Contributions of the LG modules and furin processing to laminin-2 functions.

The alpha2-laminin subunit contributes to basement membrane functions in muscle, nerve, and other tissues, and mutations in its gene are causes of congenital muscular dystrophy. The alpha2 G-domain modules, mutated in several of these disorders, are thought to mediate different cellular interactions. To analyze these contributions, we expressed recombinant laminin-2 (alpha(2)beta(1)gamma(1)) with LG4-5, LG1-3, and LG1-5 modular deletions. Wild-type and LG4-5 deleted-laminins were isolated from medium intact and cleaved within LG3 by a furin-like convertase. Myoblasts adhered predominantly through LG1-3 while alpha-dystroglycan bound to both LG1-3 and LG4-5. Recombinant laminin stimulated acetylcholine receptor (AChR) clustering; however, clustering was induced only by the proteolytic processed form, even in the absence of LG4-5. Furthermore, clustering required alpha(6)beta(1) integrin and alpha-dystroglycan binding activities available on LG1-3, acting in concert with laminin polymerization. The ability of the modified laminins to mediate basement membrane assembly was also evaluated in embryoid bodies where it was found that both LG1-3 and LG4-5, but not processing, were required. In conclusion, there is a division of labor among LG-modules in which (i) LG4-5 is required for basement membrane assembly but not for AChR clustering, and (ii) laminin-induced AChR clustering requires furin cleavage of LG3 as well as alpha-dystroglycan and alpha(6)beta(1) integrin binding.

The laminin ␣2-chain, a subunit of laminin-2 (␣ 2 ␤ 1 ␥ 1 ) and laminin-4 (␣ 2 ␤ 2 ␥ 1 ), is expressed in the basement membranes of skeletal muscle, peripheral nerve, brain, and placenta (1-3). ␣2-Laminins, similar in molecular morphology and functional attributes to ␣1-laminin, are thought to play important roles in basement membrane assembly and the maintenance of the neuromuscular axis (reviewed in Ref. 4). Mutations in the laminin ␣2-subunit are a cause of a human congenital muscular dystrophy typically characterized by early onset, severity, and involvement of peripheral nerve and brain (5,6). Some of these dystrophies have mutations within the part of the gene coding for LG modules (7). The dystrophic phenotype in mouse models of the disease is one of defective basement membranes in muscle and nerve, muscle necrosis with poor regeneration, patchy peripheral nerve dysmyelination, and decreased complexity of infoldings and post-junctional membrane lengths of the NMJ motor endplate (6, 8 -12).
Studies on cultured myotubes have revealed that ␣1and ␣2-laminin polymerization and G-domain contributions drive laminin assembly on the cell surface and direct a redistribution of interacting cytoskeletal components (13). The ability of laminin to induce such cytoskeletal reorganizations is thought to reflect an important receptor-dependent property of laminin during early steps in assembly of a basement membrane. It has also been shown that laminins can induce the clustering of the acetylcholine receptor (AChR), 1 a property shared with agrin (14). However, while agrin is secreted by terminal neurons and triggers this differentiation in a MuSK-dependent and topographically site-specific manner, the activity of laminins is MuSK-independent and topographically diffuse (8, 14 -17). The laminin activity has been found to depend on binding contributions for ␣ 7 ␤ 1 integrin and dystroglycan and appears to synergistically enhance the role of agrin (16,17).
Although the laminin ␣2 chain is closely related to the embryonic-type laminin-␣1 chain, and although both laminins are capable of polymerization mediated through short-arm domains (18), it is different in that the ␣2-chain becomes proteolytically cleaved within G-domain and has a different distribution of dystroglycan-and heparin-binding sites. Laminin's biologically relevant functions are thought to derive from coordinated contributions from different domains of all three subunits (4). To analyze the LG subdomain dependence of these functions in ␣2-laminins, we generated recombinant laminins of the subunit composition ␣ 2 ␤ 1 ␥ 1 bearing three deletions of the LG modules. These laminins were proteolytically processed in LG-3 by a furin-like convertase, and both unprocessed and processed forms were isolated for study. Using these reagents, we investigated their effects on cultured myotubes and laminin ␥1-null embryoid bodies, the latter a model system with which to evaluate basement membrane assembly. We report that laminin-2 binds to cell surfaces, mediates basement membrane assembly, and induces AChR clustering. While all activities depend on LG1-3, AChR clustering requires only LG1-3. Furthermore, furin processing is needed for AChR clustering but not for basement membrane assembly.
Lm-␥1 FLAG -pRc/CMV-The ␥1-chain cDNA (21) in pVL941 was modified to contain a nucleotide sequence encoding the FLAG epitope. Laminin ␥1 cDNA with a 3Ј-terminal FLAG sequence was prepared from the pVL941 plasmid by replacing a SacI-SnaBI DNA fragment containing the 3Ј end of cDNA with a SacI digested PCR product synthesized using primers 5Ј-CAACTGTCCTACTGGCAC-3Ј and 5Ј-C-GGGATCCTACTTGTCATCGTCGTCCTTGTAGTCGGGCTTTTCAAT-GGACGG-3Ј. The second primer contained the 3Ј end of ␥1 cDNA and FLAG sequence preceded the stop codon. PCR was accomplished with Vent polymerase (New England Biolabs). The 5.5 BamHI DNA fragment from the plasmid was excised, blunted, and ligated into the expression vector pRc/CMV (Invitrogen, neo resistance) previously digested with HindIII and blunted.
