite the uniform expression of dpp in the entire ectoderm, Dpp activity levels appear to be only mildly increased in the ventrolateral region since only low-level (zen) or intermediate- level (pnr) target genes are ectopically expressed, causing a reduction in the size of the nervous system and ventral epidermis accompanied by an expansion of dorsal epidermis. Peak levels of Dpp in dorsalmost positions appear to be normal, judging from both target gene expression and cell type differentiation.

The question arises as to why uniform expression of dpp in the ectoderm is compatible with the substantial degree of DV polarity exhibited by brk mutant embryos. In wild type, Dpp activity is polarized by sog expression in the ventrolateral region of the embryo (Fig. 2B) such that ventrolateral Dpp activities are reduced and a peak of activity is established centered on the dorsal midline (Zusman et al., 1988; François et al., 1994; Biehs et al., 1996). Embryos mutant for sog show a reduction of ventrolateral fates, albeit to a weaker degree than brk embryos (Fig. 1G,H; see legend for details). In contrast to brk embryos, they differentiate only a small number of scattered amnioserosa cells (Fig. 1I,). The lateral fate shift is not accompanied by strong expansion of dpp transcription as seen in brk embryos (Fig. 2C), and pnr expression is not as greatly expanded as in brk embryos (Fig. 3L).

In brk embryos, sog is still expressed in the ventrolateral domain of the ectoderm (Fig. 2E). To test whether this expression accounts for the DV polarity in the ectoderm of brk embryos, we constructed brk embryos that were also mutant for sog. In contrast to either single mutant, the ectoderm of brk sog embryos forms only dorsal-type cuticle hairs and completely lacks ventral denticles (Fig. 1M). During germ band extension, neuroblast expression of sna cannot be detected; instead sna expression in the peripheral nervous system (PNS) precursors, normally restricted to dorsolateral regions, expands into the ventral region of the embryo (Fig. 1N). This suggests a complete deletion of the ventral neurogenic region. We confirmed this observation using a number of different molecular markers for the neuroectoderm including msh (D’Alessio and Frasch, 1996), the achaete-scute transcript l′sc (Cabrera et al., 1987) and the Enhancer of split transcript m5 (Knust et al., 1992). In all cases the early neuroectodermal expression of these genes was eliminated in double mutant embryos, although expression was only slightly affected in each of the single mutants (data not shown). This indicates that Dpp activity is elevated in the ventrolateral regions of brk sog double mutants to reach levels that completely suppress neurogenesis. Despite this, there is no corresponding increase in Dpp activity at the dorsal side of double mutant embryos. This can be seen by comparing sog plus extra copies of dpp+ (Ferguson and Anderson, 1992b; Biehs et al., 1996) with brk sog embryos. In the former case Dpp levels are increased along the entire embryonic circumference, which leads not only to a strong reduction of the ventral ectoderm (Fig. 1J,K), but also to the rescue of the amnioserosa defect of sog. Moreover, this rescue is accompanied by a large increase in the number of amnioserosa cells compared to wild type (Fig. 1L). In contrast, in brk sog double mutant embryos the amnioserosa defect of sog is not rescued (Fig. 1O). This demonstrates that brk action is confined to the ventrolateral region and that despite ectopic expression of dpp in brk mutants there is no detectable increase in Dpp activity at the dorsal side even in a situation where Dpp activity levels are limiting.

Taken together, brk and sog have both overlapping and distinct roles in shaping the Dpp activity gradient of the Drosophila embryo. While sog has an important function in providing peak levels of the gradient necessary for amnioserosa development, brk and sog together are essential to limit the ventral extension of the anti-neurogenic activity of dpp. Interestingly, brk sog double mutants do not completely eliminate all polarity of the ectoderm. This can be seen from the expression of pnr, which in the double mutant has uniformly high ectodermal levels except in a 5-cell wide stripe bordering the mesoderm (Fig. 3P). Thus, in addition to brk and sog, other ventrally localized factors provide patterning information for the ectoderm. Candidates for this function are members of the spitz group of genes (Schweitzer et al., 1995; Golembo et al., 1998).

