Integrins modulate Sog activity in the Drosophila wing

Vein differentiation is controlled by groups of genes that act in two developmentally distinct phases. During mid-third larval instar and early prepupal stages, the positions of vein territories are defined within the monolayer of wing imaginal disc cells (reviewed by Bier, 2000). In the second phase of vein development, during pupal stages, the vein versus intervein cell fate choice is resolved among cells in broad vein competent domains of a bilayered wing primordium. This refinement step is mediated in part by lateral inhibitory signals elaborated by presumptive vein cells. At the same time that lateral inhibition limits the width of veins, vein continuity signals act along the axis of the vein to promote their formation in straight continuous lines. Genes such as net (Brentrup et al., 2000) and blistered(Roch et al., 1998), which govern intervein development, are also required for restricting vein development to appropriate cells. In addition, the activity of intervein genes, which are required for intersurface adhesion, such as those coding for the integrins, must be excluded from veins to permit the non-adherent strips of cells in the veins to form open channels between the two wing surfaces(Brown, 2000). Interestingly,some combinations of integrin mutants have been found to generate ectopic veins; however, the mechanisms that underlie this phenotype are not understood(Brower and Jaffe, 1989; Brown et al., 1989; Zusman et al., 1993; Zusman et al., 1990).

During early pupal development, Decapentaplegic (Dpp), the Drosophila homolog of vertebrate Bmp2/Bmp4, and the Bmp-binding protein Short gastrulation (Sog), the homolog of vertebrate chordin, function antagonistically to ensure the formation of straight continuous veins(Haerry et al., 1998; Lecuit et al., 1996; Sturtevant and Bier, 1995; Yu et al., 1996; Zecca et al., 1995). Throughout this period, dpp is expressed in vein primordia, while sog is expressed in a complementary intervein pattern(Yu et al., 1996). Sog also opposes Bmp signaling during dorsoventral patterning of the early embryo,which involves zygotic (Biehs et al.,1996; Francois et al.,1994; Marques et al.,1997) as well as maternal(Araujo and Bier, 2000)functions of this pathway. sog encodes a secreted molecule with domains resembling thrombospondin and procollagen(Francois and Bier, 1995; Francois et al., 1994) and has been shown to bind Dpp (Ross et al.,2001). It has been suggested that regulated cleavage of Sog generates different forms with distinct activities(Yu et al., 2000). Cleavage at three sites by the metalloprotease Tolloid (Tld) inactivates Sog(Marques et al., 1997), while alternative cleavage at a different site, which occurs in the presence of the co-factor Twisted Gastrulation (Tsg), results in the production of truncated forms of Sog referred to as Supersog, which antagonize a broader spectrum of Bmp activities than intact Sog (Yu et al.,2000). Chordin is also subject to proteolysis by Xolloid, the vertebrate homolog of Tolloid (Piccolo et al., 1997). Because all of these molecules interact extracellularly, it is important to understand how the activities and localization of these factors are regulated in the extracellular milieu.

Binding of growth factors to specific proteins or to the ECM is one type of mechanism for regulating the availability or dispersion of growth factors in different developmental contexts. ECM proteins may sequester growth factors in an inactive form, as well as modulate cellular responses to them(Streuli, 1999). Several ECM proteins such as Collagen, Fibronectin, Thrombospondin, Noggin and Chordin bind to TGFβ or to members of the bone morphogenetic protein (Bmp)subfamily (Piccolo et al.,1996; Taipale and Keski-Oja,1997; Zimmerman et al.,1996). Such binding may activate or reduce growth factor activity and/or availability. Several extracellular matrix molecules and their receptors have been described in Drosophila (for a review, see Brown, 2000). Among these proteins, integrins are expressed during embryogenesis, larval and pupal stages and perform functions including muscle attachment, morphogenesis of the midgut, and adhesion between the two surfaces of the wing(Brabant et al., 1996; Brower et al., 1995a; Brower et al., 1995b; Martin-Bermudo and Brown,1996; Roote and Zusman,1995; Wilcox et al.,1989). During pupal development, integrins are expressed in intervein cells and perform a central role in regulating apposition (i.e. alignment and adhesion) of the dorsal and ventral surfaces of the wing(Brabant et al., 1996; Fristrom et al., 1993; Wilcox et al., 1989). Three integrin subunits are known to be required for adhesion between the two wing surfaces: βPS integrin, encoded by the myospheroid(mys) gene, is expressed on both wing surfaces during pupal development; and two α-integrins, αPS1, encoded by the multiple edematous wing (mew) gene, and αPS2, encoded by the inflated (if) gene, are expressed on the dorsal and ventral wing surfaces, respectively, during early wing development(Brabant et al., 1996; Brower et al., 1995b). Functional integrin molecules are composed of one β-subunit combined with one of the α-subunits, and consist of a large extracellular domain and a small cytoplasmic tail. Mutations in any of these integrin genes cause blisters in the adult wing, a phenotype characteristic of a lack of apposition between the wing surfaces (Brower et al.,1995b; Brown et al.,1989; Wilcox et al.,1989; Zusman et al.,1990).

