Genetic Interactions Between the RhoA and Stubble-stubbloid Loci Suggest a Role for a Type II Transmembrane Serine Protease in Intracellular Signaling During Drosophila Imaginal Disc Morphogenesis

Corresponding author: Laurence von Kalm, 4000 Central Florida Blvd., University of Central Florida, Orlando, FL 32816-2368., lvonkalm{at}mail.ucf.edu (E-mail)

Communicating editor: K. ANDERSON

	ABSTRACT

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

The Drosophila RhoA (Rho1) GTPase is essential for postembryonic morphogenesis of leg and wing imaginal discs. Mutations in RhoA enhance leg and wing defects associated with mutations in zipper, the gene encoding the heavy chain of nonmuscle myosin II. We demonstrate here that mutations affecting the RhoA signaling pathway also interact genetically with mutations in the Stubble-stubbloid (Sb-sbd) locus that encodes an unusual type II transmembrane serine protease required for normal leg and wing morphogenesis. In addition, a leg malformation phenotype associated with overexpression of Sb-sbd in prepupal leg discs is suppressed when RhoA gene dose is reduced, suggesting that RhoA and Sb-sbd act in a common pathway during leg morphogenesis. We also characterized six mutations identified as enhancers of zipper mutant leg defects. Three of these genes encode known members of the RhoA signaling pathway (RhoA, DRhoGEF2, and zipper). The remaining three enhancer of zipper mutations interact genetically with both RhoA and Sb-sbd mutations, suggesting that they encode additional components of the RhoA signaling pathway in imaginal discs. Our results provide evidence that the type II transmembrane serine proteases, a class of proteins linked to human developmental abnormalities and pathology, may be associated with intracellular signaling required for normal development.

METAMORPHOSIS in Drosophila is a remarkable developmental process mediated by the steroid hormone 20-hydroxyecdysone (ecdysone; ANDRES and THUMMEL 1992 ; FRISTROM and FRISTROM 1993 ; RIDDIFORD 1993 ). Ecdysone regulates several distinct developmental programs during metamorphosis. The early stages of metamorphosis are characterized by programmed cell death and histolysis of most larval tissues. Concurrently, the precursors of adult structures—the imaginal discs, imaginal rings, imaginal islands, and histoblast nests—undergo dramatic morphogenetic changes. The imaginal discs develop from small clusters of epithelial cells set aside during embryogenesis (COHEN 1993 ; FRISTROM and FRISTROM 1993 ). During the larval period imaginal disc development is characterized by rapid cellular proliferation to form a highly folded sac attached via a stalk to the larval epithelium. The rapid increase in ecdysone titer associated with the onset of metamorphosis triggers unfolding of the imaginal disc epithelium and dramatic changes in tissue shape. Within a few hours the discs form structures resembling the adult appendages, which evert to the outside of the presumptive imago.

Leg imaginal discs are particularly useful for the study of epithelial tissue morphogenesis. Leg imaginal disc morphogenesis has been extensively characterized at a cell biological level, providing a solid foundation for further genetic and molecular analyses (CONDIC et al. 1991 ; FRISTROM and FRISTROM 1993 ). Morphogenesis of the leg begins early in the 12-hr prepupal stage, defined as the period between puparium formation and head eversion, and is completed later during the pupal period. Prepupal leg development is characterized by a morphogenetic process collectively referred to as evagination: elongation of the presumptive leg as the disc epithelium unfolds and concurrent eversion to the outside of the animal. The prepupal elongation of leg discs to form presumptive leg segments is of particular interest because it is driven primarily by apical cell shape changes (CONDIC et al. 1991 ).

The current model describing the mechanical forces driving cell shape changes and elongation in leg discs is based on studies in vertebrate and invertebrate epithelia (reviewed in FRISTROM and FRISTROM 1993 ). Cells within the epithelium of leg imaginal discs are linked via subapically localized adherens junctions. The cytoplasmic side of each adherens junction is connected to an actin-nonmuscle myosin II contractile ring. Genetic studies demonstrate that the myosin motor protein is essential for the shape changes that occur during disc evagination (EDWARDS and KIEHART 1996 ; HALSELL and KIEHART 1998 ; HALSELL et al. 2000 ). At the onset of metamorphosis nonmuscle myosin II-driven contractility is activated coincident with the rise in ecdysone titer. The head domains of activated myosin dimers bind to and slide actin filaments past each other within the contractile ring. The result is a "purse string" effect whereby contraction of the actin-myosin ring causes the apical surface of the disc cell to adopt an isometric shape (CONDIC et al. 1991 ; see also BEMENT et al. 1999 ; KIEHART 1999 and references therein). The contractile force in one cell is transmitted to neighboring cells via adherens junctions, causing all of the cells of the epithelial sheet to change shape in a concerted manner and coordinating elongation of the presumptive leg (ODELL et al. 1981 ; FRISTROM 1988 ).

A number of genes affecting prepupal leg disc elongation have been identified (see Table 1). Genetic or experimental disruption of these genes is associated with a characteristic malformed leg phenotype in which some or all leg segments are thicker and shorter than normal and often are twisted or kinked (KISS et al. 1988 ; Fig 1, A–D). Confocal microscopic analysis of the leg imaginal disc cells of Sb-sbd and broad (Broad-Complex) mutants shows that these cells fail to undergo proper cell shape changes at the beginning of metamorphosis, suggesting that inability to control cell shape changes underlies the malformed leg phenotype (CONDIC et al. 1991 ; VON KALM et al. 1995 ). Thus the malformed leg phenotype has proven to be a valuable tool to genetically identify gene products that affect cell shape changes during tissue morphogenesis.

View larger version (32K): In this window In a new window Download PPT slide   Figure 1. Defects in leg and wing morphogenesis associated with second-site noncomplementation between RhoA and Sb-sbd mutations. RhoAE3.10/+ (A and E) and Sb63b/+ (B and F) heterozygotes showing wild-type leg and wing morphology. Note the long, slender shape of a normal femur (arrow in A) and tibia (arrowhead in A) from the third leg pair of an adult (A and B). Both Sb and sbd alleles behave as recessive mutations in the context of prepupal leg development. Note the dominant bristle phenotype in the trochanter (open arrowhead in B). Mildly (C and G) and severely (D and H) malformed third legs and wings taken from an animal of the genotype RhoAE3.10/+; Sb63b/+. Note the abnormally short and thick femur and tibia and indentation in the femur (D) and crumpled wing (H).

Genes required for cell shape changes during leg elongation can be divided into two classes on the basis of their response to ecdysone. The Sb-sbd, broad, E74, crooked legs, vulcan, and bancal loci are all transcriptionally induced in response to ecdysone at the onset of metamorphosis (BURTIS et al. 1990 ; ANDRES et al. 1993 ; APPEL et al. 1993 ; BAYER et al. 1996A ; D'AVINO and THUMMEL 1998 ; GATES and THUMMEL 2000 ), whereas the RhoA, zip, sqh, and DRhoGEF2 loci are not induced in response to ecdysone (WARD et al. 2003 , this issue; A. HAMMONDS and J. FRISTROM, unpublished data). Thus, control of cell shape changes during prepupal leg imaginal disc elongation requires the coordinated activities of both ecdysone-inducible and -noninducible proteins.