In contrast to the embryo, in wing imaginal discs brk mutant cells ectopically express Dpp target genes without affecting dpp expression itself. Furthermore, this ectopic target gene expression does not require the components of the Dpp signaling pathway. This indicates that brk is a negative transcriptional regulator of Dpp target genes in wing imaginal discs that is counteracted by Dpp signaling (Campbell and Tomlinson, 1999; Jaźwińska et al., 1999; Minami et al., 1999). Embryonic and imaginal disc data can be reconciled if dpp in the embryo is a target gene of Dpp signaling, directly or indirectly, as has been suggested earlier (Rushlow and Roth, 1996; Biehs et al., 1996). To test this possibility rigorously, we injected mRNA encoding an activated version of the Dpp type I receptor Thick veins (Tkv*, Hoodless et al., 1996) into embryos from Toll^9Q heterozygous mothers. Toll^9Q is a dominant ventralizing mutation which causes high nuclear levels of Dl protein to be uniformly present in the ectodermal region. This results in repression of dpp in the entire trunk region of the embryo (Fig. 4A; Anderson et al., 1985; Ray et al., 1991). At the site of Tkv* mRNA injection, expression of dpp was observed (Fig. 4B), confirming that activation of the Dpp signaling pathway can induce expression of dpp. We then asked whether the same effect can be produced by removal of brk in a Toll^9Q background. This was the case; brk embryos derived from Toll^9Q mothers show uniform expression of dpp throughout their entire ectoderm (Fig. 4C). This suggests that ectopic dpp expression in brk mutants does not require prior expression of the Dpp ligand. This conclusion was confirmed by double mutant analysis using an amorphic dpp allele (dpp^H46), which in early embryos produces normal amounts of dpp transcripts (Fig. 4D; Wharton et al., 1993). To identify the dpp mutant embryos we used a triple mutant chromosome which, in addition to dpp, carries sna and twi mutations. dpp sna twi triple mutant embryos can be recognized by the lack of ventral furrow formation (Fig. 4D). brk; dpp sna twi quadruple mutant embryos (see Materials and methods) show expanded dpp and zen expression (Fig. 4E,G) and secrete cuticles exhibiting a partial transformation of ventral into dorsal epidermis (data not shown). Thus, the brk phenotype in the embryo, as in imaginal discs, does not depend on endogenous dpp activity.

We further confirmed that neither receptor activity nor SMAD function are required for target gene expression in the absence of brk. To that end, we constructed double mutants of brk with mutations in the Dpp receptor tkv (Terracol and Lengyel, 1994; Nellen et al., 1994; Brummel et al., 1994; Penton et al., 1994) or in the cytoplasmic signal transducer Medea (Hudson et al., 1998; Das et al., 1998; Wisotzkey et al., 1998). Both brk; tkv and brk; Medea double mutant embryos (see Materials and methods) showed expanded dpp expression (Fig. 4F; data not shown for tkv). These observations are most significant for Medea since the allele we used (Medea¹) is predicted to cause a truncation that completely eliminates Medea function (Das et al., 1998). Thus, brk could function as a specific transcriptional repressor of Dpp target genes, as previously suggested for imaginal discs. However, the experiments presented so far do not rule out the possibility that in the embryo brk acts as a corepressor of Dl, since the target genes studied here, dpp, tld and zen, are all targets of Dl repression, and since the expansion of pnr in brk embryos (Fig. 3O) might be a secondary consequence of the expansion of dpp. Therefore, we tested whether pnr expression, which is absent in dpp mutants (Winick et al., 1993), is restored in brk; dpp sna twi quadruple mutant embryos. Indeed, the quadruple mutants show uniform expression of pnr along the DV axis at levels corresponding to those found in lateral region of brk embryos (Fig. 4H). These data strongly support the view that brk acts as specific repressor of low and intermediate-level Dpp target genes. High level target genes are not affected by brk and are consequently not expressed in brk; dpp sna twi quadruple mutant embryos as shown for ush (Fig. 4I).

In imaginal discs brk is expressed complementary to regions of Dpp signaling. The same principle applies for its expression during embryonic development. In late syncytial and in cellular blastoderm embryos brk is expressed in a ventrolateral stripe, which is initially 9-10 nuclei wide and expands to encompass 18 nuclei shortly before gastrulation (Fig. 5A,B). A narrow gap separates the initial stripe dorsally from the expression domain of dpp (data not shown). At the ventral side expression stops one cell row short of the snail domain (Fig. 5J). Thus, the brk domain is largely overlapping sog expression, in agreement with our finding that brk and sog together are required for the establishment of the ventral neurogenic ectoderm (Fig. 5K). Furthermore, the domain also corresponds to the region of the embryo showing ectopic dpp, zen and tolloid misexpression in the absence of brk. This expression pattern is thus consistent with the view that brk may directly repress the transcription of these genes.