Lm-␣2 Constructs-A synthetic DNA fragment coding for the HA epitope was inserted following a BM40 signal sequence connected inframe with the mature ␣2 cDNA, the last previously joined from two cDNAs coding for the full protein sequence (2,22). For this purpose a primer 5Ј-CCCGATATCATGAGGGCCTGGATCTTCTTTCTCCTTTGC-CTGGCCGGGAGGGCTCTGGCAGCCCCGCTAGCTTACCCTTACGA-CGTGCCTGACTACGCCCAGCAAAGAGGTTTATTCCCT-3Ј containing an HA-TAG and BM40 (SPARC) signal peptide and 5Ј end of ␣2 cDNA sequences was used together with a second primer 5Ј-CTGTC-CTCCTACTGCTGGG-3Ј to synthesize a PCR product with pCIS-␣2 plasmid DNA. The DNA fragment obtained was cleaved with EcoRV and BssHII and inserted into pCIS-␣2 vector DNA to replace the original 1837-bp EcoRV-BssHII DNA sequence.
Lm-␣2 Deletion Constructs-In the first step, two PCR reactions were carried out with two pairs of primers. One primer (first pair) corresponded to the 5Ј end and another primer (second pair) corresponded to the 3Ј end of the deletion. At the second step, PCR was carried out with a mixture of the products from the first reactions. The following sets of primers were used for the first pair for: (a) deletion of LG1-5: 5Ј-CCTCACTGGTCCGCTGCCT-GC-3Ј (number 1) with 5Ј-CC-CCTTAATTAATCAGTCACCTCCTGAAGACACAGATACTTTG-3Ј and 5Ј-TGAGGGGCGTTCAACCTGTATCATGC-3Ј with 5Ј-CAAGTTAACG-CGGCCGCTC-3Ј (number 2); (b) for deletion of LG1-3: number 1 with 5Ј-GGGCGTAGGAAAGGCTGGGGTGGGAACTGGGTCACCTCCTGA-AGACACAGATACTTTG-3Ј and 5Ј-CCAGTTCCCACCCCAGCCTTTC-CTACGCCC-3Ј with number 2; (c) for deletion of LG4 -5: number 1 with 5Ј-GCATGATACAGGTTGAACGCCCCTCACTCAGGCTGGATAACTA-TTTC-3Ј and 5Љ-TGAGGGGCGTTCAACCTGTATCATGC-3Ј with number 2. For all mutations, primers 1 and 2 were used for the second step PCR. The PCR products were then cleaved with Bsp120I and PacI and inserted into pCIS-␣2 vector to replace the original Bsp120I-PacI fragment.

Transfection and Cloning of Mammalian 293 cells
Transfection of human adenovirus-transformed 293 cells (ATCC CRL 1573) was carried out by calcium phosphate precipitation as previously described (23). Samples from conditioned media and cell lysates were screened for recombinant protein using anti-laminin-2/4 antibody. cDNAs coding for the three chains of recombinant laminin-2 (all of human origin) were transfected into 293 cells in a manner similar to that used to generate recombinant laminin-1 (24) but with different expression vectors for the ␤1 and ␥1 chains (␣2 in vector pCIS, ␤1 in pCEP4-Pu {puromycin} and ␥1 in pRC/CMV {G418}) with the FLAG tag DNA (protein sequence DYKDDDDK) inserted 3Ј to, and in-frame with, ␥1. The first step was to transfect and select for stable clones of 293 cells expressing the ␥1 chain. The second step was to co-transfect these cells with the ␣2 and ␤1 vector constructs and to select for cells stably expressing all three chains under puromycin and G418 selection. Laminin ␥1 and ␤1 chain expression was confirmed in immunoblots with laminin-2/4 and E4 specific antibodies, the latter recognizing epitopes of the N-terminal moiety of the ␤1 chain. The ␣2 chain was identified using affinity purified antibodies to whole laminin-2 and laminin-␣2G domain (18,25). Human ␥1 chain expression was found to be much greater compared with mouse-␥1 (24), a fraction of the protein appearing in medium. Cells were maintained in either 1 g/ml puromycin and/or 500 g/ml G418 following transfection with resistant colonies were isolated and expanded.

Purification of Recombinant Laminins
Cells stably expressing recombinant laminin were grown to confluency in tissue culture flasks (150 cm 2 ) maintained with G418 and puromycin. Conditioned media was collected and centrifuged to remove debris, adjusted to 1.5 mM phenylmethylsulfonyl fluoride and 0.1% Tween 20, and incubated with FLAG-specific monoclonal M2 antibodyagarose beads (Sigma, 1-ml beads/100 ml medium) at 4°C overnight. Beads were collected by centrifugation, washed, and protein was eluted with a 0.1 mg/ml solution of FLAG peptide (Sigma/2.5 ml). The eluates were purified by HPLC affinity chromatography using a heparin-5PW (0.8 ϫ 7.5 cm; Toso-Haas) equilibrated in 50 mM Tris-HCl, pH 7.4, 1 mM EDTA. The column was eluted with a shallow NaCl gradient and fractions from the separated peaks were collected.

