Crystal Structure and Cell Surface Anchorage Sites of Laminin α1LG4-5*

The laminin G-like (LG) domains of laminin-111, a glycoprotein widely expressed during embryogenesis, provide cell anchoring and receptor binding sites that are involved in basement membrane assembly and cell signaling. We now report the crystal structure of the laminin α1LG4-5 domains and provide a mutational analysis of heparin, α-dystroglycan, and galactosylsulfatide binding. The two domains of α1LG4-5 are arranged in a V-shaped fashion similar to that observed with laminin α2 LG4-5 but with a substantially different interdomain angle. Recombinant α1LG4-5 binding to heparin, α-dystroglycan, and sulfatides was dependent upon both shared and unique contributions from basic residues distributed in several clusters on the surface of LG4. For heparin, the greatest contribution was detected from two clusters, 2719RKR and 2791KRK. Binding to α-dystroglycan was particularly dependent on basic residues within 2719RKR, 2831RAR, and 2858KDR. Binding to galactosylsulfatide was most affected by mutations in 2831RAR and 2766KGRTK but not in 2719RKR. The combined analysis of structure and activities reveal differences in LG domain interactions that should enable dissection of biological roles of different laminin ligands.

Laminin-111, recently renamed from laminin-1 to better reflect its ␣1␤ 1␥1 subunit composition, is one of the first two laminins to be expressed during embryonic development, appearing in the peri-implantation period in the basement membrane of the embryonic plate along with laminin-511 (laminin-10) and in the absence of other laminins in Reichert's membrane (1,2). Later in development, laminin-111 is strongly expressed in placenta, liver, kidney, and testis, where it is thought to play a role in organogenesis (3). In the adult, the laminin ␣1 chain has very limited expression and is largely supplanted by the laminin ␣5 chain. Targeted inactivation of the LAMA1 gene coding for the ␣1 chain was found to result in a failure of Reichert's membrane with developmental arrest by embryonic day 6.5 in the mouse (4). In-frame deletion of the mouse laminin ␣1 exons corresponding to laminin G-like (LG) 3 domains 4 and 5 was found to result in similar stage lethality accompanied by defective epiblast differentiation without loss of basement membrane (5), the last possibly a result of the partially redundant expression of laminin-511 (4).
The ␣1 subunit provides most of the unique characteristics of laminin-111. The N-terminal LN domain participates in polymerization by interacting with the LN domains of the ␤1 and ␥1 chains. The C-terminal LG domains, LG1-3, bind to the ␣ 6 ␤ 1 integrin, whereas LG4-5 bind to heparin, sulfated glycolipids, and ␣-dystroglycan (␣-DG). The polymerization and cell-anchoring activities are thought to act in concert to assemble a functional basement membrane on a cell surface (6).
Our earlier understanding of the laminin ␣1 LG1-5 structure was based on crystal structures of LG4 and LG4-5 from the related ␣2 chain (ϳ40% sequence identity). The LG domain fold was revealed as a multistranded ␤-sandwich with one bound calcium ion. In the LG4-5 pair, the two domains are connected in a V-shaped arrangement, in which LG5 is disulfide-bonded to the linker preceding LG4 (7,8). Although the structure of LG1-3 remains to be elucidated, it is thought that these three domains form a closed arrangement with similar angles between domains and separated from the LG4-5 pair by a hinge-like region (9).
Analysis of contributions of ␣1LG4-5 to laminin interactions with Schwann cells, myotubes, and developing epithelia has led to a model in which these domains provide the major anchoring activity of laminin, an initiating event of basement membrane assembly that leads to alterations of the cell cytoskeleton accompanied by signaling (6,10,11). Basic residues within the LG4-5 pair of several laminins mediate key interactions with three types of molecules: heparan sulfate chains that are attached to perlecan, agrin, collagen, and syndecan core proteins of basement membranes and cell surface; the glycoprotein ␣-DG that is a component of a larger transmembrane and submembrane complex associated with dystrophin and utrophin; and sulfated glycolipids, in particular the sulfatides, that can be present in the outer leaflet of the cell plasma membrane. Earlier mutagenesis analyses of recombinant laminin ␣1 and ␣2 LG fragments revealed that there are binding similarities and differences between the two laminins. Heparin, representing the highly sulfated regions of glycosaminoglycan chains, ␣-DG, and galactosyl-3-sulfate ceramide (galactosyl sulfatide) all bind to an extensive basic surface region between the calcium sites of the laminin ␣2 LG4-5 domain pair (12,13). In contrast, a smaller topographical region confined to LG4 appears to bind the same cell surface components in the laminin ␣1 chain (12,14).
