Generation of Monoclonal Antibodies to Integrin-associated Proteins

The α3β1integrin forms complexes with other cell-surface proteins, including transmembrane-4 superfamily (TM4SF) proteins (e.g. CD9, CD53, CD63, CD81, and CD82). To identify additional cell-surface proteins associated with α3β1 integrin, a monoclonal antibody selection protocol was developed. Mice were immunized with integrin α3β1-containing complexes isolated from HT1080 fibrosarcoma cells, and then 712 hybridoma clones were produced, and 95 secreted antibodies that recognized the HT1080 cell surface. Among these, 12 antibodies directly recognizing integrin α3 or β1 subunits were eliminated. Of the remaining 83, 16 co-immunoprecipitated proteins that resembled integrins under non-stringent detergent conditions. These 16 included 15 monoclonal antibodies recognizing EMMPRIN/basigin/OX-47/M6, a 45–55-kDa transmembrane protein with two immunoglobulin domains. The EMMPRIN protein associated with α3β1 and α6β1, but not α2β1 or α5β1, as shown by reciprocal immunoprecipitation experiments. Also, association with α3β1 was confirmed by cell-surface cross-linking and immunofluorescence co-localization experiments. Importantly, EMMPRIN-α3β1 complexes appear not to contain TM4SF proteins, suggesting that they are distinct from TM4SF protein-α3β1 complexes.

To identify and characterize additional cell-surface proteins that might form complexes with integrins, we designed a systematic approach based on a three-step monoclonal antibody screening protocol. After immunizing mice with integrin complexes, we screened for mAbs to integrin-associated proteins by (i) flow cytometry to select mAbs reactive with the cell surface, (ii) flow cytometry to eliminate anti-integrin mAbs, and (iii) high and low stringency immunoprecipitation. By applying this approach we have discovered novel interactions between two integrins (␣ 3 ␤ 1 and ␣ 6 ␤ 1 ) and the EMMPRIN/basigin/OX47/M6 protein. The latter molecule is a widely distributed cell-surface protein with two immunoglobulin-like domains (25)(26)(27)(28) that may play a role in regulating matrix metalloproteinase production (29).
Immunoprecipitation-HT1080 cells were surface-labeled with NHS-LC-biotin (Pierce) or Na 125 I according to established protocols and lysed in immunoprecipitation buffer (1% Brij 96, 25 mM HEPES, pH 7.5, 150 mM NaCl, 5 mM MgCl 2 , 2 mM phenylmethylsulfonyl fluoride, 10 g/ml aprotinin, 10 g/ml leupeptin) for 1 h at 4°C. Immune complexes were collected onto Sepharose 4B beads (Pharmacia, Uppsala, Sweden) that were pre-bound with mAb, followed by four washes with the immunoprecipitation buffer. For immunoprecipitation under stringent conditions, the immunoprecipitation buffer was supplemented with 0.2% SDS. Immune complexes were eluted from beads with Laemmli sample buffer and resolved by 10% SDS-PAGE under non-reduced conditions. 125  Re-immunoprecipitation experiments were performed as described earlier (11). Briefly, protein complexes were immunopurified using anti-EMMPRIN (8G6) or anti-integrin mAb-conjugated Sepharose 4B from nonstringent (without 0.2% SDS) Brij 96 lysates of surface-biotinylated HT1080 cells. After five washes, the protein complexes were dissociated for 30 min at 4°C with Brij 96 buffer containing 0.2% SDS. The eluates were subsequently reprecipitated with mAbs directly coupled to Sepharose 4B.
Antibody Production-Protein complexes were purified from HT1080 cells using anti-␣ 3 mAb A3-IVA5-coupled Sepharose 4B. After washing, ␣ 3 ␤ 1 -containing protein complexes immobilized on antibody-coupled Sepharose beads were used for immunization of an RBF/DnJ mouse. After three injections (each time with complexes derived from 5 ϫ 10 7 -10 8 cells), mouse serum was collected and tested by immunoprecipitation to verify antibody production. Four days after the fourth injection, hybridoma clones were produced as described previously (39). Hybridoma supernatants were then analyzed as described below.
