Recombinant Human Laminin-5 Domains

Human laminin-5 fragments, comprising the heterotrimeric C-terminal part of the coiled-coil (CC) domain and the globular (G) domain with defined numbers of LG subdomains, were produced recombinantly. The α3′ chain with all five LG subdomains was processed proteolytically in a manner similar to the wild-type α3 chain. Conditions were established under which the proteolytic cleavage was either inhibited in cell culture or was brought to completion in vitro. The shorter chains of the laminin-5CCG molecule, β3′and γ2′, produced in a bacterial expression system associated into heterodimers, which then combined spontaneously with the α3′ chains in vitro to form heterotrimeric laminin-5CCG molecules. Only heterotrimeric laminin-5CCG with at least subdomains LG1–3, but not the single chains, supported binding of soluble α3β1 integrin, proving the coiled-coil domain of laminin-5 to be essential for its interaction with α3β1 integrin. The N-glycosylation sites in wild-type α3 chain were mapped by mass spectrometry. Their location in a structural model of the LG domain suggested that large regions on both faces of the LG1 and LG2 domains are inaccessible by other proteins. However, neither heterotrimerization nor α3β1 integrin binding was affected by the loss of N-linked glycoconjugates. After the proteolytic cleavage between the subdomains LG3 and LG4, the LG4–5 tandem domain dissociated from the rest of the G domain. Further, the laminin-5CCG molecule with the α3′LG1–3 chain showed an increased binding affinity for α3β1 integrin, indicating that proteolytic processing of laminin-5 influences its interaction with α3β1 integrin.

As a member of the laminin superfamily (for review, see Refs. 1 and 2), laminin-5 is characterized by the rod-like ␣-helical coiled-coil (CC) 1 domain (3) of all three laminin-5 chains, ␣3, ␤3, and ␥2, and by the C-terminal globular or G domain of the ␣3 chain, which consists of five homologous LG subdomains (4).
LG domains consist of two sheets of six and seven antiparallel ␤-strands, which form the convex and concave face of the domain (5,6). Within the G domain, the LG1-3 subdomains form a three-bladed propeller, from which the LG4 -5 tandem domain protrudes (4). In the elongated spacer sequence connecting LG3 and LG4, a proteolytic cleavage occurs. Although plasmin, bone morphogenetic protein-1 (BMP-1) and its homolog mammalian tolloid (7)(8)(9)(10) are putative candidates, the enzymes involved in this processing have not been identified. Nevertheless, this proteolytic modification of laminin-5 has been proposed to play a key role in keratinocyte cell migration during wound healing (11) as well as in tumor progression and metastasis (12,13).
Laminin-5 differs from the other members of the laminin superfamily not only in its distinct chains, ␣3, ␤3, and ␥2, and its biochemical processing, but also in its tissue distribution, its subcellular localization, and its cellular receptors. Laminin-5 is the major component of anchoring filaments, which are an essential part of the dermal-epidermal junction (14). They link epithelial cells to the underlying connective tissue via the basal membrane. This sheet-like extracellular matrix consists of a lamina densa, a dense meshwork containing laminins other than laminin-5 and type IV collagen, which are interconnected by nidogen (15). Laminin-5 plays an important part in the assembly of the basal membrane (2,16). Genetically or immunologically induced defects in laminin-5 function cause severe disorders in the basal membrane, manifested as skin blisters, which may be lethal (1,2,17).
In a previous paper, we described the production of a soluble ␣ 3 ␤ 1 integrin, a specific receptor for laminin-5 (27). Here, we report the production of recombinant laminin-5CCG molecules, which comprise 15 heptad repeats of the CC domain and the G domain with different numbers of LG subdomains. Using these laminin-5 fragments, the molecular interaction of ␣ 3 ␤ 1 integrin with its high affinity ligand was studied, bringing new evidence for the location of the ␣ 3 ␤ 1 integrin binding site and its confor-thentic signal sequence, the HA epitope sequence, a (Gly) 3 spacer, and the C-terminal 15 heptad repeats (amino acids 1074 -1193) of the human ␥2 chain (30) was generated by standard PCR using vector pPCR1.3-52 (provided by G. Meneguzzi) and two pairs of partially overlapping primers 5Ј-GTG GCT AGC ACC ATG CCT GCG CTC TGG  CTG GGC TGC TGC CTC TGC TTC TCG CTC CTC CTG-3Ј (forward  primer LC2-F1 with the NheI site underlined) and 5Ј-GGG ACG TCG  TAT GGG TAG GGC CTG GCC CGG GCT GCG GGC AGG AGG AGC  GAG AAG CAG AGG-3Ј (reverse primer LC2-R1), as well as 5Ј-CCC  TAC CCA TAC GAC GTC CCA GAC TAC GCT GGC GGC GGC CAG  ATG GTG ATT ACA GAG GCC-3Ј (forward primer LC2-F2) and 5Ј-CAC GGG CCC TCA CTG TTG CTC AAG AGC CTG GGT ATT GTA GCA GCC-3Ј (reverse primer LC2-R2 with the ApaI site underlined). The ␤3Ј and ␥2Ј chains encoding cDNA fragments were inserted together with a NotI-NheI-flanked IRES sequence into the eukaryotic expression vector pRC-CMV (Stratagene) leading to the vector pL5B3-IRES-L5C2.