Biochemical Assays and Rotary Shadowing
Dystroglycan Binding Assay-96-Well polystyrene plates were incubated overnight at 4°C with 0.5 g/well of either purified laminins or bovine serum albumin (Sigma) diluted in 50 mM Tris-HCl, pH 7.4, 150 mM NaCl, and 1 mM EDTA. Laminin-coated wells were blocked for 2 h at room temperature with 1% bovine serum albumin in 50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1 mM CaCl 2 , 1 mM MgCl 2 followed by a 2-h incubation with varying concentrations of iodinated ␣-dystroglycan diluted in blocking solution. Wells were washed with blocking solution, removed, and bound ␣-dystroglycan radioactivity was measured. Nonspecific binding of labeled ␣-dystroglycan to bovine serum albumincoated wells was subtracted (28).
Co-polymerization-This assay, used to detect laminin-specific selfassembly with small amounts of test laminin, depends on the ability of the two proteins to co-aggregate in a concentration-dependent manner with polymerization driven by the EHS-laminin (18). Briefly, aliquots of EHS-laminin in polymerization buffer were incubated with a fixed concentration of recombinant laminin at 37°C. Samples were centrifuged and the supernatant and pelleted fractions analyzed by SDS-PAGE under reducing conditions. SDS-PAGE was carried out in 3.5-12% linear gradient gels (26) and stained with Coomassie Blue R-250 or transferred to nitrocellulose membranes for immunoblotting. Immunoblotting of proteins was performed as described (26).
Rotary Shadowing-Laminin (25-50 g/ml in 0.15 M ammonium bicarbonate, 60% glycerol) was sprayed onto mica discs, evacuated in a Balzers BAF500K unit, rotary shadowed with 0.9 nm Pt/C at an 8°a ngle, backed with 8 nm carbon at a 90°angle, and viewed in an electron microscope as otherwise described (26). Electron micrograph images are shown contrast-reversed.

Cell Culturing and Analysis
Mouse C2C12 and rat L8E63 myoblasts were cultured in medium containing 10% fetal calf serum. The former were converted to myotubes in 5% horse serum, and analyzed 3-5 days post-fusion as described (13). For quantitation of myoblast adhesion, cells labeled with [ 3 H]thymidine were incubated (70 m) in tissue culture wells (3 ϫ 10 4 cells/well) pre-treated with nitrocellulose and coated with laminins as described (29) followed by determination of bound radioactivity. Fused myotubes were incubated at 37°C with laminins (10 g/ml) suspended in Dulbecco's modified Eagle's medium F-12 containing 0.5% bovine serum albumin. Unattached protein was removed by washing with PBS containing 1 mM CaCl 2 . Blocking antibodies and fragments in 100-fold molar excess, when used, were added to the culture medium prior to addition of laminin at the following final concentrations: IIH6 (1/10 dilution), Ha2/5 and GoH3 (10 g/ml), and E1Ј (500 g/ml). Wild type R1 and laminin-␥1 null ES cells (30) were grown on feeder layers of mitomycin-treated (100 g/ml, 2 h) SNL STO cells in ES medium (Invitrogen) supplemented with 15% ES-grade fetal calf serum (Invitrogen), 0.1 mM nonessential amino acids, 0.1 mM ␤-mercaptoethanol, 1 mM sodium pyruvate, 100 g of penicillin/ml, 100 g/ml streptomycin, and 1000 units/ml leukemia inhibitory factor (LIF, Invitrogen). To culture EBs, subconfluent ES cells were dispersed with 0.25% trypsin, 0.53 mM EDTA and plated onto gelatin-coated dishes (3 h) to allow feeder cells to attach. Nonadherent ES aggregates were dispersed and cultured on bacteriological Petri dishes in ES medium without LIF.

Microscopy
Following incubation, myotubes were fixed with 3% paraformaldehyde in PBS for 10 min at room temperature, blocked in PBS containing 0.5% bovine serum albumin and 5% normal goat serum, and incubated with primary antibodies diluted in the same buffer. Following several washes, the cells were incubated with FITC-or Cy3-conjugated secondary antibodies for 1 h at room temperature, washed, and overlaid with DABCO, mounted, and examined with an Olympus IX-70 inverted fluorescent microscope fitted with a Princeton Instruments 5 mHz Micromax cooled CCD camera. Myotube surface coverage by laminins was determined from digitalized images with IPLab version 3.0 (Scanalytics) using segmentation parameters adjusted to detect laminin-covered areas from background. The automatically marked areas were quantitated and the average area and S.E. were determined (n ϭ 6 -10 fields). Rat monoclonal anti-laminin ␥1 (clone A5; Upstate Biotechnology, Lake Placid, NY), rabbit anti-mouse type IV collagen antibody (Rockland Immunochemicals, Gilgertville, PA), and rat anti-mouse perlecan monoclonal antibody (Chemicon, Temecula, CA) were used for immunostaining of EBs at 1, 2.5, 2 and 2.5 g/ml, respectively.
