Crossveinless 2 is an essential positive feedback regulator of Bmp signaling during zebrafish gastrulation

Signaling by secreted Bone morphogenetic proteins (Bmps) regulates numerous processes of embryogenesis and organogenesis during vertebrate and invertebrate development (Hogan,1996). Bmps are members of the transforming growth factor β(Tgfβ) superfamily of proteins, which signal through transmembrane receptor complexes consisting of type I and type II serine/threonine kinases. Activation of the receptors leads to phosphorylation of type I by type II receptors, followed by phosphorylation of intracellular R-Smad proteins by type I receptor kinases. Upon their phosphorylation, R-Smads dissociate from the receptors, form complexes with Co-Smads, like Smad4, and translocate to the nucleus to participate in the transcriptional regulation of target genes(Baker and Harland, 1997; Heldin et al., 1997).

Given the high sensitivity of developing cells to Bmp signals, the activity of Bmps is under tight spatial and temporal regulation. One important level of regulation occurs in the extracellular space, mediated by various secreted Bmp-binding proteins (Balemans and Van Hul,2002). Some of these have rather complex functions. Chordin for example binds to Bmps and prevents the interaction of Bmps with their receptors, accounting for Bmp inhibition(Piccolo et al., 1996). In addition, Chordin can have subtle long-range pro-Bmp effects, as demonstrated for the Drosophila Chordin and Bmp homologues Sog and Dpp: Sog/Dpp complexes seem to diffuse within the extracellular space, with Sog carrying Dpp away from the Sog source (Ashe and Levine, 1999; Decotto and Ferguson, 2001). By cleaving Bmp/Dpp-bound Chordin/Sog, Tolloid metalloproteases can then release biologically active Bmps from the complex(Piccolo et al., 1997),resulting in increased Bmp/Dpp levels at the sites of Tolloid activity(Ashe and Levine, 1999; Blader et al., 1997; Decotto and Ferguson, 2001). Another factor with a similar dual effect on Bmp signaling is Twisted gastrulation, which promotes binding of Chordin to Bmps (anti-Bmp effect), but also the cleavage of Chordin by Tolloid (pro-Bmp effect)(Chang et al., 2001; Larrain et al., 2001; Oelgeschlager et al., 2000; Ross et al., 2001; Scott et al., 2001; Shimmi and O'Connor, 2003). These examples show that Bmp-binding proteins cannot be generally categorized as anti- or pro-Bmp factors, but rather can have differential functions depending on the molecular context.

In the early zebrafish and Xenopus embryo, ventrally expressed Bmps and dorsally expressed Bmp inhibitors establish a ventral-to-dorsal gradient of Bmp activity, which is required for patterning of the dorsoventral axis (De Robertis et al.,2000; Hammerschmidt and Mullins, 2002; De Robertis and Kuroda, 2004; Schier and Talbot, 2005). Accordingly, the zebrafish mutants bmp2b/swirl,bmp7/snailhouse, alk8/lost-a-fin (a type I Bmp receptor), smad5/somitabun and tolloid/minifin display dorsalized phenotypes (Bauer et al., 2001; Connors et al., 1999; Dick et al., 2000; Hild et al., 1999; Kishimoto et al., 1997; Mintzer et al., 2001; Nguyen et al., 1998; Schmid et al., 2000). By contrast, the chordin/dino mutant is ventralized(Fisher et al., 1997; Schulte-Merker et al., 1997),with additional subtle dorsalized traits in ventral-most derivatives,reflecting the dual anti- and pro-Bmp function of Chordin described above(Hammerschmidt and Mullins,2002; Wagner and Mullins,2002). The effects of loss of Tsg function on dorsoventral patterning of the zebrafish embryo are less clear. Although according to one report, Tsg is a Bmp antagonist (Ross et al., 2001), other studies come to the opposite conclusion,indicating a pro-Bmp effect of Tsg (Little and Mullins, 2004; Xie and Fisher, 2005).

The Bmp binding of Chordin is mediated by CR repeats that are characterized by ten cysteine residues with a conserved spacing pattern(Larrain et al., 2000). CR domains are also present in Crossveinless 2 (Cvl2 or Cv-2). Although in vertebrates, the role of Cvl2 had not been analyzed via loss-of-function studies as yet, analyses of Drosophila mutants indicate that it is required to enhance Bmp signaling during wing vein development(Conley et al., 2000). cvl2 gain-of-function experiments in vertebrate cell culture systems and Xenopus embryos led to conflicting results, showing either pro-or anti-Bmp effects (Binnerts et al.,2004; Kamimura et al.,2004; Moser et al.,2003; Coles et al.,2004).

Here, we describe the first loss-of-function study for a vertebrate crossveinless 2 homologue. We show that zebrafish cvl2 acts as a Bmp signaling-promoting factor during early embryonic development and that this activity has Chordin-dependent and -independent aspects. We also show that proteolytic cleavage can convert Cvl2 from an anti- to a pro-Bmp factor and modifies its interaction with the extracellular matrix. These findings might help to explain the controversies about the role of Cvl2 reported in previous gain-of-function studies.

