Mutational analysis of the Drosophila tolloid gene, a human BMP-1 homolog

Dorsal cell fates in the early Drosophila embryo are mediated by a signal transduction pathway utilizing a member of the transforming growth factor-β (TGF-β) superfamily, the decapentaplegic (dpp) gene (Padgett et al., 1987). The TGF-β superfamily comprises a large family of processed, secreted ligands that bind to specific cellular receptors. Among family members, dpp shows the most homology to the bone morpho- genetic proteins 2 and 4 (BMP-2 and BMP-4) (Wozney et al., 1988; Padgett et al., 1993), exhibiting 75% identity in the ligand domain. Substitution of the human ligand domain from BMP-4 for the dpp ligand domain results in viable larvae, reversing the extreme ventralization of dpp null embryos, rescuing them from lethality (Padgett et al., 1993). The data obtained from these rescue experiments suggest that the same TGF-β-like signal is interpreted differently in cells with alter- native developmental potential. These results also suggest that these molecules, and their mode of signal transduction, are highly conserved through evolution.

Several ‘accessory’ proteins are known to interact with TGF-β-like proteins to modulate their activity. The cleaved precursor domain of TGF-β1, the TGF-β1 binding protein, and the proteoglycan decorin have been shown to form complexes with the bioactive TGF-β1 ligand to regulate its activity (Miyazono et al., 1988, 1991; Wakefield et al., 1988; Gentry and Nash, 1990; Kanzaki et al., 1990; Yamaguchi et al., 1990; Border et al., 1992). Activin, another TGF-β family member, can be isolated bound to follistatin, an activin binding protein (Nakamura et al., 1990) that inhibits the normal action of activin. However, the mechanism of action of these ‘accessory’ proteins remains largely unknown.

dpp is a member of a secreted growth factor family (TGF- β superfamily; Padgett et al., 1987) and therefore a ligand for a cell signalling system. There exists a dorsal to ventral activity gradient of dpp in the early Drosophila embryo, which is capable of specifying at least three distinct cell fates (Ferguson and Anderson, 1992b). Some of the other members of the zygotic ventralizing group (tolloid, screw, shrew, twisted gas- trulation, short gastrulation, and zerknüllt (zen), Jurgens et al., 1984; Nüsslein-Volhard et al., 1984; Wakimoto et al., 1984; Wieschaus et al., 1984; Zusman and Wieschaus, 1985) might be involved in this signal transduction pathway, functioning similarly to the proteins that modulate the activity of TGF-β, or may be a member of the TGF-β family. dpp defects confer the strongest mutant phenotype of this group, suggesting it plays the central role. Homozygotes do not complete germ band extension, and develop to first instar larvae lacking all dorsal epidermal derivatives. The resulting strongly ventral- ized phenotype is characterized by ventral denticle bands encir- cling the entire embryo (Irish and Gelbart, 1987). Recessive embryonic lethal mutations in dpp are partial loss of function alleles, where weak alleles delete only the amnioserosa and stronger alleles delete the amnioserosa and some dorsal epidermal structures. Because the tolloid mutant phenotype resembles the weak dpp recessive phenotype (Jurgens et al., 1984; Irish and Gelbart, 1987), this gene is a good candidate for a modulator of dpp, and therefore was more closely examined at a molecular and genetic level. Several lines of evidence further support the hypothesis that dpp and tolloid may operate in the same pathway. Both genes affect similar mitotic domains, acting in a concerted manner to alter the fate of the amnioserosa and dorsal ectoderm (Arora and Nüsslein-Volhard, 1992). Additional evidence comes from genetic analysis of tolloid and dpp (Ferguson and Anderson, 1992a): certain tolloid alleles act as dominant enhancers of dpp. In most cases, however, animals that possess heterozygous recessive alleles of dpp and tolloid are viable. Only certain tolloid alleles cause lethality with some dpp recessive alleles. These recessive alleles of dpp do not interact with tolloid deficiencies to produce lethality. Thus, the tolloid alleles that interact in a dominant manner are antimorphs, because they result in a more severe mutant phenotype than does a deficiency of the tolloid gene.