Protein Purification and Amino-terminal Sequencing-For 8G6 antigen/EMMPRIN purification, two human placentas (ϳ400 g each) were blended in a Waring blender and then solubilized in 2 liters of RIPA buffer (50 mM Tris-HCl, 150 mM NaCl, 1% Triton X-100, 0.5% deoxycholate, 0.1% SDS, 2 mM phenylmethylsulfonyl fluoride, 1 g/ml aprotinin, 1 g/ml leupeptin). Insoluble material was removed by centrifugation (30 min, 65,000 ϫ g). To remove background bead-binding material, the lysate was sequentially preincubated in batch with Sepharose beads coupled to protein A, Sepharose 4B beads conjugated with BSA, and beads coupled to an irrelevant mAb. The lysate was then incubated in batch with Sepharose 4B beads coupled to mAb 8G6 for 2 h. The beads (2 ml packed volume) were collected and washed with 100-bed volumes of RIPA buffer, and protein was eluted with elution buffer (0.1 M glycine, pH 2.7, 150 mM NaCl, 0.2% Triton X-100). Eluted fractions were immediately neutralized with 0.1 volume of 1.0 M Tris-HCl, pH 8.0, and then analyzed by SDS-PAGE, and proteins were visualized by silver staining. The fraction containing the most antigen was then subjected to SDS-PAGE, and proteins were transferred to a polyvinylidene difluoride membrane. The 45-55-kDa EMMPRIN protein was visualized by Ponceau S staining and then excised. Aminoterminal sequencing was carried out using an Applied Biosystems 470A gas-phase sequenator equipped with a 120A phenylhydantoin amino acid analyzer (Harvard Microsequencing Facility, Cambridge, MA).
Flow Cytometry-Cells were incubated with control mAb P3 or specific mAb, washed twice, and then labeled with fluorescein isothiocyanate (FITC)-conjugated goat anti-mouse immunoglobulin. Stained cells were analyzed using a FACScan (Becton Dickinson, Mountain View, CA).
Immunofluorescence-For immunofluorescence analyses, HT1080 cells were grown for 12-16 h on glass coverslips, fixed for 15 min with 2% formaldehyde, in PBS containing 5% sucrose and 5 mM MgCl 2 , permeabilized for 2 min with 1% CHAPS in PBS, and blocked for 1 h with 20% FCS in PBS. Cells were then stained with mAb to integrin ␣ 2 (VIIC6) or MHC class I (W6/32) diluted in 20% FCS in PBS. Finally, cells were stained with FITC-conjugated goat anti-mouse serum (Tago Co.) before the coverslips were mounted with FluorSave (Calbiochem) and immunofluorescence examined using a Zeiss Axioscop. For colocalization experiments, paraformaldehyde-fixed and permeabilized HT1080 cells were first incubated with a combination of mouse anti-EMMPRIN mAb, 8G6, and rabbit polyclonal sera (against human ␣ 3 integrin cytoplasmic domain or anti-Tyr(P)). Staining was further visualized with FITC-conjugated goat anti-mouse serum and rhodamineconjugated goat anti-rabbit serum. Fig. 1 shows that from 1% Brij 96 extracts of Na 125 I-labeled HT1080 cells, subunit-specific anti-integrin mAbs co-immunoprecipitated not only the expected ␣ and ␤ subunits of ϳ150 and 130 kDa but also additional surface-labeled proteins. For example, at least 7-8 additional 125 I-labeled proteins were obtained using mAb to ␣ 3 (lane h) and ␤ 1 (lane k). These other proteins were either absent or less prominent in precipitations using mAb to ␣ 2 (lane g), ␣ 5 (lane i), or ␣ 6 (lane j). Also, most of the additional proteins were not present if more stringent solubilization and immunoprecipitation conditions (1% Brij 96, 0.2% SDS) were utilized (Fig. 1, lanes a-e). A control antibody yielded only a few background proteins (lanes f and l). Integrin co-immunoprecipitation experiments from several other cell lines (data not shown) yielded cell-surface proteins with sizes similar to those co-immunoprecipitated in Fig. 1.

