Hemicentin Assembly in the Extracellular Matrix Is Mediated by Distinct Structural Modules*

Hemicentins are conserved extracellular matrix proteins characterized by a single von Willebrand A (VWA) domain at the amino terminus, a long stretch (>40) of tandem immunoglobulin domains, multiple tandem epidermal growth factors (EGFs), and a single fibulin-like carboxyl-terminal module. In Caenorhabditis elegans, hemicentin is secreted from muscle and gonadal leader cells and assembles at multiple locations into discrete tracks that constrict broad regions of cell contact into adhesive and flexible line-shaped junctions. To determine hemicentin domains critical for function and assembly, we have expressed fragments of hemicentin as GFP tagged fusion proteins in C. elegans. We find that a hemicentin fragment containing the VWA domain can target to multiple assembly sites when expressed under the control of either endogenous hemicentin regulatory sequences or the muscle-specific unc-54 promoter. A hemicentin fragment containing the EGF and fibulin-like carboxyl-terminal modules can co-assemble with existing hemicentin polymers in wild-type animals but has no detectable function in the absence of endogenous hemicentin. The data suggest that the VWA domain is a cell binding domain whose function is to target hemicentin to sites of assembly and the EGF/fibulin-like carboxyl-terminal modules constitute an assembly domain that mediates direct interactions between hemicentin monomers during the hemicentin assembly process.

Extracellular matrix (ECM) 2 proteins form networks with distinct geometries that provide cells with regulatory and structural information to guide the morphogenesis and histogenesis of complex three-dimensional tissues. ECM networks act as scaffolds for cell adhesion and migration and direct assembly of the cytoskeleton. Network geometry varies from tissue to tissue and is determined by the availability of ECM components in addition to the mechanism of their assembly. Several common themes have emerged in the assembly of ECM molecules like collagens, laminins, and fibronectins (1)(2)(3)(4)(5). ECM molecules can be assembled into polymers inside the cell and activated before or after secretion as a result of proteolytic processing or interaction with cell surface receptors. Activation changes the structure of specific domains and results in increased homophilic interactions or heterophilic interactions with other ECM proteins as molecules assemble into higher order structures.
Hemicentins are a family of secreted ECM proteins first identified in Caenorhabditis elegans with two orthologs in most vertebrate genomes including human and mouse (6). Hemicentins are characterized by a single highly conserved von Willebrand A (VWA) domain near the amino terminus, followed by a long stretch (Ͼ40) of tandem immunoglobulin (Ig) domains, multiple tandem epidermal growth factors (EGFs), and a single fibulin-like carboxyl-terminal (FC) domain (6 -8).
In C. elegans, hemicentin is secreted by bodywall muscle and gonadal leader cells and assembles at multiple distant locations, where it forms an oriented track-like geometry whose apparent function is to constrict broad regions of cell contact into discrete line-shaped junctions (6). There is nothing known about the mechanism of hemicentin recruitment to cell surfaces or about its assembly there.
Hemicentin tracks anchor mechanosensory neuron and uterine cells at hemidesmosome-mediated attachments and are required for cellularization of germ cells and migration of somatic male gonadal cells (6). Hemicentin expression by the proximal gonadal cell in hermaphrodites (i.e. the anchor cell) is controlled by the fos-1 transcription factor, and accumulation of hemicentin on the anchor cell surface promotes anchor cell invasion into vulval epithelium to form a functional connection between the uterus and vulva (9).
To understand the role of individual hemicentin structural domains in hemicentin function and assembly, we have expressed hemicentin and hemicentin fragments as green fluorescent protein (GFP)-tagged fusion proteins in C. elegans and present data that suggest the function of the VWA domain is to target hemicentin to sites of assembly. The carboxyl-terminal EGF/FC domain contains a motif sufficient to bind existing hemicentin polymers in wild-type animals, suggesting that this is likely to constitute an assembly domain. Based on this data we suggest a multistep model for hemicentin assembly similar to that of other extracellular matrix components, involving cell binding, activation, and propagation (1). * This work was supported by the National Institutes of Health Grant GM65184. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. □ S The on-line version of this article (available at http://www.jbc.org) contains supplemental Fig. 1S

MATERIALS AND METHODS
Deletion Constructs-Deletion constructs were made using AvrII, KpnI, BamHI, NotI, and ApaI restriction sites to linearize a full-length GFP::hemicentin gene (6), SOE-PCR to splice together fragments from different regions of the hemicentin gene (10) and homologous recombination in Escherichia coli to combine overlapping linear DNA fragments into a circular construct expressing ampicillin resistance (11).
