Conjugation of LG Domains of Agrins and Perlecan to Polymerizing Laminin-2 Promotes Acetylcholine Receptor Clustering*

Neuromuscular junction (NMJ) assembly is characterized by the clustering and neuronal alignment of acetylcholine receptors (AChRs). In this study we have addressed post-synaptic contributions to assembly that may arise from the NMJ basement membrane with cultured myotubes. We show that the cell surface-binding LG domains of non-neural (muscle) agrin and perlecan promote AChR clustering in the presence of laminin-2. This type of AChR clustering occurs with a several hour lag, requires muscle-specific kinase (MuSK), and is accompanied by tyrosine phosphorylation of MuSK and βAChR. It also requires conjugation of the agrin or perlecan to laminin together with laminin polymerization. Furthermore, AChR clustering can be mimicked with antibody binding to non-neural agrin, supporting a mechanism of ligand aggregation. Neural agrin, in addition to its unique ability to cluster AChRs through its B/z sequence insert, also exhibits laminin-dependent AChR clustering, the latter enhancing and stabilizing its activity. Finally, we show that type IV collagen, which lacks clustering activity on its own, stabilizes laminin-dependent AChR clusters. These findings provide evidence for cooperative and partially redundant MuSK-dependent functions of basement membrane in AChR assembly that can enhance neural agrin activity yet operate in its absence. Such interactions may contribute to the assembly of aneural AChR clusters that precede neural agrin release as well as affect later NMJ development.

Laminin G-like (LG) 2 domains are ␤-sandwich globular structures present in the C-terminal moieties of basement membrane laminins, agrins, and perlecan (1,2). They serve as ligands for ␣-dystroglycan and integrins (3,4) and in the case of laminins can be anchored to the cell surface through sulfated glycosphingolipids (5). Proteins bearing these domains are found in the basement membrane of the motoneuron endplate of skeletal muscle. With the exception of neural agrin that is released by pre-synaptic neuronal processes, they are substantially secreted by the skeletal muscle and deposited on the sarcolemmal surface.
The basement membrane of the neuromuscular junction (NMJ) is enriched in laminins containing the ␣2 subunit, with predominance of ␣2␤2␥1 in the mature endplate (3). Laminins are critical for formation of basement membrane, self-assembling into cell-adherent polymers that serve as scaffolds for the tethering of agrins, nidogens, and perlecan, and are connected to a type IV collagen network (reviewed in Ref. 4). Laminins have also been found to play a role in NMJ assembly, with different subunits contributing to their size, pre-and post-synaptic alignment, infoldings, and degree of ensheathment (4). The ␣1and ␣2-laminins have been found to induce AChR clustering in vitro (6 -9), an effect resulting in tyrosine phosphorylation of AChR subunits (6,9,10). Loss of the ␣2-laminin subunit is associated with reduced depths and complexity of the post-synaptic infoldings of the NMJ accompanied by defects of myelination and muscular dystrophy (3).
Agrins are heparan sulfate proteoglycans coded by a single gene and expressed as several variant forms because of alternative RNA splicing (11)(12)(13). Agrins that possess an N-terminal NtA domain ("LN-agrins") bind with high affinity to the coiled-coil domain of ␥1-laminins (12, 14 -16). Alternative splicing of agrin at two loci within the laminin-type LG domain coding sequence gives rise to isoforms with different properties (11). The LN-agrins that possess a 4-amino acid residue insert within the chick A site (mammalian y site) confer heparin binding activity and a sequence of 8 -19 residues within the chick B site (mammalian z site) affecting MuSK-dependent AChR clustering activity. The isoforms with the insertions are selectively released by the motor nerve termini to induce the coalescence and alignment of large AChR postsynaptic aggregates during NMJ assembly (11,17). The agrin variant lacking these insertions (A0B0/y0z0, muscle agrin), whose functions are poorly understood, is widely distributed in the basement membranes of non-neuronal tissues including the muscle sarcolemma (18).
Perlecan is another basement membrane heparan sulfate proteoglycan found in the NMJ whose C-terminal domain V contains three LG domains that interacts with ␣2␤1 integrin and dystroglycan (19,20). The perlecan IG3 repeat in domain IV is the principal locus that binds to the G2 domain of nidogen-1. Nidogen-1 in turn binds to the laminin ␥1-subunit, providing a bridge between the two proteins (21,22). Perlecan mutations result in chondrodysplasia and myotonia, the latter because of its role in binding to acetylcholinesterase in the NMJ (23).
In this study we address the role of basement membrane components in AChR receptor assembly, examining relationships among neural agrin, non-neural agrin, perlecan, laminin-2 ("laminin-211"; Ref. 24), and type IV collagen. For this study we have used "minigene" versions of non-neural (nNA) and neural agrin (NA) that consist of the N-terminal (NtA) laminin-binding domain and the C-terminal LG domains that have been shown to possess biological activity (25,26) and that were chosen to simplify the analysis of domain contributions. We have also evaluated full-length perlecan and a protein consisting of the NtA domain fused to the LG domains of perlecan (NP). We report that the minigene version of non-neural agrin and that full-length and fusion versions of perlecan induce AChR clustering when bound to laminin-2. Clustering induced by these proteins is MuSK-dependent and requires laminin polymerization. Finally, type IV collagen stabilizes those clusters requiring laminin conjugation. Collectively, the findings reveal new roles of basement membrane in the promotion and stabilization of AChR clusters that may be relevant for understanding neuromuscular development.