The cDNA fragments encoding the signal sequence-free ␤3Ј and ␥2Ј chains with an N-terminally fused GST domain were generated by standard PCR using the vector pL5B3-IRES-L5C2 and primer pairs containing the restriction sites that were necessary to insert the constructs into the bacterial expression vector pGEX-4T-3 (Amersham Biosciences). The primers for the pGEX-GST-␤3Ј-construct were 5Ј-GTC GAC TTC GAA GCC CAG AAG ATG G-3Ј (forward primer LB3-F3 with the SalI site underlined) and 5Ј-GGC TCG AGT CAC TTG CAG GTG GCA TAG-3Ј (reverse primer LB3-R3 with the XhoI site underlined). The primers for the pGEX-GST-␥2Ј-construct were 5Ј-GTC GAC TAC CCA TAC GAC GTC CCA G-3Ј (LC2-F3 with the SalI underlined) and 5Ј-CTC GAG TCA CTG TTG CTC AAG AGC C-3Ј (reverse primer LC2-R3 with the XhoI site underlined).
Establishing ␣3Ј Chains Producing 293 Cell Clones-Human epithelial kidney cells (HEK) 293 were grown in Dulbecco's modified Eagle's medium and F-12 medium (Invitrogen) supplemented with 10% fetal calf serum, 100 units/ml penicillin G, 0.1 mg/ml streptomycin sulfate, and 1.75 mM L-glutamine. The pIRES-neo2 vectors containing the ␣3Ј chain-encoding cDNAs were transfected into 293 cells using the Trans-Fast TM transfection reagent (Promega, Mannheim, Germany) according to the manufacturer's instructions. Stably transfected clones were selected in the presence of 0.5 mg/ml Geneticin. Positive clones were established by screening for their production of ␣3Ј chains in a sandwich ELISA.
Isolation of Recombinantly Expressed ␣3Ј Chains-The 293 cell clones producing the different ␣3Ј chains were grown in serum-free medium for 2 days. To avoid processing of the ␣3ЈLG1-5 chain, 25 M BMP-1 isoenzyme inhibitor-1 (Fibrogen) and 1 g/ml aprotinin were added. Cell supernatants were harvested and supplemented with 1 mM phenylmethylsulfonyl fluoride (ICN Biomedicals) and 1 g/ml each aprotinin, leupeptin, and pepstatin. ␣3Ј chains were precipitated with 60% ammonium sulfate, dialyzed against Tris-buffered saline, pH 8.0, and loaded onto a Ni 2ϩ -nitrilotriacetic acid superflow column (Qiagen). After washing the resin with 20 mM Tris, 300 mM NaCl, 20 mM imidazole, pH 8.0, the ␣3Ј chains were eluted with an imidazole steps of 40, 100, and 250 mM in the same buffer. After dialysis against PBS, pH 7.4, protein concentration was determined by the BCA method (Pierce). Purity of the ␣3Ј chains was evaluated in SDS-PAGE and Coomassie Brilliant Blue staining.
Isolation of GST-␤3Ј and GST-␥2Ј Chains-Bacteria transformed with the expression vectors pGEX-GST-␤3Ј or pGEX-GST-␥2Ј were grown at 37°C and induced for 4 h with 1 M isopropyl-1-thio-␤-Dgalactopyranoside (BioTech Trade & Service GmbH, St. Leon-Rot, Germany). Bacteria were then centrifuged, resuspended in PBS containing 1 g/ml each aprotinin, leupeptin, and pepstatin, and lysed by freeze-thaw cycling, by incubation with 1 mg/ml lysozyme (Serva, Heidelberg, Germany) for 30 min, and by sonication three times for 30 s. The GST-␤3Ј-or GST-␥2Ј-containing inclusion bodies were centrifuged and dissolved in PBS containing 8 M urea, 30 mM ␤-mercaptoethanol, and 1 g/ml each aprotinin, leupeptin, and pepstatin. After centrifugation, the solubilized GST-tagged laminin chains were dialyzed against PBS, pH 7.4. They were isolated by affinity chromatography on glutathione-Sepharose (Amersham Biosciences) according to the manufacturer's instructions. Protein concentration and purity were tested by BCA (Pierce) and SDS-PAGE.
Enzymatic Modifications of the Laminin-5CCG Fragments-The ␣3ЈLG1-5 chain was processed proteolytically with plasmin (Roche Diagnostics) at an enzyme:substrate ratio of 20 units/mg in 50 mM Tris-HCl buffer, pH 8.5, at 26°C overnight. Deglycosylation of the ␣3Ј chains was achieved with PNGase F (Roche Diagnostics) at an enzyme: substrate ratio of 5 units/g at 37°C overnight. To remove the GST moiety, the GST-␥2Ј chain was treated with thrombin (Sigma) for 1 h at 26°C at an enzyme:substrate ratio of 10 units/mg. The digestion was stopped by addition of 1 mM phenylmethylsulfonyl fluoride. The GSTfree ␥2Ј chain was found in the flow-through of a GSH column.
Mapping the Sites of N-Glycosylation within the Laminin-5 ␣3 Chain-The wild-type laminin-5 was isolated from supernatant of SCC 25 cells as described previously (27). After electrophoretic separation of the three laminin-5 chains, the ␣3 chain was excised from the gel and subsequently digested with PNGase F and trypsin as described in Ref. 31. The N-glycans and tryptic peptides were extracted from the gel and analyzed by MALDI and/or electrospray mass spectrometry.