Immunoprecipitation of Integrin-associated Proteins-
Production and Initial Characterization of mAb to ␣ 3 ␤ 1 -associated Proteins-To identify additional integrin-associated cell-surface proteins (such as those seen in Fig. 1, lane h), a strategy for systematic selection of mAbs was developed. We focused on proteins complexed with ␣ 3 ␤ 1 , because of the apparent abundance of diverse cell-surface proteins associated with that integrin. Beads coated with antibody A3-IVA5 were used to purify ␣ 3 ␤ 1 -containing protein complexes from the HT1080 human fibrosarcoma cell line, and mice were immunized with integrin complexes directly immobilized on the beads. Next, 712 hybridoma clones were prepared, and then mAbs were screened by a three-step protocol. In step 1, flow cytometry was utilized to show that 95 of the 712 hybridoma clones secreted antibodies that recognize cell-surface structures on HT1080 cells. To eliminate antibodies directly recognizing human ␣ 3 or ␤ 1 subunits (step 2), mAb clones were next tested for staining of CHO, CHO-␣ 3 , CHO-␤ 1 , or CHO-␣ 3 ␤ 1 transfectants. Of the 95 mAbs from step 1, 12 stained ␣ 3 or ␤ 1 transfectants, but not untransfected CHO cells, and thus are likely to be anti-␣ 3 or -␤ 1 integrin antibodies. The remaining 83 antibodies (after steps 1 and 2) are candidates to recognize cell-surface structures distinct from but potentially associated with the ␣ 3 ␤ 1 integrin. An example of one of these is hybridoma clone 8G6, selected in step 1 because it stained the surface of HT1080 cells (Fig. 2, top  right). The 8G6 mAb was retained after step 2, because it did not show preferential binding to a CHO-␣ 3 ␤ 1 transfectant (Fig.  2, bottom right) compared with CHO cells (Fig. 2, middle right). The presence of human ␣ 3 integrin on HT1080 cells and CHO-␣ 3 ␤ 1 transfectants, but not CHO cells, is confirmed in the left panels of Fig. 2. In step 3, non-stringent immunoprecipitation experiments were carried out to determine which mAbs might co-precipitate integrin-like proteins. Of the 83 mAbs tested, 16 mAbs coimmunoprecipitated integrin-like proteins of 120 -150 kDa, together with several other proteins. Examples of these are mAb 5C11 (Fig. 3, lane d) Table I.
Identification of ␣ 3 ␤ 1 -associated Proteins-The 5C11 mAb precipitated fewer total proteins under stringent conditions (Fig. 3, lane c) compared with non-stringent conditions (Fig. 3,  lane d). A protein of 27-29 kDa was seen in lane c, but the presence of several other proteins made it difficult to determine which protein was directly recognized by mAb 5C11. The associated protein defined by this antibody will be more fully described elsewhere.
Under non-stringent conditions, multiple biotin-labeled proteins were precipitated by mAb 8G6 (Fig. 3, lane j). In contrast, only 45-55-kDa protein bands were seen using 8G6 under stringent conditions (Fig. 3, lane i). Notably, under stringent detergent conditions, 14 other mAbs (all selected as in Table I) co-precipitated 45-55-kDa proteins nearly identical to that obtained using 8G6. To characterize the 45-55-kDa protein(s)  further, we utilized 8G6-coated Sepharose beads to purify ϳ400 pmol of protein from ϳ800 g of human placenta. Purified material, visualized by silver staining, is shown in Fig. 4A,  lanes 3 and 4). Amino-terminal amino acid analysis revealed a perfect match, at 22 of 24 positions (Fig. 4B), between the purified protein and the amino terminus of the human EMMPRIN/M6 antigen (25,28). Confirming conclusively that mAb 8G6 recognizes human EMMPRIN, CHO cells transfected with EMMPRIN cDNA gained the 8G6 epitope as detected by flow cytometry (Fig. 4C). Three additional mAbs (6C6, 7E7, and 4F10) that recognize the 45-55-kDa protein also showed strong reactivity toward the EMMPRIN-transfected CHO cells (not shown).