In this report, we show that in addition to their well established adhesive function, integrins play another role during pupal vein development to modulate Bmp signaling. This modulation of Bmp activity may be mediated, at least in part, by integrins binding and/or regulating the activity or diffusion/distribution of the Sog protein. Genetic analysis indicates that the role of integrins in modulating Bmp signaling involves the well studiedβPS and αPS1 subunits, and another less extensively characterizedαPS3 subunit (Grotewiel et al.,1998; Stark et al.,1997), which we show is also expressed in dorsal cells of the pupal wing. We find that Sog diffuses into provein domains from adjacent intervein cells, but does so only on the dorsal surface. Moreover, we find that this dorsal specific diffusion of Sog into provein regions depends on the activity of both βPS and αPS1 integrin subunits. We discuss these results in light of current models for regulated Sog processing and recycling.

The following mutant alleles were used in this study. collagenIV a1: DCg1²³⁴, Cg25c^GDB; collagenIV a2: vkg^BLK, vkg^SAL, vkg^ICO, vkg^RML, vkg¹⁷⁷, vkg²²⁸; laminin: lamA^6-36, lamA^9-32, lamA²¹⁶, lamA²⁵; integrins: if³, if^B2, if^k27e, mew^M6, mew⁴⁹⁸, mys¹, mys^ts1, mys^XB87, mys^nj42, mys^XR04, scb¹, scb², vol¹, vol²; dally: dally^gem, dally^P2, dally^δP527; dachsous: ds^33k, ds¹; stranded at second: sas¹⁵; decapentaplegic: dpp^shv; glass bottom boat: gbb-60A¹, gbb-60A⁴; thick veins: tkv¹, tkv⁸; saxophone:sax⁸; tolloid: tld^68-62; tolloid-related: tlr^δ-41;

Df 3R slo³, which deletes both tld and tlr. Detailed characteristics of all alleles can be found in FlyBase. Enhancer piracy sog lines (e.g. sog^EP2, sog^EP3, sog^EP7, sog^EP8, sog^EP9, sog^EP11) are described elsewhere(Yu et al., 1996).

Eggs were collected for 24 hours and aged for 48 hours before the heat shock in order to generate predominantly small clones (<100 cells). First instar larvae were heat shocked for 15 minutes at 37°C. Unmarked clones generated with the same mys^XB87 FRT line produced similar phenotypes. Twenty-seven dorsal clones, four ventral clones, and four dorsal and ventral clones were analyzed in detail.

Eggs were collected and aged as described above. First instar larvae were heat shocked for three 1 hour intervals at 37°C, with 30 minute recovery periods in between. Twenty-seven dorsal clones, seven ventral clones, and five dorsal and ventral clones were analyzed in detail.

Eggs were collected and aged as described above. First instar larvae were heat shocked for three 1 hour intervals at 37°C, separated by 45 minute recovery periods. Eighteen dorsal clones, nine ventral clones, and five dorsal and ventral clones were analyzed in detail.

In situ hybridization was performed using digoxigenin-labeled antisense RNA probes and visualized as a blue alkaline phosphatase precipitate(O'Neill and Bier, 1994). Immunohistochemistry was performed as described by Sturtevant et al.(Sturtevant et al., 1993). Sog protein was detected using polyclonal 8B as primary antibody (1:500)(Srinivasan et al., 2002),anti-rabbit HRP as secondary antibody (1:2000, Jackson Laboratories), and visualized using the rhodamine TSA kit (NEB). For Sog and Integrin double labels, Sog protein was detected with anti-8B antiserum as above, CF.6G11 monoclonal antiserum was used for βPS integrin (1:500) and DK.1A4 monoclonal antiserum was used for αPS1 (1:500)(Brower et al., 1984), and detected with secondary anti-mouse Alexa 488 (Molecular Probes). Images were analyzed either on a Zeiss Axiovert 135, collected digitally with Axiocam or on a LSM 510 Meta Zeiss Confocal Microscope.