Nano-electrospray was carried out using a quadrupole time-of-flight mass spectrometer (Micromass, Manchester, UK) in positive ion mode (ESI(ϩ)). A Z-spray atmospheric pressure ionization source was used with the source temperature set to 80°C and a desolvation gas (N 2 ) flow rate at 75 liters/h. Peptide preparations were dissolved in 5-10 l of 0.1% trifluoroacetate and acetonitrile (1:1) and were applied to nanospray capillaries. The capillary tip was set to a potential of 1.1 kV, and the cone voltage was 40 V. For collision-induced dissociation experiments double and triple charged precursor ions were selected in the quadrupole analyzer and fragmented in the collision cell using collision gas (argon) at 2.5 ϫ 10 Ϫ5 mbar and collision energies of 20 -30 eV. Acquisition and analysis of data were performed with the MassLynx Windows NT PC data system. NaI was used as mass standard for calibration, and mass accuracy of all measurements was within 0.1 atomic mass unit. Peptide and glycan maps by MALDI mass spectrometry were obtained using Reflex III mass spectrometer (Bruker, Bremmen, Germany) and 2,5-dihydroxybenzoic acid as matrix.
Sandwich ELISA to Detect Laminin-5 ␣3Ј Chains and Heterotrimeric Laminin-5CCG Molecules-To screen ␣3Ј chain-producing 293 cell clones and to prove the heterotrimeric nature of the self-assembled laminin-5CCG molecules, sandwich ELISAs were performed. The oligohistidine-tagged ␣3Ј chains or the GST-bearing GST-␤3Ј/␥2Ј heterodimers were captured with immobilized antibodies against the oligohistidine tag and the GST moiety, respectively, which were coated at 5 g/ml to microtiter plates at 4°C overnight. After blocking with 1% bovine serum albumin in PBS, ammonium sulfate-concentrated supernatants of transfected clones, 25 mM octyl-␤-D-glycopyranoside-extracted cell pellets, or solutions of heterotrimeric laminin-5CCG molecules at 2 g/ml were added to the wells. After washing the wells, bound ␣3Ј chains and heterotrimeric laminin-5CCG molecules were detected by 10 g/ml biotinylated BM165 followed by 1:1000 alkaline phosphatase-conjugated extravidin (Sigma). Enzymatic conversion of p-nitrophenyl phosphate by alkaline phosphatase was measured at 405 nm in an ELISA reader (Dynatech, Alexandria, VA).
Binding of ␣ 3 ␤ 1 Integrin to Laminin-5CCG Molecules-Microtiter plates were precoated with capturing antibodies against the oligohistidine tag or against GST (10 g/ml in TBS, pH 7.4, 2 mM MgCl 2 , washing buffer) at 4°C overnight. After two washing steps with the washing buffer and after blocking with 1% bovine serum albumin solution in washing buffer, laminin-5CCG molecules or its individual chains (2 g/ml in 1% bovine serum albumin and washing buffer) were captured with precoated antibodies. Soluble ␣ 3 ␤ 1 integrin was added in 1% bovine serum albumin in washing buffer, supplemented with 1 mM MnCl 2 and a 3-fold molar excess of the integrin-activating mAb 9EG7, or in the presence of 10 mM EDTA (nonspecific background) for 2 h at room temperature. After two washing steps with 50 mM HEPES/NaOH, pH 7.5, 150 mM NaCl, 2 mM MgCl 2 , 1 mM MnCl 2 , bound ␣ 3 ␤ 1 integrin was fixed with 2.5% glutaraldehyde solution in the same buffer and detected with a rabbit antiserum against the integrin ␤ 1 subunit and an alkaline phosphatase-conjugated anti-rabbit IgG antibody, as described previously (27).

Recombinant Production of ␣3Ј
Chains-cDNAs were generated encoding laminin-5 ␣3Ј chains, which share the authentic signal sequence, an oligohistidine tag, and the C-terminal 15 heptad repeats of the CC domain but differ in the numbers of LG subdomains. Thus, the constructs contained the subdomains LG1-5, called ␣3ЈLG1-5, the subdomains LG1-3, called ␣3ЈLG1-3, the subdomains LG1-2, called ␣3ЈLG1-2, or LG1 only, called ␣3ЈLG1, with the prime indexing our shorter recombinant constructs as opposed to the wild-type ␣3 chain (Fig.  1A). All ␣3Ј chains were produced and secreted by transfected HEK 293 cells (Fig. 1B), which do not produce endogenous ␤3 and ␥2 chains. The amount of secreted ␣3Ј chains in the culture medium drastically declined with decreasing numbers of LG subdomains (Fig. 1B), this being particularly evident for the constructs ␣3ЈLG1-2 and ␣3ЈLG1. When protease inhibitors were added to the culture medium of cells transfected with these ␣3Ј constructs, the yields of proteins in the supernatants clearly increased (Fig. 1B, inset), indicating that they were chains were not in adequate amounts for further characterization. Hence, only the isolation of the larger ␣3Ј chains, ␣3ЈLG1-5 and ␣3ЈLG1-3, was pursued (Fig. 2). Similar to the wild-type ␣3 chain, the G domain of the ␣3ЈLG1-5 construct was specifically cleaved, albeit not completely (lane ␣3ЈLG1-5 in Fig. 2), presumably within the sequence connecting the LG3 and LG4 subdomains. This proteolytic processing of the ␣3ЈLG1-5 chain was only prevented entirely by a combination of the serine protease inhibitor aprotinin and the BMP-1 isoenzyme inhibitor-1, which is a selective inhibitor for a small family of astacin-like metalloproteases including BMP-1 and the mammalian tolloids. Addition of both protease inhibitors allowed the isolation of the non-processed ␣3ЈLG1-5 chains, referred to as np-␣3ЈLG1-5, with an apparent molecular mass of 135 kDa (lane np-␣3ЈLG1-5 in Fig. 2). On the other hand, the partial processing of ␣3ЈLG1-5 chain in the cell culture system (lane ␣3ЈLG1-5 in Fig. 2) could be brought to completion by in vitro digestion of the isolated chain by plasmin, resulting in the fully processed ␣3ЈLG1-5 chain, called fp-␣3ЈLG1-5, with an apparent molecular mass of 90 kDa (lane fp-␣3ЈLG1-5 in Fig.  2). After proteolytic cleavage in the spacer sequence between LG3 and LG4, the LG4 -5 tandem domain dissociated from the rest of the ␣3Ј chain (see below and Fig. 7). After its dissociation the tandem domain could be detected in the plasmin digest of the ␣3ЈLG1-5 chain by immunoblotting with a specific polyclonal antiserum against LG4 -5 domain (lane fp-␣3ЈLG1-5 in Fig. 2C), whereas it seemed to be completely degraded under cell culture conditions (lane ␣3ЈLG1-5 in Fig. 2C), showing that proteases other than plasmin are responsible for the degradation of the dissociated LG4 -5 tandem repeat in the cell culture medium, although plasmin cleaves the ␣3ЈLG1-5 chain in a way similar to the set of proteases that are responsible for ␣3ЈLG1-5 chain processing in vivo. The ␣3ЈLG1-3 chain lacking the LG4 -5 tandem domain a priori was as stable as the ␣3ЈLG1-5 construct. Its C terminus Gln-1337, being designed to be the natural cleavage site of laminin-5, is located within the spacer sequence between the LG3 and LG4 subdomains (7) (Fig. 1A).