DNA Constructs and Recombinant Proteins
Chick Minigene Neural-Chick minigene neural (containing the A4B8 insertions) agrin (pCepN25C95 4,8 His) and non-neural (A0B0) (pCepN25C95 0,0 His) agrin plasmids were used to express protein with an N-terminal sequence containing a His 6 tag, NtA domain, and the first follistatin repeat fused to the C-terminal neural or non-neural three G-repeats (11,26).
Perlecan and NtA-Perlecan-Domain V Fusion Protein-cDNA encoding mouse perlecan domain V (i.e. the three LG domains with intervening epidermal growth factor-like repeats) was fused to the C-terminal end of NtA-FS (16) by PCR using a full-length cDNA encoding mouse perlecan (gift of Dr. T. Sasaki, MPI for Biochemistry, Martinsried, Germany) as a template with the sense primer 5Ј-GTCA-GATCTCTGCAGGTGCCAGAGCGA-3Ј and antisense primer 5Ј-GTAGGATCCTGAGGGGCAGGGCCGTGT-3Ј. The resulting expression construct encoded a signal sequence, NtA, and the FS domain of chick agrin followed by domain V (nucleotide 9518 until the poly(A) signal including the 3Ј untranslated region). Positive 293 cell clones were selected and serum-free media of the clone was collected and concentrated by a factor of 30 by ultrafiltration with a 100,000 MWECO filter (Millipore).

Antibodies and Native Basement Membrane Proteins
Polyclonal antibodies for laminin-2/4 and specific for the laminin ␣2-G domain were used as described (7). Rabbit antiserum specific for chick agrin was used at a dilution of 1:300 for immunostaining and immunoblotting. Rabbit antiserum specific for MuSK was used for immunoprecipitation of protein and immunoblotting at a 1:500 dilution. Rabbit antiserum specific for perlecan was used for immunoblot-ting and solid-phase binding experiments. Fab fragments were prepared by treatment of IgG with papain-immobilized beads (28).

Protein Assays
Protein concentrations were determined by densitometry of Coomassie Blue-stained acrylamide gels. The concentration of NP was determined by comparing immunoblots of aliquots with known amounts of recombinant NtA with anti-agrin polyclonal antibody.
A solid-phase assay was used to detect laminin-2 binding to the NtA domain of agrin. nNA was coated onto plastic wells at 10 g/ml, and unbound sites were blocked with 1 mg/ml bovine serum albumin in PBS containing 0.06% Triton X-100 (blocking buffer). Recombinant Lm2 was incubated in the blocking buffer on the mini-agrin plate for 1 h at room temperature. For competition, recombinant Lm2 was mixed with dilutions of NtA. Binding was detected with anti-␣2G laminin polyclonal antibody (5 g/ml) followed by incubation with a 1:1000 dilution of protein A-horseradish peroxidase conjugate (Sigma). Absorbance (492 nm) was measured on a Spectrafluor (Tecan) plate reader after color development with o-phenylenediamine.
Binding of perlecan and NP to laminin-2 was evaluated in a solid phase assay in which Lm2 was coated on 96-well plates as above, blocked with 10 mg/ml bovine serum albumin in PBS with 0.05% Tween 20, and incubated with perlecan or NP for 1 h at room temperature. Bound proteins were labeled with rabbit anti-perlecan antibody (5 g/ml) and detected with anti-rabbit IgG horseradish peroxidase conjugate.

Myoblast Culturing and Analysis
Mouse C2C12 (6) and MuSK Ϫ/Ϫ myoblasts (24) were maintained and fused as described. Myotubes were incubated with recombinant proteins diluted into fusion medium after 3-5 days (6). AChR clusters were labeled by adding 5 mg/ml fluorescein isothiocyanate/␣-bungarotoxin (Molecular Probes) to live myotubes for 1 h, washed, fixed with 3% paraformaldehyde in PBS (15 min at room temperature), and immunostained as described (6). Digital fluorescent images of myotubes were recorded with a 12-bit CCD camera from a ϫ20 objective microscope using 2 ϫ 2 binned fields (corresponding to 650 ϫ 515 pixels after binning; 0.672 m/pixel) using IPLab 3.5 (Scanalytics, Inc.). AChR clusters were counted if they contained more than a 3 ϫ 3 group of contiguous pixels with a minimum mean pixel intensity at least 3 times background intensity. Clusters were counted from 10 field images for determination of average density and mean Ϯ S.E. unless otherwise indicated. Cluster density was expressed as the number of clusters per (437 ϫ 233 m) imaged field for C2C12 myotubes (which were densely distributed and filled the fields) or as the number of clusters per m 2 of projected myotube surface for MuSK Ϫ/Ϫ myotubes (which were more sparsely distributed). MuSK Ϫ/Ϫ clusters were additionally evaluated for immunostaining to the AChR ␣-subunit (mAb 398, Chemicon) and counted when BTX staining co-localized with that of the antibody.
