Glycosylation Modulates Melanoma Cell α2β1 and α3β1 Integrin Interactions with Type IV Collagen*

Background: The influence of collagen glycosylation on integrin binding has not been studied previously. Results: Glycosylation affected α3β1 integrin binding more strongly than α2β1 integrin binding. Conclusion: Glycosylation modulated integrin/collagen interactions. Significance: If changes in collagen glycosylation occur in malignancy, then metastasis may be altered by these changes. Although type IV collagen is heavily glycosylated, the influence of this post-translational modification on integrin binding has not been investigated. In the present study, galactosylated and nongalactosylated triple-helical peptides have been constructed containing the α1(IV)382–393 and α1(IV)531–543 sequences, which are binding sites for the α2β1 and α3β1 integrins, respectively. All peptides had triple-helical stabilities of 37 °C or greater. The galactosylation of Hyl393 in α1(IV)382–393 and Hyl540 and Hyl543 in α1(IV)531–543 had a dose-dependent influence on melanoma cell adhesion that was much more pronounced in the case of α3β1 integrin binding. Molecular modeling indicated that galactosylation occurred on the periphery of α2β1 integrin interaction with α1(IV)382–393 but right in the middle of α3β1 integrin interaction with α1(IV)531–543. The possibility of extracellular deglycosylation of type IV collagen was investigated, but no β-galactosidase-like activity capable of collagen modification was found. Thus, glycosylation of collagen can modulate integrin binding, and levels of glycosylation could be altered by reduction in expression of glycosylation enzymes but most likely not by extracellular deglycosylation activity.

Despite the continuous advances made, patients suffering from advanced stage melanoma still face a rather bleak prognosis. Melanoma remains unpredictable in its biological behavior, with a high risk of recurrence and a 50% chance to develop metastases in lymph nodes after recurrence (1). To support advances in treatment and detection, attention has turned to locating and identifying key elements that could be utilized either as possible therapeutic targets or as potential markers for the different stages of the disease. Metastasis occurs via a series of linked steps (2,3). Tumor cells need to extravasate through the endothelium of the lymph node or blood vessel and become anchored in the local extracellular matrix (ECM) 3 to initialize secondary growth formation. The principal proteins of the ECM are laminins and collagens, the latter of which serves as a lattice network to create the cellular microenvironment. Type IV collagen is the most important structural component of the basement membrane (BM), which is a specialized form of the ECM.
Metastasis requires a subtype of tumor cells capable of enduring release from the primary tumor site and traveling through the lymphatic or vasculatory system while evading killer cells and/or platelet aggregation. This process requires an altered phenotype, which allows cells to quickly adhere to and release from the BM to promote a faster migration. These phenotypic changes are most easily defined by changed expression profiles of transmembrane receptors, such as integrins, responsible for the rolling motion cells display during migration (4). Integrins are the foremost contributors in mediating cell-cell and cell-ECM adhesions. Interactions between integrins and ECM proteins, such as collagen, are crucial for adherence, migration, and invasion of tumor cells (5).
Integrins are heterodimers of noncovalently associated ␣ and ␤ subunits. In vertebrates, there are 18 ␣ and 8 ␤ subunits that can assemble into 24 different receptors with unique binding properties and tissue distributions (6,7). Based on the structural characteristics of their ␣ and ␤ subunits, integrins are classified as either an I-domain or a non-I-domain, which signals a fundamentally different association mechanism between the two groups of receptor types and their respective ligands (6, 8 -12). I-domain-containing integrins preferentially bind to ligands via their I-domain, which is located on the ␣ subunit, providing a more approachable binding site and a more relaxed spatial arrangement, whereas non-I-domain integrins carry out binding partly by another portion of the ␣ subunit and partly by the ␤ subunit, which sterically places the ligand in a more confined space and makes the binding site less approachable and possibly less favorable (10,12). The I-domain contains a conserved MIDAS that binds divalent metal cations. Ligand binding alters the coordination of the metal ion and shifts the I-domain from a closed, resting state to an open, active conformation, which results in increased ligand affinity and promotes subsequent integrin activation (13). Four I-domain ␣ subunits (␣1, ␣2, ␣10, and ␣11) associate with ␤1 and form a distinct collagen-binding subfamily. The structural basis of the interaction of these integrins with their ligand is a Glu residue within a collagenous Gly-Phe-Hyp-Gly-Glu-Arg motif, providing the cation coordination (9).
