Evidence for a Direct Interaction between the Tumor Suppressor Serpin, Maspin, and Types I and III Collagen*

Maspin (mammary serineprotease inhibitor) was originally identified as a tumor suppressor protein in human breast epithelial cells and is a member of the serine proteases inhibitor (serpin) superfamily. It inhibits tumor cell motility and angiogenesis, and although predominantly cytoplasmic, it is also localized to the cell surface. In this study we have investigated the use of the yeast two-hybrid interaction trap to identify novel maspin targets. A target human fibroblast cDNA library was screened, and the α-2 chain of type I collagen was identified as a potential interactant. Binding studies with isolated proteins showed interaction between recombinant maspin and types I and III collagen but not other collagen subtypes, a profile strikingly similar to mouse pigment epithelium-derived factor (caspin), which is similarly down-regulated in murine adenocarcinoma tumors and is a potent inhibitor of angiogenesis. Kinetic analysis using an IAsys resonant mirror biosensor determined the dissociation constant of maspin for collagen type I to be 0.63 μm. Further two-hybrid interactions with maspin truncation constructs suggest that collagen binding is localized to amino acids 84–112 of maspin, which aligns with the collagen-binding region of colligin. A direct interaction between exogenous or cell surface maspin and extracellular matrix collagen may contribute to a cell adhesion role in the prevention of tumor cell migration and angiogenesis.

Mammary serine protease inhibitor (Maspin) 1 was identified by subtractive hybridization as a candidate tumor suppressor protein in normal mammary epithelial cells (1). A number of findings support its role as a tumor suppressor: levels of maspin expression show an inverse correlation with progression of breast cancer; mammary carcinoma cells transfected with maspin showed reduced tumor growth and metastasis in nude mice (1,2); the addition of recombinant maspin (rMaspin) decreased the migration potential of breast and prostate tumor cells across a reconstituted basement membrane (3); and more recently, maspin has been shown to inhibit angiogenesis by blocking in vitro migration of endothelial cells and by in vivo inhibition of rat cornea neovascularization (4).
Maspin belongs to the serpin (serine proteases inhibitor) superfamily of proteins and can be included in the ovalbumin subfamily, which appear to be mostly intracellular (5,6). The maspin gene has been localized to chromosome 18q21. 3 where it is clustered with PAI-2 and the squamous cell carcinoma antigens (7). Initial studies suggested that maspin may be a non-inhibitory serpin (8), and although an unusual biphasic inhibitory and activatory effect with single chain tissue-type plasminogen activator has been reported (9), this does not appear to be a target for maspin in mammary gland extracts (10). The predicted P1 residue of maspin is arginine, and trypsin cleavage of the reactive site loop demonstrated that the loop sequence was necessary for inhibition of cell invasion (3). In contrast, inhibition of angiogenesis was found to be independent of reactive site loop integrity (4).
Tissue distribution studies have shown expression of maspin in other organs including placenta, prostate, and small intestine. Subcellular localization studies show maspin to be predominantly a soluble cytoplasmic protein (95%), but it is also associated with secretory vesicles and is present at the cell surface (11). Functional studies suggest that this surfacebound maspin is responsible for inhibition of cell invasion (12). The addition of rMaspin to MDA-MB-435 breast cancer cells was found to alter the integrin profile by inducing ␣ 5 and ␣ 3 integrins and down-regulating other integrins. This was accompanied by a decreased ability to migrate through a fibronectin matrix and may provide a mechanism whereby maspin can suppress the invasive phenotype of these cells (13). More recently, studies of the human cornea suggest a role for maspin in regulation of stromal cell adhesion to the extracellular matrix. Late-passage stromal cells that had lost the ability to produce maspin responded to exogenous recombinant maspin as measured by increased cell adhesion to fibronectin, type I collagen, type IV collagen type 1, and laminin (14).
An increasing number of serpins have been found to bind ligands other than target proteases. Interaction with heparin resulting in enhanced serpin inhibitory activity has been well documented for antithrombin III and heparin cofactor II (15). Inhibition of cell migration by PAI-1 is mediated by specific binding of vitronectin (16), which competes with vitronectin VNR ␣ v ␤ 3 receptor binding and blocks smooth muscle cell migration. This binding is independent of plasminogen activator inhibition, but formation of a complex with uPA decreases affinity for vitronectin and restores cell migration (17). The dual localization of some partially secreted ovalbumin-type serpins such as PAI-2 and maspin suggests the possibility of more than one function or ligand. This is borne out in the case of PAI-2, which in addition to inhibiting uPA can protect cells against tumor necrosis factor-mediated apoptosis (18) and can also bind annexins via the C-D interhelical region (19).