Custom Antibodies-A cDNA fragment encoding hemicentin amino acids 1-433 was amplified by reverse transcription-PCR and cloned into the pET-41 EK/LIC expression vector (Novagen). The clones were transformed into E. coli BL21(DE3) and induced with 0.5 mM isopropyl 1thio-␤-D-galactopyranoside for 3 h. at 25°C. The GST fusion protein was purified with GST-binding resin (Novagen) according to the manufacturer's instructions. Polyclonal antisera were produced in rabbits immunized with hemicentin::GST fusion protein by Proteintech (Chicago, IL).
Western Blots-Worms were collected from a 60-mm plate in M9 buffer and boiled in SDS sample buffer. After electrophoresis through (3-8%) NuPAGE-Tris acetate gels (Invitrogen), proteins were transferred to a polyvinylidene difluoride membrane (Bio-Rad). The membrane was blocked with 5% nonfat milk in PBST (phosphate-buffered saline with 0.05% Tween 20) (pH 7.4) and incubated with rabbit primary anti-GFP antibodies (Molecular Probes) followed by a polyclonal goat anti-rabbit horseradish peroxidase conjugate (Pierce). After washing, bound antibody was detected by Supersignal West Pico Chemiluminescent Substrate (Pierce).

RESULTS
Full-length C. elegans hemicentin comprises 5198 amino acids and contains a VWA domain followed by 48 tandem Ig modules, 3 EGF modules, and a FC module. To test the function of these domains in hemicentin assembly, we designed several hemicentin based constructs with various domains deleted or mutated ( Fig. 1) and introduced them into hemicentin containing wild-type animals and hemicentin-deficient (him-4(rh319)) mutant animals. Each construct contains a signal sequence followed by GFP and is under the control of endogenous him-4 5Ј-regulatory sequences (F15G9 8056 -13,471) with the exception of unc-54::1-433, which is under the control of the muscle-specific unc-54 promoter (13). Constructs 1-433, unc-54::1-433 and D42A contain nucleotides encoding the initiator methionine up to but not including the first Ig module. Construct 1-1492 includes nucleotide sequences from the initiator methionine to Ig11. Construct 434 -5198 includes all hemicentin sequence encoding Ig1 onward. Construct 1-4837 is missing the sequence that encodes amino acids following the last Ig (Ig48) and 4838 -5198 contains only sequence encoding amino acids that follow Ig48 (Fig. 1).
Expression of each of these constructs is detected by Western blot with anti-GFP antibodies and observed sizes match predicted sizes (Fig. 2). The observed size of full-length hemicentin::GFP matches the predicted molecular weight based on amino acid sequence of ϳ600 kDa, suggesting that the most abundant form of hemicentin present in wild-type animals is a full-length monomer that does not appear subject to either extensive proteolysis or post-translational modification. All constructs contain a signal sequence SI and GFP followed by hemicentin amino acid sequences as indicated to the left of each schematic. D42A is identical to 1-433 with the exception that conserved metal-ion dependent adhesion site amino acid 42 is mutated from aspartate to alanine. To the right of each schematic is a summary of the activity of each construct in the ability to bind to flexible tracks that connect pharynx to anterior bodywall muscle and the ability of each construct to bind correct cell surfaces.
As shown previously, hemicentin can assemble into fine, regularly spaced tracks that pivot and flex during foraging. These tracks are not associated with cell surfaces other than at attachment points at the pharyngeal basement membrane at one end and between lateral and medial bodywall muscle cells at the other end (6). Full-length GFP::hemicentin can assemble into tracks in the presence or absence of endogenous hemicentin ( Fig. 3; Ref. 6). When we expressed constructs 434 -5198 and 4838 -5198 we found that these constructs are sufficient to bind to existing hemicentin tracks found in wild-type animals but cannot assemble into tracks in him-4 (rh319) mutant animals that lack endogenous hemicentin (Fig. 3, B, C, E, and F). Constructs 1-433 and 1-1492 lack this region and are not able to bind to existing hemicentin tracks in wild-type animals or form tracks on their own in a him-4 (rh319) mutant background (Fig. 3, D  and G). A larger hemicentin construct that is missing only this region (1-4837) retains the ability to bind to tracks in a wild-type background but cannot assemble into tracks in the absence of existing hemicentin (Fig. 3, H and I), indicating that the sequences 4838 -5198 are sufficient, but not necessary, for hemicentin track binding activity. Constructs that subdivided this region by expressing the EGF and FC modules individually as GFP fusion proteins were also tested but these did not localize to specific structures and instead were found in the fluid of the pseudocoelomic (body) cavity (not shown). Interestingly, the localization of constructs to hemicentin tracks was most striking in younger animals (L2-L3 larvae) and decreased in older animals. The expression of all constructs in wild-type animals did not result in detectable dominant negative phenotypes.