Because leg morphogenesis is both triggered by ecdysone and dependent on RhoA signaling, we inferred that the activities of the RhoA signaling pathway must be regulated by one or more ecdysone-inducible gene products at the onset of morphogenesis. Conceptually, the ecdysone-inducible Sb-sbd locus, which encodes an apparent type II transmembrane serine protease (APPEL et al. 1993 ), could play a role in RhoA signaling during leg morphogenesis. First, the Sb-sbd locus is transcriptionally induced by ecdysone in leg imaginal discs at the onset of metamorphosis. Second, Sb-sbd mutations interact genetically with mutant alleles of broad, which encodes a family of transcription factors, and zipper, which encodes the heavy chain of nonmuscle myosin II (hereafter referred to as myosin; DIBELLO et al. 1991 ; GOTWALS and FRISTROM 1991 ; YOUNG et al. 1993 ). Broad and zipper mutations also interact genetically with RhoA mutants during leg development (this article; HALSELL et al. 2000 ; WARD et al. 2003 , this issue). Third, upon induction by ecdysone, the Sb-sbd protein is localized to the apical membrane of leg disc epithelial cells (VON KALM et al. 1995 ), potentially placing it upstream of RhoA in a signaling hierarchy. Consistent with the possibility that RhoA and Sb-sbd act in a common signaling pathway, we find significant genetic interaction between Sb-sbd mutations and RhoA mutations and between Sb-sbd mutations and mutations in other RhoA signaling pathway members during leg imaginal disc morphogenesis. Similar genetic interactions are also observed during wing imaginal disc morphogenesis. We also find that a reduction in the dosage of RhoA reduces the penetrance of a leg malformation phenotype associated with overexpression of the Sb-sbd transmembrane protease.

To identify additional genes encoding products that participate in the hypothetical Sb-sbd/RhoA signaling pathway during leg and wing imaginal disc development, we analyzed mutations in six genes on the second chromosome and show that they interact robustly with RhoA, Sb-sbd, and zipper mutants in second-site noncomplementation assays. Three of these enhancer of zipper [E(zip)] genes have been identified as new alleles of the RhoA, DRhoGEF2, and zipper loci that are known to have important roles in RhoA signaling. The remaining three as yet uncharacterized genes are therefore good candidates for key players in the RhoA-regulated mechanochemistry of actomyosin contractility or in the regulation of actin cytoskeletal dynamics (or both) during imaginal disc morphogenesis.

	MATERIALS AND METHODS

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Stocks:
Stocks tested and the molecular nature of mutant lesions (where known) are described in Table 2.

Genetic complementation analysis:
To ensure that the viability of emerging progeny classes was not distorted by overcrowding, the following crossing regimen was established empirically in pilot experiments. In each cross four virgin females (1–7 days old) were mated to five males. All crosses were set up in triplicate on standard corn meal medium. Cultures were incubated for 3 days at 25° followed by transfer of adults to fresh medium. Subsequent cultures were incubated for 2 days at 25°, generating a total of nine cultures for each cross. The legs of all emerging F₁ progeny classes were scored for malformation. An animal was scored as malformed if either the left or the right leg of a pair was malformed. F₁ progeny were scored for 19 days from the date parental adults were first placed in each vial.

Second-site noncomplementation (SSNC) assays were conducted by mating animals heterozygous for the mutant genes of interest. Leg malformation was scored in the doubly heterozygous F₁ progeny class. Genotypes generated to test for dominant interactions (i.e., */+; sbd¹/sbd²⁰¹) were obtained by first mating animals carrying the mutant of interest, e.g., */CyO, to Sco/CyO, P{sevRas1.V12}FK1; red sbd²⁰¹ e/TM6B, Tb Hu e animals. F₁ male progeny heterozygous for both mutations, e.g., */CyO, P{sevRas1.V12}FK1; red sbd²⁰¹ e/+, were mated to virgin female sbd¹ ro e ca homozygotes. Leg malformation was then scored in the relevant F₂ progeny class. For both SSNC and dominant interaction tests, sibling progeny classes were also scored, e.g., +/+; sbd¹/sbd²⁰¹, and the highest percentage of leg malformation observed in any sibling class was subtracted from the percentage of malformation observed in progeny classes being tested for interaction. Typical background leg malformation in sibling progeny classes ranged from 2 to 4%.

Cloning and heat-induced induction of hs-Stubble transgene:
A full-length Sb-sbd cDNA including 838 and 523 nucleotides of 5'-(UTR) and 3'-UTR, respectively, was excised from plasmid pBSSK) - c10h7 with BamHI and Asp718 and cloned into BamHI/Asp718-digested pLitmus 29 (New England Biolabs, Beverly, MA). The resulting clone, pLitmus29 c10h7, was digested with BamHI and SpeI to excise the full-length Sb-sbd cDNA and cloned into BglII/XbaI-digested pCaSpeR-hs (THUMMEL and PIRROTTA 1992 ), correctly orienting the 5' end of the Sb-sbd cDNA with respect to the pCaSpeR-hs promoter. The resulting construct hs-Stubble was injected into w¹¹¹⁸ embryos and transgenic lines were subsequently generated.

Thirty Rho⁷²⁰/CyO or 30 Rho⁷²⁰/CyO, P{sevRas1.V12}FK1 females were crossed to 20 w¹¹¹⁸; hs-Stubble males in bottles at 25° and turned into fresh bottles every 3 days. The progeny were picked as 0-hr white prepupae and transferred to food vials. They were either immediately subjected to a 1-hr heat shock by immersion in a 37° water bath [heat shock 0 hr after pupariation (HS 0 hr AP)] or allowed to develop for 3 hr at 25° before being heat-shocked (HS 3 hr AP). Treated animals were allowed to continue development at 25°. Eclosed animals were sorted into two progeny classes, CyO/+; hs-Stubble/+ or Rho⁷²⁰/+; hs-Stubble/+, and scored for leg malformation. The induction of hs-Stubble at either of these times in development affects second and third leg pairs differently, so the second and third pairs of legs of each animal were scored separately.

Enhancer of zipper mutations:
Mutations in eight second-chromosome genes that behave as second-site noncomplementors of zipper mutant leg and wing defects were isolated as follows. b If males were mutagenized with EMS (LEWIS and BACHER 1968 ) and mass mated to b pr cn sp zip^Ebr/SM5 virgin females. A total of 16,670 b * If/b pr cn sp zip^Ebr G₁ progeny were scored for malformed leg or wings and 365 b * If/b pr cn sp zip^Ebr malformed G₁ animals were observed and backcrossed to b pr cn sp zip^Ebr/SM5 virgin females or males to verify transmission of the malformed phenotype. Crosses in which >25% of the b * If/b pr cn sp zip^Ebr G₂ progeny exhibited malformation were retained for further analysis (eight lines retained). b * If/b pr cn sp zip^Ebr males from each of the eight lines were mated to b pr cn sp zip^Ebr/SM5 virgin females. All G₃ lines exhibited >25% penetrance of the malformed phenotype and were established as permanent stocks. One line was subsequently lost.

