The decapentaplegic gene: a TGF-β homologue controlling pattern formation in Drosophila

The TGF-β family comprises a diverse set of 12 polypeptides having profound (generally inhibitory) effects on the growth and/or differentiation of many cell types. Typically, there are no obvious common physiological themes which would have predicted the relationship of these polypeptides to one another, but rather their homology has been revealed through searches of available protein data bases. In the four years following the reported sequencing of the prototypical member of the family, now called TGF-β1 (Derynck el al. 1985), the family has grown to include one polypeptide identified in the amphibian Xenopus laevis [a vegetal pole specific maternal transcript, Vgl (Weeks and Melton, 1987)], 9 other polypeptides identified in humans and other mammals [a Vgl cross-homologous gene, Vgr-1 (Lyons et al. 1989), TGF-β2 (Marquardt el al. 1987), TGF-β3 (Dijke et al. 1988), the inhibin aand ft subunits (Mason et al. 1985; Mayo et al. 1986), Müllerian inhibiting substance, MIS (Cate et al. 1986), three bone morpho genesis proteins BMP-2A, BMP-2B and BMP-3 (Wozney et al. 1988)], and the decapentaplegic product in Drosophila [dpp (Padgett et al. 1987)].

The mammalian polypeptides are processed to generate secreted dimeric proteins which are thought to bind to cell surface receptors on target cells (Cheifetz et al.1986). The first functional demonstration of a biologically active receptor for TGF-β1 and TGF-β2 has recently been reported (Boyd and Massagué, 1989). No definitive information is available on signal transduction systems which are activated in target cells in response to ligand-receptor binding. Several members of the TGF-β family have been implicated as key factors in determinative or differentiative decisions, and therefore, this family of intercellular signalling molecules is of special interest to developmental biologists. In this report, I will briefly review the current state of information regarding the TGF-β family, and will then focus on our current view of the one member of the family thus far identified in invertebrates: the decapentaplegic gene in Drosophila.

In general, the members of the TGF-β family have been isolated on the basis of specific in vitro assays: growth inhibition, differentiation inhibition, asymmetric distributions of transcripts. Upon further study, however, it has been found that at least some of these polypeptides have profound and singular effects on the growth or differentiation of many other cell types. Furthermore, by in situ localization of transcripts or by immunohistochemistry, it has been shown that at least some TGF-β family gene products are localized in many discrete sites during vertebrate development. However, in most cases, specific in vivo roles remain uncertain.

TGF-β1 was discovered as one of two proteins (the other being TGF-α, which is a member of the EGF family of growth factors) present in conditioned medium of transformed mammalian cell lines which would cause phenotypic transformation of the immortalized NRK (normal rat kidney) cell line, such that it would now exhibit anchorage-independent growth in soft agar (Roberts et al. 1985). Since its initial discovery based on this phenotypic transformation assay, TGF-α1 activity and responsiveness have been found in many normal cell types in vivo and in cell culture. It is especially abundant in bone and platelets (Roberts et al. 1981; Assoian et al. 1985). Its physiological functions in vivo have not been established. In addition, its effects in vitro have proven quite wide ranging. Under many conditions of treatment, it turns out to be growth inhibitory rather than growth promoting (Roberts et al. 1985). In other circumstances, it leads to the differentiation of particular cell lines in culture (Ignotz and Massagué, 1985). In part, the effects of TGF-β1 seem to depend upon the profile of other factors which are added along with it (Assoian et al. 1984). Many of the actions of TGF-β1 can be explained as the indirect consequences of changes in the architecture of the extracellular matrix (Ignotz and Massagué, 1986; Allen-Hoffmann et al. 1988; Montesano and Orci, 1988).

Recent observations have uncovered two proteins (TGF-β2 and TGF-β3) which are close relatives of TGF-β1. Both TGF-β1 and TGF-β2 homodimers have been identified. They are somewhat different in their binding to putative TGF-β receptors (Cheifetz el al. 1987), suggesting that they might have different biological activities. TGF-β3 was recovered on the basis of its cross-homology to TGF-β1 and TGF-β2 (Dijke et al.1988). No information is available on its structure or biological effects.