Co-immunoprecipitation was based on procedures described by Brower(Brower, 1984), with minor modifications. Wings of pupae taken 20-24 hours after puparium formation (APF)were rapidly dissected from the pupal cases, homogenized with a pestle in ice cold lysis buffer (10 mM Tris pH 8.1/75 mM NaCl/0.5 mM MgCl₂/0.5 mM CaCl₂/0.25% NP40/0.25% BSA/0.01% NaN₃/1 mM PMSF and protease inhibitor cocktail – Complete, Boehringer Mannheim), and left for 30 minutes on ice with occasional rocking. After brief centrifugation (10 minutes at 10,000 g) to pellet non-homogenized tissue,supernatants were incubated overnight at 4°C with protein A Sepharose bound integrin antibodies. Unbound supernatants where collected as `unbound'sample and 4× SDS sample buffer was added. Beads were washed twice with lysis buffer and stripped of bound proteins by two rounds of acid elution with 200 mM glycine (pH 3.0) generating `bound 1' and `bound 2' samples to which 4×SDS sample buffer was added. All samples were boiled before running in 10% SDS-PAGE gels and transferred by electroblotting to nitrocellulose membranes. Membranes were blocked in Tris/NaCl/0.3% BSA 0.1% Tween 20 and incubated in primary antibody (anti-Sog 8A at 1:500 dilution) followed by incubation in HRP-conjugated anti-rabbit secondary antibody (Sigma, at 1:5000 dilution) and developed using Supersignal (Pierce) according to manufacturer's instructions. For detection of co-immunoprecipitated integrins used as control, membranes were stripped of antibodies using 200 mM glycine (10 minutes at room temperature), incubated in biotin-conjugated Concanavalin A,treated in Vectastain AB system and visualized by chemiluminescence as above. The 8A anti-Sog antiserum was raised against a small peptide fragment that included CR1 and the first part of the stem and the antibody recognizes a sog construct on western blots that contains the stem but not CR1.

Antibodies were covalently attached to protein A Sepharose beads by incubating them with beads overnight at 4°C. After three washes in cold PBS and two washes in 2 M sodium borate (pH 9.0), antibodies were crosslinked to the beads by addition of 5 mg/ml DMP. Beads were incubated for 30 minutes at room temperature, washed in 0.2 M triethanolamine pH 8.0, and incubated for 2 hours. After equilibrating in PBS, beads were stored at 4°C with 0.01%sodium azide. Beads were washed in PBS before use. Antibodies used were:CF.6G11 (monoclonal for βPS integrin); DK.1A4 (monoclonal for αPS1)(Brower et al., 1984); CF.2C7 monoclonal (Wilcox et al.,1981) for αPS2; and polyclonal αvol(volado, also known as scab) for αPS3(Grotewiel et al., 1998).

Wings from adult flies were dissected in isopropanol and mounted in Canada Balsam mounting medium.

N-terminal microsequencing was used to ensure the identity of the protein band co-immunoprecipitated with αPS1 integrin antibodies. Immobilon instead of Nitrocellulose membranes were used, and BSA was omitted in the homogenization buffer. Protein bands were transferred from SDS-PAGE onto Immobilon membranes, followed by N-terminal microsequencing by Edman degradation. Microsequencing revealed contaminating proteins in the band recognized by the Sog antibody; however, we were able to detect the sequence GV(X)EGR(X)H(XX)L(XX)EE(X). A Blast search for short sequences aligning to this sequence found that it aligns to the N-terminal region of the Sog protein(GVTEGRRHAPLMFEES).

Ectopic expression of sog in the wing results in the truncation of longitudinal veins and/or crossveins as a result of inhibition of Bmp signaling during pupal stages (Yu et al.,1996). We have previously described a line of flies(sog^EP7) in which a sog transgene is expressed in pupal vein primordia (Yu et al.,1996) as a consequence of the transgene bearing P-element having inserted next to a genomic enhancer [an effect we have termed enhancer piracy(Noll et al., 1994)]. sog^EP7 flies have truncated L4 and L5 wing veins, a meandering L2 vein, and/or ectopic vein material in the vicinity of L2(Fig. 1C). Consistent with previous analysis of sog during pupal development, mutant alleles of dpp and tkv can modify sog^EP7 phenotypes(Yu et al., 1996)(Table 1).

In the course of screening for additional mutations that modified the effect of ectopic sog expression, we identified interactions with several genes encoding cell-cell or cell-extracellular matrix adhesion molecules. Based on these initial findings, we tested for transheterozygous genetic interactions between sog^EP7 and mutants in components of the extracellular matrix or their receptors to identify candidate ECM proteins that might regulate Sog activity or diffusion(Table 1). Among the mutants scored, alleles of α and β-integrins showed consistent enhancement or suppression of sog^EP7 phenotypes(Table 1, Fig. 1C-F). Strong alleles of myospheroid (mys) enhanced sog^EP7phenotypes (Fig. 1D), whereas alleles of multiple edematous wing (mew) suppressed these phenotypes (Fig. 1E). Importantly, the peak period of interaction between a heat shock mysconstruct and sog^EP7 is between 20 and 28 hours after puparium formation (apf) (Table 1), which is the same time window for interaction between sog and the dpp^shv allele(Yu et al., 1996). By contrast, no interactions were observed with alleles of inflated(if). The mys and mew interactions were also tested with several other sog^EP lines as well as for several integrin alleles (Table 1). Alleles of laminin A, described as an extracellular ligand forαPS1 (encoded by mew), also strongly enhanced sog^EP vein truncations. In addition, we found that alleles of scab (or volado), which encodes an αPS3 subunit,enhance sog^EP phenotypes(Fig. 1F). Other extracellular modifiers of sog^EP phenotypes included alleles of genes encoding Drosophila Collagen IV and Selectin. The basis for these latter interactions will not be considered further in this report.