Prior to testing enhancer of zipper mutations for interactions with Sb-sbd and RhoA mutants, each of the remaining seven mutant lines was outcrossed to Canton-S flies for three generations to remove any recessive lethal mutations linked to but independent of E(zip). To ensure that free recombination occurred between mutant and wild-type chromosomes, E(zip) mutations were not selected during these three generations. Animals carrying E(zip) mutations were recovered by crossing individual females carrying the outcrossed chromosomes to Sco/CyO males and recovering putative E(zip) chromosomes over CyO. These stocks were then assayed for their ability to generate viable adult homozygotes lacking the balancer chromosome. Six lethal E(zip) mutations were recovered. However, one of these lethal mutants failed to interact with zip^Ebr following outcrossing and was not used in subsequent studies. A seventh mutant, 31-6, was semilethal with 25% of 31-6 homozygotes exhibiting malformed legs. All six remaining outcrossed chromosomes carrying E(zip) mutations were subsequently balanced over SM5, CyO, or CyO, P{sevRas1.V12}FK1. We estimate that 90% of each original E(zip) chromosome was exchanged with wild-type chromosomal DNA in this outcrossing protocol. While this outcrossing scheme does not formally eliminate the possibility that two mutations mapping close together contribute to the second-site noncomplementation of the E(zip) mutants, it is likely that each E(zip) mutant chromosome is associated with a mutation in a single gene.

Mounting of adult legs and wings:
Adult flies were fixed in 3:1 (vol:vol) 70% ethanol:glycerol for at least 24 hr. Legs and wings were dissected from the preserved animals and mounted in a drop of 80% glycerol under a coverslip and then sealed with nail polish. Mounted legs and wings were observed by brightfield microscopy using a x4 Acroplan objective on a Zeiss Axiophot MC80 microscope. Images were acquired using a Kodak DC290 Zoom digital camera and processed using Adobe Photoshop 4.0 software.

Molecular lesions associated with RhoA and Sb-sbd mutant alleles used in this study:
The RhoA mutations used in these studies have been characterized at a molecular level (Table 2). RhoA^E3.10 is a CAAX box missense mutation (C to Y), while RhoA^J3.8 is a deletion of the 13 C-terminal amino acids, including the CAAX box (HALSELL et al. 2000 ). These alleles are likely to be severe mutations since the CAAX box motif is required for RhoA association with the plasma membrane (ZHANG and CASEY 1996 ; SEABRA 1998 ). The RhoA^72F and RhoA^72O mutations are P-element-excision mutants and are also likely null mutations (STRUTT et al. 1997 ). Df(2R)Jp8 is a small deficiency that uncovers RhoA but on the basis of genetic analysis does not uncover the nearby myosin light chain kinase gene (HALSELL et al. 2000 ).

The Sb and sbd mutant alleles used in this study have been characterized to varying degrees at a molecular level and all of these mutations are likely to reduce proteolytic activity. The Sb^63b, Sb⁷⁰, and Sb¹ mutations are all associated with transpositional insertions in the same region of the gene and each of these mutations is associated with a truncated Sb-sbd transcript (APPEL et al. 1993 ; ABU-SHUMAYS 1995 ; A. HAMMONDS and J. FRISTROM, unpublished observations). The sbd¹⁰⁵ mutation is associated with a large deficiency that removes the entire Sb-sbd locus and flanking genomic DNA (APPEL et al. 1993 ). Finally, the sbd² mutation is not associated with coding region defects and is presumed to be regulatory in nature (A. HAMMONDS and J. FRISTROM, unpublished observations).

Nomenclature:
In this analysis of genetic interactors, we look at both SSNC and genetic enhancement. Formally, SSNC is the heterozygous interaction between otherwise recessive alleles at two different loci that in combination give a mutant phenotype—in this case, malformed legs or wings. In contrast, genetic enhancement is the ability of a second locus to enhance an existing phenotype that is due to a dominant allele, homozygous alleles, or trans-heterozygous alleles (at the same locus). To simplify the presentation of our results, and because the underlying molecular basis of the observed genetic interaction is likely the same, we frequently refer to both second-site noncomplementation and genetic enhancement as an interaction that "enhances" the malformed phenotype.

	RESULTS

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

RhoA mutations show robust second-site noncomplementation and dominant genetic interactions with Sb-sbd alleles:
Mutations in Drosophila RhoA exhibit second-site noncomplementation (SSNC) with mutations in the heavy chain of nonmuscle myosin (zipper), with 90–100% of doubly heterozygous mutant animals (i.e., RhoA +/+ zip) exhibiting a malformed leg phenotype (HALSELL et al. 2000 ). The observation that Sb-sbd is also a genetic enhancer of zipper with respect to leg malformation (GOTWALS and FRISTROM 1991 ) led us to investigate the possibility that Sb-sbd mutations interact genetically with RhoA mutations in elongating leg imaginal discs. In Table 3 and subsequent tables we define interactions as strong when at least 50% of animals of the relevant genotype have at least one malformed leg, moderate when 25–49% of such animals exhibit malformation, and weak when 5–24% malformation is observed. In almost all cases, background leg malformation in siblings heterozygous for only one mutant allele (e.g., Sb/+) ranged from 0 to 4%. Thus, a moderate interaction indicates that animals have malformation frequencies on average 10-fold above background, while a strong interaction indicates that the frequency is >10-fold above background. In an earlier publication (HALSELL and KIEHART 1998 ), moderate interactions were defined as malformation frequencies falling into the range of 25 to 74%. However, on the basis of subsequent experiments we conclude that malformation frequencies >50% are sufficiently rare and robust to be classified as strong genetic interactions.

Genetic interaction data for Sb-sbd and RhoA mutant alleles demonstrate that certain RhoA and Sb-sbd allele combinations result in a high incidence of leg malformation (Table 3). The Sb^63b and Sb⁷⁰ alleles interact very strongly with RhoA mutant alleles and the RhoA deficiency Df(2R)Jp8 in SSNC assays, with the frequency of animals with leg malformation ranging from 59 to 95% (see also Fig 1, A–D). The interactions with the RhoA^J3.8 and RhoA^E3.10 mutations are particularly strong. In the cases where RhoA^J3.8, RhoA^E3.10, and Df(2R)Jp8 were tested for interactions with Sb^63b and Sb⁷⁰ many individual animals exhibited malformation of both the second and the third legs, a phenotype we associate with particularly strong genetic interactions. Typically, malformation is limited to the third (most posterior) pair of legs. In contrast, only weak SSNC interactions are observed between RhoA alleles and the Sb¹ or Sb^spike. The Sb¹ and Sb^spike alleles have been reported previously to interact weakly in SSNC assays with the zipper allele zip^Ebr and the broad allele br¹ (BEATON et al. 1988 ; GOTWALS and FRISTROM 1991 ).