Mutant strains used were: swirl^ta72; minifin^tc263; lost-a-fin^tm110(Mullins et al., 1996) dino^tt250; ogon^tm305(Hammerschmidt et al., 1996a)and ogon^m60(Solnica-Krezel et al., 1996). The generation of hsp70::bmp2b and hsp70::noggin3 transgenic fish lines will be described elsewhere (F.R. and M.H., unpublished).

Conditions and PCR primers used for cloning of zebrafish cvl2, for the generation of expression constructs, for temporal RT-PCR analyses and for genomic mapping of cvl2 are available from the authors upon request. cvl2-N contains amino acids 1-355, in cvl2-CM residues 350-355 (VFGDPH) are replaced by LVPRGS.

The sequences of the used cvl2 antisense morpholino oligonucleotides are: TTA CTG GAG GAG ACA GAC ACA GCA T (ATG-MO=MO1)corresponding to nucleotides +1 to +25 of the cDNA; CTA AAT TCG CTC CAG ACG CAC GGG (UTR-MO=MO2) corresponding to nucleotides -25 to -2 in the 5′UTR. Sequences of the respective mismatch control MOs were: ACa GGA cGA GAC AGA CAg AGC tTC C (-2 to +23); CTA AAT TCc CTg CAc ACG gAC cGG. Unless stated otherwise, 1 nl containing 2.4 ng MO was injected per embryo.

In situ hybridization and whole-mount immunostaining was performed as described previously (Hammerschmidt et al., 1996a). Staining for phosphorylated Smads was done using an anti-pSmad1/5/8 antibody (Cell Signaling Technology) at a concentration of 1:200 and a Cy3-coupled secondary antibody (Jackson ImmunoResearch Labs). Graphical illustrations of staining intensities were generated from Photoshop images of vegetal views, using ImageJ software(http://rsb.info.nih.gov/ij/).

For Western blot analyses, dechorionated midgastrula embryos were dissociated in Ca-free Ringer's solution; cells were separated from yolk via centrifugation, and protein extracts were prepared by standard procedures(Westerfield, 1994). Extracts were separated on 8% polyacrylamide gels. Myc-tagged fusion proteins were detected with anti c-myc antibody 9E10 (Santa Cruz Biotechnology),anti-pSmad1/5/8 antibody (Cell Signaling Technology) was used at a concentration of 1:1000.

A detailed description of the expression and purification of recombinant proteins, microsequencing of the Cvl2 fragments and interaction assays is available from the authors upon request. Recombinant Chd protein was obtained from R&D systems (Minneapolis, MN).

Plasmids pCS2-cvl2-CM, pCS2-cvl2-WT and pCS2-cvl2-N were transfected into HEK293 cells with the calciumphosphat method, efficiency was controlled by co-transfection of pCS2-gfp. Transfected cells at ∼60% confluence were incubated with serum-free medium for 48 hours. Separation of supernatant, cells and ECM was carried out as described previously(Novikova et al., 2000). Briefly, the supernatant was collected and the cells were washed off with cold PBS. The ECM attached to the culture dish was washed off with 95°C sample buffer (50 mM Tris, pH 6.8, 1% SDS, 4% glycerol).

Heparin binding assays were carried out as described by Jasuja et al.(Jasuja et al., 2004), using proteins from cell extracts obtained in the transfection experiments.

We isolated the complete coding region of a zebrafish homologue of crossveinless2, consisting of a signal peptide, five chordin-like cysteine rich domains (CR-domains, also called von Willebrand Factor type C domains) and a partial von Willebrand Factor type D (vWFd) domain(Fig. 1A; GenBank Accession Number AY847871; see Fig. S1 in the supplementary material for alignments). RT-PCR analysis at different stages of embryonic development failed to detect maternally supplied cvl2 transcripts. cvl2 expression commenced at late blastula stages, ∼1 hour before the onset of gastrulation, and continued during all later developmental stages investigated(Fig. 1B).

Whole-mount in situ hybridization revealed that during blastula stages, cvl2 was broadly expressed at rather low levels (not shown). During gastrula stages (shield, 60% epiboly, 80% epiboly), cvl2 transcripts were predominantly found on the ventral side of the embryo, with from ventral to dorsal decreasing mRNA levels (Fig. 1C,D,I). Additionally, transcripts became detectable around nuclei of the yolk syncytial layer (YSL, Fig. 1E,K) and in the prechordal plate (arrow in Fig. 1D). At the tailbud stage(end of gastrulation), cvl2 was most prominently expressed in the lateral mesoderm along the entire anteroposterior length of the embryo(Fig. 1E,F). During segmentation stages, this domain became restricted to the anterior half of the embryo (Fig. 1G,H) and to the tailbud mesoderm (see Fig. S2A in the supplementary material). Interestingly,all these expression domains correspond to areas of bmp expression: bmp2b, bmp4 and bmp7 are expressed ventrally, bmp2bin the YSL, and bmp4 in the prechordal plate(Dick et al., 2000; Nikaido et al., 1997; Schmid et al., 2000).