N-Glycoconjugates of the Laminin-5 ␣3Ј Chain-The ␣3Ј chains produced by HEK 293 contained N-linked glycoconjugates, which were enzymatically removed by treatment with PNGase F (Fig. 3). Deglycosylation did not increase sensitivity of the ␣3ЈLG1-5 chain to plasmin (Fig. 3). To map the Nglycosylation sites, the ␣3 chain of wild-type laminin-5 was separated by SDS-PAGE. The excised band was subsequently digested with PNGase F and trypsin. The released N-glycans and tryptic fragments were analyzed by mass spectrometry. Because the PNGase F cleaves the N-glycosidic bond yielding a glycosamine and aspartic acid, tryptic fragments containing a formerly occupied N-glycosylation site showed a mass shift of 0.99 Da. Thus, we identified the following Asn residues as anchor sites for N-glycan side chains: Asn-645 (CC domain), Asn-882 and 964 (LG1), Asn-1108 and 1131 (LG2). No peptides could be found for the two missing potential N-glycosylation sites at Asn-745 (CC domain) and Asn-1325 (spacer sequence between LG3 and LG4) (Fig. 4). Conspicuously, the LG3 subdomain lacks any N-glycosylation. A MALDI mass spectrometry glycan map (not shown) demonstrated that the N-glycans of the ␣3 chain were heterogeneous, belonging to both the complex and the high mannose types.
Production of ␤3Ј and ␥2Ј Chains of Laminin-5CCG and Their Heterodimerization-Lacking glycosylation sites, both ␤3Ј and ␥2Ј chains of laminin-5CCG were produced in a bacte- FIG. 3. The ␣3LG1-5 chain is specifically cleaved but not degraded by plasmin, irrespective of its N-glycosylation. Glycosylated and PNGase F-digested ␣3ЈLG1-5 chains were incubated without or with plasmin as detailed under "Materials and Methods." 1-g samples were separated by SDS-PAGE in a 7.5-15% polyacrylamide gel and transferred to a nitrocellulose membrane. On the Western blot membrane, proteins were detected with mAb BM165 and a peroxidaseconjugated anti-mouse IgG secondary antibody, using a chemoluminescent substrate. The two major bands indicate the unprocessed (135 kDa) and processed (90 kDa) a3Ј chain and their deglycosylated forms at 112 and 72 kDa, respectively. The molecular masses of standard proteins are marked.
rial expression system as GST fusion proteins. The N-terminal GST moiety was fused via a thrombin-cleavable spacer sequence to a tag sequence and the C-terminal 15 heptad repeats of the human laminin ␤3 and ␥2 chains. As tag sequence, we utilized the biotin ligase recognition sequence and the HA epitope in the ␤3Ј and ␥2Ј chains, respectively. In the transformed bacteria, the GST-␤3Ј chain was found exclusively in inclusion bodies, whereas a minor fraction of the GST-␥2Ј chain was also found in the cell lysate. Both chains were extracted from inclusion bodies under denaturing conditions and isolated by affinity chromatography on a glutathione column. Because each of the two chains contains a single cysteine residue, purification yielded not only monomeric GST-␤3Ј and GST-␥2Ј chains with apparent molecular masses of 42 and 44 kDa, respectively, but also (GST-␤3Ј) 2 and (GST-␥2Ј) 2 homodimeric molecules with apparent molecular masses of 84 and 88 kDa (Fig. 5A). Presumably because of a higher resistance against proteolysis, the GST-␥2Ј chain was obtained in higher yield (15 mg/liter) than the GST-␤3Ј chain (10 mg/liter).
To avoid steric hindrance of chain assembly, attempts were made to remove the GST moieties by thrombin cleavage. Although thrombin cleaved the GST-␥2Ј chain at the designated site between the GST domain and the HA-tagged ␥2Ј-chain (20 kDa) (Fig. 5A), digestion of the GST-␤3Ј chain resulted in degradation of the ␤3Ј chain (data not shown). Therefore, the GST-␤3Ј chain was left uncleaved, and the GST moiety was used as tag sequence for the ␤3Ј chain.