The present study describes the crystal structure of laminin ␣1LG4-5 in conjunction with a site-directed mutagenesis and in vitro analysis of the binding sites for heparin, ␣-DG, and sulfatides. We report that the backbone of the ␣1LG4-5 domain pair is very similar to that of ␣2LG4-5, but there are significant differences in the distribution of charged residues on the protein surface. Binding of ␣1LG4-5 to heparin, ␣-DG, and sulfatides was found to be dependent upon partially overlapping basic amino acid residue clusters. Our results have led, we believe, to an improved understanding of the amino acid residues involved in each type of interaction and will aid in defining the biological roles of different laminin ligands.

EXPERIMENTAL PROCEDURES
Laminin ␣1LG4-5 Vector Constructs-An expression construct containing the wild-type (WT) mouse laminin ␣1LG4-5 sequence (coding for residues 2666 LHREH . . . PGPEP 3060 of the mature laminin ␣1 chain (i.e. the numbering scheme used here omits the 24-residue signal peptide of SwissProt entry P19137)) was created by amplifying cDNA from laminin ␣1 pCIS (15) utilizing three successive PCRs. Three overlapping 5Ј sense primers were used to place a 5Ј-terminal NheI site followed by the 5Ј-untranslated region and signal sequence of human BM-40 (cleaved by the signal peptidase) and a FLAG epitope tag (DYKDDDDK), whereas the 3Ј primer placed a KpnI site downstream of the STOP codon at the 3Ј terminus of the amplified product (oligonucleotide sequences are listed in supplemental Table 1). Pfx polymerase (Invitrogen) was used along with a PTC-100 thermal cycler (MI Research) to amplify the DNA, which was then purified after each reaction (UltraClean PCR DNA purification kit; MoBio). The NheI and KpnI sites were used to clone the PCR product into the analogous sites in the pcDNA3.1ϩ/zeo vector (Invitrogen). The mutated ␣1LG4-5 DNAs were constructed by strand overlap extension PCR using the same three 5Ј upstream sense primers and downstream 3Ј antisense oligonucleotide in conjunction with internal primers introducing the desired mutations. Escherichia coli DH5␣ cells (Invitrogen) were transformed with the plasmids, and plasmid DNA was purified by alkaline lysis and spin columns (Ultra-Clean Standard Mini Plasma Prep Kit, MoBio). All generated plasmids were completely sequenced.
Recombinant ␣1LG4-5 Protein Production and Purification-Expression constructs were linearized with BglII (New England Biolabs) and transfected into the human kidney fibroblast cell line 293 (ATCC) using Lipofectamine 2000 (Invitrogen), and stable clones were selected under zeocin for secretion of ␣1LG4-5 protein. Cells were grown in Dulbecco's modified Eagle's medium with high glucose (Invitrogen) supplemented with 10% fetal bovine serum, 50 units/ml penicillin G, 50 g/ml streptomycin, and 100 g/ml zeocin (Invitrogen). Once cells had reached confluence, the growth medium was replaced with fresh medium minus zeocin and then collected 72 h later. The genomic DNA was isolated from the cells after medium harvesting (Exact-N-Amp kit; Sigma) and sequenced to verify the identity of the various ␣1LG4-5 proteins. The conditioned media were passed through a gravity column (Bio-Rad) packed with anti-FLAG M2-agarose resin (Sigma), and recombinant ␣1LG4-5 proteins were eluted with FLAG peptide (Sigma) in 90 mM NaCl, 1 mM CaCl 2 , 50 mM Tris-HCl, pH 7.4 (TBS50/Ca) at 4°C. The eluted proteins were then loaded onto a heparin 5PW column (TosoHass) on an Ä kta FPLC system (Amersham Biosciences), where the FLAG peptide (unbound component) was recovered for reuse, and the ␣1LG4-5 proteins were eluted using a 1 M NaCl gradient. The eluted ␣1LG4-5 proteins were concentrated, and the buffer was exchanged into TBS50/Ca at room temperature via centrifugal filtration with Amicon Ultra spin filters (Millipore). The recombinant proteins were further dialyzed against TBS50/Ca buffer with several buffer changes for 2 days using dialysis cassettes (Slide-A-Lyzer; Pierce). Deglycosylated WT ␣1LG4-5 protein was produced either by isolating WT ␣1LG4-5 from the medium of stably transfected cell lines grown in 2 g/ml tunicamycin (Sigma) for 24 h or by treating 50 g of purified ␣1LG4-5 with 1000 units of peptide N-glycosidase F (New England Biolabs) in TBS50/Ca for 1 h at 37°C.