Although constructs located near the amino-terminal were not able to bind to hemicentin tracks (Fig. 3, D  and G), we observed that they did have the ability to bind to cell surfaces where hemicentin tracks normally assemble in wild-type animals including mechanosensory neurons, uterine attachments, and germ cell surfaces (Fig. 4). This was true for constructs containing the first 1492 amino acids (Fig. 4, J-L) as well as constructs limited to the first 433 amino acids of hemicentin (Fig. 4, D-F). Even in hemicentin mutant (him-4 (rh319) and him-4 (hd99)) backgrounds, the first 433 amino acids were still able to bind to uterine attachments and germ cell surfaces (Fig. 4, H and I) suggesting that these hemicentin fragments were not binding to endogenous hemicentin but binding directly to cell surface molecules or extracellular molecules associated with those cell surfaces. 1-433 and 1-1492 constructs were expressed under control of endogenous him-4 5Ј-regulatory sequences and under the control of the muscle-specific unc-54 promoter to determine whether the accumulation of 1-433 on mechanosensory neurons, uterine attachments and germ cell surfaces was the result of targeting of secreted hemicentin from muscle cells or low levels of expression in those tissues. We found that accumulation of the 1-433 construct on these tissues was still strong even under the muscle-specific unc-54 promoter (Fig. 4, M-O).
To determine whether the VWA motif or other sequences within hemicentin 1-433 were important for cell surface localization, we introduced a mutation in the conserved metal iondependent adhesion site (14). The conserved sequence DXTXS was mutated by substituting an alanine codon for the aspartate codon. When expressed in transgenic animals, the ability of this construct (D42A) to bind cell surfaces was dramatically reduced (cf Fig. 4, D-F and P-R), although similar protein levels were detected in both strains (Fig. 2).
Because the first 433 amino acids of hemicentin seemed to be sufficient for targeting to cell surfaces, we deleted this domain to determine whether this domain was necessary for hemicentin targeting. We found that a construct without the VWA domain (434 -5198) retained the ability to target cell surfaces in the presence or absence of endogenous hemicentin, suggesting the presence of an additional cell binding domain in hemicentin (Fig. 4, S-U).

DISCUSSION
Extracellular matrix proteins assemble into networks with a variety of geometries that provide structural and regulatory information to cells to pattern tissue histogenesis and morphogenesis. In C. elegans, hemicentin assembles into discrete tracks that contract broad regions of cell contact into discrete lineshaped junctions. To understand the mechanism of hemicentin assembly, we expressed fragments of hemicentin as GFP fusion proteins in wild-type and hemicentin-deficient animals. Below, we discuss our findings and present a model for the assembly of hemicentin that incorporates this data.
The EGF/FC Modules Constitute an Assembly Domain of Hemicentin-A construct containing GFP fused to the EGF/FC modules of hemicentin (4838 -5198) binds to hemicentin containing structures that are present in wild-type animals (Fig. 3). Hemicentin is found in flexible structures that extend from bodywall muscle cell surfaces to the pharynx basement membrane and is not associated with cell surfaces other than at these attachment sites. Because hemicentin and fibulin are the only proteins that have been identified in these structures (15), this observation raises the possibility that one function of the EGF/FC modules are to mediate direct hemicentin-hemicentin or hemicentin-fibulin interaction.