The sbd¹⁰⁵ mutation exhibits moderate levels of SSNC interaction with the RhoA^J3.8 and RhoA^E3.10 alleles and interacts weakly with RhoA^72F and Df(2R)Jp8 (Table 3). No interactions were observed with the sbd²⁰¹ allele. Therefore we asked if RhoA mutants interact dominantly with sbd alleles. Because sbd¹⁰⁵ homozygotes are inviable and sbd²⁰¹ homozygotes have reduced viability and exhibit 100% malformation, we tested the trans-heterozygous combinations sbd²⁰¹/sbd¹ and sbd²⁰¹/sbd², which exhibit background malformation rates of 10 and 1%, respectively. The RhoA^E3.10, RhoA^J3.8, and Df(2R)Jp8 alleles all show very strong dominant interactions with sbd²⁰¹/sbd¹ trans-heterozygotes. In contrast, weak-to-moderate interactions are observed between sbd²⁰¹/sbd² trans-heterozygotes and the RhoA^E3.10, RhoA^J3.8, and Df(2R)Jp8 alleles. Therefore, in subsequent tests for dominant interactions with sbd alleles we used the sbd²⁰¹/sbd¹ combination. Collectively, our data show that Sb-sbd mutations interact strongly with mutations in RhoA and raise the possibility that the membrane-associated Sb-sbd protease plays a role in RhoA signaling during leg morphogenesis.

Sb-sbd mutants interact with RhoA signaling pathway mutants:
We extended our analysis by asking if Sb-sbd mutations interact with mutations in other known components of the RhoA signaling pathway. Mutations in the gene encoding the guanine nucleotide exchange factor DRhoGEF2 have been previously reported to affect RhoA-mediated cell shape changes during Drosophila gastrulation (BARRETT et al. 1997 ). These DRhoGEF2 mutants show weak-to-moderate interactions in leg discs with zip^Ebr in SSNC genetic assays (HALSELL et al. 2000 ) and moderate-to-strong interactions with RhoA mutations (C. BAYER and L. VON KALM, unpublished observations). Three DRhoGEF2 alleles, DRhoGEF2^1.1, DRhoGEF2^4.1, and the P-element-insertion mutant DRhoGEF2⁰⁴²⁹¹, show mostly weak interactions with Sb-sbd alleles in SSNC and dominant interaction assays (Table 4). While these findings might suggest that the DRhoGEF2 protein might not have a major role in a hypothetical Sb-sbd-RhoA signaling pathway in leg imaginal discs, we have identified a new allele of DRhoGEF2 that does interact in the context of leg morphogenesis. This new allele, DRhoGEF2^11-3, was tested with Sb-sbd alleles in SSNC and dominant interaction assays (Table 4). DRhoGEF2^11-3 interacts moderately with Sb^63b and Sb⁷⁰ and sbd²⁰¹/sbd¹ trans-heterozygotes in a dominant interaction assay. The allele-specific nature of genetic interactions in our leg malformation assays has also been reported by others (e.g., BEATON et al. 1988 ; GOTWALS and FRISTROM 1991 ; HALSELL and KIEHART 1998 ). A striking example of an allele-specific interaction is the observation that null and hypomorphic alleles of zipper fail to enhance broad or Sb-sbd mutant leg defects in SSNC assays, whereas high percentages of animals are malformed when the zip^Ebr allele is tested in similar assays with broad or Sb-sbd mutants. Thus, although only one of the DRhoGEF2 mutants tested interacts significantly with Sb-sbd alleles in our assays, the strength of the interactions between Sb-sbd mutants and DRhoGEF2^11-3 clearly implicates DRhoGEF2 function in leg imaginal disc morphogenesis.

Previous reports have indicated that the RhoA effector kinase, Rho kinase, plays an important role in actin cytoskeletal dynamics in vertebrates via phosphorylation of the myosin regulatory light chain (AMANO et al. 1996 ; BURRIDGE and CHRZANOWSKA-WODNICKA 1996 ; RIDLEY 1996 ). A recent report demonstrates that the Drosophila homolog of Rho kinase (Drok) is also a RhoA effector kinase in wing imaginal discs (WINTER et al. 2001 ). We therefore tested two mutant alleles of drok for interactions with Sb-sbd mutations (Table 4). drok¹ and drok² show strong interactions with Sb⁷⁰ and weak-to-moderate interactions with Sb^63b. In contrast, both drok mutants show weak dominant interactions with the sbd²⁰¹/sbd¹ trans-heterozygote. Nonetheless, the strength of the SSNC interaction of both drok alleles tested with the Sb⁷⁰ mutation suggests that Rho kinase and Sb-sbd function together to control cell shape changes in elongating leg imaginal discs.

The second-chromosome deficiency Df(2R)Jp1 uncovers the cytogenetic interval 51C3-52F8/9 and therefore should delete the myosin light chain kinase (MLCK) gene located at 52D11/E1 (KOJIMA et al. 1996 ; TOHTONG et al. 1997 ; CHAMPAGNE et al. 2000 ; http://www.Flybase.org). In vertebrates the MLCK functions in conjunction with other kinases such as Rho kinase to activate myosin via phosphorylation of the myosin regulatory light chain (reviewed in TAN et al. 1992 ). We find that Df(2R)Jp1 shows moderate interactions with Sb^63b and Sb⁷⁰ in SSNC assays and interacts strongly in a dominant genetic interaction test with sbd²⁰¹/sbd¹ trans-heterozygotes (Table 4). Df(2R)Jp1 complements the embryonic lethal phenotype of the RhoA alleles used in these studies but it also interacts genetically with these alleles giving rise to the malformed leg phenotype (HALSELL et al. 2000 ). Further evidence that Df(2R)Jp1 uncovers a RhoA signaling pathway member distinct from RhoA comes from our observation that SSNC genetic interactions between Sb-sbd mutants and Df(2R)Jp4, a deficiency predicted to uncover both RhoA and MLCK, are stronger than those observed for Sb-sbd and Df(2R)Jp1 or Sb-sbd and Df(2R)Jp8, which uncovers RhoA (data not shown). Thus, our observation of significant genetic interactions between Sb-sbd mutants and Df(2R)Jp1 are consistent with the possibility that Sb-sbd mutants interact genetically with loss-of-function MLCK alleles. We will be able to test this possibility directly when loss-of-function MLCK alleles become available.

We also tested mutant alleles of spaghetti squash (sqh) for interactions with Sb-sbd alleles because of the known physical interaction of the myosin regulatory light chain encoded by sqh with the myosin heavy chain encoded by zipper and experimental data showing that Sqh is required for leg morphogenesis (EDWARDS and KIEHART 1996 ). We observe no interaction between Sb-sbd alleles and the strongly hypomorphic sqh¹ or amorphic sqh² (data not shown; KARESS et al. 1991 ; EDWARDS and KIEHART 1996 ). Likewise, sqh null alleles also fail to interact with zipper mutations including zip^Ebr in SSNC assays even though these two proteins assemble into the high-affinity complex that is the native myosin protein. Presumably, one dose of the wild-type myosin regulatory light chain gene in sqh/+; zip/+ animals produces sufficient protein to drive cell shape changes in elongating leg imaginal discs. In addition, the sqh mutant alleles we tested do not interact with the RhoA^J3.8 and RhoA^E3.10 mutations even though RhoA is a known regulator of myosin regulatory light chain phosphorylation (data not shown; see Table 10 for an overview of all genetic interactions described in this study). Thus, we are not surprised by the failure of strong loss-of-function sqh mutations to interact with Sb-sbd mutants.