Laminin ␣1LG4-5 Vector Construct for Crystallography-DNA coding for residues 2682 QPELC . . . PGPEP 3060 of the mature mouse laminin ␣1 chain was obtained by PCR amplification from laminin ␣1 pCIS (15). The PCR primers added NotI and NheI sites at the 5Ј end and a STOP codon followed by XhoI and BamHI sites at the 3Ј end. The PCR product was cloned into pBluescript II KSϩ using NotI and BamHI, and four mutations (N2714Q, N2811K, N2900Q, and C3014S) were introduced by strand overlap extension PCR. The sequence-verified insert was cloned into the pCEP-Pu vector (16) using NheI and XhoI. After cleavage of the BM-40 sequence signal, a vectorderived APLA sequence remains at the N terminus of the secreted recombinant ␣1LG4-5 protein.
Crystal Structure Determination-The ␣1LG4-5 quadruple mutant protein was concentrated to 19 mg/ml in 10 mM Na-HEPES, pH 7.5. Crystals were obtained by hanging drop vapor diffusion using 20% polyethylene glycol 8000, 100 mM Tris-HCl, pH 8.5, 200 mM MgCl 2 as precipitant. Crystals were frozen in liquid nitrogen in mother liquor supplemented with 20% glycerol. Diffraction data to 1.9 Å resolution were collected at 100 K on beamline 9.6 at the SRS Daresbury ( ϭ 0.87 Å). The crystals belong to space group P2 1 , a ϭ 70.53 Å, b ϭ 55.81 Å, c ϭ 100.99 Å, ␤ ϭ 98.48°. There are two ␣1LG4-5 molecules in the asymmetric unit, resulting in a solvent content of ϳ45%. The diffraction data were processed with MOSFLM (available on the World Wide Web at www.mrc-lmb.cam.ac.uk/harry/mosflm) and programs of the CCP4 suite (17). The structure was solved by molecular replacement with PHASER (18), using the laminin ␣2LG4-5 structure (8) as a search model; the LG domains had to be placed individually to obtain a solution. The structure was rebuilt with O (19) and refined with CNS (20) without noncrystallographic symmetry restraints. Data collection and refinement statistics are summarized in Table 1. The figures were made with PYMOL (available on the World Wide Web at www.pymol.org).