The ␤ subunit plays an important role in ligand binding when ␣ subunits lack the I-domain. Integrin ␤ subunits contain an ion binding site homologous to MIDAS with a sequence motif of Asp-Xaa-Ser-Xaa-Ser. Mutation of any of these ion-coordinating residues within the ␤1, ␤2, ␤3, or ␤5 subunits ablated ligand binding to the respective integrins (14 -16). In ␣␤ integrin heterodimers, ligands bind to a crevice in the head domain between the ␣␤ subunit interface. In many cases the ligand interacts with the metal ion-occupied MIDAS located within the ␤ subunit and the propeller domain of the ␣ subunit (17). The ␣3␤1 integrin is a non-I-domain integrin that binds collagen and laminin (18 -20). More specifically, the ␣3␤1 integrin binds type IV collagen (21) and contributes to melanoma cell migration on this ligand (22,23). Thus, collagen receptors include both ␣ I-domain and non-I-domain integrins, implying differing binding mechanisms at play.
The ␣1(IV)531-543 sequence promotes adhesion and spreading of melanoma and other cell types (34). The ␣3␤1 integrin was identified as the receptor that binds to type IV collagen via this sequence (35). Interestingly, peptides promoted adhesion of melanoma cells in single-stranded and triple-helical conformation (34), thus providing the first evidence for existence of triple-helix-independent integrin binding sites within the collagenous domain.
An essential characteristic of native type IV collagen is the high level of Lys hydroxylation and subsequent Hyl glycosylation present in each ␣ chain. These post-translational modifications are carried out on almost all Lys residues present in type IV collagen, compared with the relatively low level (ϳ10%) of modification that is present on types I and II collagen (36,37). Prior research conducted in our laboratory indicated an altered affinity of a cell surface proteoglycan, CD44, toward binding sites in type IV collagen based on Hyl glycosylation (38). However, prior studies have not considered how Hyl glycosylation impacts integrin recognition of collagen. To specifically examine the possible modulation of integrin function by glycosylation, THPs with Lys substituted by glycosylated Hyl for Lys 543 and Lys 540 from the human ␣1(IV)531-543 gene sequence (␣3␤1 integrin-specific) and Lys 393 from the human ␣1(IV)382-393 gene sequence (␣2␤1 integrin-specific) were synthesized. These ligands were utilized to compare the promotion of melanoma cell adhesion, to observe the effects of ligand glycosylation. Cellular integrin concentrations were quantified utilizing immunocytochemistry. Alternative receptors were examined for recognition of glycosylated collagen. We also tested the possibility of melanoma cell modulation of collagen glycosylation by examining extracellular ␤-galactosidase-like activity.

MATERIALS AND METHODS
All chemicals were molecular biology or peptide synthesis grade and purchased from ThermoFisher Scientific (Waltham, MA) or Sigma-Aldrich.
Removal of side chain protecting groups and peptide-resin cleavage were carried out as reported previously (41) for 3 h in an atmosphere of ambient gas (Ar) using 7 ml of cleavage mixture (5% H 2 O, 5% thioanisole, 2.5% phenol, and 2.5% 1,2-ethanedithiol in TFA). Cleaved (glyco)peptides were precipitated in cold methyl tert-butyl ether, centrifuged, and lyophilized. Crude glycopeptides were subjected to Ac protecting group removal from sugar moieties using 0.1 M NaOH solution for 15 min. After this time the glycopeptide solution was neutralized with HCl and lyophilized.
Crude (glyco)peptides were purified using RP-HPLC on an Agilent 1260 Infinity series preparative HPLC equipped with a Vydac C18 column (15-20 m, 300 Å, 250 ϫ 22 mm). The elution gradient was 5-50% B in 60 min (where A was 0.1% TFA in H 2 O, and B was 0.1% TFA in acetonitrile), with a flow rate of 10 ml/min and detection at ϭ 220 and 280 nm. The HPLC fractions were combined, frozen, and lyophilized.