In this report we have investigated the yeast two-hybrid LexA interaction trap (20) as a potential method for detecting novel serpin targets and have found an interaction between maspin and type I collagen. Verification with protein-protein interactions indicated that maspin preferentially binds collagen subtypes I and III. The similarity of this profile to that of the anti-angiogenic serpin, pigment epithelium-derived growth factor (PEDF) (21), the expression of which correlates inversely with metastatic potential in colon adenocarcinoma cells (22), suggests that collagen association may be a factor in serpin inhibition of angiogenesis.

EXPERIMENTAL PROCEDURES
Materials-All components of the two-hybrid system including vectors pEG202 and pJG4 -5, WI-38 fetal fibroblast cDNA "target" library in pJG4 -5, Saccharomyces cerevisiae EGY48, Escherichia coli strain KC8, and the polyclonal anti-LexA antibody were kindly donated by Luke O'Neill (Trinity College Dublin). E. coli strains XL-1 Blue and BL21(DE3) were obtained from Stratagene. The bacterial expression vector pRSETC was obtained from Invitrogen. Q-Sepharose, iminodiacetic acid (metal affinity matrix), collagen-agarose, collagen subtypes I, II, III, IV, and V, and horseradish peroxidase-linked goat anti-mouse IgM secondary antibody were purchased from Sigma. The IgM mouse monoclonal antibody to maspin was obtained from Transduction Laboratories. Immunoblots were developed by enhanced chemiluminescence (Roche Molecular Biochemicals). All oligonucleotide primers were obtained from Genosys (Cambridge, UK).
PCR Amplification and Cloning of Serpins into pEG202-Full-length human maspin cDNA was amplified by reverse transcription-PCR. Total RNA was isolated from HeLa cells using a phenol/guanidinium thiocyanate extraction method (23), and first strand cDNA was synthesized from total RNA (1 g) using Moloney murine leukemia virusreverse transcriptase and random hexanucleotide primers. PCR amplification of first strand cDNA was performed with the following primer pair specific to the 5Ј and 3Ј ends of the maspin gene: primer A, 5Ј-GGGGAATTCATGGATGCCCTGCAACTA-3Ј (sense oligonucleotide corresponding to nucleotides 1-18 of maspin coding sequence and incorporating an EcoRI restriction site) and primer B, 5Ј-AAAAGTC-GACTGCCACTTAAGGAGAA-3Ј (antisense oligonucleotide corresponding to nucleotides 1119 -1128 of maspin coding sequence plus nucleotides 1-6 of the 3Ј untranslated sequence and incorporating a SalI restriction site). Amplification was carried out with 1.5 mM MgCl 2 for 35 cycles of 94°C ϫ 1 min, 45°C ϫ 1 min, and 72°C ϫ 2 min. The product was cloned into the yeast two-hybrid bait vector pEG202 using EcoRI and SalI restriction sites and propagated in E. coli XL-1 Blue cells. Dideoxy sequencing was performed to verify correct in-frame insertion.
Immunoblot Analysis of pEG202 Constructs-Yeast (S. cerevisiae EGY48) was transformed with the pEG202 plasmid containing maspin using a lithium acetate protocol (24). Following overnight growth of transformants at 30°C in complete minimal medium with supplemented amino acids, fresh cultures were inoculated at A 600 of 0.15 and grown in enriched media YEPD (1% yeast extract, 2% peptone, and 2% dextrose) to an A 600 of 0.6 -1.0. Yeast total protein extract was prepared, and the expression of the maspin-LexA fusion protein was examined by immunoblotting using a mouse polyclonal antibody to LexA (1:1500) and a peroxidase-linked goat anti-mouse secondary antibody (1:2000).