Analysis of ␣1LG4-5 Binding to ␣-Dystroglycan-␣-DG was purified from rabbit muscle as described (22). Equal aliquots of ␣-DG (1 g) were loaded into the slots of SDS-polyacrylamide gels and electrophoresed under reducing conditions. The protein bands were then electroeluted onto nitrocellulose membranes, blocked in phosphate-buffered saline containing 5% nonfat dry milk for 1 h at room temperature, and assessed for binding to each ␣1LG4-5 protein (1 g/ml) using a previously described overlay assay (23). Binding of the ␣1LG4-5 proteins was detected with 1 ml of 1.1 g/ml horseradish peroxidasecoupled anti-FLAG antibody M2 (Sigma) per 3.5 cm 2 of blot membrane. The solid phase assay was performed in 96-well microtiter plates with ␣-DG bound to the plate (100 l of 1 g/ml per well) and incubated with various concentrations of ␣1LG4-5 proteins as previously described (24), except that horseradish peroxidase-linked monoclonal FLAG antibody M2 (100 l of 1.1 g/ml per well) was used for detection followed by color development with 3,3Ј,5,5Ј-tetramethylbenzidine (Bio-Rad). Color development was quantitated at 655 nm using a Molecular Dynamics Spectramax 340 UV-visible microplate reader (25). We verified that the signal readout was linear to OD Ͼ 3 by collecting kinetic data on color development. Estimates of half-maximal binding (K D ) and binding capacity (B max ) were determined by curve fitting of the binding data of WT LG4-5 using a single-site model (fitted values ϭ B max ϫ L/(K D ϩ L), where L is the molar ligand concentration), with the calculated B max value used for subsequent determinations of all other half-maximal binding, an approach employed to minimize the errors inherent in estimating binding from plots that are low and nearly linear over the concentration range evaluated.
Analysis of ␣1LG4-5 Binding to Galactosyl Sulfatide and Other Lipids-The ammonium salt of HSO 4 -3Gal␤1-1Јceramide (brain sulfatides; Avanti Polar Lipids) was dissolved in methanol, and 10 g was added per immulon-1B microtiter plate well (ThermoLabsystems). The plate was dried at 37°C for 2 h, and the wells were washed four times with 200 l of enzyme-linked immunosorbent assay wash buffer (EWB; 1% bovine serum albumin in TBS50/Ca) at room temperature. The wells were then blocked for 1 h at room temperature with 200 l of EWB, followed by three 200-l washes of EWB. ␣1LG4-5 proteins in varying concentrations in EWB were added to each well and incubated for 1.5 h at room temperature. The wells were then washed four times with 200 l of EWB, and horseradish peroxidase-linked monoclonal FLAG antibody (Sigma) in EWB was added. After 1 h at room temperature, the wells were washed four times with EWB, and 150 l of substrate solution (4 mM o-phenylenediamine (Sigma), 50 mM citric acid, 100 mM Na 2 HPO 4 , 0.012% H 2 O 2 ) was added. The developing color reaction was stopped after 2-10 min by the addition of 60 l of 2 M H 2 SO 4 , followed by 50 l of ethanol, and the plates were read in a TECAN SpectraFluor microtiter spectrophotometer at 492 nm. Samples with OD Ͼ 2 were diluted and remeasured to correct for any deviation from linearity. A molecular mass of 44.3 kDa was used to calculate molar ␣1LG4-5 concentrations. Inhibition studies were performed in the presence of either 10 g/ml low molecular weight heparin (Sigma) or 5 mM EDTA. The assay was also performed with several other lipids: galactosyl ceramide (Avanti Polar Lipids and Sigma), sphingomyelin (Avanti Polar Lipids), phosphatidic acid (Avanti Polar Lipids), cholesterol 3-sulfate (Sigma), and GM1 ganglioside (Avanti Polar Lipids). Half-maximal binding was estimated in the same manner as described for ␣-DG.
Electron Microscopy-Pt/C rotary shadowing of proteins was performed by deposition of 0.9-nm metal at an 8°angle as previously described (21). Images are shown with reversed contrast.

RESULTS
Crystal Structure of Laminin ␣1LG4-5-The C-terminal LG domain pair of the laminin ␣1 chain contains an unpaired cysteine (Cys 3014 ) and three predicted N-linked glycosylation sites (Asn 2714 , Asn 2811 , and Asn 2900 ). Because we found that WT ␣1LG4-5 preparations always contained a small fraction of disulfide-linked dimers (data not shown), we mutated Cys 3014 to serine. An ␣1LG4-5 C3014S construct with an N-terminal His tag failed to crystallize, as did several other constructs with additional mutations of asparagine residues modified by glycosylation. Eventually, crystals could be obtained of an untagged ␣1LG4-5 quadruple mutant (N2714Q/N2811K/N2900Q/ C3014S) devoid of any N-linked carbohydrate. The crystal structure of this mutant, hereafter termed simply ␣1LG4-5, was refined at 1.9 Å resolution to R free ϭ 0.261 (Table 1).