The tolloid gene product is highly homologous, in sequence and in structure, to another human bone morphogenesis protein, bone morphogenetic protein 1 (BMP-1; Wozney et al., 1988; Shimell et al., 1991). These proteins possess several structural motifs; the N-terminal portion of the molecule has a zinc-binding metalloprotease domain similar to the astacin protease from the crayfish Astacus fluviatilis (Zwilling and Neurath, 1987). The C-terminal region contains EGF repeats, and repeats found in the complement proteins C1r and C1s (Journet and Tosi, 1986; Mackinnon et al., 1987). Both EGF and C1r/s motifs are strongly implicated in protein-protein interactions (Villiers et al., 1985; Weiss et al., 1986; reviewed by Davis, 1990).

We have analyzed the nature of the antimorphic tolloid alleles to better understand the relationship between tolloid and dpp. Our analysis shows that five of the seven antimorphic mutations reside in the protease domain, where they presum- ably abolish enzymatic function. The interacting EGF and C1r/s repeats remain functional and capable of forming a complex containing DPP. This effectively sequesters DPP in an inactive state. In contrast, analysis of tolloid alleles which do not interact identified missense mutations in the C-terminal EGF and C1r/s repeats, or termination codons resulting in the deletion of these repeats, which would render them non-func- tional. These tolloid mutants are unable to form a complex with DPP, which remains free to function. We propose that the tolloid protein functions in the DPP pathway by interacting, via its C-terminal EGF and C1r/s repeats, with a complex con- taining DPP. Further, the tolloid protease activity is needed, either directly or indirectly, for the activation of the DPP complex. This model is further supported by the fact that the mammalian homologs of dpp and tolloid, (BMP-2 and BMP- 1, respectively), copurify from bone extracts (Wozney et al., 1988), suggesting a physical interaction.

tolloid alleles were isolated in two independent screens, which sought alleles that failed to complement a tolloid deficiency or a tolloid mutation, using standard EMS mutagenesis procedures (Lewis and Bacher, 1968). From these screens, we recovered the following tolloid alleles: tld^E4, tld^E5, tld^E6, tld^E7, tld^E8, tld^E9, tld^E10, tld^E11, tld^E14, and tld^E15. The tolloid alleles tld^7M, tld^7H, tld^6B, tld^7O, tld^6P, and tld^10E were isolated in a screen for mutations on the third chromosome that produce zygotic patterning defects (Jurgens, et al., 1984; Tearle and Nüsslein-Volhard, 1986). The dpp^hr27 and dpp^hr4 alleles were induced in F₂ lethal screens designed to identify mutations that failed to com- plement dpp^disk alleles (Spencer et al., 1982), dpp^hr90 (formerly known as dpp^IC) was isolated in a screen for embryonic patterning mutants (Nüsslein-Volhard et al., 1984), dpp^hr56 is described in Irish and Gelbart (1987) and dpp^hr92 was isolated by R. Blackman and W. Gelbart (Wharton et al., 1993). All other genes and balancer chro- mosomes are described in Lindsley and Zimm (1992).

Cuticle preparations were done according to Wieschaus and Nüsslein-Volhard (1986). The severity of the tolloid mutant pheno- types was determined primarily by observing the extent of development of the filzkörper of the mutant cuticles. The tolloid alleles we examined can be ordered in an allelic series based on the severity of the mutant cuticles: weak alleles: tld^E4, tld^E6, tld^E8 ; moderate alleles: tld^E5, tld^E10, tld^E14, tld^E15 ; and strong alleles: tld^E7, tld^E9, tld^E11.

The tld^γ1 deficiency was generated by irradiating males containing the transposon P[(w⁻, ry⁺)G]3 (Levis et al., 1985) with 4000 rads. This transposon was determined to be in the tolloid region (96A17- 20) by chromosomal in situ hybridizations using biotin labeled DNA as a probe on larval chromosomes. rosy offspring were selected and balanced using TM3, Sb. Each of these rosy stocks was tested for failure to complement the tolloid gene, resulting in the identification of tld^γ1. Mutations were induced on the tld^γ1 deficiency, which was cytologically normal, to obtain some of the tolloid mutations described above.