Yeast Two-hybrid Library Screening-Maspin fused to the LexAbinding domain was used as a bait to screen a human fibroblast cell cDNA library fused to the activation domain of LexA. The bait-containing vector pEG202 and the library-containing target vector pJG4 -5 were sequentially transformed into the two-hybrid yeast strain EGY48 and grown on medium lacking leucine to select for potential interactors. Approximately 2 ϫ 10 6 clones representing one-third of the library equivalent were screened against the maspin bait. Transformants with the ability to grow on leucine-deficient media were subsequently screened for ␤-galactosidase reporter gene expression. Yeast colonies passing the dual selection screen were carried forward as potential interactor clones. Unique clones within this pool were detected by PCR amplification of the target inserts using the following primers to the vector arms of pJG4 -5: (primer C, 5Ј-CCAGCCTCTTGCTGAGTG-GAGATG-3Ј, and primer D, 5Ј-GACAAGCCGACAACCTTGATTGGAG-3Ј). Dideoxy sequencing was performed on products of unique insert size and/or restriction pattern to identify the target maspin interacting protein in each case. Plasmid DNA was prepared from the clone(s) of interest and transformed into E. coli KC8, a bacterial strain that is auxotrophic for tryptophan and therefore selectively propagates pJG4 -5. Interaction studies of target and bait vectors against nonrecombinant pEG202 and pJG4 -5 were conducted to eliminate false positives, which have an activator function (␤-galactosidase activity) in the absence of an interacting partner.
Expression and Purification of Recombinant Maspin Protein in E. coli-Full-length maspin cDNA was amplified by PCR from recombinant pEG202/maspin (described above) using the following primer pair: primer E, 5Ј-GGGGAATTCAATGGATGCCCTGCAAC-3Ј (sense oligonucleotide containing nucleotides 1-16 of maspin coding sequence and incorporating an EcoRI restriction site), and primer F, 5Ј-AAAAAGCTTTTAAGGAGAACAGAATT-3Ј (antisense oligonucleotide containing nucleotides 1112-1128 of maspin coding sequence and incorporating a HindIII restriction site). The PCR was carried out with 1.5 mM MgCl 2 for 30 cycles of 94°C ϫ 1 min, 48°C ϫ 1 min, and 72°C ϫ 2 min. The amplified product was subsequently cloned into the expression vector pRSETC at the EcoRI and HindIII restriction sites.
Recombinant maspin was overexpressed from pRSETC (Invitrogen) in E. coli. BL21(DE3), producing a recombinant fusion protein with a polyhistidine metal-binding tail at the N terminus. Cells grown to an A 600 ϭ 0.6 were induced with 1 mM isopropyl-1-thio-␤-D-galactopyranoside and grown for another 5 h. Following harvesting by centrifugation (12,000 ϫ g for 15 min), cells were resuspended and sonicated in 30 ml of 50 mM Tris-HCl, pH 8, 50 mM NaCl, and 5 mM ␤-mercaptoethanol. The cell lysate was centrifuged at 19,000 ϫ g for 20 min, and the soluble fraction was applied to a Q-Sepharose anion exchange column. Column elution was performed with a 50 -200 mM NaCl gradient in 50 mM Tris-HCl, pH 8. Maspin-containing fractions were further purified by immobilized metal affinity chromatography using a fast-flow chelating Sepharose column precharged with 5 mg/ml nickel chloride and equilibrated in 500 mM NaCl and 40 mM sodium phosphate, pH 8. Column elution was performed stepwise by decreasing pH, but maspin remained tightly bound to the column at pH 4.5 and was subsequently eluted with 50 mM EDTA in 50 mM Tris-HCl, pH 8. Removal of the nickel and EDTA from the maspin fraction was achieved by ultrafiltration and dialysis.
Binding of Maspin to Collagen-Agarose-Type I collagen immobilized on agarose was equilibrated with 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, and 0.05% Brij35. rMaspin (0.1 mg/ml) was applied, and following extensive column washing with the above buffer, bound material was eluted with 125 mM Tris-HCL, pH 6.8, 2% SDS, and 5% ␤-mercaptoethanol. Control experiments examining binding of rPAI-2 and rOvalbumin, which also contained polyhistidine N-terminal tags (25), were performed under identical conditions. Native and SDS-PAGE Immunoblot Analysis of Interaction-Maspin (4 g) was incubated with collagen type I (0.75-4 g) for 30 min at 37°C. The samples were electrophoresed on a 10% native gel in addition to a 10% SDS-polyacrylamide gel. Following transfer to nitrocellulose, membranes were incubated for 2 h with an IgM monoclonal antibody to human maspin (1:1000) followed by peroxidase-linked anti-IgM secondary antibody (1:2000). Membranes were developed by enhanced chemiluminescence and autoradiography.