The asymmetric unit of the crystals contains two crystallographically independent ␣1LG4-5 molecules, A and B. We observed clear electron density for both molecules, with the exception of residues 2987-2990, 3032-3034, and 3060 of molecule A and residues 2682-2684 and 3060 of molecule B. Molecules A and B are very similar in their LG4 and LG5 domain structures (root mean square deviation 0.36 and 0.58 Å, respectively, for all C␣ atoms) but differ substantially in their respective domain arrangements. When the molecules are superimposed on their LG4 domains, a rotation by 14.5°is required to bring their LG5 domains into superposition; the pivot point of this rotation is in the interdomain linker, near Tyr 2871 (Fig. 1A). The following description of the structure is based upon the more complete molecule B.
The ␣1LG4-5 structure consists of two canonical LG domains (9), LG4 and LG5, connected by a short linker and interacting through a small interface near the domain termini (Fig. 1B). Each LG domain folds into a curved ␤-sandwich built from two antiparallel sheets and contains a single disulfide bond near the C terminus. A third disulfide bond tethers the segment preceding LG4 to an ␣-helical turn in LG5. The interface between LG4 and LG5 is water-filled and predominantly polar, and the different conformations of molecules A and B are likely to be due to the paucity of specific interactions in the LG4-LG5 interface.
Both LG4 and LG5 contain one bound metal ion, located on the rim of the ␤-sandwich opposite the interdomain linker. These ions have been modeled as magnesium, given their coordination geometry and the high magnesium concentration in the crystals, but we assume that the binding sites are occupied by calcium under physiological conditions (7). Magnesium ion 1 is coordinated octahedrally by the side chains of Asp 2747 and Asp 2816 , the main chain carbonyl oxygens of residues 2764 and 2814, and two water molecules; the average metal-ligand distance is 2.17 Å. Magnesium ion 2 is coordinated octahedrally by the side chains of Asp 2923 and Asp 2996 , the main chain carbonyl oxygens of residues 2940 and 2994, and two water molecules; the average metal-ligand distance is 2.15 Å. The unpaired cys-teine of ␣1LG4-5, Cys 3014 , is located in the convoluted loop that occupies most of the concave face of LG5. Two predicted N-linked glycosylation sites are located in LG4 (Asn 2714 and Asn 2811 ) and one in LG5 (Asn 2900 ). Asn 2811 is close to the metal ion binding site in LG4 (Fig. 1B).
LG4 of laminin ␣1 and ␣2 can be superimposed with a root mean square deviation of 0.91 Å for 148 C␣ atoms; the major differences are concentrated in the spatially adjacent B-C and L-M loops and in the edge ␤-strand J, which is irregular in ␣1LG4-5 (Fig. 2B). The LG5 domains can be superimposed with a root mean square deviation of 0.59 Å for 153 C␣ atoms; the major differences are again concentrated in the B-C and L-M loops (Fig. 2C).
The relative arrangement of LG4 and LG5 in ␣1LG4-5 and ␣2LG4-5 is also similar, with ␣2LG4-5 more closely resembling molecule B than molecule A of ␣1LG4-5 (not shown). In terms of their interdomain angles, the two crystallographically independent ␣1LG4-5 molecules are, in fact, more different than ␣2LG4-5 is from molecule B of ␣1LG4-5. Notably, only a few contacts in the LG4-LG5 interface are conserved in the two laminin isoforms. Near the Residues in most favored, additionally allowed, generously allowed, and disallowed regions (33). In both crystallographically independent molecules, two residues assume unfavorable main chain conformations: Lys 2791 , which is part of the heparin binding site, and Arg 2896 , whose peptide carbonyl oxygen receives a hydrogen bond from a buried lysine.  equivalent protein residues acting as metal ligands ( Fig. 2A) and water molecules observed in equivalent positions in both structures. Thus, we expect that calcium would readily occupy the metal sites of ␣1LG4-5 under more physiological conditions than those used for crystallization.