Acetylcholine Receptor Clustering-C2C12 myotubes were incubated for 18 h in the presence of either different concentrations or 10 g/ml recombinant or other laminins, without or with blocking reagents. Myotubes were then incubated for 1 h with 5 g/ml FITC-␣-bungarotoxin (Molecular Probes) to label clusters of AChR's, washed 4 times with PBS, 1 mM CaCl 2 to remove unattached protein, and fixed. Myotubes were incubated with antibody IIH6 (1/10 dilution from conditioned medium), washed, and then treated with goat anti-mouse Cy3 (1:100). Fluorescent micrographs of fixed cells were analyzed in IPLab to determine the number of clusters per field. Only AChR clusters that co-localized with dystroglycan were scored.
Electron Microscopy-Electron microscopy of thin sections was carried as described (31).

Expression of Recombinant Laminin Subunits in 293
Cells-We followed a two-step transfection protocol (24) to express recombinant heterotrimeric laminin. 293 cells stably expressing laminin ␥1 and selected with the antibiotic G418 were transfected with the ␤1-expression vector and ␣2-expression vectors coding for wild-type protein or protein bearing deletions within the LG modules (Fig. 1). Clones expressing protein without epitope tag were generated in parallel with clones expressing ␥1 bearing a C-terminal FLAG tag. The latter proved far more successful for the preparation of heterotrimeric laminins and furthermore, permitted purification from serum-containing medium. For the ␣2 deletion mutants and a wild-type control, a hemagglutinin (HA) tag was placed Nterminal to the mature sequence. Wild-type (WT) recombinant protein and all protein bearing LG module deletions (⌬G1-5, ⌬G1-3, and ⌬G4 -5) were secreted into the culture medium.
Heterotrimeric laminin expression was initially detected by the ability of ␣2 chain and ␤1 chain-specific antibodies to react with the respective laminin subunits first immunoprecipitated with FLAG antibody (data not shown). The ␣2 chain bands reacted with both antibodies specific for laminin-2/4 and laminin-␣2-G domain and the ␤1 chain was identified under nonreducing conditions (necessary for antibody reactivity) with E4-specific antibody.
Heparin Interactions and Identification of Unprocessed and Processed ␣2 Chains-For purification (Fig. 2), recombinant laminins in conditioned medium were initially bound to anti-FLAG antibody beads and eluted with FLAG peptide, facilitating subsequent steps by eliminating contaminating serum and secreted proteins. The eluted fraction was bound through its ␣-subunit to a heparin-5PW column and eluted with a salt gradient (Fig. 2a). Two wild-type and rLm2-⌬G4 -5 protein peaks each were eluted at 0.16 and 0.33 M NaCl, with relative peak sizes varying among different media collections. The different fractions were analyzed by reducing SDS-PAGE (Fig. 2,  b-d). The more weakly bound wild-type laminin fraction was found to consist of laminin-2 similar to that obtained from placenta and consisting of a ϳ275-kDa (processed, "p") ␣2 band (previously designated ␣2m in Ref. 18) and ϳ75-kDa fragment band (known as a C-terminal moiety) with the ␤1 and ␥1 chains (Fig. 2b), while the more strongly bound fraction consisted of a larger ␣2-subunit (␣2, unprocessed "u") with no observed additional ␣-chain component. It has previously been shown in recombinantly expressed mouse laminin ␣2-LG1-3 fragment that two closely overlapping furin-type recognition sites exist in the LG3 module with cleavage between Arg-2575 and Gln-2576 generating a blocked N terminus (32,33). We therefore predicted that deletion of LG modules 4 and 5 should leave a smaller cleavable C-terminal fragment of ϳ19 kDa plus carbohydrate. A fragment of ϳ22 kDa was identified in the low salt, but not the high salt, fractions of rLm2⌬G4 -5 (Fig. 2c), in good agreement with the fragment generated from ␣2-LG1-3 by Talts et al. (33) and paralleling the salt-elution behavior seen with wild type recombinant protein. The main ␣2 band of this high salt fraction appeared to migrate marginally slower than LG modules that comprise the ␣2-G domain project beyond the coiledcoil domain.
LG4 is connected to LG3 through a spacer sequence (hinge region) that is disulfide bonded to the LG5 terminus, positioning LG5 next to LG1-3. Laminin-2 extracted from several sources bears proteolytic cleavage in the LG modules with the distal 75-80-kDa fragment noncovalently attached to the N-terminal portion of the ␣2 chain. ␣2-Subunit deletions were ⌬G1-5, ⌬G1-3, and ⌬G4 -5 as shown. that of the low salt fraction (predicted mass difference is ϳ7%). When the FLAG-containing fractions of rLm2-⌬G1-3 and rLm2-⌬G1-5 conditioned media were subjected to heparinaffinity chromatography, each yielded a single low salt-eluting peak detected by reducing SDS-PAGE (Fig. 2, a and d). When N-terminal HA-tagged wild-type protein was evaluated (Fig.  2e), both the more slowly and faster migrating ␣2 bands were found to possess HA epitope by immunoprecipitation, further supporting the conclusion that the high salt fraction corresponded to laminin-2 bearing an unprocessed ␣2 chain while the low salt fraction corresponded to laminin-2 bearing the typical-sized G-domain processed ␣2 domain. Finally, the ϳ75-kDa species reacted with ␣2-G-specific antibody (blot not shown). Thus heparin affinity chromatography resolved the different laminin species and revealed that high-affinity heparin binding is lost upon LG3 cleavage.