A portion of the biotinylated glycopeptide ␣1(IV)382-393(Gal) (2.5 mg) was subjected to selective N-acetylation of the Hyl residue. Briefly, the glycopeptide was dissolved in 1 ml of 50 mM ammonium carbonate solution, and 100 l of acetic anhydride was added. The reaction progress was monitored by RP-HPLC and MALDI-TOF MS, and upon completion the reaction mixture was frozen and lyophilized.

CD Spectroscopy
Peptides were dissolved in 0.5% acetic acid and equilibrated at 4°C (Ͼ24 h) to facilitate triple-helix formation. Peptide concentrations were determined using a Thermo Scientific NanoDrop 1000 (Waltham, MA) via absorbance at ϭ 280 nm, ⑀ Tyr ϭ 1490 M Ϫ1 cm Ϫ1 . Triple-helical structure was evaluated by near UV CD spectroscopy using a Jasco J-810 spectropolarimeter (Easton, MD) with a path length of 1 mm. Thermal transition curves were obtained by recording the molar ellipticity ([]) at ϭ 225 nm with an increase in temperature of 20°C/h in a range of 5-80°C. Temperature was controlled by a JASCO PTC-348WI temperature control unit. The THP melting temperature (T m ) was defined as the inflection point in the transition region (first derivative). The spectra were normalized by designating the highest [] 225 nm as 100% folded and the lowest [] 225 nm as 0% folded.

Cell Culture
The M14#5 human metastatic melanoma cell line was generously provided by Dr. Barbara Mueller (Torrey Pines Institute for Molecular Studies, La Jolla, CA). The WM-115 (primary melanoma), WM-266-4 (metastatic melanoma), and SK-MEL-2 (metastatic melanoma) cell lines were obtained from Ameri-can Type Culture Collection (Manassas, VA). For cell adhesion assays, cells were grown in EMEM with L-Gln (American Type Culture Collection) supplemented with 10% fetal bovine sera (HyClone), 50 units/ml penicillin, and 0.05 mg/ml streptomycin using 175-cm 2 flasks. At ϳ80% confluency cells were subcultured (1:3 ratio for SK-MEL-2, 1:6 for the other cell lines). For cell detachment 0.25% trypsin-EDTA solution was used (Invitrogen). Flasks were kept in a humidified incubator containing 5% CO 2 , and cells were passaged only eight times to avoid genetic drifts and other variations.

Immunocytochemistry
Biotinylated mouse antihuman integrin ␣3 subunit (CD49c, clone IA3, catalog number BAM1345) and biotinylated mouse anti-human integrin ␣2 subunit (CD49b, clone HAS3, catalog number BAM1233) mAbs were purchased from R&D Systems (Minneapolis, MN). Biotin-SP-conjugated ChromPure mouse IgG, whole molecule (product code 015-060-003), and mouse serum and the Cy3-conjugated (indocarbocyanine) streptavidin (product code 016-160-084) were purchased from Jackson ImmunoResearch Laboratories, Inc. (West Grove, PA). Immunocytochemistry experiments were carried out according to the direct detection method employing working concentrations based on the manufacturer's recommendation. The specificities of the integrin antibodies were previously established (42,43). Briefly, cells were plated in a 96-well Costar plate (Corning No. 3632; Fisher Scientific) at 20,000 -40,000 cells/ 100 l growth medium and allowed to become 70 -80% confluent overnight. Growth medium was removed, and cells were rinsed with 1ϫ PBS. Cells were fixed by incubation with 4% paraformaldehyde/PBS for 20 min at room temperature. The plate was blocked against nonspecific binding with 1% mouse serum, 1% BSA/PBS for 1 h at room temperature. Biotinylated anti-integrin ␣3 or ␣2 mAbs were diluted in 1% mouse serum, 1% BSA/PBS to 25 g/ml, added to the cells, and incubated for 1 h at room temperature. As a negative control, biotin-SP-conjugated mouse IgG was utilized, diluted to 25 g/ml in 1% mouse serum, 1% BSA/PBS. Following incubation with the antibodies and/or mouse IgG, cells were rinsed three times with 1 ϫ PBS and subsequently incubated for 1 h at room temperature in the dark with Cy3-conjugated streptavidin diluted to 2 g/ml in 1% mouse serum, 1% BSA/PBS. Background fluorescence was established by incubating the cells with the Cy3streptavidin solution for 1 h at room temperature in the dark. The plate was washed with 1ϫ PBS, and bound Cy3 was detected using the red filter on an Olympus IX70 inverted fluorescence microscope camera. Semi-quantitative image analyses were carried out using the Quantity One v.4.2.2 software (Bio-Rad) on quadruplets of 16 cell/image areas. Cells were counted in selected areas using photos taken at bright light at a visible range wavelength, and then, after switching on the red filter, the selected areas were photographed to determine fluorescence. Exposure times were 250, 400, or 666 ms. Because of either cellular accumulation or entrapped dye, some areas indicated artificially high levels of fluorescence. Those "bright spots" were not utilized for quantification purposes.