Characterization of Recombinant Laminin ␣1LG4-5 Proteins-To evaluate the contributions of basic residues to ligand binding by laminin ␣1LG4-5, we prepared WT ␣1LG4-5 as well as six mutants, in which selected basic residues within LG4 were replaced by alanine ( Table 2). For comparison, we also prepared the proteolytic E3 fragment from mouse Engelbreth-Holm-Swarm sarcoma tumor laminin-111, which has the same sequence as the recombinant WT ␣1LG4-5 construct (26). Recombinant ␣1LG4-5 proteins were purified from the conditioned medium of transfected 293 cells by anti-FLAG and heparin affinity chromatography. The typical yield of recombinant ␣1LG4-5 was greater than 4 g/ml harvested culture media. Little or no degradation was detected by SDS-PAGE (Fig. 3A), and the FLAG tag could be cleaved by enterokinase treatment (Fig. 3, B-D). Rotary shadow electron micrographs were prepared of representative recombinant proteins (WT and mutants A and G) and showed a monomeric appearance similar to that of the E3 fragment (Fig. 3E). A doublet structure, thought to reflect the LG domain pairs, was frequently appreciated in the images (Fig. 3E, insets). Size exclusion chromatography of the recombinant proteins showed a single peak eluting at the same position as E3 (data not shown). Laminin ␣1LG4-5 has three potential N-linked glycosylation sites. Treatment of secreted protein with peptide N-glycosidase F (an amidase that cleaves the glycosidic bond between the modified asparigine residue and the first GlcNAc moiety) or treatment of transfected cells with tunicamycin (an inhibitor of N-acetylglucosamine transferase) produced similar increases in migration on SDS-PAGE, reflecting loss of N-linked carbohydrate mass (data not shown).
Elution of ␣1LG4-5 Proteins from a Heparin Affinity Column-To determine relative binding affinities for heparin, we evaluated the NaCl elution behavior of the different ␣1LG4-5 proteins from a heparin affinity column. Fragment E3, recombinant WT ␣1LG4-5, enterokinase-treated WT ␣1LG4-5, and deglycosylated WT ␣1LG4-5 all eluted at 0.25 M NaCl ( Fig. 4; data not shown). All mutant ␣1LG4-5 proteins eluted at lower NaCl concentrations, in the following order: WT Ͼ J Ͼ I Ͼ D Ͼ A Ϸ G Ͼ A2. We note that the elution value of 0.25 M NaCl, which holds identically for recombinant and native ␣1LG4-5 protein, corresponds to an apparent K D of 22 nM as determined by the method of affinity co-electrophoresis (27).
Dystroglycan Binding-The ability of the ␣1LG4-5 proteins to bind ␣-DG was evaluated using two different assay formats. In a blot overlay assay (Fig. 5A), all ␣1LG4-5 mutants showed  Table 2). The E3 fragment purified from Engelbreth-Holm-Swarm sarcoma tumor laminin-111 is included for comparison. Recombinant   reduced binding to immobilized ␣-DG compared with wild type protein, with mutants A and J showing no detectable binding. A very similar result was obtained using an enzyme-linked immunosorbent assay-based solid phase assay with immobilized ␣-DG (Fig. 5B). All mutations decreased ␣-DG binding, in the following order: WT Ͼ A2 Ϸ G Ͼ D Ͼ I Ͼ A Ϸ J; mutants A and J had very low ␣-DG binding activity. The finding that A2 bound ␣-DG with higher affinity than A was surprising, given the greater loss of charge of the former compared with the latter. The sequences of A and A2 were reconfirmed by PCR of genomic DNA isolated from cells secreting the two laminin fragments (data not shown). We cannot offer a mechanistic explanation for the unexpected behavior of mutants A and A2 in the ␣-DG binding assays. Sulfatide Binding-Binding of ␣1LG4-5 proteins to immobilized galactosyl sulfatide was also examined. The interaction appeared to be specific for lipids bearing a sulfated sugar residue (Fig. 6A), since no binding was detected with the nonsulfated galactosyl-ceramide or with lipids bearing sulfated charges in the absence of a sugar moiety (sulfated cholesterol) or bearing phosphate (e.g. phosphatidyl serine) or sialic acid (GM1 ganglioside) moieties. Recombinant ␣1LG4-5 and its mutants bound to immobilized sulfatide with different halfmaximal binding values (Fig. 6B), reflecting substantial differences in affinities, in the following order: WT Ϸ A Ͼ A2 Ϸ G Ϸ J Ͼ D Ͼ I. Mutant I had almost no sulfatide binding activity. We fitted the sulfatide binding data with the simplest (i.e. a