Association of EMMPRIN with Integrins-To confirm that EMMPRIN is indeed associated with ␣ 3 ␤ 1 integrin, and to determine which other integrins might be associated, we carried out three sets of experiments. First, by Western blotting we detected ␤ 1 integrins within an EMMPRIN immune complex (Fig. 5A, lane b) but not in an MHC class I complex (Fig.  5A, lane a). Notably, anti-␤ 1 antibodies reacted with both immature (pre-␤ 1 ) and mature forms of ␤ 1 in the whole cell lysate (Fig. 5A, lane c), but only mature ␤ 1 was associated with EMMPRIN (Fig. 5A, lane b). Second, in re-immunoprecipitation experiments, both the ␣ 3 ␤ 1 and ␣ 6 ␤ 1 integrins but not ␣ 2 ␤ 1 or ␣ 5 ␤ 1 could be recovered from mAb 8G6 immunoprecipitates of EMMPRIN (Fig. 5B, lanes a-d). Third, mAb 8G6 re-precipitated EMMPRIN only from ␣ 3 and ␣ 6 but not ␣ 2 or ␣ 5 immunoprecipitates (Fig. 5C, lanes a-d). Notably, the size of the EMMPRIN associated with integrins (Fig. 5C) appears a little smaller than the total EMMPRIN as seen in Figs. 3 and 4A. Possibly, a smaller form of EMMPRIN could selectively associate with integrins. Alternatively, excess glycosylation in the larger forms of EMMPRIN could block access of tyrosines to iodination (used in Fig. 5), while not blocking availability for silver staining (used in Fig. 4A) or access of amino groups to biotin (used in Fig. 3).
Integrin-EMMPRIN Complexes on the Cell Surface-To establish whether integrin-EMMPRIN complexes may be present on the cell surface, intact HT1080 cells were treated with the cleavable cross-linker dithiobis(succinimidyl propionate). The mAb 8G6 was then used to immunoprecipitate EMMPRIN under stringent conditions to disrupt non-covalent interactions with integrins, and precipitated proteins were subsequently analyzed under reducing conditions to disrupt covalent crosslinker bonds. As indicated, without cross-linker mAb 8G6 precipitated only EMMPRIN (Fig. 6, lane b), but in the presence of 1 mM dithiobis(succinimidyl propionate), additional proteins resembling integrins were present (lane d). These proteins co-migrate with ␣ 3 ␤ 1 (lane e) but likely contain ␣ 6 ␤ 1 as well as ␣ 3 ␤ 1 . Besides ␣ 3 ␤ 1 , a small amount of 45-55-kDa protein comigrating with EMMPRIN was also present in the ␣ 3 immunoprecipitate prepared from the cells treated with cross-linker (lane e). In contrast, proteins resembling EMMPRIN are not typically seen in stringent ␣ 3 immunoprecipitations in the absence of cross-linking (e.g. see Fig. 1, lane h).
Co-localization of ␣ 3 ␤ 1 Integrin and EMMPRIN in Cell-Cell Contacts-In HT1080 cells grown on coverslips for 24 -72 h, EMMPRIN and ␣ 3 ␤ 1 were found to strongly co-localize at cellcell contact sites, as seen by immunofluorescent double staining (Fig. 7, A and B). In control experiments, mAb to MHC class I (Fig. 7C) and integrin ␣ 2 ␤ 1 (Fig. 7D) showed even staining of cell surfaces, and some intracellular staining, but no appreciable staining of cell-cell contact sites.
When HT1080 cells are plated on laminin substrates, the ␣ 3 ␤ 1 , ␣ 6 ␤ 1 , and TM4SF proteins localize into focal complexes (24), a type of complex that contains phosphotyrosine and is assembled at the periphery of spread cells (45). However, EMMPRIN does not appear to localize in focal complexes. In HT1080 cells plated on a laminin-5-containing ECM substrate, focal complexes (at the cell periphery) were prominently stained with anti-phosphotyrosine mAb (Fig. 8B). In marked contrast, double staining with mAb 8G6 showed that EMMPRIN was evenly distributed on the cell surface, with elevated levels at cell-cell contacts, but was essentially absent from focal complexes (Fig. 8A).