Inhibin, a secretion found in ovarian fluid, suppresses the release of follicle-stimulating hormone (FSH) by the pituitary gland (Ramasharma et al. 1984), thereby interfering with ovulation. It is an α/ βheterodimer in which both subunits bear structural homologies to the TGF-βs (Mason et al. 1985). In pigs, two genes encoding related βchains have been found (Mason et al. 1985). β-chain homodimers, or heterodimers of these β-chains, have been found to have the opposite effect of promoting FSH release; these proteins have been termed ‘activins’ (Ling et al. 1986; Vale et al. 1986).

Müllerian inhibiting substance (MIS) is a homodimer secreted by the fetal testis that contributes to conversion of the indifferent urogenital system into the male reproductive system, by causing regression of the Müllerian ducts (Picon, 1969). It is thus a key factor in the determination of the somatic sexual phenotype of the mammalian fetus. Molecular analysis has revealed it to be a divergent member of the TGF-β family (Cate et al. 1986).

These proteins were identified as products that have the ability to induce bone growth, using the ability to induce infiltration and differentiation of osteoblasts in demineralized bone as an assay system (Wang et al. 1988). Two of the proteins, BMP-2A and BMP-2B, are highly conserved (and hence their names), while BMP-3 is more divergent (Wozney et al. 1988). No information is available on the in vivo roles of these three proteins.

The sequencing of Vgl cDNAs has revealed that the protein encoded by Vgl transcripts is a member of the TGF-β family (Weeks and Melton, 1987). Vgl RNA was first identified as the only transcript that was highly concentrated at the vegetal pole of unfertilized Xeno-pus oöcytes (Melton, 1987; Pondel and King, 1988). As the embryo begins development, the maternal Vgl RNA becomes incorporated into vegetal cells, which will eventually form the endoderm. At this stage of gastrulation, endodermal cells induce ectoderm to form mesoderm. ATGF-β-like molecule has been implicated in this induction process (Slack et al. 1987; Kimelman and Kirschner, 1987; Rosa et al. 1987), using TGF-β1 and TGF-β2 polypeptides from mammalian sources. Given the distribution of Vgl RNA, it has been suggested that Vgl protein is a strong candidate to be the natural mesoderm-inducing factor (Weeks and Melton, 1987). Characterizations of Vgl and of mesoderminducing factor are the subjects of other reports in this volume.

Recently, using Vgl probes, Lyons et al. (1989) were able to isolate a cross-homologous mammalian gene from a mouse cDNA library. This gene, called Vgr-1 (Vgl-related), is expressed in numerous tissues throughout development. No information is available on the structure of the Vgr-1 protein, or on its biological role.

In every case in which the protein has been characterized, TGF-β family members are active as secreted dimers. The subunits of these dimers are about 110–130 amino acids in length and derive from longer precursors, ranging in size from approximately 300–600 amino acids. These precursors share several structural features. They all have N-terminal secretion signal sequences and several internal potential N-glycosylation sites; both of these features are consistent with the secreted nature of the polypeptides. No transmembrane domains are apparent in the protein sequences. The precursors all contain several basic residue pairs (arginine-arginine, arginine-lysine, lysine-lysine), at least some of which are the sites of cleavage in the processing of the precursor (e.g. Derynck el al. 1985). The amino acid conservations that characterize the TGF-β family are confined to the C-terminal 110 amino acid residues. As noted above, it is this C-terminal region that dimerizes to form the mature secreted product. In general, the extensive regions TV-terminal to the mature product are poorly conserved between different family members. Nonetheless, if the same member is examined in different vertebrate species, strong homologies in more N-terminal regions are identified as well (e.g. Derynck et al. 1986). In the case of TGF-/J1, a cleavage product of the N-terminal domain remains noncovalently associated with the C-terminal dimer in a latent form of the secreted molecule (Lyons et al. 1988).