Flies carrying tolloid alleles were crossed to flies carrying dpp recessive alleles to determine if there was a genetic interaction between the two genes. First, flies containing either a tolloid defi- ciency, a presumed tolloid null mutation or an interacting tolloid allele were crossed to a series of dpp alleles, and the adult flies were scored and counted from each cross. Between 500 and 1000 progeny were obtained from each cross. The percentage survival was determined from the ratio of the number of progeny carrying the dpp and tolloid alleles in a heterozygous state compared to the number of their siblings carrying a balancer chromosome. If there is no genetic inter- action, the number of heterozygotes should roughly equal the number of sibling flies carrying one of the balancer chromosomes. In several cases, the number of flies heterozygous for tolloid and dpp mutant alleles was greater than the number of sibling flies containing a balancer chromosome, and is noted in the tables as ‘>100%’. This set of crosses was used to determine which dpp allele was most useful in indicating a potential genetic interaction with tolloid. We chose the dpp^hr92 allele to test the remaining tolloid alleles for genetic interactions.

To isolate tolloid genomic DNA from the isogenic strain st e, a λDASH II (Stratagene) library was constructed from size selected Sau3A partials. One million phage were packaged and amplified on plates. Overlapping phage containing the tolloid genomic region were isolated from this library. Southern blots of digests generated with several different restriction enzymes were probed with tolloid sequences to identify a large fragment that contained the entire tolloid gene. It was determined that the tolloid gene was contained within a 14 kb HindIII fragment.

Mutant tolloid sequences were cloned from genomic DNA as follows: DNA from each mutant strain was isolated from about 50 adult flies by homogenization with a Kontes grinder in a microfuge tube (modified from Rubin Lab Methods Book, unpublished). DNA was digested with HindIII and cloned into λDASH II (Stratagene). tolloid mutant HindIII fragments were distinguished from the analogous balancer HindIII fragments by an EcoRI restriction poly- morphism. Because we planned to use the tolloid mutant clones in other experiments, we opted not to clone the alleles via polymerase chain reaction (PCR), which might introduce random mutations into the amplified DNA.

The sequence of the tolloid alleles was determined by the dideoxy chain termination technique, using Sequenase version 2.0 (United States Biochemical). Since all the tolloid mutations were induced on isogenic chromosome strains, we could directly compare the mutant and wild-type tolloid sequence. A set of 20 oligonucleotides spanning the gene was used to sequence the 17 kb plasmids containing the tolloid mutant clones, using double stranded sequencing methodologies. DNA was sequenced in microtitre trays essentially as described (Bankier et al., 1987). The samples were either run on buffer gradient gels (Biggin et al., 1983), or run on sodium acetate gradient gels (Sheen and Seed, 1988). Computer analysis of sequences was done using GCG programs, TFASTA (Devereux et al., 1984; Pearson and Lipman, 1988) and BLAST (Altschul et al., 1990).

Deficiencies in tolloid were produced by three methods. First, the rosy gene in the P[(w⁻, ry⁺) G]3 (Levis et al., 1985) construct was mutated using γ-rays, and progeny were examined for a rosy phenotype. Six lines were established and tested for loss of tolloid function. One line, tld^γ1, was cyto- logically normal and likely represents no more than a 100 kb deletion. Since the γ-ray mutagenesis indicated that the P element was physically very close to tolloid, we surmised that it would be possible to generate lesions in tolloid using P[ry⁺, Δ 2-3](99B) and abortive jumps of P[(w⁻, ry⁺)G]3 (Robertson et al., 1988). rosy alleles were generated using P[ry⁺, Δ2- 3](99B), and each rosy line was tested for lethality over the deletion tld^γ1 to determine which lines mutated to a lethal com- plementation group by an abortive jumping event. These lines, which occurred once in every fifteen crosses, were tested for failure to complement the tolloid point mutation, tld^10E. Three lines were recovered, at a rate of one in 450 crosses. Defi- ciencies containing tolloid were also generated by γ-rays using standard techniques, looking for lesions that failed to complement a tolloid point allele.