Immunoblot Analysis of Maspin Interaction with Collagen Subtypes-Collagen subtypes I, II, III, IV, and V, boiled collagen, and gelatin were applied at varying concentrations to a nitrocellulose membrane. Dot blots were blocked overnight at 4°C with 4% BSA and then incubated for 2 h with a solution of recombinant maspin (20 g/ml in 150 mM NaCl and 50 mM Tris, pH 7.6). Following stringent washing with 150 mM NaCl and 50 mM Tris-Cl, pH 7.6, to remove unbound ligand, membranes were subsequently immunoblotted for maspin as described above.
IAsys Kinetic Analysis-IAsys affinity sensor analysis experiments were conducted on the IAsys Plus apparatus (Affinity Sensors Ltd., Saxon Hill, Cambridge, UK). Binding reactions were carried out in an IAsys resonant mirror biosensor at 25°C using planar aminosilane surfaces. Collagen type I was chemically immobilized to the surface according to the manufacturer's instructions. Collagen was made up in 10 mM sodium acetate, pH 5.0, and applied to the cuvette at 1 mg/ml, until saturation was achieved (ϳ300 arc s).
Interaction analysis was conducted with varying concentrations of maspin ranging from 0.25-3 M and made up in phosphate-buffered saline. In all experiments, BSA (30 M in phosphate-buffered saline) was used as a control. Regeneration of the cuvette was achieved by repeated washes with 100 mM HCl. Results were analyzed with Fastfit software from IAsys.
Yeast Two-hybrid One-on-One Interactions-To identify the region of the maspin protein that interacts with collagen, truncated constructs of maspin cDNA were fused to LexA in pEG202 and tested for positive interaction against the collagen-containing pJG4 -5 plasmid. In total, seven maspin cDNA fragments were amplified from recombinant pEG202/maspin. Four of these were amplified using a common forward oligonucleotide, primer A (described previously), in conjunction with each of the reverse primers G, H, I, and J (see Table I).
The remaining three maspin fragments were amplified using a common reverse oligonucleotide, primer J, in conjunction with each of the forward primers K, L, and M (see Table I). Reverse primers G, H, I, and J contain flanking BamHI restriction sites, and forward primers K, L, and M contain flanking EcoRI restriction sites. In addition, reverse primers G, H, I, and J contain reverse complimentary stop codons immediately preceding the BamHI restriction site. Amplification was carried out with 1.5 mM MgCl 2 for 30 cycles of 94°C ϫ 1 min, 55°C ϫ 1 min, and 72°C ϫ 3 min. Amplified products were cloned by standard procedures into the EcoRI and BamHI restriction sites of pEG202. Recombinant pEG202 containing maspin truncations were each cotransformed with the collagen-containing pJG4 -5 into S. cerevisiae EGY48. Collagen-maspin interactions were detected by both ␤-galactosidase and Leu-2 reporter gene expression.

Expression of Maspin in the Yeast Two-hybrid System-Ex-
pression of maspin cloned into the pEG202 vector was expected to yield a LexA fusion protein of 64 kDa. However, immunoblot analysis of transformed S. cerevisiae EGY48 cells with anti-LexA antibody indicated a fusion protein of ϳ49 kDa (Fig. 1). Sequence analysis of the entire pEG202 insert ruled out the possibility of an incorporated stop codon during or subsequent to the initial PCR amplification step. The presence of protease inhibitors during preparation of yeast protein extracts also failed to produce full-length bait on immunoblots. This processing problem, which is most likely caused by intracellular proteolysis, was not resolved and would result in a fusion with 43% of the C terminus, including the reactive site loop, absent from the bait. However, additional work showing the expression of a fully intact LexA-PAI-2 fusion in S. cerevisiae EGY48 indicates that truncation of serpin baits within the LexA yeast two-hybrid system is not necessarily a general serpin phenomenon.
Detection of a Maspin Interaction with Collagen Using the Yeast Two-hybrid Interaction Trap-Despite the apparent lack of full-length maspin protein, we proceeded with a yeast twohybrid screen to detect possible interactants. Prior to screening, the truncated maspin bait was shown to be transcriptionally inert using the ␤-galactosidase assay (result not shown). A target human fibroblast cDNA library in the pJG4 -5 plasmid was screened for expression of the Leu2 reporter gene and initially produced 50 positive interactor yeast clones. Of these, 10 failed to show transcription of the second reporter gene lacZ upon selection with 5-bromo-4-chloro-3-indolyl-␤-D-galactopyranoside (X-gal). The remaining positive clones were amplified from the target vector by PCR using primers specific to the vector arms. Four different sized inserts were selected, and dideoxy sequencing of the target plasmids was performed to identify the maspin interacting partner in each case (Fig. 2).