Drosophila Pkn binds specifically to GTP-activated RhoA and is required for epidermal cell shape changes during dorsal closure in the embryo (LU and SETTLEMAN 1999 ). We therefore investigated whether or not Pkn mutants interact with Sb-sbd alleles. We find that Pkn³ and the P-element-insertion mutant Pkn⁰⁶⁷³⁶ fail to interact with Sb^63b or Sb⁷⁰ in SSNC assays and exhibit weak dominant enhancement of leg malformation with sbd²⁰¹/sbd¹ trans-heterozygotes (data not shown). Thus we conclude that our data do not provide evidence for a major role for Pkn in Sb-sbd-RhoA-mediated cell shape changes during leg morphogenesis.

Reduced RhoA gene dosage suppresses leg malformation associated with overexpression of Sb-sbd:
In experiments aimed at characterizing and rescuing Sb-sbd mutations we generated transgenic flies carrying a full-length, wild-type Sb-sbd cDNA placed under the control of a heat-inducible promoter in the transposon pCaSpeR-hs (THUMMEL and PIRROTTA 1992 ). We used this hs-Sb-sbd transgene (hs-Stubble) to rescue leg defects associated with homozygous sbd²⁰¹ mutations by inducing the transgene with a 1-hr 37° heat shock in staged 0-hr or 3-hr prepupae. Surprisingly, we discovered that induction of hs-Stubble in wild-type 0-hr or 3-hr prepupae results in a high percentage of severe leg malformation in adults involving second pair as well as third pair legs (Table 5). As a control, heat treatment of 3-hr prepupae taken from the w¹¹¹⁸ recipient strain used to generate hs-Stubble resulted in 0–1% malformation of second and third legs (n = 287). Induction of the malformed phenotype does not occur at later stages as overexpression of Sb-sbd either 12 hr before or 10 hr after pupariation does not affect leg morphogenesis. Thus, the leg malformation associated with overexpression of Sb-sbd is highly specific to a critical period of leg development between 0 and 10 hr after pupariation.

We took advantage of the leg defect associated with Sb-sbd overexpression at the beginning of the prepupal period to test the hypothesis that RhoA and Sb-sbd act together in a common signaling pathway in elongating leg discs. We asked if reducing the gene dosage of RhoA to one copy could suppress the leg phenotype associated with overexpression of Sb-sbd. We find that in RhoA^72O/+; hs-Stubble/+ animals heat-shocked as 3-hr prepupae, second leg malformation is reduced 2.7-fold, from 69 to 25% (Table 5). When RhoA^72O/+; hs-Stubble/+ animals are heat-shocked as 0-hr prepupae, second leg malformation is reduced 2-fold, from 17 to 8%. A similar reduction in the level of leg malformation is also seen for third legs (58 to 29%; Table 5). Results similar to those shown in Table 5 were obtained in experiments using the RhoA^E3.10 allele (data not shown). The observation that hs-Stubble-induced leg malformation can be suppressed by reducing RhoA gene dosage is consistent with the possibility that RhoA and Stubble function in a common signaling pathway regulating myosin as cells change their shape during leg elongation. However, these data do not rule out the possibility that the Sb-sbd transmembrane protease and RhoA function in parallel pathways.

Enhancer of zipper mutations interact genetically with RhoA and Sb-sbd mutations during leg elongation:
To identify additional genes encoding products that participate in an Sb-sbd/RhoA signaling pathway in elongating leg imaginal discs, we analyzed mutations in six genes on the second chromosome that exhibit second-site noncomplementation with the zip^Ebr mutation. These mutations were induced with EMS and identified on the basis of the malformed leg phenotype. Five of the six E(zip) mutations are homozygous lethal. The one exception is 31-6, which is semilethal with approximately one-quarter of viable animals exhibiting leg malformation. Of the six mutants, three are alleles of known genes. One mutant, 11-3 (described in Table 4), fails to complement DRhoGEF2^1.1, DRhoGEF2^4.1, DRhoGEF2⁰⁴²⁹¹, and Df(2R)P803-Delta15, which uncovers DRhoGEF2, suggesting that 11-3 is a new allele of DRhoGEF2. Another mutant, 12-6, fails to complement RhoA^J3.8 and RhoA^E3.10 mutants and Df(2R)Jp8, which uncovers RhoA. Thus we conclude that 12-6 is likely to be a new RhoA allele. A third mutant, 33-1, fails to complement zipper null mutations and Df(2R)ES1, which uncovers zipper, and is therefore likely to be a new allele of zipper (zip^33-1) as opposed to an enhancer of zipper (S. HALSELL and J. FRISTROM, unpublished observations). zip^33-1 is semiviable over zip^Ebr and these animals have severe leg malformation (see Table 6); however, recombination analysis indicates that these mutants are separated by less than one map unit (J. FRISTROM, unpublished observations). Of course it is possible that one or more of the failure-to-complement results we have observed with these three E(zip) mutations are due to lethal SSNC interactions rather than to noncomplementation between two alleles of the same gene. Rescue experiments with full-length cDNAs or cosmids carrying genomic clones for each of the genes involved are needed to confirm our genetic complementation data.

With the exception of the weakly interacting 31-6 mutant, the E(zip) alleles and zip^33-1 show mostly strong interactions with zip^Ebr (Table 6). We find that RhoA^12-6 interacts very strongly with zip^Ebr with reduced viability (i.e., RhoA^12-6 +/+ zip^Ebr animals are significantly underrepresented relative to sibling classes). Similar results were reported by HALSELL et al. 2000 for interactions between RhoA^J3.8 or RhoA^E3.10 and zip^Ebr. In contrast, the putative DRhoGEF2^11-3 mutant is considerably more interactive with zip^Ebr than are the previously reported DRhoGEF2⁰⁴²⁹¹, DRhoGEF2^1.1, and DRhoGEF2^4.1 alleles (HALSELL et al. 2000 ). The combination zip^33-1/zip^Ebr exhibits 97% malformation with greatly reduced viability.

The E(zip) mutations exhibit universally strong SSNC interactions with RhoA alleles and Df(2R)Jp8, the deficiency that removes RhoA. The strongest genetic interactions are seen between RhoA mutants and DRhoGEF2^11-3 or zip^33-1, which exhibit 93–100% malformation and in the case of DRhoGEF2^11-3 greatly reduced viability in combination with all RhoA alleles tested. DRhoGEF2^11-3 also exhibits reduced viability in combination with Df(2R)Jp8. The uncharacterized E(zip) mutations, 12-5, 18-5, and 31-6, also interact strongly with RhoA mutants. The viability of animals doubly heterozygous for 12-5 and RhoA^J3.8 or RhoA^E3.10 is also reduced.