In addition, we initiated a genetic screen for tolloid point mutations. Ten alleles were generated in F₂ lethal screens and tested for failure to complement a tolloid deficiency or a tolloid point allele. As indicated, this screen was done without bias for alleles that interact with dpp. We also tested several previ- ously generated tolloid alleles isolated in an F₂ screen for homozygous embryonic mutations (Jurgens et al., 1984; Tearle

To characterize the interaction of tolloid and dpp, we tested representative tolloid alleles against a series of graded dpp recessive alleles (Table 1). We tested two strong tolloid alleles, tld^6P4 and tld^10E (Tearle and Nüsslein-Volhard, 1986), tld^γ1 (a small deficiency for tolloid) and four point alleles from our mutant screens, for potential interactions with a series of dpp recessive alleles, dpp^hr4, dpp^hr27, dpp^hr92, dpp^hr90 and dpp^hr56. The number of trans heterozygotes (tld/+; dpp/+) was compared to the number of adults containing only one of the two alleles (+/dpp; +/TM3, Sb) and expressed as a percentage of that class. We find that with any tolloid allele, a genetic interaction is only observed between the two genes when a strong dpp allele is used. This suggests that the interaction is not allele specific or domain specific, but rather depends on reduced activity of the dpp protein. In contrast to these results, crosses with our tolloid deficiency resulted in no interaction with any of the dpp recessive alleles.

Based upon these results, we selected a strong recessive dpp allele, dpp^hr92, to test other tolloid alleles. Eight alleles from the Tübingen stock center and ten alleles from our screens were tested for genetic interaction with dpp (Table 2). In these crosses, we find that seven tolloid alleles act as dominant enhancers of the dpp phenotype, as evidenced reduction in the number of flies het- erozygous for tolloid and dpp (<50% of expected). Two of the tolloid alleles that we generated for this study (tld^E11 and tld^E15) are antimorphic, indicating that roughly one-sixth of the tolloid alleles mutate to a form that genetically interacts with dpp.

To determine the amino acid changes encoded in these tolloid alleles, we made a λDASH II library of the isogenic st e strain, on which many of the tolloid mutations were induced. A genomic clone was chosen because the sequence of genomic DNA would allow detection of point mutants that affected splicing of the tolloid gene. tolloid-containing phage were isolated from a genomic walk of the 96A region of chromosome three (R.W.P., unpublished data). The sequence of the genomic region of tolloid is shown in Fig. 1 (GenBank accession no. U04239). It differs from the sequence published by Shimell et al., (1991) by only five nucleotides, none of which change an amino acid.

The transcription unit of tolloid is simple, consisting of six introns, the longest of which is 144 bp. At the 5′ end, about 600 bp separate tolloid from the adjacent gene, tolkin (also referred to as tolloid related-1). The tolloid cis-regulatory sequences must be present within about 6 kb upstream of the tolloid coding sequences, because a transposon containing these sequences rescues embryos mutant for tolloid (Shimell et al., 1991). Most or all of tolloid’s cis- regulatory elements probably lie in the 600 bp between these two genes, since the 5′ end of tolkin is located several kb upstream (R. W. P., unpublished data). It is unlikely that the introns contain important cis-regulatory information because of their short length.

Sequence analysis of six antimorphic alleles identified missense mutations in the metalloprotease domain (tld^E11, tld^E15, tld^6P4, tld^7M, and tld^10E) (Fig. 2). In most cases, these lesions occur at conserved sites and are non-conservative mutations (Fig. 3). The antimorphic phenotype resulting from these amino acid changes in the protease domain strongly suggests that tolloid is an active protease that functions in the dpp pathway by proteolytic cleavage of one or more molecules involved.

Sequence analysis of two of the antimorphic alleles, tld^5H and tld^9Q, revealed non-conservative amino acid changes in the C2 repeat domain. These mutations allow a full length protein to be synthesized. Without structural data of tolloid, we do not know how the protease and the C2 repeat are positioned. However, we can speculate that the function of the C2 domain may be different from that of the more C-terminal EGF and C1r/s repeats, which may mediate protein-protein interactions with the DPP complex. This N-terminal C2 repeat may play a structural role in allowing both functional domains of tolloid (protease and protein-interacting domains) to function properly. Alternatively, the C2 domain may have an active role in mediating interactions with other molecules, perhaps the tolloid protease substrate, or the protease that activates the tolloid zymogen. In these models, these two antimorphic alleles would have mutations outside the protease domain that prevent the proper functioning of the catalytic enzyme, even if the enzyme is capable of activity.