Analysis of cvl2 expression in mutants with altered Bmp activity revealed a positive regulation of cvl2 by Bmp signaling. In chordin/dino mutants, ventral expression of cvl2 was expanded to the dorsal side (Fig. 1I,J). By contrast, in bmp2b/swirl mutant embryos,ventral expression was absent (Fig. 1K). These data suggest that cvl2 expression is under the positive control of Bmp signaling. The same appears to be true for later stages of zebrafish development, when cvl2 was expressed in a variety of different tissues, including the otic vesicles and the pharyngeal pouches(Fig. 1L and Fig. S2 in the supplementary material). Using transgenic zebrafish lines expressing bmp2b or the Bmp inhibitor noggin3 under the control of a heatshock-inducible promoter (Halloran et al., 2000) (F.R. and M.H., unpublished), we obtained complete loss of this late cvl2 expression upon blockage of Bmp signaling(Fig. 1N). However, bmp2b overexpression strongly increased cvl2 transcript levels without inducing cvl2 expression at ectopic sites(Fig. 1M).

To study whether Cvl2 is required for early zebrafish development, we carried out loss-of-function experiments, using antisense morpholino oligonucleotides (MO) (Nasevicius and Ekker, 2000). Used MOs efficiently inhibited translation of cvl2 transcripts, as revealed via western blotting of protein extracts from embryos co-injected with cvl2 MOs and mRNA encoding a Cvl2-Myc fusion protein (Fig. 2A). Phenotypically, such MO-mediated loss of Cvl2 function led to reduced Bmp signaling and moderate dorsalization of the embryo.

To detect and measure Bmp signaling activity, we examined the phosphorylation state of Smad1/5 proteins, as a direct readout of Bmp type I receptor kinase activity. Western blotting(Fig. 2B) and whole-mount immunostaining (Fig. 2C,D) at midgastrula stages revealed a significant reduction in the levels of phosphorylated Smad1/5 of cvl2 morphant embryos compared with embryos injected with mismatch control MO. Graphical illustrations of pSmad1/5 signal intensities along the dorsoventral axis of the midgastrula embryo showed that in cvl2 morphants Bmp signaling was reduced over the entire ventrolateral domain (Fig. 2E,F). Consistently, cvl2 morphants displayed reduced expression of ventral marker genes such as eve1 (ventral mesoderm)(Joly et al., 1993)(Fig. 2G,H), cvl2itself (Fig. 2I,J), gata2 (ventral ectoderm) (Detrich et al., 1995) (Fig. 2K,L) and tfap2a (ventral ectoderm)(Knight et al., 2003) (see Fig. S3E,F in the supplementary material) at midgastrula stages, and a reduced number of ventral cell types such as blood precursors (marked by expression of gata1) (Detrich et al.,1995) (Fig. 2O,P)at early segmentation stages. By contrast, expression of tbx6, a marker for the entire non-axial mesoderm largely under the control of Wnt8(Szeto and Kimelman, 2004; Ramel et al., 2005), was only moderately affected in midgastrula cvl2 morphants (Fig. S3C,D), while the expression domains of dorsal markers like otx2(Li et al., 1994)(Fig. 2M,N), which mark anterior neuroectoderm, were ventrally expanded. Accordingly, at segmentation stages, cvl2 morphants displayed an expansion of dorsal and paraxial cell types, as apparent by a broadening of hindbrain rhombomeres 3 and 5,revealed by krox20 expression(Fig. 2Q,R), and by a broadening of the anterior somites, revealed by myoD expression (see Fig. S3G,H in the supplementary material). These data indicate that Cvl2 is required for ventral specification both in the mesoderm and the ectoderm. Similar results were obtained with two different cvl2 MOs, but not with the corresponding mismatch control MOs. The effect of used MOs was specific, as shown by the rescue of the morphant phenotype to near wild-type condition upon co-injection of low doses of mRNA encoding zebrafish Cvl2 (for the UTR-MO) or Drosophila Cvl2 (for both the UTR-MO and the ATG-MO; Fig. 2S,T; and data not shown). The rescue with Drosophila cvl2 mRNA further indicates that the function of Cvl2 proteins has been conserved across invertebrate and vertebrate species.

To further analyze the functional interaction of cvl2 with the Bmp pathway, we injected suboptimal amounts of cvl2 MO into mutants with weakly reduced Bmp activity, such as tolloid/minifin. minifinembryos, as well as embryos injected with low doses of cvl2 MO (0.8 ng per embryo), display a very subtle expansion of the neuroectoderm, as shown by krox20 staining (Fig. 2U-W). Injection of the same low doses of cvl2 MO into minifin embryos, however, resulted in strongly enhanced dorsalization(Fig. 2X). The same enhancement was obtained upon co-injection of low doses of cvl2 MO with low doses of chordin, noggin or twisted gastrulation RNA, or upon injection into weakly dorsalized alk8/lost-a-fin^tm110mutants (confirmed by genotyping; data not shown). Together, these results strengthen the notion that during early dorsoventral patterning, Cvl2 has an essential ventralizing function and functionally interacts with Bmp signaling.