Self-assembly of Laminin-5CCG Molecules-When the GST-␤3Ј/␥2Ј heterodimer and any of the ␣3Ј chains were mixed under nondenaturing conditions, heterotrimeric laminin-5CCG molecules formed spontaneously. The presence of all three laminin-5 chains within the self-assembled molecule was proven by sandwich ELISAs, in which a signal was only obtained when all three chains were associated in one molecule. In the first type of sandwich ELISA, the ␣3Ј chains were captured with a mAb against the oligohistidine tag, and subsequently, the GST-␤3Ј or HA-tagged ␥2Ј chains of the trimeric laminin-5CCG molecules were detected with pAbs against the GST moiety and with a mAb against the HA epitope, respectively (Fig. 6A). Detection signals for the HA-tagged ␥2Ј chain were less prominent than for the GST-␤3Ј chain, presumably because a mAb instead of a polyclonal antiserum was used.

FIG. 4. Localization of N-glycosylation sites within the wild-type ␣3 chain sequence of laminin-5 (28), as determined by mass spectrometry.
After in-gel PNGase F treatment and trypsin digestion of the wild-type ␣3 chain, its peptides were identified either by MALDI mass spectrometry (dark gray shaded), nano-electrospray mass spectrometry (frame with white background), or both methods (light gray shaded). N-Glycosylation sites proven to be occupied are marked by arrows above the primary sequence of the ␣3 chain (28). Recombinant ␣3Ј chains of laminin-5CCG start at Ile-689. Domain borders of LG1, LG2, and LG3 are also marked.
In the second sandwich ELISA, heterotrimerization was proven by capturing the GST-␤3Ј/␥2Ј heterodimer with pAb against the GST moiety. As shown in Fig. 6, B and C, the ␣3Ј chains in the heterotrimeric laminin-5CCG molecules were detected by the biotinylated mAb BM165, recognizing a sequence epitope within the LG1 domain (25). A specific signal was only observed when an ␣3Ј chain has associated with a GST-␤3Ј/␥2Ј heterodimer to the trimeric laminin-5CCG. Neither the GST-␤3Ј/␥2Ј heterodimer alone (Fig. 6B) nor the ␣3Ј chains alone (data not shown) gave signals. However, the signals for the laminin-5CCG molecules containing different ␣3Ј chains varied considerably (Fig. 6B). Presumably, this is caused by differing accessibility of the epitope of the mAb BM165 in the various ␣3Ј chains.
After enzymatic deglycosylation, the ␣3ЈLG1-5 chain associated with the GST-␤3Ј/␥2Ј heterodimer to the same extent as glycosylated chain (Fig. 6C), showing that the N-glycoconjugates of the ␣3Ј chains do not alter the self-assembly of heterotrimeric laminin-5. An additional sandwich ELISA, in which laminin-5CCG molecules with different ␣3Ј chains were captured and detected with pAb against the LG4 -5 domain, showed that the fp-␣3ЈLG1-5 chain after plasmin cleavage is not recognized by pAb against the LG4 -5 domain similar to the ␣3ЈLG1-3 chain, proving that proteolytic cleavage between the LG3 and LG4 subdomains of laminin-5 results in the dissociation of the LG4 -5 tandem domain from the rest of the laminin-5 G domain (Fig. 7). ␣ 3 ␤ 1 Integrin Binds to the Self-assembled Laminin-5CCG Heterotrimers-Binding assays with soluble ␣ 3 ␤ 1 integrin were performed using laminin-5CCG molecules or their single chains captured with precoated antibodies against their tags. Capturing the laminin-5CCG chains by their N-terminal tags located their G domain in a highly accessible orientation toward the soluble ␣ 3 ␤ 1 integrin. No binding of soluble ␣ 3 ␤ 1 integrin was observed when the single ␣3Ј chains or the GST-␤3Ј/␥2Ј heterodimer was used as substratum (Fig. 8). In contrast, when all three chains came together in the heterotrimeric laminin-5CCG molecules, binding signals reached values comparable with the wild-type laminin-5 control. The integrin binding was specific and dependent on divalent cations because EDTA reduced the binding signals to nonspecific background levels. The integrin-activating mAb 9EG7 was added in the binding tests to increase the affinity of ␣ 3 ␤ 1 integrin toward the laminin-5CCG molecules (Fig. 8). In the absence of 9EG7, ␣ 3 ␤ 1 integrin showed similar, albeit weaker binding (data not shown). The laminin-5CCG molecule with the three subdomains LG1-3 was sufficient for ␣ 3 ␤ 1 integrin binding because it showed a binding comparable with that of the laminin-5CCG molecule with the ␣3ЈLG1-5 chain, indicating that the ␣ 3 ␤ 1 integrin binding site must be located within the first three LG subdomains of the laminin-5CCG. Furthermore, heterotrimerization of the ␣3Ј chain with the ␤3Јand ␥2Ј chains is indispensable for ␣ 3 ␤ 1 integrin binding to the laminin-5CCG molecule.