The sequences identified were transketolase (U55017), metallothionein (BC008408), human elongation factor (BC018641), and type I collagen (Z74616). Of these, type I collagen was investigated further because collagen-binding serpins have been reported previously. The collagen insert in the target vector pJG4 -5 consisted of 216 amino acids in the collagen ␣2(I) chain, and this was in the correct reading frame as a fusion with the transactivating domain of LexA. This sequence corresponds to bases 2261-2909 of prepro-␣2(I) collagen (accession number Z74616), and the corresponding amino acids (708 -923) are within the repetitive (Gly-X-Y) n sequence of the collagen chain.
Protein-Protein Analysis of the Maspin-Collagen Interaction-To further investigate the relevance of a collagen interaction with maspin, recombinant full-length maspin was produced for direct protein-protein interaction studies. Recombinant maspin produced in E. coli has been shown previously to retain its function as an inhibitor of tumor cell migration in vitro (3).
The maspin cDNA open reading frame sequence was cloned into the bacterial expression vector pRSETC, and the protein was subsequently produced in E. coli with an N-terminal polyhistidine tag. Maspin was purified by anion exchange and metal ion affinity chromatography (Fig. 3A).
Binding of rMaspin to type I collagen immobilized to agarose was demonstrated (Fig. 3B). rMaspin bound to this affinity matrix in the presence of 150 mM NaCl, whereas recombinant PAI-2 produced in the same expression system (25) and also containing a polyhistidine tail did not bind under identical conditions (Fig. 3C). Recombinant ovalbumin produced similarly also failed to bind (results not shown).
Native and SDS-polyacrylamide gels were performed following incubation of maspin with collagen type I. Immunoblot analysis showed a complex formation on native gels where maspin was significantly retarded in the presence of collagen (Fig. 4). No complexes were seen on SDS-PAGE analysis however, suggesting that the interaction is noncovalent.
To investigate binding to other collagen subtypes, dot blots were performed by applying decreasing amounts of collagen subtypes I, II, III, IV, and V to nitrocellulose. The membrane was incubated with recombinant maspin at a concentration of 0.44 M, and an anti-maspin IgM antibody was used to detect collagen-bound maspin. Collagen subtypes I and III were found to bind recombinant maspin, with 250 ng of each binding detectable amounts of maspin, whereas 2 g of immobilized types II, IV, and V failed to bind maspin (Fig. 5). In addition, binding to thermally denatured collagen and gelatin (acid-denatured collagen) was examined, but in both cases maspin failed to bind (result not shown).
Kinetic Analysis of the Maspin-Collagen Interaction Using IAsys Technology-The interaction between maspin and type I collagen was analyzed by fast association kinetics using IAsys affinity sensor technology. Collagen type I was coupled covalently to an aminosilane cuvette surface, and varying concentrations of recombinant maspin (0.25-3 M) were applied. In a parallel control experiment, BSA (30 M) was added to a collagen-coated aminosilane cuvette. All binding events were monitored in real time, and the rate of association of maspin to type I collagen increased with increasing maspin concentration (Fig. 6A). To determine a dissociation affinity constant (K D ) for the maspin-collagen type I interaction, a plot of k on against ligand concentration was drawn (Fig. 6B). The slope of the line corresponds to the association rate constant k a (3.2 ϫ 10 3 M Ϫ1 s Ϫ1 ), and the y-intercept corresponds to the dissociation rate constant, k d (0.002 s Ϫ1 ). The K D is described by k d /k a and is calculated to be 0.63 ϫ 10 Ϫ6 M.