We next analyzed the functional interaction of cvl2 with Chordin,performing epistasis analyses by injecting cvl2 MOs into homozygous chordin mutants to generate double-deficient embryos. chordin/dino mutants display elevated Bmp signaling and are ventralized, as morphologically indicated by a reduction of the brain, a derivative of the dorsal ectoderm, and an enlargement of the blood islands, a derivative of the ventral mesoderm(Hammerschmidt et al., 1996a; Hammerschmidt et al., 1996b). The phenotype in the ventral tail fin, a derivative of the ventral ectoderm,is more complex. Although other ventralized mutants such as sizzled/ogon (Martyn and Schulte-Merker, 2003; Wagner and Mullins, 2002; Yabe et al., 2003) display a duplication of the ventral tail fin over its entire length (see Fig. S3J in the supplementary material), a variable percentage of chordin mutants show only duplications in posterior-most regions of the ventral tail fin, whereas more anteriorly,ventral tail fin tissue is missing (Fig. 3A-D). This feature reflects the additional subtle pro-Bmp activity of Chordin in regions of the ventral ectoderm that require peak Bmp levels (see Introduction).

To quantify the effects of cvl2 MOs in the ventral tail fin, we classified chordin mutants according to the length of the fin gap(class I: no gap, Fig. 3B;class II: partial gap, Fig. 3C;class III: complete gap, Fig. 3D), reflecting the degree of reduction of Bmp activity in this tissue. Injection of low doses of cvl2 MO (0.8 ng/embryo), which had no effect in wild-type siblings, significantly enhanced the loss of ventral tail fin tissue of chordin mutants (see numbers in Fig. 3B-D). However,ventralized traits at other sites of chordin mutants appeared less sensitive to Cvl2 levels. Thus, the blood islands of mutants even injected with highest doses of cvl2 MOs (3 ng/embryo) remained their increased size, as also confirmed by gata1 staining at the 10-somite and 24 hpf stage (data not shown). Similarly, injection of highest doses of cvl2MOs into dino mutants or chordin morphants only led to a very subtle alleviation of the reduced krox20 expression(Fig. 3E). The unresponsiveness of these phenotypic traits of chordin mutants to loss of Cvl2 indicates that here, Chordin is epistatic to Cvl2, suggesting that Cvl2 fulfills part of its pro-Bmp function via an inhibitory effect on Chordin. However, the enlarged gap in the ventral tail fin of chordin mutants injected with cvl2 MO suggests that Cvl2 in addition promotes Bmp signaling independently of a putative Chordin-antagonizing function.

In contrast to chordin mutants, sizzled/ogon mutants displayed a more widespread response to cvl2 MOs. Thus, injected ogon mutants not only showed a rescue of the ventral tail fin duplication (Fig. S3I-K), but also a significant lateral expansion of the krox20 expression domains (Fig. 3F), suggesting that Cvl2 acts largely independent of Sizzled.

Based on the reduction of Bmp activity in the loss-of-function assays, gain of function would be expected to yield increased Bmp activity and ventralization. However, upon overexpression of cvl2 mRNA we observed a small fraction of embryos with signs of mild ventralization, while a larger fraction displayed weak to moderate dorsalization, both by morphological(Fig. 4D; Table 1) and marker expression criteria (Fig. 4H,L). The fraction of dorsalized embryos was even higher, when more RNA was injected(Table 1). To better understand this unexpected and seemingly inconsistent result, we investigated the protein products deriving from the injected cvl2 mRNA.

Injection of mRNA encoding C-terminally myc-tagged Cvl2 into zebrafish embryos, and anti-Myc western blotting of protein extracts from injected embryos at late gastrula stages, revealed a C-terminal fragment (∼45 kDa including the Myc-tag) in addition to full-length Cvl2 protein (∼90 kDa including the Myc-tag) (Fig. 4A, left lane). The smaller fragment was obtained only under reducing electrophoresis conditions, whereas under non-reducing conditions,only one band corresponding to full-length protein was visible(Fig. 4A, right lane). The size of full length Cvl2 is slightly larger than the calculated mass (86 kDa including the Myc-tag), presumably owing to glycosylation of the N-terminal part (see Kamimura et al.,2004). Together, these data indicate that a large region of zebrafish Cvl2 is cleaved into two fragments that remain associated via disulfide bonds, similar to the situation recently reported for human Cvl2(Binnerts et al., 2004).

The same Cvl2 protein pattern was obtained when zebrafish cvl2 was expressed in SF9 cells, using the baculovirus transfection system(Fig. 6A). After purification of Cvl2 protein from the supernatants of transfected SF9 cells(Fig. 6A), microsequencing the ends of the Cvl2 fragments identified the Cvl2 cleavage site as VFGD_353-354PHYN, located between the fifth CR repeat and the vWFd domain of the full-length protein (see Fig. S1 in the supplementary material).