With the recent expansion of the membership of the TGF-β family, it has been possible to discern subgroups on the basis of conserved or divergent structural features of the C-terminal portion of the polypeptides. The conservations among the members of the family generally reside in three domains surrounding six of the seven conserved cysteines present in each polypeptide. TGF-β1, TGF-β2, TGF-β3 and inhibin-βhave two additional conserved cysteines, placing these molecules in a different subfamily from the others. MIS and inhibin-o-show only a low level of conservation to the other polypeptides and to each other (∼20–25%). BMP-2A, BMP-213, BMP-3, Vgl, Vgr-1 and dpp exhibit at least 45% conservation (at the level of amino acid identities), and appear to be in their own subfamily. Below, we will discuss evidence that BMP-2A, BMP-2B and dpp are very closely related, and may actually represent the dipteran and mammalian derivatives of the same ancestral gene. The relationships of the different TGF-β family members to one another are demonstrated by a comparison of primary amino acid sequences of their C-termini (Fig. 1).

In summary, in the course of the four years since the first of these sequences - TGF-β1-was published, this vertebrate polypeptide family has come to include a diverse and physiologically unrelated set of factors with a range of very intriguing developmental effects. The homology of the decapentaplegic (dpp) polypeptide, which will be described in the next section, demonstrates that at least one ancestor of this gene family predates the divergence of arthropods from vertebrates, and provides yet another set of important developmental functions controlled by a representative of the TGF-β family.

We have now characterized several cDNAs corresponding to 5 different classes of overlapping dpp transcripts (Padgett et al. 1987; St. Johnston et al. 1989; Padgett et al. in preparation). Each transcript appears to encode the same polypeptide. We have not addressed the questions of possible heterogeneity among the transcripts in terms of stability or translational efficiency. The structures of the cDNAs that correspond to these transcripts are depicted in Fig. 2. These five transcript classes have three exons apiece, and they differ only in their first (j.e. 5’) untranslated exon. These untranslated exons appear to be initiated by different promoters and range in size from about 150 bases up to 2 kb. Each transcript class utilizes the same splice acceptor site in the second exon. There is only a single pattern of splicing between the second and third exons.

The first AUG in the open reading frame (ORF) of each transcript is located near the beginning of the common second exon. The ORF extends through the middle of the third exon. If translation initiated with the first AUG, the polypeptide that was produced would be 588 amino acids in length (Padgett et al. 1987). We think it likely that the first AUG of the ORF does indeed represent the actual translation initiation codon because (a) it conforms to the consensus sequence for Drosophila initiation codons and (b) unlike other AUGs in the ORF, it is followed by a consensus N- terminal signal secretion sequence (Watson, 1984), as expected for a secreted TGF-)5-like polypeptide.

Several potential glycosylation sites can be found within the dpp ORF, consistent with the view that the dpp polypeptide is itself a secreted protein. Several dibasic residues are found within the ORF, suggesting that the primary polypeptide product might be proteolytically cleaved to generate a smaller mature product. The positions of these sites are diagrammed in Fig. 3.

The C-terminal ∼110 amino acids of the dpp polypeptide show extensive similarity to the comparable regions of the vertebrate TGF-β family members (Fig. 1). Three arginine-arginine dipeptides reside just A-terminal to this region of high conservation; these potential proteolytic cleavage sites are thus positioned analogously to the known cleavage sites in most of the other family members. The dpp polypeptide contains the same seven cysteines in its putative C-terminal fragment as are shared by the vertebrate family members. The locations of the similarities between the dpp polypeptide and the vertebrate members of the TGF-β family conform to the regions of similarity among the vertebrate family members themselves. These conservations reside in three domains surrounding six of the seven C-terminal cysteines. We presume this pattern of conservation reflects a common structural motif in this family of molecules.