Molecular Shape of Recombinant Laminins-The different recombinant fractions were analyzed by electron microscopy following rotary Pt/C replication of glycerol spreads on mica (Fig. 3). Wild-type processed recombinant laminin-2 had a typical laminin shape consisting of three short arms with small terminal and mid-arm globules, and one long arm ending in the large elliptical structure corresponding to the LG modules (arrows, upper left panel). The long axis of the LG modules varied with respect to that of the lower portion of the long arm coiled coil domain. The most striking difference in the deletion mutants was observed with rLm2⌬G1-5 (upper middle panel and extreme right lower two panels) which completely lacked the globular domain of the long arm. The long arm globules of rLm2-⌬G1-3 and rLm2-⌬G4 -5 were similar in appearance, characterized by a more spherical shape, on average smaller than the elliptical globule of wild-type protein. Processed and unprocessed forms of wild-type laminin-2 could not be distinguished from each other in the electron micrographs.
Furin Processing-While both unprocessed and processed ␣2-subunit forms were secreted by WT-transfected 293 cells, only unprocessed laminin subunit was detected in the cell lysates, either in Western blots of transiently transfected cells (Fig. 4a) or FLAG-epitope protein from stably transfected cells (data not shown). When stably transfected cells expressing rLm2-WT protein were treated with different concentrations of the furin protease-inhibitor d-RVKR-cmk, proteolytic cleavage of the laminin was prevented (Fig. 4b). When rLm2-WTp was added to the medium of nontransfected 293 cells, C2C12 myotubes, and rat DRG Schwann cells, laminin in the conditioned media (collected after 2 days) was partially processed from the 293 and Schwann cell cultures unless treated with d-RVKRcmk, but minimally if at all from the C2C12 myotube cultures (Fig. 4c). Thus cultured 293 and Schwann cells can process extracellular laminin in a furin-dependent fashion while myotubes do not to any appreciable degree.
Cell Adhesion, Dystroglycan Binding, and Polymerization Activities-Mouse myoblasts adhered to and spread on both processed and unprocessed wild-type protein with 50% maximal coat concentrations of about twice that of laminin-1 (Fig.  5). Cells also adhered to and spread on both processed and unprocessed recombinant laminin bearing deletions of LG4 -5 with only a small reduction of coat concentration dependence. However, cells adhered poorly to, and did not spread on, recombinant laminin bearing deletion of LG1-3. Deletion of all LG modules abolished cell adhesion over the concentration range studied, indicating that G-domain is required for myoblast cell adhesion. Adhesion contributions were detected from ␣ 6 ␤ 1 and ␣ 7 ␤ 1 integrins for the myoblasts. Furthermore, this interaction was largely dependent upon LG1-3 but not LG4 -5 and was independent of LG processing.
A solid-phase assay was used to evaluate ␣-dystroglycan binding (Fig. 6). Radioiodinated rabbit skeletal muscle ␣-dystroglycan was incubated at different concentrations in wells coated with equimolar coatings of different recombinant laminins. Dystroglycan bound to wild-type rLm2, rLm2⌬G4 -5, and rLm2-⌬G1-3, but not to rLm2-⌬G1-5, in a concentration-dependent manner. Binding to all active laminin-2 proteins was salt-sensitive but heparin-insensitive. The data were fitted by nonlinear regression analysis with single-and two-class binding algorithms, and the dystroglycan concentrations at halfmaximal binding (indicated in parentheses) were determined by inspection from the fitted curves. rLm2-⌬G4 -5p (18 nM) and rLm2-⌬G1-3 (15 nM) could be well fit by the single-class algorithm while rLm2-WTp (average 43 nM), WTu (38 nM), and ⌬G4 -5u (62 nM) were better fit by the two-class algorithm. The levels of binding to wild-type protein at a given concentration was greater than twice that observed for either rLm2-⌬G4 -5p or rLm2-⌬G1-3 alone, suggesting that at least two dystroglycan-binding sites are present in the full complement of ␣2-LG modules. Dystroglycan bound to rLm2-WTu and rLm2-⌬G4 -5u, both unprocessed proteins, as well. The rLm2-WTu binding plot was similar to that of rLm2-WTp while the rLm2-⌬G4-5u binding was at least 2-fold higher compared with that of rLm2-⌬G45p, suggesting that there may be two sites within LG1-3 and that cleavage of G-domain decreases dystroglycan binding. However, rLm2-WTu was not observed to achieve higher binding levels compared with rLm2-⌬G45u as would be expected from the summation of rLm2-⌬G1-3 and rLm2-⌬G4 -5u contributions. Both the apparent inaccessibility of more than three sites and two-class binding behavior may be due to steric hindrance occurring at adjacent G-domain-binding sites occurring at high dystroglycan concentrations. Since dystroglycan binding was not reduced in the unprocessed state, and since alanine mutagenesis of the furin-recognition site caused a substantial decrease of dystroglycan affinity for recombinant ␣2LG1-3 (32), its seems likely that the furin-recognition site overlaps with the cleavage site.