Even with the EGF/FC modules missing, other hemicentinhemicentin binding sites are still present that can bind to EGF/FC modules present in endogenous hemicentin molecules that are found in the flexible tracks of wild-type animals. It is not surprising for an extracellular matrix molecule to have multiple self-assembly domains, because a minimum of two self-assembly domains are required for assembly into a polymer larger than a dimer. Fibronectin, for example, assembles into large, complex fibrillar structures and has at least four self-assembly domains (2). We speculate that the large number of Ig domains found in hemicentin reflect not only homophilic interactions but interactions with other binding partners. The Ig domains in perlecan, for example, are known to be involved in direct protein-protein interactions with multiple partners including nidogen and fibulin-2 (16).
Interestingly, the combination of EGF and FC modules seems to form a single functional unit, because we could not subdivide this region into smaller functional domains. This combination is also found in the two vertebrate hemicentins as well as all five fibulin family members (8). This striking similarity has resulted in occasional reference to human hemicentin-1 as fibulin-6. A C. elegans fibulin-1 ortholog co-assembles with hemicentin on mechanosensory neurons, uterine attachments, and flexible tracks (15). It is tempting to speculate that like hemicentin, fibulin uses the EGF/FC modules to interact directly with hemicentin as the 2 molecules co-assemble at these sites.
A mutation in the last EGF of human hemicentin-1 was found linked to the disease phenotype in a familial case of agerelated macular degeneration (17). Based on data presented here, we suggest that the effect of this dominant mutation is likely to interfere with the hemicentin assembly process.
The VWA Domain of Hemicentin Is a Cell-targeting Domain-VWA domains are known to be frequently involved in mediating divalent cation-dependent interactions between proteins assembled into multiprotein complexes. A VWA domain containing fragment of hemicentin (1-433) targets to uterine attachments and germ cell surfaces in the presence or absence of endogenous hemicentin (Fig. 4). Targeting occurs when 1-433 is expressed under the control of endogenous hemicentin regulatory sequences or when expressed under the control of the muscle-specific unc-54 promoter. It is likely that this is a key domain in targeting hemicentin to different sites of assembly and that hemicentin binds to cell surface receptors or other extracellular matrix protein found at assembly sites. The binding is likely to be dependent on divalent cations because the cell-targeting activity of 1-433 is dramatically reduced when a point mutation is introduced into the highly conserved divalent cation binding sequences of the VWA domain metal-ion dependent adhesion site.
Structurally, the VWA domain is the most highly conserved region of hemicentin. This raises the possibility that the VWA domain has a similar function, that of targeting hemicentins to cell surfaces in vertebrate organisms as well. Another molecule, G7c or NG37, has been identified in vertebrate genomes that has a VWA motif very similar to the VWA found in hemicentins (6,7,18) and may indicate that this domain targets the secreted G7c/NG37 molecule to hemicentin binding cell surfaces as well.
The observation that a hemicentin construct where amino acids 1-433 have been deleted still binds to uterine attachments and germ cell surfaces even in the absence of endogenous hemicentin suggests the possibility of a second cell binding site located elsewhere in the molecule. This is not unexpected for a large extracellular matrix molecule, many of which have multiple cell binding domains. Fibronectin, for example, has multiple cell binding sites that bind to distinct members of the integrin family of adhesion receptors (1).
Model for Hemicentin Assembly-We have summarized the results and incorporated these observations into a model of hemicentin assembly (Fig. 5).
Step 1 (secretion): in C. elegans, hemicentin is secreted from muscle and gonadal leader cells. Potential intramolecular interactions between the carboxylterminal EGF/FC modules and other domains within hemicen-tin may result in newly secreted hemicentin forming compact "blocked" structures where the EGF/FC modules are sequestered, preventing hemicentin assembly in ectopic locations. Steps 2 (nucleation) and 3 (activation): the blocked molecule encounters a cell surface or ECM molecule that binds to the VWA and/or sites in the Ig domain, causing the monomer to unfold, exposing homophilic binding sites in the carboxyl-terminal EGF/FC modules and elsewhere in the molecule.
Step 4 (propagation): additional hemicentin molecules are added to the nascent structure as carboxyl-terminal EGF/FC modules bind to homophilic binding sites. At this point, we cannot determine whether additional hemicentin molecules are added in a parallel or anti-parallel orientation. The next step in the assembly of hemicentin is likely to involve direct interactions with other hemicentin molecules and possibly other ECM components (Step 5). An obvious candidate to interact with hemicentin is fibulin, which, as mentioned above, also contains EGF/FC motifs and has been shown recently to co-assemble with hemicentin in flexible tracks and hemidesmosome-mediated uterine and mechanosensory neuron attachments (15).