With the recent report of the primary sequences of human BMP-2A and BMP-2B (Wozney et al. 1988), we now have two very strong candidates for the vertebrate analogues of dpp. The C-termini of BMP-2A and BMP-2B are each 75 % identical in amino acid sequence to the corresponding region of the dpp polypeptide. Moreover, within the N-terminal precursor region, there is 25–30% amino acid identity. The only other polypeptides in the family which show any significant conservation in the N-termini are between BMP-2A and BMP-2B themselves, and among TGF-β1, TGF-β2 and TGF-β3. Based on these considerations, we think it very likely that the dpp gene of Drosophila and the BMP-2A and BMP-2B genes of humans are the modern highly conserved derivatives of a gene which was present in an ancestor common to the arthopod and vertebrate lineages.

Numerous observations on the structure and function of the dpp gene have coalesced into a clear picture of its organization. In brief, the dpp gene (map location 2–4.0; 22F1,2) is a rather large genetic unit (>55kb), which produces a single TGF-β-homologous polypeptide (St. Johnston et al. 1990; Padgett et al. in preparation). This polypeptide is encoded by at least 5 overlapping transcripts (St. Johnston et al. 1990; Padgett et al. in preparation). These transcripts are initiated by different promoters and contain different 5’ untranslated exons. Each transcript has its own temporal (and presumably spatial) pattern of expression (Gelbart et al. 1985; St. Johnston and Gelbart, 1987; St. Johnston et al. 1989; Posakony et al. 1990). The bulk of the genetic information contained in the >55 kb of dpp DNA consists of c/s-regulatory elements controlling the timing, location and quantity of dpp transcription (Gelbart et al. 1985; St. Johnston et al. 1990; Blackman et al. in preparation). These c/s-regulatory elements must drive a very complex pattern of expression, as the dpp gene is required for numerous developmental events, including determination of dorsal ectoderm along the dorsoventral axis of the early embryo (Irish and Gelbart, 1987; St. Johnston and Gelbart, 1987), proper morphogenesis of the adult appendages derived from the imaginai disks (Spencer et al. 1982; Posakony et al. 1990; Spencer et al. in preparation), and other, less well-defined events necessary for proper development of the larva and adult (Segal and Gelbart, 1985).

Some mutations in the dpp gene interfere with the production of an active dpp polypeptide product. They are identified as lesions that affect all developmental events controlled by dpp (Irish and Gelbart, 1987). Null (amorphic) mutations are called dpp^Hin alleles, because they are Haplo-insufficient that is, dpp^Hin/ + animals are lethal, apparently because there is too little dpp product in the early embryo to produce proper determination along the dorsoventral axis. Most of these are likely to act by altering the structure of the dpp polypeptide, while the remainder probably derange the structure of all of the transcripts. Some dpp^Hin alleles have lesions that can be identified by restriction mapping; these lesions cluster in a region of about 6 kb which we term the Hin region (St. Johnston et al. 1990). This region also contains the two exons encoding the open reading frame for the dpp polypeptide. Leaky (hypomorphic) mutations are not haplo-insufficient; because they map to the Hin region but are recessive, they are called dpp^hin-ralleles. All dpp^hin-r alleles have normal restriction maps in the Hin region. We think it likely that some of the dpp^Hin and many of the dpp^hin-rlesions are point mutations altering the amino acid sequence of the dpp polypeptide.

Other dpp mutations affect only a subset of the diverse array of dpp functions. We think that these mutations (virtually all of which are inversions, translocations or deletions: Spencer et al. 1982; Segal and Gelbart, 1985; St. Johnston et al. 1990) represent lesions disrupting batteries of cfs-regulatory elements necessary for the proper spatial and temporal deployment of dpp transcripts. This is best exemplified by a consideration of the disk region, which is a >25 kb untranscribed region 3′ to the polyadenylation sites of all dpp transcripts.

Mutations in the disk region are recessive lesions that disrupt proper dpp expression in the imaginai disks, leading to the production of distally incomplete appendages. Several classes of such mutations have been described, with the most severe removing all but the most proximal regions of each appendage. All mutations of the disk region are recessive to wild-type.