Laminin polymerization is a property of full-sized laminins including laminins-1, -2, and -4 that is dependent upon interactions among the short arms (18,26). A laminin co-polymerization assay, developed to measure the polymerization potential of test laminins in small amounts, was employed to evaluate the recombinant proteins (18). Wild-type processed protein was found to co-precipitate with laminin-1 in a concentration-dependent manner (Fig. 7). The fraction of polymeric recombinant laminin was lower than (lagged behind) that of laminin-1 as observed previously for placental laminin-2 (18). The co-polymerization of rLm2⌬G1-5, which possessed an N-terminal ␣2 HA epitope tag was compared with that of wild-type protein lacking an HA tag and was found to possess similar activity, i.e. neither the HA tag nor loss of G interfered with polymerization.
Laminin-2 Binding to Myotube Surfaces and Co-localization with Dystroglycan and Integrin-Recombinant wild-type and deleted ␣2-laminins were added to the culture medium of fused C2C12 myotubes for 2 or 4 h and then evaluated for the distributions of laminin, ␣-dystroglycan, and ␤ 1 -integrin on the myotube surfaces by indirect immunofluorescence microscopy (Fig.  8). By 2 h, rLm2-WTp,u and ⌬G4 -5p,u had decorated the myotube surfaces in a patchy honeycombed pattern (Fig. 4, a-d) with maximal coverage achieved by 2-4 h (quantitation not shown). The myotubes were co-stained for laminin/␣-dystroglycan (Fig. 8, a and b) and laminin/␤ 1 -integrin (c and d). By 2 h incubation, rLm2-WTp and WTu were noted to co-localize with ␣-dystroglycan (aЈ and bЈ) and laminin co-localization with ␤ 1 -integrin was detected at 4 h following treatment with rLm2-WTp and WTu (Fig. 4, cЈ and dЈ). Dystroglycan and integrin co-localization was also observed with ⌬G45p at 4 h (images not shown); however, coverage was reduced. The relative amounts of coverage by the different laminins were determined at 2 h (Fig. 8e). Wild-type protein covered a much greater area compared with laminin deleted in LG4 -5. Furthermore, coverage was only slightly above background for laminin deleted in LG1-3 or LG1-5. We evaluated the ability of blocking antibodies, and heparin to perturb rLm2-WTp and WTu accumulation on myotube surfaces (Fig. 8f). Inhibition of ␣-dystroglycan (IIH6) and heparin-heparan sulfate interactions each separately caused a modest decrease in the surface area coverage with inhibition of all three activities preventing most binding. We also evaluated ⌬G4 -5 (Fig. 8g) and found that with the processed form of this deletion mutant, nearly complete inhibition was achieved following treatment with IIH6, Ha2/5, or heparin alone under conditions that produced only modest reductions with wild-type protein. In contrast, significantly less inhibition was observed following treatment of the unprocessed deletion mutant.
Contribution of Laminin ␣2-LG Modules to AChR Clustering-When recombinant laminin-2 was incubated with myotubes for longer periods, a substantial (5-10-fold) increase in AChR clusters (detected with bungarotoxin) was observed over that detected in the absence of exogenous laminin (Fig. 9). These clusters co-localized with ␣-dystroglycan that was concentrated at these sites. Maximal induction was achieved by about 10 g/ml (Fig. 9h). Surface laminin co-localized with the bungarotoxin as well but was also present in areas away from the clusters (Fig. 9aЉ, inset). Of all the recombinant laminins tested, only rLm2-WTp and ⌬G4 -5p induced AChR clustering above background levels ( Fig. 9, b-g and i). Both antibodies IIH6 and Ha2/5 blocked AChR clustering induced by either of the laminins (Fig. 9, k and n). Since GoH3, an ␣ 6 -integrinspecific blocking reagent, also inhibited clustering of these two proteins (l and n), ␣ 6 ␤ 1 was implicated in mediating clustering. Heparin (1 mg/ml) inhibited clustering induced by rLm2-⌬G4 -5p (n), but not that induced by WTp (j). This observations seems likely to be related to the ability of heparin to nearly completely inhibit 2-4 h accumulation of rLm2-⌬G4 -5p, but only to partially inhibit accumulation of WTp. In comparison, laminin-1-induced AChR clustering was substantially inhibited with heparin as well as anti-␣-dystroglycan (m). Laminin polymerization, a process mediated by the three short arms, was previously found to be selectively inactivated following treatment with AEBSF (13). To determine whether polymerization contributes to laminin-induced AChR clustering, myotubes were incubated overnight with 10 g/ml recombinant laminin-2 (WTp) either alone or in the presence of laminin-1 fragment E1Ј which inhibits polymerization unless treated with AEBSF (o). rLm2-WTp induced AChR clustering was inhibited by 1 M (0.5 mg/ml) E1Ј, but not by AEBSF-E1Ј.