We also cloned and sequenced eleven non-interacting tolloid alleles. Most tolloid alleles that do not genetically interact with dpp contain either point mutations in EGF#2 or C1r/s repeats C3, C4, and C5, or encode proteins that truncate these domains (Fig. 2). tld^E7 contains a nonsense mutation five amino acids into the first C1r/s repeat. This mutation indicates that the protease domain alone is insufficient to produce an interaction, and further supports the observation that mutations in the inter- acting alleles must interfere with protease activity, but do not alter the protein-interacting EGF and C1r/s repeats. Three other non-interacting alleles contain mutations that result in termination codons. They encode mutant proteins containing some of the more N-terminal EGF and C1r/s and interacting domains, yet they are apparently not able to interact with the other proteins in the DPP complex, as evidenced by their mutant phenotype. This lends further support to the postulation that the more C-terminal EGF and C1r/s repeats (namely EGF#2, C3, C4, C5), are involved in mediating protein-protein interactions with components of the DPP complex, while the other EGF and C1r/s repeats may play a different role.

Most of the antimorphic tolloid alleles contain mutations in the protease domain. However, we expected that some non- interacting alleles would also contain mutations in the protease domain. These mutations might disrupt the proper folding or stability of the tolloid protein, resulting in the inactivation of the interacting domains, as well as inactivating the protease. In these cases, the tolloid protein would be incapable of interacting with the DPP complex to generate the antimorphic phenotype. We identified two alleles that fit this model. One allele, tld^E5, results in a non-conservative amino acid change in the protease domain, and does not interact genetically with dpp. Another non-interacting allele, tld^E8, contains a 24 bp in- frame deletion occurring in the protease domain.

In this study, we have examined eighteen mutations in the tolloid gene. These mutations were selected for failure to complement a tolloid deficiency, for failure to complement a strong tolloid allele, or for failure to complement as homozygotes. The correlation of the location and type of tolloid mutations with their observed genetic interaction with dpp is informative; those alleles that genetically interact to worsen the dpp phenotype mostly contain missense mutations in the protease domain, leaving the EGF and C1r/s protein-interacting domains intact and functional. These data suggest that tolloid protease activity is necessary for activation or modification of the DPP complex, while the C-terminal protein-interacting domains are required for binding to the complex. In support of this model, the copurification of the human homologs of tolloid and dpp, BMP-1 and BMP-2 in bone extracts, suggests a physical association between these two proteins. However, based on these experiments, we cannot state conclusively that TOLLOID and DPP interact directly, or that TOLLOID interacts with other proteins in the DPP complex, although this seems to be the simplest explanation. Alternatively, TOLLOID could possibly interact with another protein that subsequently interacts with DPP.

Five of the seven antimorphic tolloid alleles were found to contain mutations in the protease domain that resulted in amino acid changes. All these changes result in non-conservative amino acid changes, except in one case where an arginine is changed to a histidine. These mutations likely abolish protease activity, without disrupting the function of the C-terminal EGF or C1r/s domains, which are implicated in forming protein- protein interactions with the DPP complex. These mutant tolloid proteins could still form a complex with DPP via these EGF/C1r/s repeats, but proteolytic activation would not be possible. The net effect is that DPP is sequestered in an inactive complex. In addition to providing evidence that the function of the C1r/s and EGF repeats is to mediate protein interactions with a DPP complex in the dpp pathway, the mutations in the protease domain also support the idea that tolloid encodes an active metalloprotease. Mutations in this region, which occur in the highly conserved amino acids of the domain, produce a mutant phenotype. Since astacin has been crystallized recently (Bode et al., 1992), it will be possible to model the structure of tolloid to astacin when the coordinates become available, and to determine the effect of the tolloid mutations on the structure and function of the protein. Upon visual inspection of the stereo-diagrams in Bode et al. (1992), we can assert the following: amino acid substitutions at positions 272 (tld^6P4) and 276 (tld^10E) are near the active position 240 (tld^E11) may affect access to the cleft. Defects caused by other mutations cannot be determined without further structural analysis.