In order to analyze the impact of this proteolytic cleavage on the biological activity and the molecular properties of Cvl2, we generated two mutant constructs of Cvl2: a full-length construct in which the cleavage site is destroyed (Cvl2-CM) and a construct representing only the N-terminal half of Cvl2 containing the five CR domains (Cvl2-N, amino acids 1-355)(Fig. 4B). These constructs,when injected into zebrafish embryos or transfected into SF9 cells, led to protein bands of the expected sizes (Fig. 4C, Fig. 6A). In contrast to wild-type Cvl2-WT, which can both ventralize and dorsalize, these two opposite effects were completely separated in the two mutant versions of Cvl2. Although the mRNA encoding cleavage-resistant Cvl2-CM caused strong dorsalization (Fig. 4F,J,N and Table 1), injection of cvl2-N mRNA led to ventralized phenotypes only(Fig. 4G,K,O; Table 1). Co-injection of cvl2-CM and cvl2-N mRNA caused intermediate phenotypes,similar to the effect of cvl2-wt(Table 1). Together, these results indicate that uncleaved Cvl2 acts as an inhibitor of Bmp signaling,which upon proteolytic processing can be converted to a Bmp signaling-promoting factor, the apparent in vivo role of Cvl2 according to the loss-of-function studies described above.

Our observation that cvl2 morphants display reduced levels of phosphorylated Smad1/5 proteins suggests that Cvl2 acts by modulating Bmp signaling. To study whether this is also true under gain-of-function conditions, we injected cvl2-N mRNA into bmp2b morphant embryos. Indeed, while sibling embryos injected with cvl2-N mRNA alone were ventralized, embryos co-injected with cvl2-N mRNA and bmp2b MO were as strongly dorsalized as control embryos injected with bmp2b MO only (Fig. 5A-D; Table 1). This indicates that Cvl2 requires Bmps to fulfill its ventralizing activity,suggesting that it is a Bmp signaling-promoting factor that acts at the level or upstream of Bmp proteins. By contrast, embryos co-injected with chordin and cvl2-N mRNAs displayed a strong reduction of dorsalization compared with sibling embryos injected with chordinmRNA alone (Fig. 5E,F; Table 1). This shows that Cvl2 can rescue the blockage of Bmp signaling by Chordin, consistent with the results of the Cvl2-Chordin epistasis analyses described above(Fig. 3).

Comparing the ventralized traits of chordin mutants in the absence or presence of Cvl2, we had concluded that Cvl2 has both Chordin-dependent and Chordin-independent pro-Bmp effects (Fig. 3). To examine the Chordin-independent effect under gain-of-function conditions, cvl2-N mRNA was co-injected with maximal amounts of chordin MOs. Nevertheless, chordin morphants co-injected with cvl2-N mRNA displayed an enhancement of all ventralized traits, including a further expansion of the blood island and a further reduction of the size of the head, as well as a loss of the gap in the ventral tail fin (Fig. 5G,H; Table 1). Although we cannot rule out that some residual Chordin protein is present in morphants, this result suggests that Cvl2 has an additional, Chordin-independent pro-Bmp function.

To gain insights into the molecular mechanisms underlying the effects of Cvl2 on Bmp signaling, we first studied the Bmp binding properties of the Cvl2 versions via Biacore analyses. Cvl2 proteins were isolated from baculovirus-transfected SF9 cells, and purified to ∼90% purity(Fig. 6A). Comparing cleavage-resistant Cvl2-CM with the CR repeats-containing Cvl2-N fragment, and with wild-type Cvl2 (Cvl2-WT, ∼80% cleaved/associated; Fig. 6A, lane 4), all three immobilized versions were found to bind Bmp2 with nearly identical high affinity (apparent K_D ∼ 1.2 nM)(Table 2 and Fig. 6B). Similar affinities were determined for Bmp4 and Bmp7, while binding to the more distantly related growth and differentiation factor 5 (Gdf5) was about 25-fold weaker(Table 2). The affinities between Cvl2 and Bmp members are similar to the affinity between Bmp2 and the type I receptor BRIA (Keller et al.,2004). Together, these data indicate that Cvl2 strongly binds to Bmps, that this binding occurs via its N-terminal CR repeats, and that binding is not affected by the proteolytic cleavage of Cvl2.

In the loss-of-function experiments, we had observed that the pro-Bmp activity of Cvl2 is strongly reduced in chordin mutants. As both Cvl2 and Chordin directly bind to Bmps, we sought to determine whether Cvl2 might function to prevent the inhibitory interaction between Chordin and Bmps. We therefore carried out competitive Biacore analyses, measuring the binding of Chordin to preformed Bmp2-Cvl2 complexes (K_D ∼ 25 nM; Fig. 6B; Table 3). We found that saturation of Bmp2 with Cvl2-WT or Cvl2-N blocked binding of Chordin to Bmp2(Fig. 6D), although Chordin readily bound to immobilized Bmp2 alone (K_D ∼ 12 nM; Fig. 6C, Table 3), consistent with previous reports (Larrain et al.,2000). This shows that Cvl2 and Chordin bind to overlapping epitopes of Bmp2, and that direct binding of Cvl2 and Chordin to Bmps is mutually exclusive.