During germ band extension brk continues to be expressed in the ventral ectoderm where it appears to be restricted to neuroectodermal cells during neuroblast segregation (Fig. 5C,D,L). New sites of expression appear in the ventral mesoderm at stage 10 (Fig. 5E,L), when Dpp-signaling from the ectoderm induces dorsal mesoderm (Frasch, 1995). During stage 11 some ectodermal cells surrounding the tracheal pits express brk (Fig. 5F). The tracheal pits lie in between the dorsolateral and the ventral stripe of dpp expression, and patterning of the invaginating tracheal cells has been shown to be influenced by dpp (Wappner et al., 1997; Vincent et al., 1997). During germ band retraction the developing midgut broadly expresses brk (Fig. 5G). In stage 13, three localized domains of expression can be seen in the visceral mesoderm and endoderm of the developing gut: a domain slightly anterior to the region of the developing gastric caeca, a domain approximately at the location of parasegment 5 and a posterior domain at the border between hindgut and midgut (Fig. 5H). These regions lie between domains of dpp expression in the visceral mesoderm where dpp mediates inductive processes between visceral mesoderm and endoderm (for a review see Bienz, 1996). Finally, brk is expressed during dorsal closure in a broad stripe of ectodermal cells localized ventrally to the leading edge cells in which dpp expression controls morphogenesis (Fig. 5I; for a review see Noselli, 1998). In summary, these expression patterns suggest a potential role of brk in all major aspects of Dpp signaling in the embryo.

The ventrolateral expression of brk in early embryos suggests that brk, like sog and rho, is a target gene of the maternal Dl protein gradient (Ip et al., 1992; François et al., 1994). In support of this notion, we found that brk expression is completely abolished in maternally dorsalized embryos (Fig. 6A). Conversely, in maternally ventralized embryos derived from the Toll^9Q heterozygous mothers (used above in Fig. 4), brk expression is initiated along the entire embryonic circumference except in the presumptive mesoderm (Fig. 6B). In sna twi mutant embryos (Fig. 6D) and in sna single mutants (data not shown) brk expression is uniform at the ventral side. Thus, as shown for rho (Ip et al., 1992), sna might be a ventral repressor of brk transcription.

The complementarity between brk expression and regions of Dpp signaling in the embryo might arise if brk is itself negatively regulated by Dpp, as previously shown in imaginal discs (Campbell and Tomlinson, 1999; Jaźwińska et al., 1999; Minami et al., 1999). To test this idea, we examined brk expression in dpp mutant embryos. Here, brk expression is normal before the onset of gastrulation, but subsequently expands towards the dorsal side of the embryo so that brk becomes uniformly expressed in the entire ectoderm (Fig. 6E). The opposite phenotype results if dpp expression expands into the ventrolateral region, as in a sog mutant embryo with extra wild-type copies of dpp (Fig. 2N; Frasch, 1995; Rushlow and Roth, 1996; Biehs et al., 1996). These embryos exhibit a strong repression of brk transcription in the ventrolateral region although a small domain of brk expression is maintained close to the border of the mesoderm (Fig. 6F). This residual expression might be responsible for the narrow stripe of neuroblasts which still forms in sog embryos with four copies of dpp+ (Fig. 1K). We wondered whether the expansion of brk expression in dpp mutants requires the previous Dl-dependent activation of brk transcription. In dl dpp double mutant embryos, brk is initially not expressed (data not shown); nevertheless, uniform brk expression is initiated during gastrulation (Fig. 6C). Thus, absence of dpp leads to derepression of brk irrespective of whether Dl is present, indicating that other mechanisms of transcriptional activation of brk exist that are normally counteracted by Dpp signaling.

Ectopic pnr expression in brk; dpp double mutant embryos shows that brk does not entirely function as a corepressor of Dl. However, the latter still could apply to brk function in regulating dpp, tld and early zen. To test whether brk can act as repressor in the absence of Dl we first expressed UAS-brk under the control of a maternally expressed GAL4 driver to achieve uniformly high-level expression of brk in early wild-type embryos. This completely suppresses the formation of the dorsal dpp domain (Fig. 7E,G). It also abolishes terminal expression of dpp, which is not subject to regulation by Dl (see dpp expression in the termini of Tl^9Q embryos in Fig. 4A). The resulting embryos secrete only cuticle with ventral denticle belts and resemble dpp mutant embryos (Fig. 7A,C). Thus, early uniform expression of brk can completely block all dpp expression. We also injected brk mRNA into dl mutant embryos that uniformly express dpp (Fig. 7F). At the site of injection dpp is repressed (Fig. 7H) and the injected embryos form cuticle with ventral denticles that resembles the cuticle from dl dpp double mutants (Wharton et al., 1993; Fig. 7B,D). However, since dpp is not uniformly repressed cuticles often show a transition from ventral epidermis to dorsal epidermis with increasing distance from the site of injection. In the region of transition dorsolateral structures are formed, such as Filzkörper (FK in Fig. 7D). This indicates that different amounts of repression by brk might lead to different ectodermal cell fates.