The mildest alleles of the disk region are defective only in wing posture. These ‘heldout’ (dpp^d-ho) mutations cause the wings to be held out laterally and perpendicular to the body axis; this phenotype is associated with the loss of specific pattern elements (a group of 25 sensory structures called the sensilla campaniformia-25 or Sc25) on the dorsal base of the wing (Spencer et al. 1982). Most mutations of the disk region have been recovered as dpp^d-ho alleles (Fig. 4). When examined in homozygotes, most have a more severe array of mutant phenotypes (Spencer et al. 1982; Irish and Gelbart, 1987; Blackman et al. 1987). These mutations are generally associated with gross chromosomal rearrangements. Mutant homozygotes lack structures derived from distal portions of the adult appendages. Mild alleles (dpp^disk-11) only affect the wing, halter and male genital derivatives. Intermediate alleles (dpp^disk-11) affect all major appendages (wing, halter, leg, eye, antenna, proboscis, male and female genitalia), but sufficient appendages remain to permit dpp homozygotes to survive to adulthood. Severe alleles (dpp^disk-11) cause such severe defects in all of the appendages that they engender death early in pupation. The defects in the adult structures are due to corresponding abnormalities in the imaginai disks (the precursors of the adult appendages). Distal structures of adult appendages are known to derive from central regions of the imaginai disks (Bryant, 1978). These central regions are defective in these dpp homozygotes. The amount of imaginai disk material lacking in these mutant individuals parallels the severity of the adult defects (Spencer et al. 1982).

We have several lines of indirect evidence that the disk region is cú-regulatory in nature, (a) When duplications covering the Hin region, but not the disk region are used to rescue pseudopoint dpp^Hin mutations past the Haplo-lethai embryonic period, dpp^Hin mutations exhibit a phenotype indistinguishable from the most severe disk region phenotype, exhibited by dpp^disk-Vhomozygotes, (b) The pattern of expression of the dpp transcripts emerging from the shv-Hin region is indistinguishable from the location of imaginai disk cells requiring normal disk region expression for proper appendage formation (Spencer et al. in preparation; Posakony et al. 1990). (c) We have no clear evidence of transcriptional activity within the disk region (St. Johnston et al. 1989). (d) The closer to the transcribed shv-Hin region that mutant breakpoints in the disk region fall, the more severe are their mutant phenotypes (Fig. 2). All of these mutations break within the disk region, and can be viewed as removing all disk region cis-regulatory DNA proximal to the breakpoint from the neighborhood of the dpp transcription unit (Fig. 4).

Based on these observations, we conclude that the disk region acts by controlling the expression patterns of shv-Hin transcripts in the imaginai disks. Similar considerations lead us to conclude that the Hin region must contain c/s-regulatory elements for the embryonic ectodermal expression of the dpp transcripts (St. Johnston and Gelbart, 1987; Irish and Gelbart, 1987; Hoffmann and Goodman, 1987), and the shv region must contain cis-regulatory elements driving still other spatial and temporal patterns of dpp expression (Segal and Gelbart, 1985; St. Johnston et al. 1990).

The nature of the dpp protein product, combined with the developmental effects of dpp mutations, has led us to speculate that dpp contributes to Drosophila development at the level of intercellular signalling of positional information (Padgett et al. 1987). By analysis of chimeric individuals containing marked dpp∼ tissue, we have made several observations (chiefly on wing disks and their derivatives) which bear on possible models of dpp’s developmental role (Spencer et al. in preparation; Posakony et al. 1990). (1) In mosaics that are mixtures of genotypically dpp⁺ and dpp^d-111 tissue, each imaginai disk solely determines the dpp phenotype of its consequent adult appendage. (2) In those mosaics in which a single imaginai disk contains a mixture of genotypically wild-type and mutant cells, the derivative adult appendages may be fully normal, even though 50 % or more of the pattern elements that would be lacking in wholly mutant appendages are of the mutant genotype. (3) Alternatively, such disks of mixed dpp genotype can sometimes give rise to appendages that are partially or fully mutant in phenotype. (4) The dpp phenotype of a genotypically mixed imaginai disk depends solely upon the genotype of the cells in the region of the disk just anterior to, and abutting, the anterior-posterior (A/P) compartment boundary. We term this region the ‘focus’ of dpp disk expression. (5) dpp transcripts accumulate in a pattern concordant with the position of the dpp focus, i.e. along the A/P compartment boundary.