plexes versus cell-cell contact sites), we hypothesized that EMMPRIN-␣ 3 ␤ 1 and TM4SF-␣ 3 ␤ 1 complexes may be distinct. To examine this issue further, we carried out immunoprecipitation experiments to determine whether CD81, one of the most predominant TM4SF proteins on HT1080 cells, was present in ␣ 3 ␤ 1 -EMMPRIN complexes. As indicated, anti-CD81 mAb precipitated an abundance of the 22-kDa CD81 molecule, together with proteins co-migrating with ␣ 3 ␤ 1 , but no proteins resembling EMMPRIN (Fig. 9, lane b). Conversely, anti-EMMPRIN mAb 8G6 precipitated an abundance of the 45-55-kDa EMMPRIN protein and proteins resembling ␣ 3 ␤ 1 but no proteins co-migrating with CD81 (lane c). From proteins precipitated using anti-EMMPRIN mAb 8G6, re-immunoprecipitation experiments clearly showed the ␣ 3 ␤ 1 integrin (Fig. 9, lane d) but no detectable CD81 (lane e). Thus, CD81 does not appear to associate with EMMPRIN. Non-stringent precipitation with an anti-␣ 3 mAb yielded a range of proteins (Fig. 9,  lane a), but those corresponding to the 22-kDa CD81 protein and the 44 -55-kDa EMMPRIN doublet were not very prominent.  8. EMMPRIN does not localize to focal adhesion complexes. HT1080 cells were plated on glass coverslips coated with laminin-5-containing matrix from A431 cells, prepared as described previously (34). After cells were fixed and permeabilized, they were doublestained using mAb 8G6 (A) and polyclonal rabbit anti-phosphotyrosine (PTYR) (B).

DISCUSSION
Identification of Associated Proteins-It is becoming increasingly apparent that full understanding of integrin functional activities may require understanding of integrin associations with other cell-surface molecules (see Introduction). Here we describe a novel approach for identifying cell-surface proteins that associate with integrins. Our screening protocol is based on well established techniques (flow cytometry and immunoprecipitation) and requires only 3-4 days to identify mAbs that recognize integrin-associated proteins. Our protocol has two important advantages over alternative approaches used previously (such as the yeast two hybrid screening or affinity chromatography with integrin fragments). First, selected mAbs recognize proteins that interact with intact integrin molecules. Thus we may avoid adventitious interactions that arise upon un-natural presentation of fragments. Second, use of immunoprecipitation early in screening provides a reliable assessment of protein-protein interactions. Furthermore, cross-linking subsequent to immunoprecipitation provides an early control for the authenticity of putative associations in the context of intact cells.
In a previous study, we immunized mice with intact cells from a mammary epithelial cell line, screened all of the antibodies by immunoprecipitation, and thus identified a novel association between integrins and the TM4SF protein CD63 (10). Our current immunization and antibody selection protocol is designed to generate fewer irrelevant antibodies upon initial immunization, and it requires many fewer time-consuming immunoprecipitations. Furthermore, it seems likely that this approach could be easily applied to identify proteins associated with other integrins or with other cell-surface proteins (e.g.. see the accompanying paper (46)).
Integrin Association with EMMPRIN-Perhaps the more novel finding is that specific integrins (␣ 3 ␤ 1 and ␣ 6 ␤ 1 ) can associate with EMMPRIN (25), a widely distributed 44 -45-kDa cell-surface protein with two immunoglobulin domains. Integrin/EMMPRIN association was established by means of reciprocal immunoprecipitation experiments, and ␣ 3 ␤ 1 / EMMPRIN association was confirmed by cell-surface crosslinking and immunofluorescence co-localization experiments.
The human EMMPRIN (25) protein is named for its extracellular matrix metalloproteinase induction (29) and is identical to the M6 leukocyte activation antigen (28). Also, EMMPRIN is highly homologous to a rat molecule named OX-47 (27) or CE9 (47), the mouse basigin (26) or gp42 (48) molecule, and the chicken HT7 (49) or neurothelin (50) molecule. Previous analyses have shown that EMMPRIN size variation results from glycosylation heterogeneity (28). Interestingly, by screening a cDNA expression library with polyclonal sera raised against partially purified integrin ␤ 1 subunit (51), Altruda and co-workers (48) cloned the mouse EMMPRIN/gp42 molecule. Although this earlier study lacks firm biochemical evidence for the EMMPRIN-integrin interactions, to some extent it foreshadows our current finding.
From co-immunoprecipitation studies we estimate that at least 2-5% of the total ␣ 3 ␤ 1 (and ␣ 6 ␤ 1 ) integrin may be associated with EMMPRIN. However, this may be an underestima-tion due to the technical limitations of immunoprecipitation. Indeed, a much larger proportion of both ␣ 3 ␤ 1 and EMMPRIN appears to be co-localized by immunofluorescence and thus at least are in proximity, if not directly associated.