In summary, the ectopic expression data show that brk repression of dpp is independent of prior repression by Dl and therefore brk functions in early embryos, like in imaginal discs, as a regulatory component of the Dpp signaling pathway.

In Drosophila Dpp acts as a morphogen in two completely different developmental contexts: in DV patterning of the early embryo and in AP patterning of the appendage primordia (Ferguson and Anderson, 1992a,b; Wharton et al., 1993; Nellen et al., 1996; Lecuit et al., 1996). In the first case, gradient formation is a fast process and occurs in an epithelium in which no cell divisions take place; in the second case gradient formation is a slow process and occurs in an epithelium that constantly grows by cell division. In each context the gradients are established by completely different molecular mechanisms: in the embryo mainly by diffusion of an inhibitor (Sog) into a region of uniform expression of dpp (François et al., 1994; Holley et al., 1996), and in the imaginal discs mainly by the spreading of Dpp protein from a local source of dpp transcription in the center of the primordium (Lecuit and Cohen, 1998). In addition, in each context Dpp signaling affects a completely different set of target genes. Despite these differences there are some remarkable similarities, which might identify mechanisms important for gradient formation and morphogen function. In both cases two different BMP-type ligands each activate a different type I receptor, which results in synergistic effects important for correct establishment of the activity gradients (Nguyen et al., 1998; Neul and Ferguson, 1998; Haerry et al., 1998; Khalsa et al., 1998). In both cases receptor expression itself is influenced by signaling which has been shown to be important for spreading of Dpp protein in the wing disc (Biehs et al., 1996; Lecuit and Cohen, 1998; Haerry et al., 1998). Finally, in both cases brk acts as negative regulator of low or intermediate level target genes, being itself negatively regulated by signaling.

In the wing disc brk mutant clones have completely cell autonomous effects and lead to cell fate transformations corresponding to low or intermediate levels of Dpp signaling (Campbell and Tomlinson, 1999; Jazźwińska et al., 1999; Minami et al., 1999). Mutant cells autonomously express the Dpp target genes optomotor blind (omb) and spalt (sal) without expressing dpp. Omb and sal each represent a target gene of a different class with regard to brk regulation. While omb shows normal levels of expression in brk clones, sal, the target gene depending on higher Dpp levels, is only weakly expressed. This indicates that sal integrates brk-dependent and brk- independent mechanisms for its activation. High-level target genes which are not affected by brk have not been found in wing discs so far; however, the cell fate transformation caused by ectopic expression of Tkv* indicate that such target genes exist (Jaźwińska et al., 1999). Since in the complete absence of MAD, which is essential for all aspects of Dpp signaling, removal of brk leads to the ectopic expression of omb and sal, we concluded that Dpp signaling acts to relieve brk repression. brk is expressed in lateral regions of the wing disc in a pattern which might largely result from transcriptional repression by Dpp. This might also be the predominant mechanism by which Dpp signaling relieves Brk’s repression of target genes.

The relationship between dpp and brk is more complex in the embryo than in the wing imaginal disc for several reasons. Firstly, brk influences dpp expression in the embryo and thus changes the Dpp activity gradient. Therefore, it is more difficult to judge whether target gene mis-expression in brk mutants is a direct result of removal of brk or of the expansion of Dpp. Secondly, three of the genes, dpp, zen and tld, whose expression is expanded in brk, are also targets of the maternal Dl gradient, and so are subject to both Dl-mediated and Brk- mediated repression in early embryos. Thirdly, brk itself is clearly an activated target of Dl when it is first required for repression of dpp, tld and zen; only later during gastrulation does Dpp negatively regulate brk. Finally, in the embryo but not in the disc, brk acts with sog in an intricate way to shape the Dpp gradient.