To investigate the effect of proteolytic processing of the laminin-5 G domain on ␣ 3 ␤ 1 integrin binding, we titrated the different laminin-5CCG molecules with soluble ␣ 3 ␤ 1 integrin (Fig.  9). Titration curves of laminin-5CCG molecules containing either nonprocessed or fully plasmin-processed ␣3ЈLG1-5 chains coincided, showing that the plasmin cleavage between the LG3 and LG4 subdomains and the subsequent dissociation of LG4 -5 tandem domain from the G domain does not affect ␣ 3 ␤ 1 integrin binding. However, the laminin-5CCG molecule containing the ␣3ЈLG1-3 chain with the natural C terminus Gln-1337 (7) showed a remarkable increase of affinity toward ␣ 3 ␤ 1 integrin. This increase of ␣ 3 ␤ 1 integrin affinity toward laminin-5CCG with the ␣3ЈLG1-3 chain was also observed when the titrations were performed in the presence of integrin-activating antibody 9EG7 (32) (Fig. 9). The binding affinities of ␣ 3 ␤ 1 integrin to the heterotrimeric laminin-5CCG molecules (data not shown) were in the same range or even slightly stronger than the affinities, which we had measured for ␣ 3 ␤ 1 integrin interaction with wild-type laminin-5 (27).
Role of N-Glycosylation in ␣ 3 ␤ 1 Integrin Binding-To study the role of N-glycan chains of laminin-5 on integrin binding, we investigated the binding of soluble ␣ 3 ␤ 1 integrin to the laminin-5CCG containing a deglycosylated ␣3ЈLG1-3 chain. The deglycosylated laminin-5CCG was recognized by the soluble ␣ 3 ␤ 1 integrin to the same extent as the glycosylated form (Fig. 10), demonstrating that its N-glycan side chains neither contribute to nor mask the ␣ 3 ␤ 1 integrin binding site. This proves that, despite being highly glycosylated, laminin-5 interacts with ␣ 3 ␤ 1 integrin by a direct protein-protein interaction. DISCUSSION To analyze the interaction of laminin-5 with one specific integrin on the molecular level without any other interfering cell surface receptors and without any cell-secreted laminin-5processing proteases, we developed a cell-free system in which the binding of soluble ␣ 3 ␤ 1 integrin to recombinant laminin-5 fragments with clearly defined G domains was assayed. Attempts to produce laminin-5 and its fragments recombinantly have been reported earlier (23,33) with both groups producing individual LG subdomains of human or rat laminin-5 ␣3 chain as GST fusion proteins in a bacterial expression system. Although a cell adhesion site within the LG2 or LG3 subdomain was described in both cases, reducing agents, which disrupt the disulfide bridge-stabilized LG domain, were used during the isolation of the GST-tagged LG subdomains. Furthermore, none of these constructs contained the heterotrimeric CC domain, which we proved here essential for ␣ 3 ␤ 1 integrin binding to laminin-5. Recombinant expression of human laminin-5 was achieved in a eukaryotic expression system, using either HT1080 cells, in which transfected ␣3 chain constructs com-bine intracellularly with endogenous ␤3 and ␥2 chains (22), or second in HEK 293 cells, which were transfected with fulllength cDNAs of all three laminin-5 chains (34). However, studies with the full-size laminin-5 molecule are difficult to interpret because this multidomain protein interacts with dif-  ␥2 heterodimer (B and C) is captured. The mAb against the oligohistidine tag (A) or the pAb against the GST moiety (B and C) was coated onto microtiter plates (5 g/ml) and incubated with the following: 5 g/ml laminin-5CCG molecule containing ␣3ЈLG1-5 chain (A), 2 g/ml laminin-5CCG molecules containing differently processed ␣3Ј chains or 2 g/ml GST-␤3Ј/␥2Ј heterodimer alone (B), or with the laminin-5CCG molecules containing either glycosylated or deglycosylated ␣3ЈLG1-5 chains (C). Bound laminin-5CCG molecules were then detected with goat pAb against GST (vertically striped bar in A) and rat mAb against the HA epitope (horizontally striped bar in A), thus proving the presence of GST-tagged ␤3Ј chain and HA-tagged ␥2Ј chain in the laminin-5CCG molecule. As secondary antibodies, alkaline phosphatase-conjugated anti-goat antibodies and anti-rat antibodies, respectively, were used. The GST-␤3Ј/␥2Ј-associated ␣3Ј chains (B and C) were detected with the biotinylated mAb BM165 and alkaline phosphatase-conjugated extravidin. The ELISA was developed as detailed under "Materials and Methods. "   FIG. 7. Proteolytic processing of the G domain results in the dissociation of the LG4 -5 domain. Laminin-5CCG molecules containing the differently processed forms of the ␣3ЈLG1-5 chains or the ␣3ЈLG1-3 chain were captured with goat pAb against the GST tag onto microtiter plates. Bound LG4 -5 domain was detected with rabbit pAb against the human LG4 -5 tandem domain and alkaline phosphataseconjugated secondary antibodies.