Adhesion Assays
The melanoma cell adhesion assay was performed as described previously (44). Peptide ligands (see Table 1) were dissolved in 40% ethanol in PBS and further diluted to desired concentrations in PBS. Pro-Bind TM 96-well plates (BD Biosciences, San Jose, CA) were coated with 100 l of desired peptide and incubated at 4°C overnight. Nonspecific binding sites were blocked with 2 mg/ml BSA in PBS for 2 h at 37°C (200 l/well). Cells were split 1-2 days before the experiment and were washed with PBS (without Ca 2ϩ and Mg 2ϩ ) and then released with Accutase (Invitrogen). Cells were then washed and resuspended to 75,000 cells/well in adhesion medium (20 mM HEPES, 2 mg/ml albumin in RPMI 1640 medium). Cell suspension was added to the plate (100 l/well), and plates were incubated for 60 min at 37°C. All nonattached cells were removed by washing three times with warm adhesion medium. Adherent cells were counted using CellTiter-Glo luminescent cell viability assay (Promega, Madison, WI) and quantitated with a Synergy H4 Hybrid multimode microplate reader (BioTek, Winooski, VT).

AlphaScreen Assay
The AlphaScreen assay was performed accordingly to recently published methodology (45). His-tagged galectin-3 (1.25 l) and biotin-ASF (1.25 l) were added to wells containing varying concentrations of ␣1(IV)382-393 and ␣1(IV)382-393(Gal) THPs (2.5 l, 0 -1 mM final concentration) in optimized assay buffer (25 mM HEPES, 100 mM NaCl, 0.05% Tween 20, pH 7.4). Because of the low solubility of the peptides at higher concentrations, final solutions contained 1% DMSO. The nonbiotinylated ASF was used as a control. The final concentration of His-tagged galectin-3 was 100 nM and biotin-ASF 5 nM. The reaction mixture was incubated for 1 h at room temperature, and then 5 l of nickel-chelate Acceptor and 5 l streptavidin-conjugate Donor beads were simultaneously added to a final concentration of 25 g/ml. Incubation proceeded for 1 h in the dark at room temperature, and the assay plate was subsequently read at 22°C in the AlphaScreen mode on a Synergy H4 Hybrid multimode microplate reader. AlphaScreen signal counts (cps) versus log [ligand] (M) were expressed as the mean of five replicate measurements. The IC 50 values were obtained by nonlinear regression analysis using the Graph Pad Prism 5.04 software.

Molecular Modeling
To generate a model of ␣1(IV)382-393(Gal) THP interacting with the ␣2␤1 integrin, a homology modeling approach was utilized. Briefly, starting structure 1DZI was used as a template (9). Collagen-like peptide residues were mutated manually in PyMOL (46) and UCSF Chimera software (47). Residues were mutated using Dunbrack backbone-dependent rotamer library (48). Charges were added using AMBER ff12SB force field, and for unknown residues (Gal) were calculated using AM1-BCC model (49). Mutated residues were subjected to minimization using the antechamber program (50) included in Chimera.
The ␣3␤1 integrin model was built using the ␣5␤1 integrin x-ray crystallographic structure (51). The ␣5 subunit was replaced with ␣3 by homology modeling using the Modeler program (52) and subsequent minimization steps of ␣3/␤1 interface residues using the antechamber module of the Chimera package. Next, docking of ␣1(IV)531-543 singlestranded peptide was performed using Autodock Vina (53). Because the geometry of the ␣1(IV)531-543 peptide backbone is unknown, we have selected three different combinations of / torsion angles within the polyproline type II family, namely / of Ϫ60°/150°, Ϫ70°/160°, and Ϫ75°/175°. Three separate docking runs were performed and compared. In each docking the peptide backbone was kept rigid, and side chains contained rotatable bonds. The docking site was chosen arbitrarily and contained the top of the ␣3␤1 interface along with the MIDAS site in the ␤1 subunit.