Despite these complexities brk’s relationship to the Dpp activity gradient in early embryos and to the Dpp gradient in wing discs is very similar. In both contexts low and intermediate-level targets are misexpressed in Brk mutants independently of Dpp signaling. Using brk; dpp double mutant embryos we demonstrated that the ectopic expression of zen and pnr in brk embryos is not a secondary consequence of the activation of dpp, but occurs even in the complete absence of Dpp signaling. pnr activation in brk; dpp double mutant embryos furthermore confirms that repression by brk is not dependent on prior repression by maternal Dl since pnr is not a direct target gene of Dl repression. pnr’s regulation by brk is strikingly similar to brk’s regulation of sal in the wing disc. pnr and sal both are intermediate-level Dpp target genes. In brk mutant cells they both show weak levels of expression compared with the expression in their normal domains. Thus, in the embryo and in the disc three types of Dpp target genes can be distinguished with regard to regulation by brk: those that are fully activated upon loss of brk, those that are weakly activated and those that are not affected. The regulation of the first group might occur entirely by antagonizing brk’s repression; the second group requires both relief of repression by brk and positive activation by Dpp; and the third group might only be subject to direct activation by Dpp.

The strongest phenotypic effects of brk mutations we have seen so far are in places where Dpp acts as a morphogen. In both the wing disc and embryo brk regulates Dpp target genes that are activated at the lower end of the gradient. In these regions special mechanisms of gradient interpretation might be required which are sensitive to small changes in Dpp activity. Target gene regulation by brk could provide such a mechanism. The full activation of low-level target genes requires only transcriptional repression of brk by Dpp. However, a more non- linear mechanism suited to establish sharp threshold responses would result if target gene promoters had both repressive brk and activating SMAD binding sites. Then, Dpp would simultaneously downregulate brk transcription and antagonize its function at the target gene promoters. Such a dual mechanism is the more likely as our data demonstrate that it operates for the intermediate targets sal and pnr. However, experimental support for such a model can only come from a detailed analysis of how brk interacts with the target gene promoters.

Originally the brk mutant phenotype in the embryo attracted our attention because of its similarity to the phenotype caused by sog with extra copies of dpp (Ferguson and Anderson, 1992b). Since both lead only to a partial deletion of the neurogenic ectoderm we speculated that Brk and Sog had parallel functions and thought that Brk might be another extracellular inhibitor of BMP, like Noggin in vertebrates. Indeed a large number of secreted molecules have been shown to act as BMP inhibitors in early vertebrate development (Cho and Blitz, 1998). However, the epistasis analysis clearly shows that Brk acts in the absence of Dpp and SMAD proteins and thus is likely to be a transcription factor. In accordance with the genetic data Brk protein has a potential DNA binding domain and harbors a motif found in many transcriptional repressors (Campbell and Tomlinson, 1999; Jaźwińska et al., 1999; Minami et al., 1999). Thus, in Drosophila the neurogenic region is established by three levels of repression: firstly, maternal Dl represses dpp and the Dpp-group genes tld and zen so that already during syncytial blastoderm stages their expression is confined to the dorsalmost 40% of the egg circumference. Secondly, Dl activates brk in ventrolateral regions. During the late cellular blastoderm stage repression by Dl alone becomes insufficient and brk, presumably acting also as a transcriptional repressor, is required to prevent the ventral expansion of dpp, zen and tld. Thirdly, Dl activates sog which, even in the absence of brk when dpp is uniformly expressed in the ectoderm, acts as a sufficiently strong inhibitor of Dpp to allow formation of neurogenic ectoderm. Only when both brk and sog are removed is the ectoderm almost apolar and early neurogenesis completely suppressed.

Expression of brk in Xenopus embryos antagonizes BMP-4 signaling and leads both to a dorsalization of the ventral marginal zone and to neural induction in ectodermal explants (Minami et al., 1999). Thus, a Brk-like activity probably exists in vertebrates that antagonizes BMP and its target genes at the intracellular level providing another means of negative regulation in addition to the large number of extracellular BMP inhibitors.

The authors would like to thank Oliver Karst for excellent technical assistance. We thank Serena Silver for help with the double in situs, Jo Jack and Tom Kline for deficiencies and duplications in the brk region and advice about them, Eric Wieschaus for making brk germline clones, Manfred Frasch and Rolf Reuter for looking at embryos with us, and Regine Terracol, Rob Ray and Daniel St. Johnston for providing fly stocks. We are grateful to Eric Wieschaus for insightful discussions and critical reading of the manuscript. A. J. is a fellow of the Gottlieb Daimler-und Karl Benz-Stiftung.