Identification of the Collagen-binding Region within Maspin-The specific protein region of maspin that interacts with collagen was investigated using yeast two-hybrid technology. A number of bait constructs were cloned in pEG202 to contain varying lengths of maspin cDNA, corresponding to amino acids 1-34, 1-112, 1-166, 1-214, 38 -214, 84 -214, and 112-214 of the maspin protein. Because the original screening bait appeared to be missing the C-terminal half of maspin, baits longer than 214 amino acids were not constructed. To perform one-on-one interactor analysis, the isolated collagen containing pJG4 -5 plasmid was co-transformed with each of the maspin fragment pEG202 vectors into the two-hybrid yeast strain EGY48. The resulting transformants were subsequently assayed for ␤-galactosidase reporter gene expression to identify a maspin-collagen interaction. As expression of the collagen target from pJG4 -5 is under the control of a Gal promotor, true interactor clones will form blue colonies when grown on galactosecontaining medium and white colonies when grown on glucosecontaining medium. In contrast, false positive interactor clones form blue colonies on glucose-containing media and denote a bait that can activate the reporter gene systems on its own. Results show that all maspin fragments except 112-214 interact with collagen (Fig. 7), and assays for the second reporter gene, Leu2, were in complete agreement (data not shown). However, the shortest maspin construct, 1-34, was found to be transcriptionally active in the absence of an interacting partner. Taken together, these results suggest that the interaction with collagen occurs within amino acids 84 -112 of the maspin protein. DISCUSSION The use of the yeast-two-hybrid protein interaction trap to identify serpin targets has not previously been reported, and the ability of the system to carry out post-translational modifications and correct folding of proteins such as serpins is uncertain. The LexA system gives relatively high expression of the binding do-

FIG. 2. Positive maspin interactants identified from a yeast two-hybrid screen of a human fibroblast library.
Positive interactor yeast clones that yielded expression of both reporter genes were isolated. Above, agarose gel of inserts amplified from the pJG4 -5 target library plasmid using oligonucleotide primers directed against vector arms flanking the multiple cloning site. Insert size and restriction analysis indicated that at least four different target inserts were contained within the pool of 26 positive interactor yeast clones. Below, identity of the four target inserts as revealed by exact data base sequence homology. main fusion, allowing detection of bait fusion, and our results suggest that there may be a lack of expression of the full-length maspin. The product detected for the maspin construct indicates that the fusion protein lacks the C-terminal 43% of the serpin including the reactive loop, and if caused by proteolysis, this cleavage site would correspond to residues in the serpin structure lying between strands 1 and 2 of ␤-sheet B. However, PI-6 has been successfully expressed in an active form in Pichia pastoris (26), and these findings may be unique to the vector or the EGY48 yeast strain used in this study. Also, although no fulllength product was detected on immunoblots, there may be a proportion of the entire fusion protein in vivo that could potentially pick up serpin reactive site loop interactions. Alternatively, the system may be useful for detecting novel protein interactions with serpin N-terminal sequences or other selected regions. The two-hybrid system has previously been used to successfully detect interactions between extracellular matrix components, demonstrating an association between collagen types IV and VI (27).
The positive result suggesting a potential interaction with collagen type I was further investigated in view of previous reports of other serpins binding to collagen. Colligin or heat shock protein 47 can bind procollagens 1 and IV and gelatin but not collagen type III (28,29). Colligin is localized to the endoplasmic reticulum and appears to act as a molecular chaperone in the protection of procollagen from intracellular degradation (30,31). Protease nexin-1 activity with thrombin is enhanced with binding of heparin sulfate glycoproteins, but its specificity is significantly altered by binding to collagen type IV, which decreases affinity for uPA and plasmin while leaving thrombin inhibition unaffected (32). More relevant is caspin (collagen associated serpin), which is identical to murine PEDF and can preferentially bind collagen types I and III (22). In mouse colon this protein is expressed at high levels in nonmetastatic clones and at low levels in high metastatic clones of adenocarcinoma cells, which is similar to the loss of maspin expression in mammary carcinoma. Human PEDF is a noninhibitory serpin with neurotrophic and neuronal survival activity (33,34), which is independent of an intact reactive site loop. It can bind glycosaminoglycans (35), and both recombinant and tissuepurified PEDF are potent inhibitors of angiogenesis (21).
The yeast two-hybrid results were corroborated with protein- FIG. 5. Dot blot analysis of rMaspin binding to collagen subtypes. Nitrocellulose membrane was spotted with decreasing amounts of collagen types I, II, III, IV, and V and blocked with 4% BSA and incubated with rMaspin (20 g/ml). Collagen-bound maspin was detected by immunostaining with anti-maspin as described.