␤-Galactosidase Activity Assessment
Isolated ␤-Galactosidase with Synthetic Substrate-Escherichia coli ␤-galactosidase (EC 3.2.1.23, grade VIII) was purchased from Sigma-Aldrich. Enzymatic assays were performed in 100 mM phosphate buffer, pH 7.2, supplemented with 10 mM MgCl 2 and 5 mM 2-mercaptoethanol (added freshly before the assay). The enzyme activity was determined using the fluorogenic substrate MUG (Sigma-Aldrich) at excitation ϭ 365 nm and emission ϭ 445 nm. ␤-Galactosidase activity was measured using the Synergy H4 Hybrid Multi-Mode Microplate Reader over a period of 1 h, with occasional shaking to assure even substrate distribution.   Table 1). The Hyl(O-Gal) residue was incorporated manually, whereas other amino acids were incorporated using an automated synthesizer under microwave conditions. The N termini of all (glyco)peptides were modified with n-dodecanoic acid to ensure triple-helical character of the (glyco)peptides and to facilitate their attachment to plastic surfaces during the adhesion assay (55,56).

Synthesis of Galactosylated Hyl
All peptides were characterized by RP-HPLC and MALDI-TOF MS (Table 1), with appropriate purity and mass values observed. The triple-helical character of the peptides was analyzed by CD spectroscopy, in the range of ϭ 250 -180 nm (Fig.  1). All peptides had characteristic triple-helical spectra, with a positive peak at ϭ 222 nm and a negative peak at ϭ 205 nm. Thermal transition curves were obtained by recording molar ellipticity ([]) at ϭ 225 nm as a function of increasing temperature (Fig. 2). The melting point (T m ) was defined as the inflection point in the transition region (Table 1). All peptides exhibited good triple-helix stability, with T m values ranging from 37 to 45°C. Galactosylation of Hyl had a destabilizing effect on the triple-helix, because the glycopeptides had T m values 6 -8°C lower than the corresponding nonglycosylated peptides. This could be caused by the presence of racemic D,L-5-Hyl used in the present study.
Immunocytochemistry-The cell surface concentrations of the ␣2 and ␣3 subunits of the ␣2␤1 and ␣3␤1 integrins were evaluated for all melanoma cell lines by immunocytochemistry. Image analysis for each cell line for both receptor subunits (Fig.  3A) provided semiquantitative numerical information on cellular integrin concentrations. Numerical values reflected relative fluorescence intensities of the same cell number, normalized by area measured, and by subtraction of the autofluorescence of the cells (Fig. 3B). All four cell lines showed abundant levels of both integrin subunits, with somewhat higher levels for the ␣2 subunit compared with the ␣3 subunit (Fig. 3B). The primary cell line (WM-115) had lower levels of the ␣2 and ␣3 subunits compared with the metastatic cell lines. Overall, integrin levels were sufficient to investigate melanoma-ligand interactions.
Galectin-3 Interaction with THPs-Galectin-3 is known to mediate cell binding to galactosylated ligands (57). Thus, we examined the possibility that glycosylation may result in a switch of receptor binding, in that galectin-3 may mediate binding to glycosylated THPs. Binding was tested using a galectin-3 AlphaScreen assay (45) with ␣1(IV)382-393 and ␣1(IV)382-393(Gal) THPs along with ASF as a control (Fig. 5). Only nonspecific binding of the THPs with galectin-3 was observed (IC 50 in the millimolar range) with no difference between glycosylated and nonglycosylated ligands. Thus, galectin-3 does not appear to be involved in binding to these ligands.
Molecular Modeling of integrin⅐THP Complexes-To further investigate the influence of glycosylation of ␣1(IV)382-393 and ␣1(IV)531-543 THPs on binding to their respective integrins, molecular modeling was performed. Models for both integrin⅐ THP complexes were prepared (Figs. 6 and 7).