Chemiluminescence-Chemiluminescence intensity was measured from films or blots using digitalized images recorded with a Chemidoc-XRS CCD camera (Bio-Rad).

Basement Membrane Binding
Interactions-Laminin-2, like other polymerizing laminins, is thought to exist as a cell surface-associated network tethered to nidogens and agrins, with nidogen bound to perlecan and (not shown) type IV collagen (Fig. 1). Recombinant laminin-2 (furin-processed form found in muscle and other tissues), minigene versions of nNA and NA, isolated NtA domain, perlecan derived from the Engelbreth-Holm-Swarm matrix, and NP were used in this study (Fig. 1b). NA has previously been shown to possess AChR clustering and NMJ assembly properties of full-length agrin, whereas nNA was found to substantially ameliorate the dystrophic muscle histology and phenotype, prolonging animal survival when overexpressed in the muscle of laminin ␣2-deficient mice (25,26).
Recombinant proteins and their binding interactions were characterized (supplemental Fig. S1). Laminin-2 binding to nNA was inhibited by excess NtA fragment. Native perlecan binding to laminin-2 depended upon the presence of nidogen-1, whereas NP binding to laminin-2 depended upon interaction through the chimeric NtA domain. Conditioned medium of cultured C2C12 myotubes contained nidogen-1 (1-2 g/ml) but otherwise contained little or none of the basement membrane components described in this work (data not shown). Thus, it is unlikely that any of the effects described here are based on laminins, agrins, perlecan, or type IV collagen released from the myotubes.
Agrin-induced Acetylcholine Receptor Clustering-Myotubes were incubated for 18 h with laminin-2, NA, and/or nNA (Fig. 2). AChR clusters were visualized with fluorescent bungarotoxin (Fig. 2, a and b). Laminin-2 induced AChR clusters severalfold above spontaneous levels (ϳ7 clusters/ϫ20 field), whereas NA (1 g/ml) induced AChR clustering to even higher levels. Whereas nNA (1 g/ml) failed to affect clustering as a single added component, a striking increase in the number of AChR clusters was observed when myotubes were treated with a mix- ture of nNA and laminin-2. The highest clustering levels were achieved when NA, nNA, and laminin-2 were incubated as a mixture. Plots of the AChR cluster density dependence on ligand concentration were empirically found to be well fitted (solid lines) using the simple binding equation substituted with terms for clustering, i.e. c ϭ (Cm ⅐ L)/(EC 50 ϩ L), where c is the cluster density, L is the ligand concentration, Cm is the maximal possible cluster density, and EC 50 is the ligand concentration that induces half-maximal AChR clustering. The average EC 50 of laminin-2 was determined to be 3.7 nM (Ϯ1.3 nM, n ϭ 4). This value did not change for laminin when NA, nNA, or a mixture of NA and nNA were added to the laminin. AChR cluster density also depended upon the concentration of agrins (Fig. 3). Maintaining laminin-2 constant at 28 nM (20 g/ml; which produces a near-saturating clustering effect), nNA was observed to substantially increase cluster density from 50 to 131/field, reaching near maximal values by 0.8 nM nNA (Fig. 3a). The EC 50 for nNA in this paradigm was 0.05 nM compared with 4 nM for laminin-2 alone. This higher activity was comparable with that observed for NA alone (0.06 nM) as well as NA in the presence of laminin-2 (0.11 nM NA, Fig. 3b). The maximal clustering observed when NA was mixed with laminin-2 (254/field at 8 nM NA, 28 nM laminin-2) was greater than the sum of the maximal laminin (50/field) and maximal NA (101/field) values, interpreted that the laminin enhancement was synergistic. Maximal clustering was achieved at NA or nNA molar concentrations substantially below that of laminin, indicating that the population of molecules in the complex likely contained many more laminins than agrins. Furthermore, we found that if Lm2⌬G1-5 and intact laminin were mixed in a 9:1 ratio, almost no loss of AChR clustering was observed despite a 10-fold dilution of cell-interactive laminin G-domains (data not shown). This suggests that the relatively high laminin concentration to mediate clustering is primarily needed to drive polymerization.
The data further reveal that NA induces clustering through both a laminin-dependent and laminin-independent mechanism but that nNA is strictly laminin-dependent, i.e. that NA possesses a unique activity because of its B/z splice insert that is laminin-independent and that is superimposed upon a laminin-dependent activity common to both agrins. To evaluate this further, the contributions of NtA domain interactions, laminin polymerization, and laminin LG domains to AChR clustering were examined (Fig. 4). nNA and NA were separately incubated with laminin-2 in the presence of either a molar excess of the NtA domain to inhibit agrin binding to laminin or with a molar excess of laminin fragment E1Ј to inhibit polymerization of laminin (7,31), or with laminin-2 with all LG domains deleted to prevent a direct interaction of laminin with the cell surface (7). There were several expectations. First, E1Ј should block clustering dependent upon laminin polymerization, i.e. all clustering induced by either laminin or nNA but only the laminin-dependent component of NA. Second, the NtA fragment should block clustering that depends upon tethering to laminin through the coiled-coil, i.e. all clustering induced by nNA but not that of laminin alone and only the laminin-dependent component of NA. Third, if laminin-mediated clustering requires laminin anchorage to the cell surface through its own LG domains, then deletion of the laminin LG domains should prevent all laminin-associated clustering.