Two of the seven antimorphic tolloid alleles contained mutations in the second C1r/s repeat, C2. These data suggest several models of a role for this repeat distinct from that of the more C-terminal repeats that mediate protein-protein interactions with the DPP complex. This is evident when comparing the locations of the mutations in the EGF and C1r/s repeats that are antimorphic, versus those which do not interact. The two antimorphic alleles contain mutations in the C2 domain, while the alleles that do not interact contain point mutations exclusively in the C-terminal repeats (C3, EGF#2, C4, C5), or are truncations, deleting these repeats. The C2 domain may interact with the protease that cleaves the tolloid zymogen, or may interact with the tolloid protease substrate. Another possibility is that the C2 domain may form protein-protein interactions with another molecule that is necessary for activation, or some modification of the DPP complex, but is not itself a component of the DPP complex.

Alternatively, the C2 domain may be a ‘spacer’ region, necessary for bringing the two functional domains (protease and C-terminal EGF and C1r/s repeats) in correct spatial proximity to allow both binding of a complex containing DPP, and cleavage of the protease substrate. Mutations here might alter conformation of this ‘spacer’ such that it affects the orientation of these functional domains.

A third possibility is that these mutations in the C2 domain alter the conformation of the more C-terminal EGF and C1r/s repeats, such that they have an enhanced affinity for the DPP complex, perhaps keeping DPP permanently bound so that it cannot function. An analogous model has been suggested for the Abruptex alleles of the Notch protein (Heitzler and Simpson, 1993), in which single amino acid changes in EGF repeats influence EGF repeats 11 and 12, which are responsible for interaction with Delta (Rebay et al., 1991). Heitzler and Simpson (1993) suggest that one possible explanation for the enhanced ability of Abruptex alleles to bind Delta is that the mutated EGFs cause EGF repeats 11 and 12 to adopt a con- formational change that renders them more accessible for inter-sites of the enzyme. The substitution at position 317 (tld^E15) is on an alpha helix positioned near the active site. These mutations may interfere with the catalytic activity of the protease. The astacin protease has an active site that is at the bottom of a deep cleft that positons the substrate for cleavage. The substitution at action with Delta. Non-conservative amino acid changes in complement domains, such as those in C2 of tld^5H and tld^9Q, may change the conformation of these repeats which might influence the structure of the neighboring C-terminal EGF and C1r/s domains, such that they interact more strongly with the DPP complex, or are more accessible for this interaction.

Phenotypic analysis of the tolloid mutant cuticles indicates that there is not a correlation between the severity of the mutant phenotype and whether it is an antimorphic allele. An anti- morphic allele may give rise to embryos exhibiting either stronger or weaker tolloid mutant phenotypes. This could be explained by postulating that tolloid interacts with more than one protein and an alteration of these interactions result in varying degrees of phenotypic changes, depending on which interaction is lost. Further biochemical studies must be done to examine this model.

It is of interest to note that animals possessing an antimorphic tolloid allele also have a wild-type tolloid allele. If the tolloid protein can form multimers with itself, some of the mutant phenotype observed with the antimorphic allele could be due to an interaction with the wild-type tolloid protein, rendering it nonfunctional.

If tolloid/BMP-1 proteases exist in complexes with TGF-β-like proteins, the possibility exists that they are responsible for processing the growth factors into the mature active C-terminal form. Precedent for this is exemplified by nerve growth factor, which forms a complex with an enzyme that is responsible for its processing (Server and Shooter, 1976). However, indirect evidence suggests it is unlikely that TOLLOID cleaves DPP or other related growth factors. A set of enzymes that is capable of cleaving the TGF-β-like molecules, the subtilisins, has already been isolated from higher eucaryotes (Barr, 1991; Roebroek et al., 1992). They cleave at dibasic residues, and are good candidates for processing TGF-β-like proteins. As a further argument against tolloid/BMP-1 protease involvement with growth factor cleavage, the potential substrates for the astacin protease contain cleavage sites that are entirely different from the known cleavage sites found in TGF-β members. As the sequence of this family of metalloproteases remains very highly conserved, it is unlikely that the substrate cleavage specificity has changed. Thus, TOLLOID probably does not function to cleave DPP, but this issue must await bio- chemical analysis.