Despite the opposing activities of cleaved and uncleaved Cvl2 in overexpression assays, we were unable to detect significant differences in their binding properties. As many secreted signaling proteins, including Bmps themselves and their inhibitors Noggin and Chordin, can be tightly associated with components of the extracellular matrix (ECM)(Ruppert et al., 1996; Paine-Saunders et al., 2002; Jasuja et al., 2004), we wanted to examine whether differential affinity to the ECM might account for the opposite effects of uncleaved and cleaved Cvl2.

For this purpose, we transfected plasmids encoding Myc-tagged Cvl2-WT or Cvl2-N into HEK293 cells, cultured cells for 48 hours and isolated proteins from cells, supernatants and ECM. Upon transfection with Cvl2-WT, which gives rise to both cleaved/associated and uncleaved Cvl2 in cell extracts (data not shown), the two forms distributed differently in the extracellular domain. Whereas uncleaved Cvl2 was found in the ECM fraction only, cleaved Cvl2 was in the supernatant (Fig. 6E, lanes 3-5). Similarly, Cvl2-N, which lacks the C-terminal half, was present only in the supernatant, but not in the ECM fraction(Fig. 6E, lanes 6,7). This shows that the C-terminus containing the vWFd domain is required for binding of uncleaved Cvl2 to the ECM, while cleavage of Cvl2 strongly reduces ECM binding, although the N- and C-terminal fragments remain associated.

As association of Bmps, Chordin and Noggin with the ECM mainly occurs via heparan sulfate proteoglycans (HSPGs)(Ruppert et al., 1996; Paine-Saunders et al., 2002; Jasuja et al., 2004), we next tested whether Cvl2 can bind to heparin-coated sepharose beads. After incubation of Myc-tagged Cvl2-N and Cvl2-CM with the beads, proteins were eluted with increasing concentrations of NaCl. The elution profiles showed that Cvl2-CM binds to heparin with much higher affinity than Cvl2-N (elution maximum for Cvl2-CM at 1.25-1.5M NaCl, for Cvl2-N at 0-0.5M NaCl, Fig. 6G), suggesting that differential binding to HSPGs might account for the difference in binding to the ECM between cleaved and uncleaved Cvl2.

To study the functional relevance of heparin binding, we sought to interfere with the heparin and ECM binding of Cvl2-CM. Cvl2 contains 67 basic amino acids (Arg and Lys) as potential binding sites for HSPGs, making it difficult to completely abolish heparin binding. We focused on one site that conforms to a consensus heparin binding site (see Hileman et al., 1998), which is located at amino acids 393-396 (RRTR), 40 residues C-terminal of the cleavage site. Removal of these four residues from Cvl2-CM reduced the affinity for heparin beads (elution maximum at 0.75-1M NaCl; Fig. 6F) and led to the release of a significant fraction of the protein into the supernatant after transfection into HEK293 cells (Fig. 6G). Importantly, injection of mRNA for this construct(cvl2-CMΔ393-396) into zebrafish embryos had a strongly reduced dorsalizing activity compared with cvl2-CM, although both proteins were synthesized at equivalent levels(Table 1; Fig. 6H). These data suggest that association of uncleaved Cvl2 with HSPGs of the ECM significantly contributes to its anti-Bmp activity, possibly by tethering Bmps at the ECM,making them inaccessible for their transmembrane receptors.

Providing the first loss-of-function study of a vertebrate crossveinless2 homologue, we show that Cvl2 is required to promote Bmp signaling during dorsoventral patterning of the zebrafish embryo. We further show that cvl2 is co-expressed with bmps on the ventral side of the gastrulating embryo, and that its transcription is positively regulated by Bmp signaling, constituting a positive feedback loop of Bmp signaling.

During gastrula stages, Bmps are expressed in a graded fashion with highest levels at the ventral pole. According to the morphogen concept, this Bmp gradient defines positional values and differential cell fates along the dorsoventral axis. Cvl2 promotes Bmp signaling at all positions of the gradient: in the presence of the Cvl2-dependent positive feedback loop, this gradient reaches higher maximum levels at the ventral pole and is steeper, as revealed by pSmad1/5 immunostaining (Fig. 2B). Clearly, an increased slope should yield a higher spatial resolution of positional values along the dorsoventral axis, thereby facilitating the translation of the gradient into differential cell fate. In addition, the Cvl2-dependent positive feedback should ensure higher gradient stability over time. A similar role of Cvl2 to stabilize territories of Bmp signaling might also occur during later processes of vertebrate development,when cvl2 displays a highly dynamic, but restricted expression at various sites that are supposed to be under the control of Bmps(Fig. 1; see Fig. S2 in the supplementary material) (Coffinier et al.,2002; Coles et al.,2004; Kamimura et al.,2004). Such a role of zebrafish Cvl2 to locally elevate Bmp signaling levels is consistent with the function of its counterpart crossveinless2 during wing development in Drosophila(Conley et al., 2000). Here,the developing crossveins of the pupal wing show increased Bmp signaling activity, revealed by increased Mad (the Drosophila Smad1/5 homologue) phosphorylation, although neither the Drosophila bmphomologs decapentaplegic and glass bottom boat, nor their type I receptor thick veins show elevated expression in this region. However, cvl2 expression is upregulated in this territory, and in cvl2 mutants, the elevated Mad phosphorylation is lost(Conley et al., 2000).