When the AChR cluster densities were compared for the different conditions, E1Ј treatment to inhibit laminin polymerization was found to reduce the laminin-2-induced AChR density to spontaneous (baseline) levels and the high nNA/laminin density to that of the baseline, but to reduce the high NA/laminin density to that of NA alone (Fig. 4a,  asterisks). This supported the first expectation. NtA treatment was found to reduce the high nNA/laminin-2 cluster density to that equivalent to laminin-2 alone, and the high NA/laminin cluster level to one between that of NA alone and laminin-2 alone (Fig. 4a, daggers). This supported the second expectation. Finally, deletion of the laminin LG domains reduced clustering induced by nNA but did not eliminate it (Fig. 4b). Examination of the concentration dependence of clustering induced by nNA/Lm2⌬G1-5 mixtures (Fig. 4c) revealed an EC 50 of 0.13 nM, similar to that observed with nNA/Lm2 (Fig. 3a). Excess NtA completely blocked AChR clustering (Fig. 4d), revealing the clustering dependence on nNA binding to laminin. These results only partially supported the third expectation. Although there was a contribution arising from the laminin LG domains, the persistence of a substantial amount of clustering in the absence of laminin domains suggested that much of the nNA activity required only its own LG domains and tethering to a laminin polymer, and that laminin exerted much of its contribution by either increasing the amount of agrin on the myotube surface or by cross-linking agrin molecules to the laminin network polymer.
To explore this issue further, the binding of nNA was examined on myotubes following treatment by nNA alone, nNA ϩ Lm2⌬G1-5, or nNA ϩ full-length laminin-2 (Fig. 4e). Laminin and nNA immunostaining were examined at 4 h (maximal time of laminin accumulation (7) but before onset of clustering). The distribution or level of laminin-2 was not affected by the presence of nNA. In contrast, Lm2⌬G1-5 was observed above background levels only when co-incubated with nNA, i.e. no independent laminin binding could be detected if the laminin LG domains were absent. nNA immunofluorescence, on the other hand, was distributed at similar intensities (confirmed by quantitation of pixel density and intensity, data not shown) on the myotube surface in the absence or presence of laminin-2 and the amount of bound nNA (evaluated by immunoblotting of detergent extracts, Fig. 4f) was found to be the same regardless of the presence or absence of laminin.
We then asked whether the laminin LG domains were required for recruitment of dystroglycan, laminin, and nNA to AChR clusters. These components were found to substantially co-localize within AChR clusters following treatment with nNA/Lm2⌬G1-5 (Fig. 4, g-h) or nNA/ laminin-2 (data not shown), similar to that previously observed with laminin-2 alone (7). Taken together, the cell surface binding of nNA in the absence of ␥1-laminins, the unchanging levels of nNA because of laminin treatment, and ability of polymerizing laminin to induce AChR clusters in the absence of the laminin LG domains argue that laminin binding to nNA does not serve to anchor the agrin to the cell surface but rather that binding to laminin is needed to incorporate the agrin into the laminin polymer network to enable laminin enhancement of AChR clustering.
The dependence of nNA-induced clustering on laminin polymerization suggested that aggregation of the nNA LG domains served as an important mechanism in the process. If this is true, we expected that aggregation of agrin LG domains independent of laminin might also induce clustering. We therefore evaluated the ability of antibody specific for the His 6 N-terminal tag of nNA to induce clustering of nNA (Fig. 5). Antibody IgG, but not its Fab fragments, induced the formation of AChR clusters with a similar morphology to that seen with laminin-2/nNA mixtures. This clustering was antibody concentration-dependent with addition of a secondary IgG further increasing clustering. If the inactive Fab fragments preparation was treated with intact IgG secondary antibody, AChR clustering was restored. Agrin-specific antibody added to nNA also induced clustering in the absence of laminin (data not shown).