FIG. 6. Kinetic analysis of maspin interaction with collagen type I using Iasys technology. A, Iasys-binding curves for the interaction of maspin to collagen type I immobilized to the surface of a planar aminosilane biosensor surface. The binding of different concentrations of maspin (0.5-3.9 M) to immobilized collagen was followed in real time for ϳ3 min at 25°C with binding of BSA (30 M) to collagen as a negative control. B, using Fastfit software, the k on of maspin for collagen type I at each concentration of maspin was determined. A plot of k on against maspin concentration yields a straight line, the slope of which corresponds to the association rate constant, k a ϭ 3.2 ϫ 10 3 M Ϫ1 s Ϫ1 , and the y-intercept of which corresponds to the dissociation rate constant, k d ϭ 0.002 s Ϫ1 . The dissociation affinity constant K D described by k d /k a is calculated to be 0.63 ϫ 10 Ϫ6 M.

FIG. 7.
Yeast two-hybrid one-on-one interactions with maspin N-terminal fragments. Maspin truncation constructs (sizes indicated in amino acids) were individually analyzed for collagen interaction. The isolated target plasmid, pJG4 -5, containing the ␣2 chain of type I collagen was cotransformed into S. cerevisiae EGY48 with pEG202 bait plasmid containing either no insert or maspin truncation constructs corresponding to amino acids 1-34, 1-112, 1-166, 1-214, 55-214, 84 -214, and 112-214. Yeast harboring a positive interaction were identified by their ability to activate the ␤-galactosidase reporter system. True interactors will show ␤-galactosidase expression when grown on a carbon source of galactose/raffinose (Gal), but not glucose (Glu), because the collagen target expression is under the control of a Gal1 promoter. All of the maspin fragments except 112-214 show ␤-galactosidase activity. The shortest fragment (1-34) possesses an intrinsic ability to activate transcription of the reporter gene in the absence of a collagen partner.
protein interactions with full-length maspin binding to type I collagen-agarose, whereas other similarly expressed recombinant serpins failed to do so. Most interestingly, binding studies with different collagen subtypes showed maspin-binding types I and III but not II, IV, and V, a profile identical to murine PEDF. Collagen types I and III are widely distributed, representing 70 and 25%, respectively, of total collagen content in normal breast tissue. They are abundant neomatrix components in invasive breast carcinomas (36) and present a physical barrier to metastasizing cells. The physiological significance of this finding is further supported by findings that maspin is present at the cell surface of mammary myoepithelial cells (11), a localization that would automatically facilitate a contact with these extracellular matrix components. The dissociation affinity constant of maspin for collagen type I was estimated in this study to be 0.63 M, which is within the range (0.3-3 M) used to show murine PEDF binding to collagen by surface plasmon resonance (22). It is also similar to the concentration of maspin (0.44 M) used to alter the integrin profile of MDA-MB-435 cells (13), to achieve inhibition of metastasis (0.17 M) in in vitro invasion assays (3), and to inhibit angiogenesis (0.2-0.3 M) (4).
The fact that the interaction was detected in the intracellular yeast system indicates that glycosylation of maspin or collagen is not required for binding, in contrast to some serpinglycosaminoglycan interactions. At this stage, using the twohybrid system, we have narrowed down the collagen-binding region of maspin to amino acids 84 -112. This corresponds with evidence that colligin binding is localized to amino acids 104 -169 (29), which aligns with residues 65-116 of maspin and suggests the existence of a general collagen-binding motif in serpins that remains to be well defined. In contrast to cell migration effects, the inhibition of angiogenesis by maspin was found to be independent of the C-terminal sequence but appeared to require the N-terminal 139 amino acids, suggesting that this function may coincide with the collagen-binding site.
The finding that maspin can alter the integrin profile of MDA-MB-435 cells provides evidence that maspin may prevent invasion by increasing the anchoring of cells to the extracellular matrix. This has been demonstrated more directly in corneal stromal cells where exogenous recombinant maspin has been found to restore cell adhesion in stromal cells as measured by increased adhesion to collagen type I and other extracellular matrix components (14). A direct adherence to both collagen and an uncharacterized cell surface receptor for maspin may be an explanation for these findings. As found with other serpinligand interactions, it is possible that collagen binding may alter the structural conformation of maspin, being necessary for either stability or for receptor binding, or for active presentation of the maspin reactive site loop.
In conclusion, using yeast two-hybrid technology, we provide evidence that the N-terminal region of maspin can bind collagen types I and III. This interaction may contribute to the function of maspin as a tumor suppressor and angiogenesis inhibitor either by direct adherence between cell surface maspin and extracellular matrix collagen or by altering the ability of maspin to interact with other proteins.