singlesite) model to obtain the values listed in Table 2 but noted that the fits seen for the strongest binders (WT ␣1LG4-5 and mutants A and G) were not good. We suspect that the poor fits are a reflection of the inherent difficulties in analyzing protein-lipid interactions in a solid phase binding assay, in which a large protein binds to a relatively small ligand presented as a densely packed array. A two-site model improved the fits (not shown), perhaps by allowing for some degree of negative cooperativity resulting from steric hindrance between proteins that limits access to the lipid head groups as saturation is approached. We suspect that such a phenomenon of dense sulfatide packing may also exist on cell surfaces that assemble basement membranes (e.g. Schwann cells).

DISCUSSION
Interactions of the distal LG4-5 portion of the laminin ␣1 chain with cellular receptors (heparan sulfate proteoglycans and ␣-DG) and plasma membrane components (sulfatides) are responsible for laminin anchorage during basement membrane assembly as well as for cell signaling (6,10,11). Previous mutagenesis studies have identified basic amino acid residues within LG4 contributing to ligand binding (14,28), but, in the absence of a ␣1LG4-5 structure, the precise structure of binding sites remained unknown. We now have determined the crystal structure of ␣1LG4-5 and analyzed the binding properties of this laminin ␣1 portion.
An electrostatic surface representation of the ␣1LG4-5 structure reveals a large, contiguous surface area of positive potential extending over both LG domains (Fig. 7A). Basic residues previously implicated in receptor binding (14,28) are clustered around the metal ion binding site in LG4 (Fig. 7B). A particularly striking feature is the spatial proximity of two basic sequences, 2719 RKR and 2791 KRK, which are located, respectively, at the start of ␤-strand C and in the H-I turn (Fig. 7C). Five of the six basic side chains are fully surface-exposed and available for receptor binding. The only exception is Arg 2792 , which donates two hydrogen bonds to main chain carbonyl groups in the long J-K loop, which in turn makes rather loose contacts with the body of the domain. Two other basic motifs, 2766 KGRTK and 2831 RAR, are located closer to the cleft between the LG4 and LG5 domains.
Regarding heparin binding, we found similar elution behavior to the earlier study of Andac et al. (14) for those mutants evaluated in common (supplemental Table 2). Hozumi et al. (28) used an energy-minimized homology model of ␣1LG4-5 to  gain insight into the interactions of LG4 with heparin/heparan sulfate proteoglycans and integrin ␣ 2 ␤ 1 , the latter a cryptic activity in intact laminin. These authors found heparin binding to LG4 to depend upon contributions from 2719 RKR and 2791 KRK and to a lesser extent from 2766 KGRTK, in full agreement with our heparin binding data.
Regarding the interactions with ␣-DG and sulfatide, there are some notable discrepancies between our data and those of Andac et al. (14) (the interaction with these molecules was not analyzed by Hozumi et al. (28)). We found that ␣-DG binding to 2766 KGRTK (mutant D) and 2791 KRK (mutant G) was only moderately reduced, whereas the same mutations led to a complete loss of ␣-DG in the study of Andac et al. (14). Conversely, we observed strong effects on sulfatide binding when we mutated 2831 RAR (mutant I) and 2858 KDR (mutant J), unlike Andac et al. (14). There are several possible reasons for these different results. First, we expressed ␣1LG4-5 proteins bearing mutations within LG4, because the tight structural linkage between the two LG domains argues that the domain pair behaves as a single structural unit. In contrast, the earlier study (14) looked at interactions of isolated ␣1LG4 proteins. Second, there were methodological differences that may have contributed to the observed differences. We found that drying and rehydrating the protein led to loss of activity and that it was preferable to detect an N-terminal epitope tag located away from the residues of interest rather than use an E3-specific antibody that could act as a blocking agent. Finally, a different ␣-DG preparation was used in our study compared with Andac et al. (14), and it is known that tissue-specific ␣-DG modifications can have an effect on laminin binding.