Perlecan Induction of AChR Clustering-Genetic analyses have not revealed a unique role of non-neural agrin in the coalescence of AChR  /ml), and the laminin polymerization inhibition fragment E1Ј (500 g/ml). b, nNA induction of clustering does not require laminin LG domains. Myotubes were treated (18 h) with nNA (30 g/ml) mixed with laminin-2 (Lm2) or laminin-2 expressed with deletion of all five LG domains (⌬G15, 28 nM) in the absence or presence of E1Ј (500 g/ml) followed by determination of AChR cluster density. c, concentration effect of G-deleted laminin-2. Myotubes were incubated with the indicated concentrations of nNA in the presence of 10 g/ml Lm2⌬G1-5 and evaluated for clustering at 18 h (EC 50 ϭ 0.127 nM nNA). d, NtA domain inhibits nNA-induced but not laminin-2-induced AChR clustering. Myotubes were incubated (18 h) in increasing concentrations of the NtA domain fragment in the presence of either 10 g/ml laminin-2 or Lm2⌬G1-5. e and f, nNA accumulates on myotube surfaces without laminin-2, whereas G-deleted laminin-2 requires nNA to accumulate on myotube surfaces. Lm2 and Lm2⌬G1-5 (20 g/ml each) were added to the conditioned medium of myotubes in the absence or presence of nNA (1 g/ml) for 4 h prior to washing, fixation, and immunostaining with antibody for laminin-2/4 and agrin. f, immunoblot (agrin antibody) of detergent extracts of nNA incubated with myotubes alone, in the presence of Lm2⌬G1-5 (⌬G) or full-sized laminin-2 for 4 h. g-i, nNA-induced clusters contain agrin, laminin, and DG. AChR clusters were induced with nNA (1 g/ml) and Lm2⌬G1-5 (20 g/ml) for 18 h, labeled with BTX, and immunostained for agrin, laminin-␣2, and ␣-dystroglycan. . Antibody-induced aggregation of nNA clusters AChRs in the absence of laminin. a, C2C12 myotubes were incubated as indicated with nNA (1 g/ml) and anti-His 6 rabbit antibody (aHis, 3 g/ml), anti-His 6 Fab fragment (3 g/ml before removal of Fc portion with protein-A beads), goat anti-rabbit IgG (10 g/ml) with nNA or nNA plus Fab, or intact primary antibody in media for 18 h followed by detection of AChR clusters. b, representative fluorescent micrographs (fluorescein isothiocyanate/BTX) shown for myotubes that were treated with nNA (1 g/ml) in the absence of laminin or nNA (1 g/ml) plus anti-His 6 antibody IgG (3 g/ml). c, myotubes were incubated (18 h) with nNA (1 g/ml) plus the indicated concentrations of anti-His 6 antibody IgG in the presence of 10 g/ml goat anti-rabbit IgG.
clusters during mouse development (17,32). The apparent difference between in vitro and in vivo findings suggests that the NMJ contributions of non-neural agrin might also be provided for by other basement membrane components with similar clustering activities, i.e. redundancy of function. Because the LG domains of both laminins and agrins exhibit AChR clustering activity, we asked whether basement membrane perlecan, which similarly has LG domains, might possess this property.
Perlecan was found to induce AChR clustering when mixed with laminin-2 (Fig. 6a). The molar EC 50 was determined by data fitting to be ϳ14 nM, i.e. higher than that of nNA. Incubation of perlecan with Lm2⌬G1-5 also resulted in AChR clustering, but with about half the cluster density observed with intact laminin-2. Clustering induced by perlecan/laminin-2 was completely inhibited by laminin fragment E1Ј but was unaffected by NtA fragment (Fig. 6b), revealing a requirement for laminin polymerization but not NtA.
If the clustering activity of perlecan resides in the LG domains and requires tethering to laminin, we reasoned that a fusion protein (NP) consisting of NtA and the LG domains of perlecan, but lacking the major nidogen-binding and other domains of perlecan, would be sufficient to induce AChR clustering. Conditioned medium containing this protein was incubated for 18 h with fused myotubes and found to induce AChR clustering in the presence of either laminin-2 (EC 50 ϭ 1.5 nM) or laminin-2⌬G1-5 (EC 50 ϭ 1 nM), the maximal level of the former exceeding that of the latter (Fig. 6c). Given that NP binds laminin-2 more strongly than perlecan does, the differences in EC 50 values may result from the differences in laminin binding. Clustering induced by NP/laminin-2 could be blocked by excess NtA when NP was incubated with either intact laminin-2 or Lm⌬G1-5 (Fig. 6d). Thus it seemed likely that the perlecan clustering activity, like that of nNA, resides in the LG domains.
Time Course of AChR Clustering-We evaluated the time course of AChR clustering (Fig. 7). Myotubes incubated with NA alone achieved maximal cluster densities by 4 h with no obvious lag. In contrast, a 4-h lag was observed with laminin-2, nNA plus laminin-2, and NP plus laminin-2,withclusteringincreasingovertheensuinghoursforalllaminindependent interactions. This difference in time of clustering distinguished the activity of NA from that of the other LG modular proteins.
MuSK Dependence and AChR Phosphorylation-NA is thought to bind to a MuSK-containing complex resulting in phosphorylation of MuSK and leading to AChR subunit phosphorylation and clustering (33). Laminin-1, on the other hand, has been thought to cluster AChRs in a largely MuSK-independent fashion because laminin also induced AChR clustering in MuSK-null myotubes (6). We asked whether nNA and perlecan-dependent contributions to clustering would contribute to MuSK tyrosine phosphorylation and depend upon the presence of MuSK (Fig. 8).
A large and rapid increase in MuSK tyrosine phosphorylation was detected following NA treatment, reaching a maximum by 1 h (Fig. 8, a  and b). No phosphorylation increase was detected at 18 h with nNA alone (data not shown). Laminin-2 alone induced a gradual and much smaller phosphorylation increase. When either nNA or NP were incubated at their maximal clustering concentrations in the presence of laminin-2, an increase in MuSK phosphorylation, particularly in the first 4 h, was observed.