FIG. 8. ␣ 3 ␤ 1 integrin binds to heterotrimeric laminin-5CCG molecules but not to the single ␣3 chains or the GST-␤3/␥2 heterodimer. Laminin-5CCG molecules, containing either ␣3ЈLG1-5 or ␣3ЈLG1-3 chains and the ␣3Ј chains alone, each at 5 g/ml, were immobilized to microtiter plates precoated with mAb against the oligohistidine tag (5 g/ml). The GST-␤3Ј/␥2Ј heterodimer was captured with immobilized pAb against GST (5 g/ml). As positive control, 5 g/ml wild-type laminin-5 was immobilized to the microtiter plates directly. Soluble ␣ 3 ␤ 1 integrin (75 nM) was added either in the presence of 1 mM Mn 2ϩ and 300 nM mAb 9EG7 (diagonally striped bars) or in the presence of 10 mM EDTA (open bars) for 2 h at room temperature. Bound integrin was detected as described under "Materials and Methods." Binding values were corrected for nonspecific binding measured in the presence of EDTA to obtain specific integrin binding (cross-hatched bars). Binding signals were normalized to ␣ 3 ␤ 1 integrin binding to wild-type-laminin-5. Means Ϯ S.D. of duplicate determinations are shown. ferent cell surface molecules, which trigger different cellular functions or affect other cell adhesion molecules. Furthermore, the previous studies were done with intact cells, which contain various receptors recognizing its G domain, e.g. the integrins ␣ 3 ␤ 1 , ␣ 6 ␤ 1 , and ␣ 6 ␤ 4 (2,18). Keratinocytes, lacking these integrins, are still able to interact with the laminin-5-rich basal membrane (36) in part because of additional cell contacts, such as those between the ␣ 2 ␤ 1 integrin and the short arm of the laminin-5 ␥2 chain (35), or syndecan-2 and syndecan-4 with the heparin binding region in the laminin-5 subdomain LG4 (21). The investigation of such cell contacts with laminin-5 is complicated further because it is processed by proteases (9,13), the concentrations and activities of which depend on the integrinmediated cell contact to laminin-5 itself (37). However, as we show here, the proteolytic processing of the G domain also changes the interaction of ␣ 3 ␤ 1 integrin and perhaps other laminin-5-binding integrins with their substrate and may therefore affect cell attachment studies.
For our assays, the chains of the laminin-5CCG were expressed individually in vivo and isolated and assembled in vitro. The ␣3Ј chains were expressed in HEK 293 cells, and although lacking endogenous ␤3 and ␥2 chains, the single ␣3Ј chain was secreted by the transfected cells into the culture medium. No self-association of the ␣3Ј chains was observed (data not shown), and compared with other models for laminin secretion (38,39), the ␣3Ј chain obviously does not need to heterotrimerize with the ␤3 and ␥2 chains to be secreted.
However, the secretion of ␣3Ј chains in transfected HEK 293 cells, which do not express laminin-5 or any of its chains endogenously, is low compared with its recombinant expression in SCC25, which also produce the endogenous molecule (39). In our system, only the ␣3Ј chains comprising at least the three subdomains LG1-3 were expressed at isolatable levels, in part this may be because of a greater stability against proteolytic degradation. This could be explained by the structural model of the laminin G domain (4 -6), where it is proposed that the first three subdomains LG1-3 form a three-bladed propeller, from which the LG4 -5 tandem domain protrudes. Although the spacer sequence connecting the subdomains LG3 and LG4 is amenable to proteolytic cleavage, the intact three-bladed propeller of LG1-3 seems to be a compact, less accessible structure. Further, loss of any of the three LG domains would conceivably destabilize the three-bladed propeller structure and result in improper folding and intracellular degradation as well as open it to extracellular proteolysis. Our constructs, ␣3ЈLG1-5 and ␣3ЈLG1-3, were comparatively stable, and the correct proteolytic cleavage of the full-length G domain without any nonspecific degradation indicated a proper folding. Several proteases have been proposed as candidates for this processing of the laminin-5 G domain in vivo (8 -10). We observed that such processing of the ␣3ЈLG1-5 chain in the HEK 293 cell culture was entirely inhibited only by a combination of inhibitors to both serine proteases and metalloproteinases but not by either of the inhibitors alone. This suggests that the cleavage of the ␣3Ј chain within the G domain involves at least one serine proteinase and one metalloproteinase. Furthermore, we demonstrated that the ␣3ЈLG1-5 chain is cleaved by plasmin, one of the candidates for G domain processing (9).
The short ␤3Ј and ␥2Ј chains were produced in a bacterial expression system. Similarly to the analogous ␤1 and ␥1 chain fragments of laminin-1 (40), the ␤3Ј and ␥2Ј chains of laminin-5 FIG. 9. Titration of laminin-5CCG molecules containing nonprocessed np-␣3LG1-5, fully processed fp-␣3LG1-5, and ␣3LG1-3 chains with soluble ␣ 3 ␤ 1 integrin. Laminin-5CCG molecules containing either nonprocessed np-␣3ЈLG1-5 (squares), fully processed fp-␣3ЈLG1-5 (triangles), or ␣3ЈLG1-3 (circles) chains were immobilized to microtiter plates that had been precoated with pAb against GST (5 g/ml). Soluble integrin was added to the plates either in the presence of 1 mM Mn 2ϩ -ions (open symbols) or 1 mM Mn 2ϩ ions along with the integrin-activating mAb 9EG7 (gray symbols), or in the presence of 10 mM EDTA (nonspecific background). Bound ␣ 3 ␤ 1 integrin was detected as described under "Materials and Methods." Nonspecific background values measured in the presence of EDTA were subtracted from all values, which were then normalized to the OD value at saturation to obtain the saturation yield. Means Ϯ S.D. of duplicate measurements are shown.