Summarizing all of the binding data in the light of our new crystal structure, we can safely assign the heparin/heparan sulfate binding site of laminin ␣1LG4 to the basic patch made up of 2719 RKR and 2791 KRK; of the 2766 KGRTK sequence, only the side chain of Lys 2766 is positioned to contribute to heparin binding (Fig. 7, B and C). The ␣-DG binding site appears to be formed by a larger, semicircular arrangement of basic side chains, with our new results assigning a key role to residues located away from the heparin binding site (i.e. 2831 RAR and 2858 KDR). The metal ion bound to LG4 (probably calcium under physiological conditions) is likely to be essential for ␣-DG, given that mutation of calcium ligands in LG4 of the related laminin ␣2 chain abolished ␣-DG binding (13) and that equivalent calcium sites in the LG domains of neurexin (29) and agrin (30) are critical for biological function. Finally, sulfatide binding in our hands is most strongly affected by mutations of basic residues on the upper face (view of Fig. 7B) of LG4. Two asparigine residues in laminin ␣1LG4, Asn 2714 and Asn 2811 , carry bulky glycan modifications, and the glycosylation status of these residues might influence ligand binding. However, whether and how glycosylation of laminin affects its interaction with cell surfaces has not been studied.  Fig. 2A and Table 2) onto a surface representation of ␣1LG4-5. The view direction is the same as in A. The locations of the metal ion binding site and two asparagines modified by glycosylation are also shown (DG, dystroglycan; hep, heparin; S, sulfatide). C, atomic details of the major basic region in LG4. A magnesium ion and selected side chains are shown as ball-and-stick models. Hydrogen bonds are indicated by dashed lines. Selected ␤-strands are labeled. Asn 2811 carries an N-linked glycan in WT ␣1LG4-5.
Laminin ␣2LG4-5 binds to the same cell surface molecules as ␣1LG4-5. Using a similar approach as in the present study, Wizemann et al. (13) found that heparin and sulfatide binding to ␣2LG4-5 were most strongly affected by mutation of 2870 KK in LG4 and 3088 KLTKGTGK in LG5, whereas ␣-DG binding was particularly dependent upon Arg 2803 and 2870 KK as well as upon the calcium ion in LG4. Remarkably, neither of these critical sequences correspond to the binding sites identified in ␣1LG4-5 ( Fig. 2A and Table 2). The lack of conservation of functionally important residues in two closely related proteins is unusual. We speculate that the predominance of electrostatic interactions in ligand binding by the laminin LG4-5 portion accounts for the poor conservation of binding sites. Evidently, it is sufficient to maintain the general basic character of the binding surfaces.
Our earlier findings do not support a role of heparin/heparan sulfates in basement membrane anchorage but instead argue for a prominent role of sulfated glycolipids with a signaling rather than an anchorage contribution arising from ␣-DG (11). Analysis of general and tissue-specific DG knockouts has revealed an essential role of DG for Reichert's membrane, but not muscle, peripheral nerve and other basement membranes, where it may function primarily as a signaling receptor (31). A recent analysis of cultured breast epithelial cells (32) revealed an anchoring activity for DG, raising the possibility that some cells employ DG as an anchor in a manner similar to sulfated glycolipids. In this regard, it will be informative to analyze cell surface assembly of heterotrimeric laminins bearing mutations that distinguish between ␣-DG and sulfated glycolipid binding (i.e. A ( 2719 RKR to AKA) and I ( 2831 RAR to AAA)). Based on the results presented here, we would predict reduced anchorage only of the latter mutant, whereas both mutants could be compromised in their ability to signal through ␣-DG.