We also observed genetic interactions between integrins and other Bmp signaling components. For example, the hypomorphic β-integrin allele mys^nj42 suppresses the thickened vein phenotype of the tkv¹ allele of the Dpp receptor thick veins(Fig. 1G,H; Table 1). Similarly, decreasing the level of scb (in scb¹tkv¹/tkv¹ flies) suppresses the tkv¹ phenotype (not shown). In addition, dpp and gbb alleles enhance sog^EP phenotypes as do reduced levels of tld and tok, which encode highly related metalloproteases (Table 1). This latter observation is interesting in light of the fact that tldand tok can collaborate to either degrade(Marques et al., 1997) or process Sog into more broadly active Bmp inhibitory forms in embryos and pupae(Yu et al., 2000).

Because decreasing the dose of mys enhanced sog^EP phenotypes, we examined the effect of complete loss of mys function in a sog^EP7 background by producing mys-null clones. Large mys^- clones generated by FLP-FRT-mediated recombination frequently induce blisters due to non-apposition of the dorsal and ventral surfaces of the wing. In small mys^- clones, however, blisters are not observed and veins appear normal although the two wing surfaces remain unapposed within the center of the clone (Brabant et al.,1996). When similar small mys^- clones are generated in a sog^EP7 background, a different phenotype is observed in which veins become ill-defined and broadened (e.g. four or five cells across compared with two or three cells in diameter in wild type)wherever the clones cross or abut longitudinal veins or the posterior crossvein (Fig. 2), and is observed in clones consisting of as few as 20 cells. The ability of mys^- clones to induce vein broadening non-autonomously in neighboring cells occurs only at very short range as clones displaced by three or more cell diameters from veins have a wild-type phenotype. Dorsal and ventral mys^- clones can induce non-apposition of the wing surfaces in both wild-type and sog^EP7 backgrounds,consistent with the fact that βPS integrin is expressed on both surfaces of the wing during larval and pupal development(Brabant et al., 1996; Brower et al., 1995a). Vein broadening in an sog^EP7 background, however, is observed only in dorsal clones, indicating that this phenotype is not simply a secondary consequence of an adhesion defect. The restriction of mys^- vein phenotypes to the dorsal surface also suggests that there is a dorsally expressed α-integrin, which acts in conjunction with β-integrin during vein development.

As in the case of mys^- clones, large null mew^- clones result in wing blisters(Brabant et al., 1996). Consistent with mew being expressed exclusively on the dorsal surface of the pupal wing, only dorsal mew^- clones produce a phenotype. Large mew^- null clones generated in a sog^EP7 background produce similar blistered phenotypes. In contrast to these adhesion defective phenotypes, smaller mew^- clones (e.g. <100 cells) generated in a sog^EP7 background alter vein formation, but do so in a different way than observed for small mys^- clones. Such mew^- clones located in the proximity of veins bend and displace the veins towards the clone, which then run along and outside the clone boundary (Fig. 3A-C). This vein shifting phenotype of mew^- clones is observed for all longitudinal veins. However, clones that cross over a vein, and therefore lack mew expression in intervein cells on both sides of the vein, generate no phenotype (Fig. 3D), neither do clones generated at a distance of three or more cells from a vein (not shown). A different vein thickening phenotype is associated with clones generated in the vicinity of crossveins(Fig. 3E). As expected, based on the dorsal specific expression of mew, ventral mew^- clones have no affect even in a sog^EP7 background (Fig. 3F). We conclude that mew^- clones act non-autonomously to promote vein development in adjacent longitudinal provein cells. These phenotypes suggest that αPS1βPS integrin may normally play a role in positioning veins by regulating the levels or activity of Sog at the vein/intervein border.

As mentioned above, the phenotypes generated by mys^-and mew^- clones in a sog^EP7 background differ with respect to their effects on vein formation. As α andβ-integrin subunits bind to form complexes, the mys^-clone phenotype may disrupt the formation of a complex composed of mew and an additional dorsally expressed integrin α-subunit. This alternative α-subunit is unlikely to be αPS2 (=if)because if is expressed only on the ventral surface of the wing and if mutant alleles do not modify the sog^EP7phenotype.

It has not yet been reported whether the α-integrin scb is expressed in the wing or functions during wing development. Because scb alleles interacted genetically with sog^EP7,we examined the pattern of scb expression during larval and pupal wing development by in situ hybridization. No scb expression was detected during larval and prepupal stages (data not shown); however, a dynamic pattern of scb expression emerges in 20-30 hour pupal wings(Fig. 4A,B). scbexpression, which is restricted to the dorsal surface of the wing at all times, is initiated most intensely in intervein regions in the vicinity of L2 and L3 and then rapidly expands to encompass all intervein cells. Subsequently, scb expression is also observed within the provein domain, initially at high levels and then tapering off after 25 hours of pupal development (Fig. 4B,F).