Based on these results, we conclude that dpp cannot act as an exocrine function, secreted from one organ to act on distant target tissues. However, within an imaginai disk primordium, we cannot determine if the primary action of dpp is strictly local, or if it acts on target cells throughout the disk.

One surprising aspect of our observations on dpp expression within the imaginai disks is its dissimilarity to the phenotypes dpp elicits in adult appendages. While the basic phenotype engendered by dpp^disk mutations are the loss of tissue in a center to peripheral (radial) direction within the disk, transcript accumulation and localization of the dpp focus occur as a stripe cutting across the entire disk at the level of the A/P compartment boundary. In other words, there is no apparent asymmetry of dpp expression along the radial axis of the imaginai disk. How then does dpp cause pattern deletions along the radial axis, leading to distally incomplete appendages?

One possible explanation is suggested by Meinhardt (1982). Based on theoretical considerations, he supposed that intersections of compartment boundaries (A/P and dorsal/ventral) would identify unique sets of cells, which would then serve as point sources of radial positional information. Given our observations on dpp expression in disks, I suggest that the strip of expression of dpp (which is constrained by the position of the A/P compartment boundary) serves as a reference line for a series of intersections which lead to the elaboration of radial positional information (detailed in Fig. 5). A gradient of dpp product within the disk may be one component of this positional signalling system, or alternatively, the localized expression of dpp along the A/P boundary may contribute to the activation of a different positional signalling molecule. In either case, it would be the integrated information from the intersection of dpp expression and the expression of other localized molecules within the disk that would lead to the elaboration of radial positional information. The wingless gene, which encodes an apparent secreted factor and which is expressed in a different but also highly localized pattern in all disks, is an intriguing candidate for another component of the positional signalling system (Cabrera et al. 1987; Rijsewijk et al. 1987; Baker, 1988).

In essence, then, I am proposing that dpp contributes to the establishment of central positions within developing imaginai disks. The establishment of these central positions would then, directly or indirectly, lead to the elaboration of radial positional information. Perhaps the establishment of central positions, together with the pre-existence of peripheral positional values, could lead to the elaboration of intermediate positions through a process of intercalation. While there are some difficulties in fitting the range of dpp adult appendage phenotypes into such a model, it does have the attractive feature of relating to observations coming from so-called ‘competence’ experiments in imaginai disk development. As first noted by Schubiger (1974) in transplantation experiments in which immature leg disks were forced to metamorphose precociously, there is a regional hierarchy in the abilities of disk cells to differentiate adult cuticular structures. In the youngest disks at all competent to elaborate adult structures, structures from the very periphery and the very center of the disk differentiate. In somewhat older disks, larger swatches of peripheral and of central tissue differentiates. As the age of the disk increases, the size of the swatches increases until finally, in mature disks, the entire radial pattern is differentiated. A similar scenario holds for the wing disk (Bownes and Roberts, 1979). It is as if the cells forming center and periphery are first specified during development, and as if further specification occurs by intercalation. Curiously, the first structures from the leg and the wing that are removed by the mildest dpp alleles affecting those tissues are precisely the first structures competent to be differentiated from the central regions of those disks (the tarsal claws of the leg in dpp^d-111homozygotes, and the Sc25 of the dorsal base of the wing in dpp^d-ho homozygotes). Thus, dpp mutations and competence experiments appear to identify the same set of structures as central within the developing disk. Whether this relationship is of importance in understanding the molecular basis of positional specification within the developing disks must await future experimentation.

This work summarizes the excellent and invaluable intellectual and technical contributions of many investigators over a ten year period, most notably (in alphabetical order) Ronald Blackman, F. Michael Hoffmann, Vivian Irish, Richard W. Padgett, Leila Posakony, Laurel Raftery, Daniel Segal, Forrest Spencer and R. Daniel St. Johnston. This work has been supported by research grants from the Public Health Service and from the American Cancer Society.