The EMMPRIN/basigin/HT7/OX-47/CE9/neurothelin molecule has been proposed to be a blood-brain barrier molecule (49), but knock out mice lacking this gene have not yet displayed alterations in blood-brain barrier function (52). Also, EMMPRIN has been suggested to be involved in cell-cell interactions. For example, a mAb against the chicken HT7/neurothelin protein caused a perturbation in cell contact-dependent maturation of glial cells in vitro (53). Indeed, our immunofluorescent studies showed that EMMPRIN co-localizes with ␣ 3 ␤ 1 integrin in cell-cell contacts and therefore may potentially interact with one another in a receptor-ligand fashion. In this regard, several transmembrane proteins in the immunoglobulin superfamily act as ligands for various integrins. For example, VCAM-1 binds to ␣ 4 ␤ 1 and ␣ 4 ␤ 7 (54); CD31/ PECAM-1 binds to the ␣ V ␤ 3 integrin (55); and ICAM-1, ICAM-2 and ICAM-3 bind to ␣ L ␤ 2 (56). However, expression of moderate to high levels of EMMPRIN in CHO cells did not result in increased adhesion to ␣ 3 -positive cells (not shown). Also, EMMPRIN-integrin co-precipitation was not diminished in the presence of EDTA, an agent known to inhibit nearly all integrin-ligand interactions (not shown). Furthermore, unlike EMMPRIN, no established integrin immunoglobulin-like ligands (VCAM-1, CD31, and ICAM) have ever been shown to co-precipitate with soluble integrins in detergent lysates. Therefore, it is possible that ␣ 3 ␤ 1 and ␣ 6 ␤ 1 associate with EMMPRIN in lateral fashion, similar to interactions of ␤ 3 integrins with CD47 (5, 6), another IgSF protein.
Several lines of evidence suggest that integrin-TM4SF and integrin-EMMPRIN complexes may represent two separate entities on the cell surface. First, immunofluorescence studies clearly showed that EMMPRIN is distributed in regions of cell-cell contact and diffusely throughout the membrane but not in focal complexes. In contrast, TM4SF proteins such as CD63 and CD81 can readily localize to focal complexes at the periphery of a spread cell, in a pattern that looks quite distinct from EMMPRIN (24). Second, ␣ 3 integrin was readily re-precipitated from an EMMPRIN immunoprecipitation, but the most prominent TM4SF protein on HT1080 cells, CD81, could not be detected. Finally, neither EMMPRIN itself nor EMMPRIN-integrin complexes appear to associate with phosphatidylinositol 4-kinase activity (not shown). In contrast, phosphatidylinositol 4-kinase activity is readily detected in association with several TM4 proteins and TM4-integrin complexes (24).
The functional relevance of integrin-EMMPRIN association is unclear. The preferential association of EMMPRIN with mature ␤ 1 integrins suggests that the interaction may not play a role during integrin chain association or biosynthetic processing. So far we have found that functions of mature integrins such as cell-cell adhesion or adhesion to extracellular matrix were not altered by either ectopic EMMPRIN expression or anti-EMMPRIN antibodies (not shown). Also, preliminary evidence suggests that ␣ 3 ␤ 1 -EMMPRIN association is not regulated by cell adhesion, since the stoichiometry of the ␣ 3 ␤ 1 -EMMPRIN interaction was comparable on K562 suspension cells and on HT1080 adherent cells. However, it is perhaps more than a coincidence that both EMMPRIN (29) and ␣ 3 ␤ 1 (57, 58) may help to regulate matrix metalloproteinase production. Future studies will be aimed at determining the extent to which the physical association between EMMPRIN and ␣ 3 ␤ 1 (or ␣ 6 ␤ 1 ) may play a role in adhesion-induced matrix metalloproteinase production.
In conclusion, here we have demonstrated a new strategy for characterization of cell-surface proteins that form complexes with integrins. By using this approach we have discovered that integrins (␣ 3 ␤ 1 and ␣ 6 ␤ 1 ) not only associate with TM4SF proteins but also with EMMPRIN, a member of the Ig superfamily. This represents the first clear demonstration that the EMMPRIN/basigin/OX47/M6 molecule can associate with other proteins.