The MuSK dependence of clustering was then evaluated with MuSK Ϫ/Ϫ myotubes (Fig. 8, c and d). A low spontaneous level of AChR clustering was observed. The spontaneous clusters, however, did not stain for ␣-dystroglycan (data not shown). Laminin-2 induced a 4 -5fold increase in AChR clusters that co-localized with dystroglycan (Fig.  8c). The laminin-induced clusters were smaller than those observed following laminin treatment of C2C12 myotubes, suggesting at least partial MuSK dependence. On the other hand AChR densities (6.0 ϫ 10 Ϫ4 /m 2 MuSK Ϫ/Ϫ versus 7.4 ϫ 10 Ϫ4 /m 2 C2C12) were the same. However, strikingly, no further increase in clustering or change in the size was observed when nNA or NP were added to laminin.
It has been reported that laminin-1-induced clustering leads to AChR subunit phosphorylation (9). We found that laminin-2 slowly induces tyrosine phosphorylation of the ␤-subunit of AChR at levels below that observed with NA and that addition of nNA to laminin-2 increased the level of phosphorylation by 4 h (supplemental Fig. S2). In contrast, NP produced little or no enhancement of phosphorylation over that observed with laminin alone.
Stability of AChR Clusters-Formation of a cluster does not ensure that the cluster will be stable, and we wondered whether there might be (a) differences in the longevity of clusters induced by the different agents and (b) specific requirements to achieve stability (Fig. 9). NAinduced clusters, whether treated for only 18 h ("pulse") or treated for as long as an additional 6 days in a prolonged "chase," decreased in number after 18 h with apparent loss of half of the clusters by 3 days of chase FIGURE 6. Perlecan and its induction of AChR clustering. a, perlecan induction of AChR clustering. Myotubes were cultured for 18 h in the presence of exogenous perlecan plus 10 g/ml laminin-2 (solid circles, EC 50 ϭ 14 nM perlecan) or 10 g/ml laminin-2⌬G1-5 (EC 50 ϭ 18 nM perlecan). b, AChR clustering was evaluated in the presence of excess E1Ј (500 g/ml, 1 M) or NtA fragment (3 g/ml, 0.1 M). Only E1Ј, an inhibitor of laminin polymerization, blocked perlecan-induced clustering. c, AChR clustering (myotubes, 18 h) induced with a fusion protein (NP) (3 g/ml) in the presence of either laminin-2 (10 g/ml, EC 50 ϭ 1.5 nM) or laminin-2⌬G1-5 (10 g/ml, EC 50 ϭ 0.8 nM). d, AChR clustering induced with NP (3 g/ml) is inhibited by excess NtA domain fragment. FIGURE 7. Time course of AChR clustering. Fused myotubes were treated with NA (1 g/ml), rLm2 (20 g/ml), nNA (1 g/ml) with rLm2 and with NP (3 g/ml), rLm2 (20 g/ml), or NP ϩ rLm2. A lag in clustering was detected with laminin-2, nNA ϩ laminin-2, and NP ϩ laminin-2, but not with NA.
time. This decay in cluster density was unaffected if the NA was replaced by medium alone, laminin-2, or NA (Fig. 9a). However, if NA was combined with laminin-2, or replaced by nNA plus laminin-2, NP plus laminin-2, or perlecan plus laminin-2 during the chase, cluster density remained high (Fig. 9b and data not shown). AChR clusters induced by nNA plus laminin-2 (Fig. 9a), NP plus laminin-2 (Fig. 9c), or perlecan plus laminin-2 without further treatment (data not shown) were unstable, decaying to spontaneous levels by 2 days of chase.
Type IV collagen, which forms a second polymer network (4), is thought to stabilize basement membrane architecture through interactions with nidogens and other components (34). We asked whether this collagen affected the induction or stability of AChR clusters (Fig. 9, d-i). The collagen, when added to culture medium, did not increase the den-sity of AChR clusters by itself, by laminin-2, or laminin-2 plus nNA. However, type IV collagen promoted the stability of clusters induced by laminin-2, nNA plus laminin-2, or NP plus laminin-2, with almost no loss of cluster density observed in a 4-day chase if the collagen was maintained with these components. In the case of NA-induced clusters formed in the absence of laminin, type IV collagen was not found to stabilize clusters. This suggests that the effect of the collagen was mediated through laminin.