Laminin Mediation of Basement Membrane Assembly-The greatly increased coverage of intact laminin on myotubes compared with all deletion mutants suggested that deletion of LG modules would have an adverse effect on the efficiency of basement membrane assembly. The myotubes produce little or no type IV collagen or other basement membrane components and we find they do not form stable basement membranes. Therefore, to examine the role of the LG modules in basement membrane assembly, we evaluated the ability of the recombinant laminins to mediate basement membrane assembly in embryoid bodies (EBs) derived from ES cells that are null for the laminin-␥1 subunit. These EBs, when treated with exogenous laminin-1, form basement membranes (34) that contain type IV collagen, nidogen, and perlecan in addition to the added laminin. rLm2-WTp, WTu (in the presence of 10 M furin inhibitor), ⌬G4 -5p, ⌬G1-3, and ⌬G1-5 laminins were incubated (50 g/ ml) with laminin-␥1-null EBs and examined by immunofluorescence microscopy (Fig. 10). Wild-type laminin-2, regardless of processing (immunoblots of a cell lysate at the end of the incubation revealed that essentially all of the ␣2 chain was still  WTu (b and d), and laminins bearing deletions were incubated (10 g/ml) with C2C12 myotubes, washed, and detected with antibody for laminin-2/4 (a-d) and either ␣-dystroglycan (IIH6, aЈ and bЈ, after 2 h incubation) or ␤ 1 integrin (Ha2/5, cЈ and dЈ, after 4 h incubation). Relative laminin surface areas measured from the immunofluorescence images are shown in panel e (n ϭ 6). rLm2-WTp, rLm2-WTu, rLm2-⌬G45p, and rLm2-⌬G45u (2 h) were also treated with ␤ 1 integrin (Ha2/5) and ␣-dystroglycan (IIH6) blocking antibodies, and heparin, as shown in panel f (n ϭ 10). The inhibitory effects on laminin surface coverage following incubation with rLm2-⌬G45p,u, the minimally active recombinant laminin, are also shown (g, n ϭ 7). in the unprocessed state, Fig. 10m), mediated formation of a subendodermal linear co-staining for laminin, type IV collagen, nidogen, and perlecan. In contrast, all deletion mutants were unable to mediate basement membrane formation. The formation of basement membrane following treatment with rLm2-WTp was confirmed by electron microscopy.

DISCUSSION
The C-terminal G-domain, composed of five LG modules, appears to be the principal cell-interactive domain of laminin. The structure deduced from the ␣2-LG4 -5 crystal is that each LG module is a ␤-sandwich with the first three modules separated from the last two by a hinge disulfide-linked to LG5 (35). Although the ␣2-laminin subunit has fairly high homology with the ␣1-subunit, it exhibits several G-domain differences. First, the ␣2-subunit becomes proteolytically processed in LG-3 by a furin-like convertase that cleaves at the end of the sequence RRKRR in laminin-2 (32). Second, ␣2-laminin G-domain possesses dystroglycan sites in both LG-3 and LG4 -5, compared with a limitation of binding to LG module-4 in laminin-1. Each dystroglycan-binding site within ␣2 LG1-5 is insensitive to heparin treatment (unlike laminin-1), accounting for the nearcomplete lack of inhibition of dystroglycan binding reported for intact laminin-2/4 (28).
Previously we found the short and long arm domains of laminin-1, through polymerization and cell binding, act in con-cert to assemble ECM and alter cytoskeleton (13). To dissect LG-modular functions in a laminin isoform yet preserve the potential for such cooperativity, we introduced LG1-5, LG1-3, and LG4 -5 deletions into the laminin ␣2 subunit DNA and expressed the respective heterotrimeric laminins. 293 cells, by virtue of a furin-like convertase, were found to cleave much, but not all, of the wild-type and LG4 -5 deleted laminin within LG3, providing us with two additional laminin states for study. Through our analysis we found that myoblast adhesion and spreading are mediated by LG1-3 through ␣ 6 ␤ 1 and ␣ 7 ␤ 1 integrins (similar to laminin-1), that ␣-dystroglycan binds separately to LG1-3 and LG4 -5 (unlike laminin-1), and confirmed that polymerization is independent of G-contributions. In our evaluation of myotubes and embryoid bodies, we found that: (a) LG1-3 is required for basement membrane and for clustering of AChR, (b) LG4 -5 is required for basement membrane assembly but not for AChR clustering, (c) furin-processing of LG-3 is required for AChR clustering, but not for basement membrane assembly, (d) ␣ 6 ␤ 1 integrin as well as dystroglycan are required for AChR clustering, and (e) polymerization of laminin is required for AChR clustering (providing the evidence for a hypothesis as described in Ref. 36). The laminin contribution to AChR clustering can be regulated by furin in ␣2-laminins but is constitutively active in ␣1-laminin.
It is conceptually useful to distinguish anchorage, interac- tions required to bind and retain otherwise free laminin on a cell surface, from receptor activities of laminin, required to mediate cellular responses. Using surface area coverage as an index of binding to compare contributions on myotube surfaces, we found that coverage was greatest when laminin possessed all of its LG modules regardless of processing, and was almost completely absent if all LG modules were deleted. Deletion of LG4 -5 caused a substantial decrease of myotube surface accumulation yet did not interfere with AChR clustering. This loss of coverage correlated with the ability of the laminin to assemble basement membrane in embryoid bodies lacking endogenous ␥1-subunit expression. Deletion of LG1-3 caused near total loss of myotube accumulation and here too the laminin was unable to assemble basement membrane. For wild-type laminin-2, inhibition of ␣-dystroglycan, ␤ 1 -integrin, and heparan sulfate/sulfatide binding each caused a modest decrease of coverage, while inhibition of all three caused a substantial decrease, suggesting that all three types of interactions con-tribute to and account for most of the anchorage. In contrast, for laminin lacking LG4 -5 (processed form), surface binding could be abolished by separate inhibition of dystroglycan, integrin, or heparin class binding. Since dystroglycan-␣2-laminin interactions are insensitive to heparin treatment, the heparin effect seen with the deleted laminin is not due to inactivation of dystroglycan binding. Taken together, these data are consistent with a model in which anchorage of laminin to the cell surface is mediated by ␤1-integrin-, ␣-dystroglycan-, and heparin-type binding to LG1-3, and a combination of ␣-dystroglycan-and heparin-type binding to LG4 -5. Since unprocessed rLm2⌬G4 -5 resisted inhibition by single blocking agents, the sum of affinity contributions is likely greater in LG1-3 if processing does not occur, contributions obscured when LG4 -5 are also present in the laminin.