We examined the requirement for scb during wing development by producing scb^- clones in an otherwise wild-type background. Small dorsal scb^- clones result in a phenotype not previously observed for PS integrin mutants in which clones that touch or cross over a vein by a few cells broaden the vein from two or three cells to four or five cells across (Fig. 5B,C). The vein thickening phenotype of scb clones is observed for all longitudinal veins, as well as for the posterior crossvein. In contrast to mys^- clones generated in a sog^EP7 background, which result in irregular broadened veins, the veins associated with scb^- clones generated in a wild-type background are well defined. In some cases, small scbclones (<20 cells) are restricted to the center of veins and can split the vein into two vein-like territories separated by a central less pigmented intervein-like area (Fig. 5D). This phenotype is accentuated when scb^- clones are generated in a sog^EP7 background(Fig. 5F) and is similar to phenotypes observed with clones of lateral inhibitory mutants(Garcia-Bellido and de Celis,1992; Sturtevant and Bier,1995). When scb^- clones of similar size are generated at a distance from veins, however, they have no effect. In the few large scb^- clones we have recovered (>200 cells),dorsal clones were associated with blisters characteristic of integrin mutants(Fig. 5A), although no phenotype was observed for any ventral scb^- clones(Fig. 5E), consistent with the dorsally restricted expression of scb.

One explanation for the observed genetic interactions between sogand integrin mutations is that Sog physically binds to an integrin(s)subunit(s). To test for physical interactions between Sog and integrins we performed co-immunoprecipitation experiments(Fig. 6). During pupal development, sog is expressed in intervein cells and produces a full-length protein species of 120 kDa as well as several lower molecular weight forms of 76, 60, 50 and 42 kDa detected by immunoblotting with the 8A anti-Sog antiserum, which detects an epitope located immediately after the CR1 domain (Yu et al., 2000). We prepared protein extracts from hand dissected pupal wings, performed immunoprecipitations with antibodies against βPS, αPS1 orαPS2 integrins, ran the precipitates on SDS-PAGE, and immunoblotted with the 8A anti-Sog antiserum. These experiments revealed that a 50 kDa Sog protein co-precipitates efficiently with αPS1 and weakly with αPS2(Fig. 6A), but not withαPS3 (not shown) or with β-integrin(Fig. 6A). A small amount of full-length Sog was also co-precipitated with αPS1. To confirm that the 50 kDa immunoprecipitated protein recognized by the Sog antibody was indeed a fragment of Sog, we isolated the reactive Sog band, subjected it to N-terminal microsequencing and found that it corresponds to an N-terminal fragment of the Sog protein (see Materials and Methods). Given the size of the band on SDS-PAGE (e.g. 50 kDa), this Sog fragment is predicted to include the epitope immediately following the CR1 domain recognized by the 8A antiserum(Yu et al., 2000) and terminate before the second cysteine repeat (CR2)(Francois et al., 1994). The basis for the binding between Sog and αPS1 may be related to the binding between a vertebrate α-integrin and thrombospondin, which are similar to Drosophila αPS1 and the CR domains of Sog, respectively(Francois et al., 1994; Guo et al., 2000; Hynes and Zhao, 2000). The failure of Sog to co-precipitate with βPS is surprising as ligand binding surfaces of integrins typically span both the α and β-subunits(Calderwood et al., 1997; Humphries, 2000; Sonnenberg, 1993). Such anα-chain-specific association could either result from an interaction between Sog and αPS1 outside of the ligand binding site of αPS1,or may reflect an indirect interaction with αPS1 mediated by another extracellular matrix protein or cell surface receptor.

Despite the clear genetic interaction between sog and scb, we have not been able to detect a physical binding between Sog and αPS3 as we have observed between Sog and αPS1. The lack of observed binding between Sog and αPS3 could result from a difficulty in detecting an alternatively processed form of Sog bound to αPS3, or may reflect an indirect mode of action of scb (e.g. by altering the abundance of interacting integrins such as αPS1βPS or by binding to a different Bmp inhibitor).