DISCUSSION
Two MuSK-dependent stages of development have been identified that lead to the formation of the mature NMJ (17). The first, apparent by mouse embryonic day (E) 14, is characterized by AChR clustering that is independent of either neural agrin or innervation (aneural clustering). The second stage, apparent by E18, is characterized by the replacement of the early AChR clusters with ones that become aligned with branching nerve process termini in large part through the action of neural agrin. Both stages of AChR assembly occur in a muscle basement mem-FIGURE 8. MuSK phosphorylation and dependence. C2C12 myotubes were incubated for the indicated times with NA (1 g/ml), rLm2 alone (20 g/ml), nNA (1 g/ml) ϩ Lm2 (20 g/ml), or fusion protein (NP) (3 g/ml) ϩ Lm2 (20 g/ml), lysed with detergent, precipitated with MuSK-specific antibody, and analyzed in immunoblots for tyrosinephosphorylated MuSK (pMuSK) and total MuSK. Immunoblots for representative data sets are shown in a, and plot of MuSK phosphorylation for NtA-agrins and laminin (average and S.E., n ϭ 5 except for rLm2 at 1 h where n ϭ 3) is shown in b. MuSK Ϫ/Ϫ myoblasts were fused and incubated with NA (1 g/ml), no agent (NT), or with laminin-2 (20 g/ml) without (Ϫ) or with nNA (1 g/ml) or NP (3 g/ml) for 18 h. Representative BTX-stained (BTX) myotubes shown for the different conditions (including colocalization of BTX with anti-AChR␣ (AR) and BTX with ␣-dystroglycan for Lm2 treatment) (c). Plot of AChR cluster density/myotube area is shown in d.  a and b), nNA plus laminin-2, or NP plus laminin-2 (panel c) for 18 h to induce AChR clusters followed by replacement with fresh medium containing either no added reagent (NT) or nNA, laminin-2, nNA ϩ laminin-2, or NP ϩ laminin-2 for the indicated number of days. Panel d, effect of type IV collagen on AChR clustering. Myotubes were treated with laminin-2 or nNA ϩ laminin-2 without or with type IV collagen for 18 h as indicated for determination of AChR clustering. Panels e-i, myotubes were treated with the indicated components for 18 h and analyzed or treated for 18 h plus an additional 4 days and analyzed for AChR cluster density. The concentrations of components used were type IV collagen (Col, 10 g/ml), laminin-2 (Lm, 20 g/ml in panels a-c and 10 g/ml in panels d-i), nNA (1 g/ml), NA (1 g/ml), and NP (3 g/ml). brane zone enriched in ␣2-laminins, nidogens, non-neural agrin, perlecan, and type IV collagen.
The cell surface anchorage and polymerization of ␥1-laminins, mediated by laminin LG and LN domains, respectively, is required for basement membrane assembly on cell surfaces (35,36). Furthermore, laminins, by tethering other basement membrane proteins through direct and indirect binding, serve to recruit and immobilize the proteins to the nascent basement membrane (4). These components in turn provide a rich array of ligands that can interact with adjacent cell surface receptors such as dystroglycan and integrins. A number of studies have shown that laminins can induce AChR clustering with a dependence on dystroglycan, ␣7␤1-integrin, and laminin polymerization (6 -8,37). The current study reveals that non-neural agrin and perlecan can promote the assembly of AChR clusters. This clustering, unlike that induced by neural agrin, requires the LG domains of these components be tethered through laminin and that the laminin exhibit polymerization activity. Clustering occurs with a lag in a MuSK-dependent manner leading to modest levels of tyrosine phosphorylation of MuSK and ␤AChR. The clusters that form are stabilized, but not induced, by type IV collagen. Non-neural agrin is particularly potent as a laminin-dependent clustering agent, exerting its induction at subnanomolar concentrations similar to those achieved with neural agrin.
We have approached the myotube AChR clustering model from the viewpoint that ECM can self-assemble on an anchoring cell surface when added in solution, mimicking assembly in vivo (5,7,35,38,39). Under these conditions the AChR clustering resulting from non-neural agrin and perlecan, like that of neural agrin, is dependent on MuSK. This contrasts with laminin-2 (this study) and laminin-1 (6) that show a limited dependence (affecting AChR cluster size but not density in our hands) upon MuSK expression. These relationships are reflected in the tyrosine phosphorylation of MuSK. Neural agrin induces a large increase in MuSK phosphorylation, whereas laminin-2 produces very little. The combination of non-neural agrin and laminin-2 (or perlecan and laminin-2) induces an increase in phosphorylation over that observed with laminin alone. Because the simplified minigene versions of non-neural agrin and perlecan used in the study consists of the laminin-binding NtA domain that has no clustering activity of its own and the LG domains, it follows that the cell-active moiety corresponds to the latter domains.
The formation of an ECM that contains interacting ligands, and not just the binding of the ligands to the myotube surface, appears to be important for AChR clustering. A body of literature has shown that the spatial arrangement of ligands and the mechanical properties (e.g. rigidity) of the attached polymers can greatly affect receptor interactions and cell behavior (40,41). The data of this study support the hypothesis that AChR clustering is induced by aggregation and/or cross-linking of existing agrin by laminin on the cell surface. Signaling may result from both clustering of LG domain ligands to an appropriate density and the formation of a gel matrix that enables anchorage-dependent interactions and whose viscoelastic properties affect activation of engaged cell surface receptors. It is interesting to note that adhesion of myotubes to laminin immobilized on rigid plastic (10) results in differences in AChR clustering compared with addition of soluble laminin to myotubes, such that it polymerizes on the cell surface. Under immobilized conditions, laminin and other ECM components were observed to induce complex ring-and pretzel-like clusters in a MuSK-dependent fashion that are similar to clusters observed in postnatal development. We have found that plastic-immobilized laminin-1 will support complex AChR cluster-ing even if laminin polymerization is chemically inhibited. 3 We suggest that the plastic immobilization bypasses the polymerization requirement by providing a pre-existing solid-phase interface of laminin LG ligands. The rigid immobilization of ligands, not obtainable with laminin polymer on cell surfaces, and not just adhesion strength, may be responsible for the complex AChR clustering phenomenon.