Laminin activity has been proposed to play a synergistic role with agrin in NMJ formation, affecting AChR clustering through dystroglycan and ␣ 7 -integrin (16,27,(37)(38)(39). From this study, we found that AChR clustering requires both laminin polymerization (a short-arm activity) as well as LG1-3 module interactions, the latter minimally dependent upon ␣-dystroglycan and ␣ 6 ␤ 1 -integrin binding in the presence of anchorage partially provided through heparin-type interactions. The previously identified ␣ 7 ␤ 1 -integrin contribution (27) and complete inhibition of clustering achieved with GoH3 suggests that ␣ 6 ␤ 1 works in concert with ␣ 7 ␤ 1 to cluster AChR.
To compare the basement membrane requirements to those of AChR induction, we employed the recently described laminin ␥1-null embryoid bodies that generate all required basement membrane components except laminin and that when treated with laminin assemble a basement membrane and then differentiate to form epiblast (34,40,41). The necessity of both LG1-3 and LG4 -5 for basement membrane assembly revealed important G-domain contributions emanating from both sides of the hinge region. However, unlike AChR clustering, several studies reveal that ␤ 1 -integrin and dystroglycan are not, at least separately, essential for basement membrane assembly. First, Schwann cell basement membranes, which contain laminin-2, do not require ␤ 1 -integrin (42). Second, basement membrane is detected in dystroglycan-free muscle of genetic chimeric mice (43,44). Third, we have found that basement membranes can assemble in dystroglycan-null and polymerizing laminin-treated ␤ 1 integrin-null embryoid bodies. 2 A conclusion suggested by the above and data described herein is that G-domain anchorage with high surface accumulation of laminin, rather than ligation of a particular integrin or dystroglycan, is critical for basement membrane assembly, while specific integrin and dystroglycan receptor ligation, rather than high accumulation, is essential for AChR clustering. This characteristic, along with the presence of an appropriate integrin-binding site and dystroglycan-binding site in ␣2-laminin LG1-3 (the latter is missing in ␣1-laminin LG1-3) can explain why LG4 -5 is superfluous for clustering. A general hypothesis is that the relative LG-module contributions for different basement membrane functions differ among laminin isoforms.
All laminins with the exception of ␣1-laminins become proteolytically cleaved, often in G-domain (see Ref. 4). These modifications may have evolved to allow for post-depositional regulation of laminin function during development and tissue remodeling. Unprocessed laminin-2 can support cell adhesion/ spreading, bind to dystroglycan, accumulate on myotube surfaces, and form basement membrane, yet strikingly cannot cluster the AChR. The only binding interaction we found to be 2 S. Li, D. Harrison, S. Carbonetto, N. Smyth, D. Edgar, R. Fä ssler, and P. D. Yurchenco, manuscript in preparation.

FIG. 10. Basement membrane assembly in embryoid bodies.
Laminin ␥1-null ES cells were cultured for 7 days in suspension as embryoid bodies with 50 g/ml rLm2-WTp (a-f), ⌬G45p (g and h), ⌬G13 (i and j), or rLm2-WTu with d-RVKR-cmk (k and l). Immunoblot (Lm-1/4 antibody) of EB conditioned media from WTu-treated EBs (7 days) without (lane 1) and with (lane 2) d-RVKR-cmk. Cryosections were prepared and immunostained for laminin-␥1 (FITC, a, g, i, and k), type IV collagen (Cy3, b, h, j, and l), perlecan (FITC, d), or nidogen (Cy3, e). Untreated (n) and rLm2-WTp-treated (o) embryoid bodies were also fixed with gluteraldehyde and embedded in plastic for electron microscopy (endodermal/inner cell mass interface shown; paired arrows indicate no basement membrane in untreated and thick basement membrane in rLm2-WTp-treated EBs; n ϭ nucleus). markedly different in unprocessed laminin is increased heparin binding affinity. However, it is not known if this difference is causal in preventing AChR clustering activity in the unprocessed state. Furin, a serine protease, is a membranebound enzyme that is found in the trans-Golgi stacks and plasma membrane and that can cleave a number of ECM components (45). In an examination of its distribution in muscle, we have detected furin in relevant locations, i.e. at the neuromuscular junction, and, to a lesser extent, in embryonic muscle. 3 We have found that extracellular laminin is processed by Schwann and 293 cells but much less efficiently by EBs and (particularly) cultured myotubes, revealing that furin can cleave laminin-2 following its secretion and that this activity can vary. The existence of laminin processing in G-domain argues for an important functional dependence of a number of isoforms on controlled proteolysis through the LG modules and leads to the attractive possibility that furin regulates laminin-2 activity.