FIG. 10. N-Glycoconjugates of laminin-5CCG do not affect ␣ 3 ␤ 1 integrin binding. Laminin-5CCG molecules containing the ␣3ЈLG1-3 chain (2 g/ml), which had been treated either with or without PNGase F, were immobilized to microtiter plates precoated with pAb against GST moiety (5 g/ml). Soluble ␣ 3 ␤ 1 integrin (12.5 nM) was added either in the presence of 1 mM Mn 2ϩ ions and 300 nM mAb 9EG7 (diagonally striped bars) or in the presence of 10 mM EDTA (open bars). Bound integrin was detected as described under "Materials and Methods." Binding signals were corrected for nonspecific binding measured in the presence of 10 mM EDTA to obtain specific binding (cross-hatched bars). Means Ϯ S.D. of duplicate measurements are shown.
showed a very strong preference toward heterodimerization in agreement with in vivo results (39). In vitro, the GST-␤3Ј/␥2Ј heterodimer is joined spontaneously by the ␣3Ј chains under native conditions to form the heterotrimeric laminin-5CCG molecule. Hence, laminin-5 is a self-assembling molecule, similar to laminin-1 (41)(42)(43). For laminin-1, Deutzmann and coworkers (41) as well as Sung and co-workers (42) have shown that the C-terminal parts of all three chains are able to form a heterotrimeric laminin-1 fragment comprising the CC and G domains. Further, all three chains of laminin-1 were required for cell attachment predominantly mediated by ␣ 6 ␤ 1 integrin. For laminin-5, we demonstrated here that the C-terminal 15 heptad repeats representing 105 amino acids from the 580amino-acid-long CC domain are sufficient to allow specific heterotrimerization. Furthermore, the formation of the coiled-coil domain is a prerequisite for ␣ 3 ␤ 1 integrin binding, an observation that helps explain the loss of cell attachment upon heat denaturation of laminin-5 (44). Hence, formation of the CC domain may change the conformation within the LG1-3 domain, thus activating the ␣ 3 ␤ 1 integrin binding site. Alternatively, the necessity of the CC domain for ␣ 3 ␤ 1 integrin binding may be explained by a direct interaction of this domain with the integrin, suggesting that ␣ 3 ␤ 1 integrin recognizes a noncontiguous binding site comprising regions of both the G domain and the CC domain. Further studies with recombinant laminin-5 fragment molecules containing even shorter regions of the CC domains will be necessary to discriminate between these possibilities.
Neither heterotrimerization nor ␣ 3 ␤ 1 integrin binding was affected by the N-glycosylation of the ␣3 chain in laminin-5. Each of the subdomains LG1 and LG2 bear two N-glycan side chains. Alignment of the amino acid sequences of the laminin ␣3 and ␣2 chains and modeling of the ␣3ЈLG modules according to the structure of ␣2LG domains (5, 6) indicated that the two N-linked carbohydrate side chains are located on each face of the two ␣3 subdomains. Considering the steric requirement of N-linked carbohydrate side chains, large surfaces of the two subdomains LG1 and LG2 should not be accessible for the ␣ 3 ␤ 1 integrin. Interestingly, there would be no steric hindrance for the ␣ 3 ␤ 1 integrin to both faces of the LG3 domain, which lacks any N-glycosylation sites. In contrast to other cell interactions with laminin-5 (21), we could show that the interaction of ␣ 3 ␤ 1 integrin with laminin-5 is merely mediated by protein-protein interaction. Although our studies do not prove N-glycoconjugates in laminin-5 to be functional in its chain assembly or in its ␣ 3 ␤ 1 integrin binding, the N-glycan side chains protect laminin-5 from proteolytic degradation (data not shown) and may regulate its proteolytic processing in vivo, thus affecting its ␣ 3 ␤ 1 integrin binding indirectly.
Here we demonstrated that the binding affinity of ␣ 3 ␤ 1 integrin to laminin-5 is regulated by the proteolytic processing of the G domain. Cleavage of the ␣3Ј chain by different proteases, among them plasmin, a potential candidate for the in vivo processing of laminin-5, leads to a dissociation of LG4 -5 domain from the rest of the ␣3Ј chain. However, because plasmin is unable to degrade the dissociated domain, other proteases must be responsible for this in vivo. Cleavage and subsequent loss of the LG4 -5 tandem domain do not affect ␣ 3 ␤ 1 integrin binding to plasmin-processed laminin-5. In contrast, the laminin-5CCG molecule containing the ␣3ЈLG1-3 chain with the natural C terminus Gln-1337 (7) was bound by ␣ 3 ␤ 1 integrin with a remarkably higher affinity. So, although plasmin maybe involved in laminin-5 processing, it does not seem responsible for the affinity-increasing effect of laminin-5 processing in vivo. This may emphasize the role of the other proteases, especially BMP-1 and its homologs (7-10) in the increase in the ␣ 3 ␤ 1 integrin affinity for laminin-5.
These observations may bring new insights into the cellular processes of wound healing. After skin injuries, keratinocytes at the wound edges secrete and deposit laminin-5 with a predominantly nonprocessed G domain (45). Its proteolytic processing and the subsequent loss of the LG4 -5 tandem domain (45) lead to an increase in binding affinity of ␣ 3 ␤ 1 integrin, altering integrin signaling and keratinocyte migration into the wound bed (9,11,46).
In conclusion, by using well characterized laminin-5CCG molecules with clearly defined numbers of LG subdomains, we demonstrated (i) that heterotrimerization of the three laminin-5 chains is a prerequisite for ␣ 3 ␤ 1 integrin binding, although the subdomains LG1-3 of the ␣3 chain contain the putative ␣ 3 ␤ 1 integrin binding site; (ii) that N-glycoconjugates of the G domain do not contribute directly to ␣ 3 ␤ 1 integrin binding, and hence, their location may sterically hinder ␣ 3 ␤ 1 integrin access to LG1 and LG2, but not to LG3; and (iii) that the proteolytic processing within the G domain increases ␣ 3 ␤ 1 integrin binding affinity. The analysis of this interaction on the molecular level allows new insights in ␣ 3 ␤ 1 integrin-mediated cellular processes, such as cell attachment and migration, which are triggered by laminin-5 and play a key role in wound healing and tumor invasion.