It has been observed that sog mRNA is confined to intervein cells during pupal development (Yu et al.,1996) (Fig. 4E). As Sog protein diffuses during early embryonic development(Srinivasan et al., 2002),however, we wondered whether Sog might also travel from its intervein site of production into the provein region during pupal development. We stained pupal wings with the 8B anti-Sog antiserum(Srinivasan et al., 2002) and observed a dynamic pattern of Sog protein distribution, which includes vein competent domains as well as intervein cells(Fig. 7). Anti-Sog staining is initially patchy (around 20 hours apf), stronger on the dorsal surface and mostly restricted to intervein cells (Fig. 7A). Shortly thereafter (22-26 hours apf), Sog staining spreads into provein cells on the dorsal surface of the wing, at which point it is excluded only from the most central vein-proper cells(Fig. 7B,E). On the corresponding ventral surface, however, Sog staining remains excluded from the entire provein region throughout pupal development (i.e. up to 34 hours apf)(Fig. 7F,H). Between 26 and 30 hours apf, Sog staining fills all the provein domains on the dorsal surface,with increased levels of staining observed at the provein/intervein border(Fig. 7C,G). At 30 hours apf,Sog staining diminishes overall and becomes restricted to intervein cells and hemocytes running in the middle of the vein(Fig. 7D). As sog mRNA is detectable only in intervein cells during the examined pupal period, we conclude that Sog protein must be delivered to cells within the provein territory on the dorsal surface by some form of passive diffusion or active transport.

Because diffusion of Sog into provein domains is restricted to the dorsal surface of the wing where integrins interact with Sog, we asked whether they play a role in regulating the distribution of Sog protein on the dorsal surface of pupal wings. We generated marked mys^- or mew^- clones in an otherwise wild-type background and examined Sog staining (Fig. 8). In control wings, double-labeling with anti-β integrin and anti-Sog antisera confirmed that the dorsally restricted pattern of reticular Sog staining extends beyond β-integrin staining into provein domains(Fig. 7I-N). By contrast, Sog staining has a patchy intracellular appearance in dorsal mys^- clones (Fig. 8A-F), and is excluded from wild-type provein cells on the dorsal surface that are adjacent to mys^- clones. In such cases where mys^- clones are located on one side of a provein domain, Sog is still able to enter the provein region from the opposite mys⁺ side of the same vein(Fig. 8A-C). These results demonstrate that mys is required for diffusion or transport of Sog into the vein competent domain. Consistent with the observations discussed above in which only dorsally located integrin^- clones can alter the course of veins, we find that only dorsal mys^- clones modify Sog distribution in the pupal wing(Fig. 8D-F). We observed similarly altered Sog staining within mew^- clones on the dorsal wing surface resulting in punctate rather than reticular staining and lack of Sog diffusion into the provein region(Fig. 8G-L). These results demonstrate that the βPS and αPS1 integrins play an important role in determining the distribution of Sog protein in the pupal wing.

In this study, we have provided three primary lines of evidence that integrins play an important role in regulating Bmp signaling in provein regions of the pupal wing. First, we showed that integrin^- clones generated on the dorsal surface of the wing alter the trajectory and/or width of adjacent veins. Second, we found that a truncated form of Sog present in pupal wings binds to αPS1. Finally, we observed that diffusion of Sog into provein domains, which is restricted to the dorsal surface of the wing,depends on integrin function. Cumulatively, these results strongly suggest that the ability of Sog to diffuse or to be transported into provein regions on the dorsal surface depends on an interaction with integrins.

Consistent with Sog interacting genetically with integrins to alter the course of veins on the dorsal surface of the wing, we found that theαPS1 and βPS-integrins are required for the diffusion or transport of Sog from dorsal intervein cells where sog mRNA is expressed into adjacent provein regions. As αPS1 binds Sog, this physical interaction may contribute to regulating the distribution of Sog. The 8B anti-Sog antiserum, which recognizes Sog protein in intervein cells and inside the provein domain, detects an epitope located near the second cystein repeat(CR2). Consequently, Sog fragments that diffuse or that are delivered to provein cells must be either full length, which weakly binds to αPS1 in co-immunoprecipitation experiments, or fragments that contain CR2. The truncated Supersog-like fragment that binds strongly to αPS1 in coimmunoprecipitation experiments should not be recognized by the 8B antiserum. Therefore, integrins may differentially regulate the distribution of Sog fragments on the dorsal surface of the pupal wing, restraining the movement of broad spectrum Bmp inhibitory Sog fragments (such as Supersog-like molecules) and allowing or mediating transport of other fragments to provein cells, such as full-length Sog, which also has a vein inhibitory function. Unfortunately, it is not possible currently to examine the diffusion of Supersog-like fragments directly because the 8A antiserum is not suitable for staining pupal wings. These findings suggest that integrins regulate the delivery or diffusion of active Sog protein from intervein cells into the vein competent domain. In contrast to the dorsally restricted functions of integrins required for vein development, the previously analyzed adhesive functions of integrins depends on subunits functioning on both surfaces of the wing.

There are several possible mechanisms by which interaction with integrins could modulate Sog activity in pupal wings. It has been previously shown that elevated sog expression results in vein truncation, while misexpression of dpp induces ectopic veins, indicating that sog restricts vein formation by opposing Bmp signals emanating from the center of the vein (Yu et al.,1996). One possibility is that such a Bmp inhibitory form(s) of Sog must interact with integrins in order to diffuse or be transported into provein domains (on the dorsal surface of the wing only). This hypothesis would be consistent with the finding that veins appear to be attracted to integrin^- clones. Such a vein repulsive form(s) of Sog would presumably act as a Bmp antagonist.