The E(zip) mutants also interact to varying degrees with Sb-sbd mutants. RhoA^12-6 and DRhoGEF2^11-3 exhibit strong and moderate SSNC with Sb^63b and Sb⁷⁰, respectively, while the remaining E(zip) mutants interact moderately with at least one of these Sb alleles. With the exception of 12-5, all E(zip) mutants show moderate-to-strong dominant interactions with sbd²⁰¹/sbd¹ trans-heterozygotes. While the E(zip) mutants show significant interactions with Sb-sbd mutants, these interactions are less robust than those observed between E(zip) and RhoA mutations. Given the strong interaction between E(zip) and RhoA mutations, it is somewhat surprising that the E(zip) mutants by and large fail to interact with DRhoGEF2⁰⁴²⁹¹, DRhoGEF2^1.1, DRhoGEF2^4.1, drok¹, drok², or Df(2R)Jp1 (data not shown). There are two exceptions to these observations. First, RhoA^12-6 exhibits 25% malformation with DRhoGEF2^4.1, 20% malformation with the drok¹ and drok² alleles, and 13% malformation with Df(2R)Jp1. Second, DRhoGEF2^11-3 is lethal with all DRhoGEF2 alleles tested, and exhibits 17% malformation with Df(2R)Jp1. As positive controls, we tested all DRhoGEF2 and drok alleles for interaction with RhoA and zipper alleles. The DRhoGEF2⁰⁴²⁹¹, DRhoGEF2^1.1, and DRhoGEF2^4.1 alleles exhibited interactions ranging from 21 to 64% malformation with RhoA^J3.8, RhoA^E3.10, and zip^Ebr in SSNC assays (data not shown and HALSELL et al. 2000 ). The drok mutants also exhibited interactions with RhoA^J3.8, RhoA^E3.10, and zip^Ebr in SSNC assays (48–68% malformation for drok¹ and 19–38% malformation for drok²). Taken as a whole these observations may indicate that the three uncharacterized E(zip) gene products (12-5, 18-5, and 31-6) and Sb-sbd act in different RhoA-regulated signaling pathways in developing leg imaginal discs.

We have tested the three uncharacterized E(zip) mutants for noncomplementation with mutations in second-chromosome genes thought to have a role in RhoA signaling and with genes reported to exhibit the malformed leg phenotype in the mutant condition. Genes tested include Drosophila Pkn (LU and SETTLEMAN 1999 ), blistered (GOTWALS and FRISTROM 1991 ), vulcan and bancal (GATES and THUMMEL 2000 ), and crol (D'AVINO and THUMMEL 1998 ). Other mutants tested include Df(2R)Jp1, which putatively uncovers the myosin light chain kinase gene; concertina, which has been proposed to act as an upstream regulator of RhoA signaling leading to cell shape changes during Drosophila gastrulation (BARRETT et al. 1997 ); and the protein 4.1 family members coracle, inscuteable, and expanded (BOEDIGHEIMER and LAUGHON 1993 ; FEHON et al. 1994 ; KRAUT and CAMPOS-ORTEGA 1996 ). The 12-5, 18-5, and 31-6 alleles are fully viable over all of these mutations and exhibit wild-type leg morphology.

The E(zip) mutants interact with each other but not with the broad¹ allele (br¹):
We asked if the E(zip) alleles interact with each other because SSNC between these mutants would be consistent with the hypothesis that these genes are active in a common signaling pathway. As seen in Table 7 there are significant genetic interactions between E(zip) alleles in these assays. Of particular interest are the uncharacterized 12-5, 18-5, and 31-6 mutants. The 12-5 mutant exhibits moderate-to-strong interactions with all E(zip) mutants tested. A very strong interaction is observed with 18-5. In addition to severe malformation, 12-5 +/+ 18-5 animals have reduced viability. A moderate interaction (35% malformation) is also observed between the 18-5 and 31-6 mutants. With the exception of DRhoGEF2^11-3, the semiviable 31-6 mutant interacts mostly at moderate levels with all E(zip) mutants tested. In addition, approximately one-quarter of 31-6 homozygotes have malformed legs, an observation fully consistent with a role for the 31-6 gene product in leg morphogenesis.

We also asked if the broad allele br¹ interacts with RhoA pathway members and the E(zip) mutants. The br¹ mutant has previously been shown to interact strongly with Sb-sbd and zipper mutations (BEATON et al. 1988 ; GOTWALS and FRISTROM 1991 ). The br¹ allele also exhibits temperature sensitivity with typically greater penetrance of leg malformation observed at 18° than at 25° (BEATON et al. 1988 ). The br genetic function of the broad locus is essential for prepupal leg development. Animals carrying amorphic mutations that affect this Broad function are unable to elongate or evert their leg imaginal discs (KISS et al. 1988 ; BAYER et al. 1997 ). We find that RhoA mutants interact dominantly with br¹ at both 18° and 25° with stronger interactions observed at 18° (Table 8). RhoA has also been recently identified in a genetic screen as an enhancer of br¹ leg malformation (WARD et al. 2003 , this issue). Weak-to-moderate interactions are also observed between br¹ and Df(2R)Jp1, which putatively uncovers the MLCK gene. However, br¹ fails to show significant genetic interactions with DRhoGEF2^1.1, DRhoGEF2^4.1, DRhoGEF2⁰⁴²⁹¹, or any of the E(zip) mutants, including DRhoGEF2^11-3, RhoA^12-6, and zip^33-1 (data not shown). In addition, br¹ fails to interact with drok mutants in SSNC assays (data not shown).

RhoA, Sb-sbd, and E(zip) mutants interact during wing morphogenesis:
We were interested to know if the interactions between RhoA, Sb-sbd, and E(zip) mutants occur in a wider developmental context than leg development. The primary lethal period for Sb-sbd, 12-5, 18-5, and 31-6 homozygotes is during larval and pupal development (data not shown). We therefore decided to explore the possibility that the interactions we have described in legs also occur during wing morphogenesis, a process known to be dependent on RhoA signaling (HALSELL et al. 2000 ). Mutations in genes required for prepupal wing imaginal disc morphogenesis frequently exhibit a characteristic crumpled or malformed wing phenotype (BEATON et al. 1988 ; GOTWALS and FRISTROM 1991 ; HALSELL et al. 2000 ; see also Fig 1, E–H). We asked if wing morphogenetic defects are observed in SSNC assays involving RhoA, Sb-sbd, or E(zip) mutations. In control experiments we determined that RhoA and RhoA pathway members (zip, DRhoGEF2, and drok) interact in developing wings (Table 9). Typically, strong interactions were observed between mutant alleles of these genes confirming that the RhoA pathway is required for wing morphogenesis. These interactions ranged from 100% wing malformation for interactions between zip^Ebr and RhoA^E3.10 to 35% malformation for the combination RhoA^E3.10 and DRhoGEF2^4.1. Strong interactions were also observed between Sb-sbd and RhoA or zip mutants (Table 9; Fig 1, E–H). In contrast, the Sb^63b and Sb⁷⁰ mutants exhibit weak-to-moderate interactions with drok¹, and Sb^63b does not interact with DRhoGEF2^4.1. The relative strength of these interactions is similar to those observed in developing legs, suggesting that Sb-sbd plays a role in RhoA signaling during morphogenesis of legs and wings.