The identity of the agrin receptor mediating MuSK activation and AChR clustering has been elusive. Whereas it has been argued that dystroglycan and integrins are not the primary receptors that mediate clustering, and whereas antibody-induced clustering of AChR mimics the activity of neural agrin, evidence to support neural agrin as a direct ligand of MuSK is not strong (reviewed in Refs. 42) and it has been suggested that neural agrin interacts indirectly with MuSK through a co-receptor (28). Because non-neural agrin and perlecan cluster induction is also MuSK-dependent resulting in MuSK phosphorylation, the two LG-domain proteins may also engage a MuSK co-receptor. One mechanism to explain the unique actions of these components is that their incorporation into polymerizing laminin forces increased ligand density, enhancing an otherwise low affinity interaction with the putative coreceptor. Areas of high density cannot be achieved simply by the LG domain binding that is responsible for the accumulation of nonneural agrin to the myotube cell surface. A difference between nonneural agrin/perlecan and neural agrin may be that the former binds weakly to the putative MuSK coreceptor unless non-neural agrin is presented at high density and cross-linked by polymer, whereas neural agrin binds strongly in a monovalent state. Although all of the LG domains can interact with dystroglycan and integrins (7,43,44), and although dystroglycan accumulates in the clusters, the effects of these receptors may be ones of promoting adhesion and modulating AChR clustering rather than serving as the primary mediators of cluster induction (45,46). An argument that favors this interpretation in the case of dystroglycan is that neural agrin can induce clustering in the absence of the dystroglycan binding region (47).
An additional property of basement membrane appears to be to stabilize AChR clusters. Type IV collagen substantially increases the survival time of laminin-, non-neural agrin-, and perlecan-induced clusters. Whereas we could not measure stabilization of neural agrin clusters induced in the presence of laminin because they persisted at their 18-h density without further treatment, we suspect that type IV collagen actually stabilizes all laminin-dependent clustering. Laminin, which is required for all clustering except that because of the agrin B/z insert, can only stabilize neural agrin-induced clusters in the absence of collagen. Together, these activities suggest that type IV collagen and laminin help preserve AChR clusters.
Genetic analysis of the developing mouse has revealed that AChR clustering first occurs in an aneural phase requiring MuSK (17). Because neural agrin is not required for induction of these early clusters, how might they be initiated? Whereas it is possible that the aneural clusters arise spontaneously as suggested (45), the clusters seen in developing mouse are at much higher density than those that form on MuSK-null or other myotubes. We also note that the clusters that form at this stage are relatively simple in shape, not unlike those observed on cultured myotubes treated with LG domain proteins. The clusters formed on myotubes can persist for at least 4 days in the presence of type IV collagen and do not require neural agrin. Based upon these findings we propose that proteins with LG domains, in particular non-neural agrin and perlecan, serve this role.
The question then arises as to whether selective genetic inactivation 3 H. Colognato and P. D. Yurchenco, unpublished observations. of non-neural agrin and perlecan reveals a relevant phenotype in mice at the aneural phase of NMJ development. First, inactivation of neural agrin with sparing of non-neural agrin produced a phenotype that has not been distinguished from the total agrin knock-out (17). Second, whereas transgenically expressed neural agrin can substantially rescue the perinatal lethality of agrin-deficient mice, with survival extending about 2 months after birth, 4 transgenic overexpression of non-neural mini-agrin in muscle causes substantial changes in the NMJ characterized by their more diffuse appearance, the persistence of AChRs at sites devoid of nerve terminals at the light microscopic level, and an increase in postsynaptic folds at the ultrastructural level. 5 Third, knock-out of the perlecan gene has revealed a role in anchoring acetylcholinesterase to the motor endplate but not failure of NMJ assembly (23). We suggest that functional redundancy of non-neural agrin and perlecan may explain why genetic analyses have failed to reveal a clear NMJ assembly phenotype for these components. The identification of a NMJ phenotype from overexpression, not seen when the component is absent, further supports this interpretation. Collectively the genetic and cell biological data are compatible with a model in which neural agrin is a critical component for the late assembly of a NMJ that cannot be adequately compensated by other LG modular components, but that the laminin-tethered LG domains of the basement membrane collaborate in the assembly, particularly during the aneural phase, of the NMJ with redundancy of function among different LG proteins. Finally, it is interesting to note that the two fusion proteins consisting of NtA domain and non-neural agrin and perlecan LG domains observed to induce AChR clustering, when transgenically overexpressed in muscle of the laminin ␣2-deficient mice, substantially ameliorated the muscular dystrophy (26). 6 This suggests that the LG modular interactions that lead to induction of AChR clustering also mediate other general basement membrane activities that affect muscle functions other than NMJ assembly. Given this correlation, the ability of a component to induce AChR clustering on myotubes appears to be a predictor of their ability to ameliorate laminin-deficient muscular dystrophies. This may provide a useful avenue to explore in the development of diagnostic and therapeutic modalities in the future.