According to the simple model in which integrins are essential for delivering a Bmp inhibitory form of Sog to provein cells, one would expect that integrin^- and sog^- clones would generate similar phenotypes in which veins deviated towards the mutant clones and/or broadened within them. However, sog^- clones induce meandering of veins (Yu et al.,1996), which show only a weak tendency to track along the outside of sog^- clones (B.N. and E.B., unpublished), in contrast to integrin^- clones, which bend or widen veins in a more dramatic fashion. One possible explanation for the differences between the sog^- and integrin^- phenotypes is that there are several different endogenous forms of Sog in pupal wings(Yu et al., 2000), which might exert opposing activities. If multiple Sog fragments exert effects on vein development, some providing repulsive and others attractive activities on vein formation, the differences between the behaviors of sog- and integrin- clones could be explained by a repulsive (Bmp inhibitory) form(s) of Sog selectively requiring an interaction with integrins. The possibility that a positive Bmp-promoting activity of Sog might also be present that acts as a vein attractant has precedent in that a positive Sog activity has been proposed to explain a requirement for Sog in activating expression of the Dpp target gene race in early embryos(Ashe and Levine, 1999). Structure/function studies of Sog have also revealed a potential Bmp promoting form of Sog, which is longer than Supersog forms (K. Yu and E.B.,unpublished). According to this model, altering the balance between repulsive and attractive Sog activities would generate different vein phenotypes. In the total absence of sog, both repulsive and attractive activities would be lost, generating a mild meandering vein phenotype in which neither attraction nor repulsion clearly dominated, as is observed in sog^- clones (Yu et al., 1996). If an interaction with integrins were required only for production or delivery of Bmp inhibitory forms of Sog into the vein, then integrin^- clones, which still contain the Bmp-activating forms of Sog, could exert a net attractive influence on veins, leading to more pronounced deviation of veins toward the clones. This hypothesis is consistent with vein phenotypes we have observed associated with integrin^-clones that cross over veins or run along both sides of the vein, such as narrowing, bending and wandering of veins which are similar to phenotypes observed in correspondingly located sog^- clones. The existence of different Sog fragments bearing opposing activities would also explain the different phenotypes we obtain upon ectopic Sog expression in some sog^EP lines (Yu et al.,1996), such as sporadic ectopic vein material between L3 and L2 and meandering L2 veins in addition to vein loss in other areas.

Another possible explanation for the differences between the sog^- and integrin^- phenotypes is that integrins may regulate the activity of extracellular signals in addition to Sog. One hint of such an activity is that when a scb^- clone falls within the provein area, the vein splits around the border of the clone in a cell autonomous fashion. As this later phenotype is enhanced by ectopic sog expression in veins (e.g. in a sog^EPbackground), αPS3 may normally promote Bmp signaling within the vein. Although the identity of such potential targets is unknown, candidates would include Bmps (e.g. Dpp or Gbb) or Bmp receptors. Further analysis will be needed to explain the basis for the different behaviors of sog^- and integrin^- clones, as well as the variations observed in different integrin^- clones.

In summary, we propose that Sog fragments with differential activities may regulate vein formation. The vein bending phenotype observed in the absence ofαPS1 would result from a remaining attractive Sog activity that outweighs the activity of a repulsive form of Sog, which can no longer be delivered from intervein cells (Fig. 9). As βPS integrin forms heterodimers with both αPS1 and αPS3 (Brower et al.,1984; Stark et al.,1997) mys would be expected to be required for the activity of both αPS chains. Consistent with this expectation, the phenotype of mys^- clones (i.e. broad poorly defined veins)resembles a hybrid of those observed for mew^- and scb^- clones.

Endocytosis has been shown to play an important role in the establishment of Bmp activity gradients. The endocytic pathway has been implicated in transport of Dpp between cells by transcytosis during larval wing development(Entchev et al., 2000; Teleman and Cohen, 2000) and for vectorial transport of Wg during mid-embryogenesis(Dubois et al., 2001; Moline et al., 1999). During early embryonic development, formation of a Sog protein gradient in dorsal regions also relies on the action of Dynamin(Srinivasan et al., 2002),although in this preblastoderm context, it has been proposed that endocytosis limits the dorsal diffusion of Sog, which is essential for the partitioning of the dorsal ectoderm into epidermis and amnioserosa. Recently, deRobertis'group has shown that vertebrate α3β Integrin binds to the Xenopus Sog counterpart Chordin in vitro, leading to endocytosis of Chordin (E. deRobertis, personal communication).