The E(zip) mutants also show significant interactions during wing development (Table 9). All E(zip) mutants interact strongly with at least one RhoA allele and zip^Ebr. The only exception is 18-5, which has 49% wing malformation in combination with zip^Ebr, just below the 50% threshold for a strong interaction. Consistent with our leg malformation data, DRhoGEF2^11-3 and RhoA^12-6 interact most strongly with the Sb-sbd mutants tested, and in general the interactions between E(zip) and Sb-sbd mutations are less robust than those observed between E(zip), RhoA, and zip^Ebr mutants. Thus we conclude that the relative strength of interactions (strong, moderate, weak) observed between RhoA, Sb-sbd, and E(zip) mutants in elongating legs is closely mirrored in developing wings, and therefore Sb-sbd and the E(zip) gene products appear to have important roles in RhoA signaling in multiple developmental contexts.

	DISCUSSION

TOP ABSTRACT MATERIALS AND METHODS RESULTS DISCUSSION LITERATURE CITED

Previous studies have clearly established a connection between RhoA signaling and regulation of actin cytoskeletal dynamics (reviewed in TAPON and HALL 1997 ; HALL 1998 ; VAN AELST and SYMONS 2002 ). Evidence from vertebrates and invertebrates indicates that RhoA signaling activates nonmuscle myosin and thereby controls the assembly and contractility of actomyosin-based structures (NUSRAT et al. 1995 ; BROCK et al. 1996 ; BARRETT et al. 1997 ; KOZMA et al. 1997 ; HALL 1998 ; HALSELL et al. 2000 ; WINTER et al. 2001 ). The RhoA GTPase plays a pivotal role in morphogenetic events during many stages of Drosophila development, including, but not limited to, head involution, dorsal closure, control of planar polarity, and imaginal disc morphogenesis (STRUTT et al. 1997 ; HARDEN et al. 1999 ; MAGIE et al. 1999 ; HALSELL et al. 2000 ; WINTER et al. 2001 ; BLOOR and KIEHART 2002 ). Although several of the molecules that act downstream of RhoA as it signals to myosin have been characterized, much less is known about membrane-associated events leading to activation of the RhoA signaling pathway.

Our genetic data implicate the Drosophila Sb-sbd type II transmembrane serine protease in RhoA signaling during leg and wing imaginal disc morphogenesis. A summary of our leg genetic interaction data is shown in Table 10. The strength of interactions between Sb-sbd and RhoA or RhoA pathway members is similar to that observed between RhoA and RhoA pathway members (Table 10). Sb-sbd mutations interact strongly with RhoA, drok, Df(2R)Jp1, and zip mutants and moderately with DRhoGEF2^11-3. Sb-sbd mutants interact weakly with other DRhoGEF2 alleles, which interact weakly in all of our SSNC assays. The absence of significant interactions between any of the DRhoGEF2 and drok mutants we tested may indicate that DRhoGEF2 and Drok act in parallel or branching RhoA pathways to regulate myosin activation. Alternatively and perhaps more likely, the absence of interactions between DRhoGEF2 and drok could be attributable to the specific nature of the mutations in these genes and an inherent ability of the RhoA pathway to compensate for reduced levels of DRhoGEF2 and Drok. In support of the latter argument, we note that zipper null mutations do not interact in leg malformation assays, whereas the zip^Ebr mutation, which may encode a weak dominant negative allele of myosin and potentially form nonproductive myosin dimers, is highly interactive in similar assays (HALSELL et al. 2000 ). The Sb-sbd and the E(zip) genes also fail to interact with Pkn, a known downstream effector of RhoA during dorsal closure (LU and SETTLEMAN 1999 ). Although these observations may indicate that Pkn has no role in RhoA signaling in imaginal discs we have not yet determined if Pkn interacts with RhoA and RhoA pathway members during leg and wing morphogenesis.

Our data also support a role for the E(zip) genes in RhoA signaling. By genetic analyses three of the E(zip) mutants encode known components of the RhoA signaling pathway (RhoGEF2^11-3, RhoA^12-6, and zip^33-1). All E(zip) mutations are strongly interactive with RhoA and, with the exception of 31-6, which is a semilethal mutation, exhibit moderate-to-strong interactions with zip^Ebr (Table 10). However, our interaction data suggest that the unidentified 12-5, 18-5, and 31-6 genes may not act in the same pathway as Sb-sbd during leg morphogenesis. First, weak interactions are observed between the 12-5, 18-5, and 31-6 mutants and RhoA pathway members such as DRhoGEF2, drok, and Df(2R)Jp1. In contrast Sb-sbd mutants exhibit robust interactions with drok and Df(2R)Jp1. Second, while interactions between E(zip) and Sb-sbd mutants are robust (all moderate), they are not as strong as those observed between Sb-sbd and the RhoA signaling pathway. Additional mutants and characterization of the gene products encoded by the 12-5, 18-5, and 31-6 genes are necessary to resolve the relationships between these genes, Sb-sbd, and RhoA signaling during leg morphogenesis.

What is the role of Sb-sbd in RhoA signaling in leg and wing imaginal discs? In addition to our genetic interaction data we have shown that leg malformation associated with overexpression of Sb-sbd during leg elongation is suppressed by reducing the gene dosage of RhoA. Taken as a whole, these data raise the possibility that Sb-sbd-mediated proteolytic events at the apical membrane are required for RhoA signaling in elongating prepupal leg imaginal discs. On the basis of the strength of the SSNC interactions we have observed between Sb-sbd and RhoA mutants and Sb-sbd and RhoA pathway members, a linear signaling pathway from Sb-sbd to RhoA is one reasonable interpretation of our data. A possible linear signaling pathway in which ecdysone-mediated induction of the Sb-sbd locus is necessary to activate and/or maintain the RhoA signaling pathway in late larval leg discs is shown in Fig 2. In this model the Sb-sbd zymogen is induced by ecdysone and then secreted and inserted into the apical membrane of leg disc cells. At the cell surface Sb-sbd is activated, possibly by autoproteolytic cleavage (APPEL et al. 1993 ), and signals to DRhoGEF2 via an unknown mechanism. Signaling to DRhoGEF2 could involve cleavage of another extracellular protein (pathway 1 in Fig 2) or signaling via the Sb-sbd cytoplasmic domain (pathway 2 in Fig 2). DRhoGEF2 cycles RhoA into the active GTP-bound form, activating downstream effector kinases that control myosin assembly, changes in actin dynamics, and additional structural changes (e.g., modifications in cadherin function) that are required for proper morphogenesis.

View larger version (15K): In this window In a new window Download PPT slide   Figure 2. Model of Sb-sbd regulation of RhoA signaling and actin cytoskeletal dynamics during prepupal leg imaginal disc morphogenesis.

The genetic interactions we observe between Sb-sbd and members of the RhoA pathway can be interpreted in ways other than a linear signaling pathway. For example, Sb-sbd and RhoA might act in parallel pathways. These pathways might converge to activate myosin or regulate different components of the contractile apparatus. Alternatively, Sb-sbd may be a transcriptional or post-transcriptional target of RhoA signaling in leg imaginal discs. This latter model is consistent with our observation that leg malformation caused by overexpression of Sb-sbd is suppressed by reducing RhoA gene dosage. Thus, an overall reduction in Sb-sbd protease synthesis or activity in animals carrying one wild-type copy of the RhoA gene might compensate for high levels of heat-shock-induced overexpression of the protease and lead to suppression of the leg malformation phenotype. Arguing against this interpretation is the observation that Sb-sbd transcription is ecdysone responsive, and to date no ecdysone-responsive member of the RhoA signaling pathway has been identified (WARD et al. 2003 , this issue).