Most of the alleles that do not interact with dpp have either non-conservative amino acid changes in the second EGF and C1r/s repeats C3, C4, and C5, or nonsense mutations in these repeats, which result in truncated proteins that delete these repeats. The mutations most likely render TOLLOID incapable of forming the proper interactions with the DPP complex, either through alteration of the interacting domains (i.e., point mutations), or truncation of some interacting domains. According to our model, DPP would form a complex only with active, functional TOLLOID, encoded by the other chromo- some, in which case an antimorphic phenotype is not generated.

One postulated function of the EGF repeat is in the mediation of protein-protein interactions (reviewed by Davis, 1990). A consensus sequence that binds calcium is found in both EGF domains of tolloid (Shimell et al., 1991; our own results). These EGF repeats may interact with each other, or regulate the conformation of adjacent domains in the same protein (reviewed by Davis, 1990; Handford et al., 1990). Calcium-binding EGF domains, bounded by C1r/s repeats also occur in the complement proteins C1r and C1s (Tosi et al., 1987). These EGF domains are thought to regulate the calcium-dependent conformation of the flanking C1r/s repeats, mediating the specific binding of the proteins to each other (Villiers et al., 1985; Weiss et al., 1986; Davis, 1990). A role for the calcium-binding EGFs in tolloid may be to regulate the calcium-dependent conformation of the flanking C1r/s repeats, which would then be capable of specific complex formation with other proteins.

The tolloid protein belongs to a family of ancient and conserved proteins from distantly related organisms, with members in nematodes, sea urchins, insects, and vertebrates. These molecules are characteristically zinc-binding metallo- proteases with attached C-terminal EGF and C1r/s repeats (Fig. 4). These repeats, of which order and number vary between individual members, are strongly implicated in forming regu- latory interactions with other proteins. There exists another member of the tolloid/BMP-1 family in Drosophila, tolkin, which has the same overall structure as tolloid, but exhibits more homology to BMP-1 (A. L. F. and R. W. P., unpublished data). Although we are uncertain of tolkin’s function, it may interact with dpp at a later developmental stage, be involved in the pathway regulating the other known TGF-β-like member in Drosophila, the 60A gene (Wharton et al., 1991; Doctor et al., 1992), regulate a heterodimer of these two growth factors, or interact with other unidentified growth factor molecules. Another possibility may be that TOLKIN interacts with TOLLOID to form a multimer. The BP10 and SpAN proteins from the sea urchin have been cloned, and exhibit the same modular organization as tolloid and BMP-1 (Lepage et al., 1992; Reynolds et al., 1992). A role for BP10 in the differen- tiation of ectodermal lineages and subsequent patterning of the embryo has been suggested (Lepage et al., 1992). A new member of this family has recently been found in C. elegans (Waterston et al., 1992; A. L. F. and R. W. P., unpublished results), but its role is undefined. Since these tolloid/BMP-1- like proteins are found in many different organisms and are conserved, it seems likely that their function will be similar. It appears that they are involved in modulating the activity of growth factors, and thus are critical for specifying cell deter- mination, developmental, and differentiation processes. The observation that these TGF-β-like molecules exhibit diverse developmental roles reflects the fact that it is the developmental potential of the cells receiving the growth factor signal that determines the ultimate consequence of cell fate.

We thank Bill Gelbart for support in the early stages of the tolloid studies, which were begun in his laboratory. Special thanks to Don Nelson and Desmond Smith for help in generating some tolloid alleles used in this study and for discussions about dorsal-ventral patterning. We thank Guy Montelione for assistance in the structural analysis.

We thank Mike O’Connor and Chip Ferguson for exchanging data prior to publication. The P[(w⁻, ry⁺)G]3 transposon line was a gift from Bob Levis. This work was supported by grants to R. W. P. from the National Institutes of Health, the American Cancer Society, the New Jersey Cancer Commission, and a Biomedical Research Support Grant. C. A. B. was supported by a New Jersey Cancer Commission Postdoctoral Grant. A. L. F. is a Benedict-Michael Graduate Fellow and T. X. is a Busch Predoctoral Fellow.