For the ␣2␤1 integrin, a model was generated using the x-ray crystallographic structure of the ␣2 I-domain in complex with a THP (PDB: 1DZI) (9). The Hyl 393 glycosylation site is located four residues away from the Glu 389 responsible for binding to the MIDAS of the I-domain. It appears that the Hyl 393 site is at the outer interface of the integrin interaction site, and thus mono-glycosylation does not impact binding significantly (Fig.  6). When the disaccharide-containing residue (Glc-Gal)Hyl is considered, glycosylation of Hyl 393 will interfere with binding to the ␣2 integrin because the binding site is masked by the sugar moiety (data not shown). Thus, the effect of glycosylation on ␣2␤1 integrin binding to type IV collagen may very well depend on whether monosaccharide or disaccharide glycosylation has occurred.
The recognition site for the ␣3␤1 integrin is much different from ␣2␤1 because there is no homology between the peptide ligand sequences. Also, the ␣1(IV)531-543 sequence does not have the classic -Gly-Xaa-Yaa-repeat required for stabilization of the triple-helix, but rather has a noncollagen-like insertion/ break region of n ϭ 7 (58). The break region within ␣1(IV)531-543 is anticipated to have some strand separation, based on prior studies of break regions within THPs (59,60). Conversely, the ␣1(IV)531-543 break region does not possess either Pro or Hyp and nor do the flanking regions, and thus the break is not anticipated to significantly affect the triple-helical structure of the remainder of the THP (61,62). In addition, the presence of hydrophobic residues within the break region further aids in the stability of the THP (62).
For molecular modeling studies, it was assumed that the ␣1(IV)531-543 region possesses a polyproline type II-like structure (Fig. 7A). Single-stranded ␣1(IV)531-543 was used for molecular docking purposes. Three different sets of polyproline type II-like torsion angles, / ϭ Ϫ60°/150°, Ϫ70°/160°, and Ϫ75°/175°, were considered, and the peptide backbone was kept rigid upon docking (Fig. 7B). Previous studies revealed the  crucial role of both Asp residues of the ␣1(IV)531-543 sequence for binding to the ␣3␤1 integrin (34). Taking this into account, another set of docking was performed where the Gly 541 and/or Gly 544 residues were allowed flexibility around the C␣ atom. In each docking result the lowest energy ligands were obtained by having the Asp 542 residue in close proximity of the MIDAS (Fig. 7C). The docking studies revealed that glycosylation of Hyl 540 and Hyl 543 occurred right in the middle of key electrostatic/metal binding interactions and thus would dramatically impact the binding of the ␣3␤1 integrin to the ␣1(IV)531-543 THP.
␤-Galactosidase Activity Evaluation-Knowing that glycosylation could negatively impact integrin binding to type IV collagen, we next examined whether melanoma cells could modulate O-glycosylation of the microenvironment. Initially, E. coli ␤-galactosidase was tested with the ␣1(IV)382-393(Gal) THP. Different enzyme concentrations (1-100 units) were compared using incubation at 37°C for up to 72 h. At certain time points aliquots were taken and subjected to RP-HPLC/MALDI-TOF MS analysis. No hydrolysis of Gal by E. coli ␤-galactosidase was observed (data not shown). Activity of the ␤-galactosidase was confirmed with MUG fluorogenic substrate (data not shown). The influence of the ⑀-amino group of Hyl on ␤-galactosidase activity was then tested. Prior studies indicated that ␤-galactosidase was effective in cleaving the Gal moiety from (Gal)Hyl only if the ⑀-amino group was acetylated (63)(64)(65). No cleavage of the sugar moiety was observed when ␣1(IV)382-393(Gal) THP, in either nonacetylated or acetylated (␣1(IV)382-393(Gal)-Ac THP) form, was treated with ␤-galactosidase (data not shown). Although this result is in contrast to the results obtained by Spiro (63)(64)(65), the prior study tested the enzyme activity on isolated (Gal)Hyl moiety only.