Although we have not directly addressed whether integrins regulate Sog endocytosis in this current study, the altered distribution of Sog within integrin^- clones is suggestive of such a role. Reticular Sog staining, which outlines the cell perimeter is lost in integrin^-clones on the dorsal surface, leaving only a punctate intracellular staining. This mis-localization of Sog implicates integrins in internalizing and/or trafficking of Sog to the cell surface. Because appropriately located integrin^- clones also block the accumulation of Sog in adjacent provein domains, the observed defects in Sog distribution between the surface and the cytoplasm may underlie the failure to deliver Sog to vein competent cells. The endocytic pathway could promote the transport of Sog to provein cells by a mechanism similar to that proposed to be involved in the transport of Dpp along the AP axis during larval stages(Entchev et al., 2000; Teleman and Cohen, 2000). Alternatively, endocytosis could function to limit Sog diffusion as is the case during embryogenesis (Srinivasan et al., 2002). According to this latter scenario, integrins would normally prevent or reduce Sog endocytosis because integrins are necessary for delivery of Sog to provein cells. Integrins have been shown to play a direct role in endocytosis of viral particles and in mediating membrane traffic through the endocytic cycle (de Curtis,2001; Triantafilou et al.,2001). Indirect mechanisms for integrin-mediated endocytosis may also exist that would not involve endocytosis of the integrin receptor itself,but of other components that regulate Sog trafficking. Further analysis will be necessary to investigate whether Drosophila integrins regulate delivery of Sog to endocytic vesicles or transport of Sog through the endocytic pathway to adjacent cells.

The modulatory effect of integrins on Sog activity described in this paper are likely to be mediated by dpp and/or gbb signaling because existing evidence indicates that Sog is a dedicated modulator of Bmp signaling. In addition, the phenocritical period for mys and sog interaction coincides with that for interaction between sog and dpp (Yu et al.,1996). On the one hand, we cannot exclude the existence of an additional role of integrins in regulating vein formation through another pathway, such as the Egf and Notch pathways, which have been shown to exert important roles on vein development (de Celis et al., 1997; de Celis and Garcia-Bellido, 1994; Garcia-Bellido and de Celis,1992; Guichard et al.,1999; Huppert et al.,1997; Martin-Blanco et al.,1999; Sturtevant and Bier,1995). On the other hand, the integrin^- clonal phenotypes described in this manuscript are observed only on the dorsal surface and all known components of the Egfr pathway promote vein development on both surfaces of the wing(Diaz-Benjumea and Garcia-Bellido,1990; Diaz-Benjumea and Hafen,1994; Guichard et al.,1999).

We also found that mys^nj42 and scb¹suppress the thickened vein phenotype of tkv¹ mutants,which raises the possibility of a direct interaction between integrins and a Bmp receptor involved in wing vein development. The vein splitting and vein thickening scb^- clonal phenotypes are reminiscent of tkv mutant phenotypes, which derive from a positive requirement for Bmp signaling for vein formation inside the vein competent domain and a negative ligand titrating function that limits the range of Bmp diffusion into the intervein territory adjacent to the provein domain(de Celis, 1997). The fact that scb is expressed in both vein and intervein territories is consistent with a dual action of scb. Additional experiments will be necessary to investigate whether scb plays a direct role in modulating Bmp receptor activity.

Diffusion of putative growth factors and the shaping of their activity gradients have been the focus of intense interest since Allan Turing formulated the concept of morphogens(Turing, 1952). Recently,several groups have described mechanisms to explain how soluble factors can create morphogen gradients. These include degradative proteolysis and a retrieval role for endocytosis in creating the early embryonic Sog gradient(Srinivasan et al., 2002),regulated endocytosis of wingless(Strigini and Cohen, 2000),extracellular transport of Wg in membrane bound argosomes(Greco et al., 2001), planar transcytosis [as is required for Dpp movement in the wing imaginal disc(Entchev et al., 2000; Teleman and Cohen, 2000)], and the formation of thin cell extensions (cytonemes) that deliver Dpp over several rows of cells (Ramirez-Weber and Kornberg, 1999). Protein-protein interactions in the extracellular milieu, such as those described here, may also be capable of modulating the magnitude and spatial pattern of Bmp activity, working independently or in conjunction with other mechanisms.

We thank Dr Susan Zusman, Bill McGinnis, Deborah Kimbrell, Corey Goodman,Scott Selleck, Kathy Mathews (Bloomington) and Karin Elkstrom (Umea) for fly stocks; Dr Ron Davis for volado antibody and cDNA; and Dr Lina Zingalli for N-terminal sequencing. We are grateful to Dr Danny Brower for fly stocks and antibodies. We also thank Dr Marcelo Santiago for help with confocal imaging and members of the Araujo and Bier laboratories and Danny Brower for helpful comments on the manuscript. This work was supported by Fogarty Fellowship, FIRCA/NIH and Rio de Janeiro State Foundation for Research Support (FAPERJ) grants to H.A., and by NIH grant # R01 NS 29870 to E.B.