Our embryological epistasis analyses revealed a significant dependence of Cvl2 activity on Chordin function. Thus, we found that although loss of Cvl2 in wild-type embryos leads to reduced Bmp signaling and a dorsalization of the embryo, loss of Cvl2 in chordin mutants had only very moderate dorsalizing effects and failed to rescue particular ventralized traits of the mutants, as revealed by the persistent enlargement of the blood islands and the persistent reduction of the neuroectoderm(Fig. 3). This indicates that in the absence of Chordin, Cvl2 cannot display its full pro-Bmp effect,suggesting that it normally promotes Bmp signaling by antagonizing Chordin. This notion of Cvl2-Chordin competition is further supported by our overexpression studies (Fig. 5), showing that application of Cvl2-N can rescue the dorsalization caused by Chordin, whereas it cannot compensate for the loss of Bmp signaling caused by knockdown of Bmps.

In addition, we could detect Chordin-independent pro-Bmp effects of Cvl2 both in gain- and loss-of-function experiments. In the loss-of-function scenario, this effect was most apparent in the ventral tail fins, the tissue supposed to derive from the ventral-most ectoderm. In chordinmutants, it is reduced, pointing to a pro-Bmp effect of Chordin in ventral tail fin tissue. Recent experiments with temporally controlled Bmp inactivation indicate that Bmp function driving ventral tail fin formation is required during post-gastrulation stages(Pyati et al., 2005), and the same might be true for Chordin. In addition, Chordin might act during gastrula stages, transporting Bmps against their gradient into ventral-most regions to reach peak Bmp levels in concert with Tolloid function(Hammerschmidt and Mullins,2002; Wagner and Mullins,2002). In any case, our findings that both in chordin and in tolloid mutants, dorsalized traits of the ventral tail fin tissue are further enhanced by loss of Cvl2 suggests that here, Cvl2 has a pro-Bmp effect independently of the Chordin-Tolloid system (Figs 2, 3 and data not shown). Consistent results pointing to a Chordin-independent pro-Bmp effect of Cvl2 were obtained in our gain-of-function experiments, showing that Cvl2-N can further ventralize chordin morphant embryos(Fig. 5).

Cvl2 contains five CR domains, which were initially characterized in Chordin, but are also present in other Bmp-regulating secreted proteins(Garcia Abreu et al., 2002). Accordingly, we found that both cleaved and uncleaved full length Cvl2, as well as its N-terminal fragment containing the CR domains, readily and specifically bind Bmps (Fig. 6). Binding affinities to immobilized Bmp2 were similar to that determined for Chordin under identical conditions (K_D ∼25 nM for Cvl2, ∼12 nM for Chordin; Table 3). We further found that direct binding of Chordin or Cvl2 to Bmp2 is mutually exclusive (Fig. 6), consistent with the opposing roles of Cvl2 and Chordin during zebrafish development. In this respect, it appears that the Chordin-dependent pro-Bmp effect found in the zebrafish embryo is due to an `anti-Chordin'function, with Cvl2 competing with Chordin for free and possibly also for Chordin-prebound Bmps. Along the same lines, the Chordin-independent Bmp-promoting effect of Cvl2 could be due to competition with other Bmp inhibitors or Bmp-binding proteins.

We also searched for differences in the molecular properties of Cvl2-Bmp versus Chd-Bmp complexes that could account for the opposite effects of Cvl2 and Chd on Bmp signaling. One possibility could be that, in contrast to Chd(Larrain et al., 2000),binding of Cvl2 to Bmp does not block the interaction between Bmps and their transmembrane receptors, similar to how it has been previously reported for the anti-Bmp factor Follistatin (Iemura et al., 1998) and the CR domain-containing pro-Bmp protein Kcp(Lin et al., 2005). To test this notion, we performed competitive Biacore analyses between Chd or Cvl2,Bmp2 and type I (BmpRIA) or type II (BmpRII) receptors. However, obtained results were inconclusive. Thus, depending on which of the components were immobilized, each of the tested proteins (Cvl-N, Cvl-CM or Chd) either blocked Bmp-BmpR interaction, or was permissive, whereas we failed to detect crucial differences between Cvl-N on one side, and Cvl-CM and Chd on the other (J.Z. and W.S., unpublished).

Alternatively, Cvl2 could act at the level of other components of Bmp-containing complexes, such as Twisted gastrulation (Tsg). In gain-of-function experiments, Tsg can display both pro- and anti-Bmp effects,as has been described for Cvl2. As Cvl2, Tsg binds Bmps, mediated by a partial CR-like domain (Oelgeschlager et al.,2003). Tsg also binds Chordin, thereby facilitating binding of Chordin to Bmps, accounting for its anti-Bmp effect(Chang et al., 2001; Oelgeschlager et al., 2000). In addition, Tsg promotes cleavage of Chordin by the metalloprotease Tolloid,resulting in a pro-Bmp effect(Oelgeschlager et al., 2000; Scott et al., 2001; Larrain et al., 2001; Shimmi and O'Conner, 2003; Little and Mullins, 2004; Xie and Fisher, 2005). In view of this, we tested whether Cvl2 might act via Tsg, but Biacore analyses indicated that Cvl2 and Tsg do not physically interact (J.Z. and W.S.,unpublished). However, we cannot rule out that Cvl2 binds to higher order complexes containing Chd and/or Tsg.