Whole cell assays were next performed. Two cell lines were selected: primary (WM-115) and metastatic (WM-266 -4) melanoma obtained from the same patient. Different cell culture media (EMEM and OptiMEM with or without HI-FBS) were also tested. In the first experiment, cells were grown for 24 or 48 h, and then MUG was added and activity monitored for 1 h (Fig. 8). There was no activity present in EMEM and OptiMEM media (Fig. 8, A and C). In contrast, media containing HI-FBS possessed some ␤-galactosidase activity (Fig. 8, B  and D). In the case of OptiMEM ϩ HI-FBS, the activity was associated with the presence of serum, because control (without cells) also exhibited this activity (Fig. 8D). However, both WM-115 and WM-266-4 exhibited some activity toward MUG   All peptides were clearly identified during RP-HPLC/ MALDI-TOF MS analysis of aliquots (Fig. 9). Under the employed conditions, neither of the glycosylated THPs was deglycosylated, as confirmed by MS analyses (data not shown). It was also possible to perform MALDI-TOF MS analyses of crude aliquots (without HPLC separation), and these results confirmed that glycopeptides were unmodified (Fig. 10). The results with cells in suspension were identical, in that (a) ␤-galactosidase activity could be observed with CUG and inhibited by phenylethyl ␤-D-thiogalactopyranoside (a selective ␤-galac-tosidase inhibitor) and (b) no degalactosylation of the THPs was found (data not shown).

DISCUSSION
Tumor cells interact with type IV collagen at the site of extravasation through distinct cellular receptors, including the ␣1␤1, ␣2␤1, and ␣3␤1 integrins. Integrins contribute to the ability of melanoma cells to migrate, invade, and metastasize to secondary sites (6,66). Because they play a pivotal role in both inside-out and outside-in signaling (67), integrins affect most aspects of cell behavior, including shape, motility, differentiation, proliferation, and survival. Thus, it is not surprising that these receptors are also known to be differentially expressed in tumors relative to normal cells, depending on tumor type and stage of progression (4,6,66,68,69).
The types and concentration of cellular receptors has long been a focus for finding indicators of disease progression. Testing of 10 different human melanoma cell lines found that the ␣2, ␣3, and ␤1 integrin subunits were expressed on all of them (70). It is interesting to note that the ␣3 subunit showed the highest expression profile, with subtle differences in regards to the invasive profile of a given cell line. The same variation was observed for the ␣2 and ␤1subunits, showing higher expression levels in more invasive tumor types, although the overall concentrations were somewhat lower than that of the ␣3 subunit. In light of these prior results, the role of the associative relationships between type IV collagen and ␣2␤1 and ␣3␤1 integrins with regards to melanoma progression was examined here.
Prior research conducted in our laboratory indicated an altered affinity of a cell surface proteoglycan, CD44, toward binding sites in type IV collagen based on glycosylation (38).
This result led to us to consider whether hydroxylation/glycosylation of Lys residues modulates ligand binding by other receptors, such as integrins. Previous studies have not considered how Hyl glycosylation impacts on integrin recognition of collagen. To specifically examine the possible modulation of integrin function by glycosylation, THPs with Lys substituted by glycosylated Hyl for Lys 393 from the human ␣1(IV)382-393 gene sequence (␣2␤1 integrin-specific), and Lys 543 and Lys 540 from the human ␣1(IV)531-543 gene sequence (␣3␤1 integrin-  specific) were synthesized. These ligands were utilized to compare the promotion of cell adhesion.
Collagen glycosylation was found to modulate integrin binding. The integrins were affected differently, with only modest inhibition of ␣2␤1 binding (with the primary melanoma cell line being least affected) and significant inhibition for ␣3␤1 interaction.
Molecular modeling studies of the ␣2␤1 integrin in complex with the glycosylated ligand ␣1(IV)382-393(Gal) THP indicated that (Gal)Hyl 393 was at the outer interface of the integrin interaction site, and thus galactosylation only slightly diminished binding (Fig. 6). However, for the cell lines tested herein, the effects of ligand glycosylation on binding varied, with the smallest effect on the primary melanoma cell line (WM-115) and the largest effect on the highly metastatic M14#5 cell line (Fig. 4, top right). Because the relative amount of ␣2␤1 integrin on each cell surface was similar (Fig. 3B), variations in activity could be due to interactions of the glycosylated ligand with different activation states of the ␣2␤1 integrin or with different cell surface complexes that incorporate the ␣2␤1 integrin.