Another interpretation of our data is that the Sb-sbd protease facilitates cell shape changes via cleavage of apical extracellular matrix, while RhoA signals to activate myosin. In support of a role for Sb-sbd in cleavage of extracellular matrix, exogenous application of trypsin to elongation-defective Sb-sbd mutant leg discs in culture leads to rapid elongation of the discs (APPEL et al. 1993 ). However, trypsin-mediated rescue of Sb-sbd mutants might occur because trypsin cleaves a cell-surface signaling molecule normally targeted by the Sb-sbd trypsin-like proteolytic domain. For example, physiological concentrations of trypsin cleave the G-protein-coupled receptor PAR2 on the cell surface of enterocytes (KONG et al. 1997 ). If trypsin is cleaving a cell-surface signaling molecule in the in vitro disc culture experiments, then activation of the RhoA pathway must occur immediately since Sb-sbd mutant discs elongate within seconds of the addition of exogenous trypsin. Such rapid activation of the RhoA signaling pathway would be possible if it is already partially activated in Sb-sbd mutants, and the addition of trypsin provides the stimulus to rapidly increase the strength of signaling through the pathway. Two reports demonstrate rapid changes in strength of signaling through Rho GTPase pathways in response to inductive signals. Stimulation of human neutrophils with chemoattractants or platelet activating factor leads to optimal activation of p21rac within 5 sec (GEIJSEN et al. 1999 ). In addition, activation of the thrombin receptor in HEK-293T cells results in an 11-fold increase in the levels of GTP-bound RhoA within 1 min (CHIKUMI et al. 2002 ). Thus, trypsin may be capable of activating the RhoA pathway within seconds of addition to the imaginal disc culture medium. Finally, a combination of these models may explain Sb-sbd behavior during leg morphogenesis. For example, the Sb-sbd protease may cleave extracellular matrix components to relieve physical constraints on morphogenesis and also contribute to intracellular signaling leading to modification of the actin cytoskeleton. Ultimately, molecular genetic and biochemical approaches will be required to determine the mechanism underlying the genetic interactions between Sb-Sbd and the RhoA signaling pathway.

If Sb-sbd is involved in activation and/or maintenance of the RhoA pathway in leg imaginal discs it is unlikely to act alone in this context. The br¹ allele of the broad locus interacts with RhoA mutations in the context of leg development (this study; WARD et al. 2003 , this issue) and leg elongation fails completely in the amorphic br⁵ mutant. Sb-sbd transcription is delayed 2 hr in br⁵ mutants and then induced to normal levels, indicating that Sb-sbd is necessary but not sufficient for leg elongation (WARD et al. 2003 , this issue). These observations raise the possibility that broad and Sb-sbd might act cooperatively to activate and/or maintain RhoA signaling. broad mutants interact genetically with both Sb-sbd and zipper mutations during leg development although the molecular basis of these genetic interactions is unknown (BEATON et al. 1988 ; GOTWALS and FRISTROM 1991 ). Sb-sbd transcription is not quantitatively affected in broad mutant animals (see above) and the zipper gene is not ecdysone responsive (A. HAMMONDS and J. FRISTROM, unpublished observations). broad may have an indirect role in regulating the timing of Sb-sbd transcription or perhaps regulate the transcription of an inhibitor or substrate for the Sb-sbd protease (see for example BAYER et al. 1996B ). While the role of the broad gene remains unclear, it clearly has a critical role in leg morphogenesis. Characterization of genes identified in a recent large genetic screen for enhancers of br¹ leg malformation (WARD et al. 2003 , this issue) will doubtless further our understanding of the role of this important master regulator of metamorphosis in prepupal leg development.

The Sb-sbd protease has been strongly implicated in the regulation of the actin cytoskeleton during bristle development where it may have a role in actin polymerization or nucleation (APPEL et al. 1993 ). However, three observations indicate that interactions between Sb-sbd and the actin cytoskeleton proceed via different mechanisms in bristles and imaginal discs. First, all Sb-sbd mutants behave as recessive alleles during leg and wing development. In contrast Sb and sbd alleles behave as dominant and recessive mutations, respectively, during bristle development. Second, the genetic interactions between Sb-sbd, broad, zipper, and RhoA described here and elsewhere (BEATON et al. 1988 ; GOTWALS and FRISTROM 1991 ; HALSELL et al. 2000 ) are observed in legs and wings but not in bristles. These differences in Sb-sbd function during imaginal disc and bristle development are not surprising given that there is no evidence for an actin-myosin contractile apparatus in bristles. Indeed, viable zip^Ebr/zip^33-1 animals have bristles of normal length even though they exhibit close to 100% leg malformation (J. FRISTROM, unpublished observations). Third, the Sb^63b mutation interacts genetically with mutations in two other genes required for bristle morphogenesis, singed (sn) and forked (f). singed and forked encode actin bundling proteins related to echinoderm fascin and mammalian espin, respectively (BRYAN et al. 1993 ; CANT et al. 1994 ; PETERSON et al. 1994 ; TILNEY et al. 1995 ; BARTLES et al. 1996 ). Compared to sn, f, or Sb-sbd mutants, sn; Sb-sbd and f; Sb-sbd double mutants have extremely reduced bristles that appear as small nubs barely extending beyond the exoskeleton. However, neither sn nor f mutants enhance Sb-sbd leg or wing malformation suggesting that actin bundling is not important for disc morphogenesis (L. VON KALM, unpublished data). Collectively, these data suggest that the Sb-sbd protease interacts with different sets of proteins in bristles and developing imaginal discs and that RhoA signaling is not required for bristle morphogenesis.

Finally, our data suggest that type II transmembrane serine proteases (TTSPs) may be involved in intracellular signaling pathways during development. Aberrant expression of vertebrate members of the TTSP family has been linked to developmental abnormalities and a variety of pathologies including congenital heart defects, blood pressure regulation, and several forms of cancer (reviewed in HOOPER et al. 2001 ). In addition to Sb-sbd, two other members of the TTSP family have been tentatively linked to "outside-in" signaling events. Enteropeptidase and MT-SP1 may participate in intracellular signaling via activation of the proteolytically activated G-protein-coupled receptor PAR2 (KONG et al. 1997 ; TAKEUCHI et al. 2000 ). These observations raise the possibility that other members of the TTSP family participate in signal transduction events and that aberrant intracellular signaling may underlie their association with pathology and developmental abnormalities. The enteropeptidase, MT-SP1, and Sb-sbd TTSPs are characterized by a variety of extracellular motifs in addition to the proteolytic domain. These extracellular domains are potentially involved in protein-protein interactions necessary for interaction with cleavage targets and may provide a basis for substrate specificity among different members of the TTSP family (HOOPER et al. 2001 ). The Sb-sbd TTSP has an extracellular disulfide knotted domain (APPEL et al. 1993 ) and in vitro studies indicate that this domain is capable of mediating protein-protein interactions (KELLENBERGER et al. 1995 ). In vivo mutational analysis of the knotted domain in the Drosophila Snake serine protease, which is secrete