Recently, it has been reported that the Bmp-antagonizing function of Chordin depends on the binding of Chordin to heparan sulfate proteoglycans(HSPGs) (Jasuja et al., 2004). HSPGs are either components of the extracellular matrix (ECM) or remain attached to the cell membrane, usually via GPI anchors. In Drosophila, the membrane-bound Glypican protein Dally has been shown to promote signaling by the Bmp homologue Dpp, acting as some kind of co-receptor (Fujise et al.,2003). In this light, both the Chordin-dependent and the Chordin-independent pro-Bmp effects could be explained by a role of Cvl2 to regulate the distribution of Bmps within the extracellular space, or the recruitment of Bmps to their cell-surface receptors. Here, we found that uncleaved Cvl2 binds to heparin with an affinity very similar to that reported for Chordin, and also to the ECM of transfected cells. Removal of a putative heparin binding site from uncleaved Cvl2 results in reduced affinity for heparin, reduced binding to the ECM and decreased anti-Bmp activity in embryos. Moreover, co-injection of exostosin1 MOs, targeting a key enzyme of HSPG synthesis, strongly reduced the dorsalizing activity of cvl2-CM mRNA (F.R. and M.H., unpublished). This suggests that similar to Chordin, uncleaved Cvl2 requires binding to HSPGs to exert its anti-Bmp activity.

By contrast, cleavage of Cvl2 into two fragments that remain associated via disulfide bonds apparently leads to a conformational change that abolishes Cvl2 binding to the ECM (Fig. 6). As Bmps usually display strong affinity to the ECM (Reddi,2002) and as cvl2 is co-expressed with bmps, binding of Bmps to cleaved Cvl2 could prevent sequestration of Bmps at the ECM, thereby helping to increase the availability of Bmps for receptor activation on target cells. Alternatively, or in addition, cleaved Cvl2 might facilitate the interaction between Bmps and their HSPG co-receptors.

In summary, our biochemical analyses indicate that Cvl2 binds to Bmps, that Cvl2 and Chordin compete for Bmp binding, and that cleaved Cvl2, in contrast to uncleaved Cvl2 and Chordin, lacks binding to the ECM. However, how this leads to the pro-Bmp effect of Cvl2 observed in vivo, remains largely unclear. In any case, our results highlight that binding of CR proteins to Bmps does not necessarily lead to an inhibition of Bmps, as in the case of Chordin(Picollo et al., 1996), but can, by contrast, be required to promote Bmp signaling, consistent with recent data obtained for the Kielin protein Kcp during renal regeneration in mouse (Lin et al., 2005).

Cvl2 gain-of-function studies in cell culture systems and in Xenopus embryos have yielded controversial results, revealing either pro- or anti-Bmp effects. Our findings that Cvl2 can be proteolytically cleaved, and that cleaved and uncleaved Cvl2 display similar affinities to Bmps, but different affinities to the ECM, might provide a biochemical explanation for this phenomenon. Our overexpression studies further indicate that Cvl2 must be present in the cleaved state to elicit its pro-Bmp effect,while the contrary phenotype obtained in our loss-of-function studies suggests that it is this cleaved version which is predominantly present during early zebrafish development to mediate the essential pro-Bmp effect of Cvl2. However, this might be different during other processes and/or in other organisms, where the uncleaved version might be present in excess, leading to a net anti-Bmp function of Cvl2. Such differences in the ratio of uncleaved versus cleaved protein could depend on the synthesis and turnover rate of Cvl2, or the abundance of Cvl2 proteases. According to our data, proteolytic activity is present in human (Fig. 6E) and insect (Fig. 6A) cell cultures, as well as in zebrafish embryos(Fig. 4A). The nature of the protease, if existent, remains unclear. We could not detect altered ratios of Cvl2 cleavage in tolloid gain-or loss-of-function experiments (data not shown), ruling out this apparent candidate. In any case, a spatial and temporal control of Cvl2 cleavage would offer an efficient and flexible system for the fine-tuning of Bmp activity, allowing context-dependent switching of a potent Bmp inhibitor to a Bmp agonist.

We thank M. Brand, J.-S. Joly, R. T. Moon, E. Weinberg and L. Zon for sending plasmids; Mike O'Connor for the Drosophila cvl-2 construct;Yukiyo Yamamoto for help with transfection experiments; Nicole Hopf for excellent technical assistance; Joachim Nickel for supplying GDF-5; and Michael Oelgeschlaeger for critical reading of the manuscript. Work in M.H.'s laboratory was supported by the Max-Planck Society and NIH Grant 1R01-GM63904,work in W.S.'s laboratory by DFG Grant KFO 103, TeilC.