The simulations of interactions of the ␣3␤1 integrin with ␣1(IV)531-543 THP indicated that the relatively flexible 531-543 region is capable of binding across the ␣3␤1 interface with Asp 542 binding to the MIDAS motif within the ␤1 subunit (Fig.  7). Because the MIDAS motif is located in the ␣␤ interface groove on ␤ subunit, complexation of the Mg 2ϩ cation by the Asp 542 side chain is a primary driving force for binding the peptide to the receptor (Fig. 7C). This model is with agreement with previously published data identifying Asp 542 as a critical residue for ␣3␤1 integrin binding to ␣1(IV)531-543 (34) and consistent with several integrin x-ray crystallographic structures including ␣IIb␤3 (71), ␣5␤1 (51), and ␣v␤3 (17). Glycosylation within the ␣1(IV)531-543 sequence results in significant inhibition of integrin binding, mostly likely because of the proximity of the galactosylated residues (Hyl 540 and Hyl 543 ) to the key electrostatic/metal binding interactions via Asp 542 . Although inhibition caused by glycosylation is an uncommon phenomenon, the presence of sialic acid on sialoglycoprotein P2B reduced the binding of tumor cells to type IV collagen (72,73).
The reduced binding of integrins caused by ligand glycosylation presents a possible "cryptic sites" mechanism by which tumor cells may invade the BM (38). In the native, glycosylated state, regions within type IV collagen may have minimal interaction with receptors such as the ␣3␤1 integrin and CD44. After tumor cells bind to type IV collagen (presumably via the ␣2␤1 integrin), cell surface or secreted glycosidases could liberate the collagen-bound carbohydrates. This process would expose cryptic sites for interaction with the ␣3␤1 integrin, CD44/CSPG, and/or other cell surface receptors.
Extracellular removal of carbohydrates could also occur under other circumstances. Numerous bacterial pathogens bind to collagen (74), with binding occurring at several sites within the triple helix (75). The collagen binding protein from Staphylococcus aureus has been identified as CNA, and its mode of binding has been determined (76). Upon binding to collagen, bacteria could secrete ␤-galactosidases that facilitate deglycosylation. Reduced glycosylation could impact integrin interactions, as well as other collagen-binding proteins. The endocytic collagen receptor urokinase plasminogen activator receptor-associated protein mediates glycosylated collagen turnover (77). DDR1 binds to type IV collagen (78), and this binding may be mediated by ligand glycosylation (79).
For dynamic modification of collagen to occur, carbohydrates would need to be removed from type IV collagen extra- cellularly. An age-dependent increase in ␤-galactosidase activity (at pH 6) has been reported (97), and a cell surface inactive ␤-galactosidase functions as an elastin and laminin receptor (98). An ␣-glucosidase that removes glucose from (Glc-Gal)Hyl has been characterized (99). However, we found no evidence that deglycosylation could be performed extracellularly, because triple-helical glycopeptides were not substrates for purified ␤-galactosidase or melanoma cells. Thus, although a deglycosylation/cryptic site mechanism provides interesting speculation, it should also be noted that glycosylation is not 100% efficient; collagen O-glycosylation sites are found as mixtures of Lys, Hyl, (Gal)Hyl, and (Glc-Gal)Hyl (36,89,100,101). Thus, receptor interaction may just occur with the subpopulation of type IV collagen that does not contain carbohydrate.
Alternatively, tumor cell binding may be mediated by differential glycosylation that is tissue-specific. For example, LH3 is found extracellularly in kidney, spleen, and muscle (91). Thus, ␣2␤1 and ␣3␤1 integrin binding may be regulated by different levels of type IV collagen glycosylation as determined by LH3 activity. It is also possible that Hyl glycosylation enzymes are decreased in cancer, in similar fashion to ␤3-N-acetylglucosaminyltransferase-1 (102), and this in turn could enhance receptor association with type IV collagen.
The present study has focused on cellular interactions with glycosylated collagen models. When one considers the BM in vivo, variations in collagen glycosylation are generally unknown. If the levels of glycosylation are indeed modulated, tumor interactions with and response to the BM may be altered or simply compensated for by adherence to other BM ligands (i.e. laminin). Future studies may consider the complexities of cell-BM interactions and the role of glycosylation within that microenvironment.