Polydom/SVEP1 Is a Ligand for Integrin α9β1*

Background: Polydom/SVEP1 is a putative extracellular matrix protein of unknown function. Results: Polydom/SVEP1 is a potent ligand for integrin α9β1 and colocalizes with the integrin in vivo. Conclusion: Polydom/SVEP1 is a hitherto unknown high affinity ligand for integrin α9β1. Significance: The identification of this high affinity ligand offers important clues toward better understanding of the consequences of integrin α9β1-mediated cell-extracellular matrix interactions. A variety of proteins, including tenascin-C and osteopontin, have been identified as ligands for integrin α9β1. However, their affinities for integrin α9β1 are apparently much lower than those of other integrins (e.g. α3β1, α5β1, and α8β1) for their specific ligands, leaving the possibility that physiological ligands for integrin α9β1 still remain unidentified. In this study, we found that polydom (also named SVEP1) mediates cell adhesion in an integrin α9β1-dependent manner and binds directly to recombinant integrin α9β1 with an affinity that far exceeds those of the known ligands. Using a series of recombinant polydom proteins with N-terminal deletions, we mapped the integrin-binding site to the 21st complement control protein domain. Alanine-scanning mutagenesis revealed that the EDDMMEVPY sequence (amino acids 2636–2644) in the 21st complement control protein domain was involved in the binding to integrin α9β1 and that Glu2641 was the critical acidic residue for the integrin binding. The importance of this sequence was further confirmed by integrin binding inhibition assays using synthetic peptides. Immunohistochemical analyses of mouse embryonic tissues showed that polydom colocalized with integrin α9 in the stomach, intestine, and other organs. Furthermore, in situ integrin α9β1 binding assays using frozen mouse tissues showed that polydom accounts for most, but not all, of the integrin α9β1 ligands in tissues. Taken together, the present findings indicate that polydom is a hitherto unknown ligand for integrin α9β1 that functions as a physiological ligand in vivo.

A variety of proteins, including tenascin-C and osteopontin, have been identified as ligands for integrin ␣9␤1. However, their affinities for integrin ␣9␤1 are apparently much lower than those of other integrins (e.g. ␣3␤1, ␣5␤1, and ␣8␤1) for their specific ligands, leaving the possibility that physiological ligands for integrin ␣9␤1 still remain unidentified. In this study, we found that polydom (also named SVEP1) mediates cell adhesion in an integrin ␣9␤1-dependent manner and binds directly to recombinant integrin ␣9␤1 with an affinity that far exceeds those of the known ligands. Using a series of recombinant polydom proteins with N-terminal deletions, we mapped the integrin-binding site to the 21st complement control protein domain. Alanine-scanning mutagenesis revealed that the EDDMMEVPY sequence (amino acids 2636 -2644) in the 21st complement control protein domain was involved in the binding to integrin ␣9␤1 and that Glu 2641 was the critical acidic residue for the integrin binding. The importance of this sequence was further confirmed by integrin binding inhibition assays using synthetic peptides. Immunohistochemical analyses of mouse embryonic tissues showed that polydom colocalized with integrin ␣9 in the stomach, intestine, and other organs. Furthermore, in situ integrin ␣9␤1 binding assays using frozen mouse tissues showed that polydom accounts for most, but not all, of the integrin ␣9␤1 ligands in tissues. Taken together, the present findings indicate that polydom is a hitherto unknown ligand for integrin ␣9␤1 that functions as a physiological ligand in vivo.
Integrin ␣9␤1 is widely expressed in both adult and embryonic tissues, including airway epithelium, the basal layer of squamous epithelium, smooth muscle, skeletal muscle, hepatocytes, and corneal epithelia (2)(3)(4). Mice deficient in integrin ␣9 expression develop chylothorax and respiratory failure and die between 6 and 12 days after birth (5). Integrin ␣9␤1 is also expressed in the endothelial cells of lymphatic valves, and endothelial cell-specific depletion of integrin ␣9 in mouse embryos results in failure of lymphatic valve development, suggesting a role for integrin ␣9␤1 in lymphatic valve morphogenesis (6). Recent studies on skin-specific knockout of integrin ␣9␤1 revealed a crucial role for the integrin in re-epithelialization during cutaneous wound healing (7).
A panel of ECM proteins, including tenascin-C, osteopontin, and the fibronectin isoform containing an EIIIA domain have been shown to bind to integrin ␣9␤1 (8 -10). Some membrane proteins (ADAM proteases and vascular cell adhesion molecule (VCAM)-1) and growth factors (VEGF and NGF) have also been reported as ligands for integrin ␣9␤1 (11)(12)(13)(14). However, it remains to be explored whether these proteins function as integrin ␣9␤1 ligands in a physiological context. For example, integrin ␣9␤1 bound to the N-terminal fragment or polymeric form of osteopontin, but not to the intact protein, and also bound to the EIIIA domain of fibronectin, but not to a longer fragment (15)(16)(17). Furthermore, the binding specificities of these ligands have often been based on cell adhesion assays using integrin ␣9-expressing cells and/or cell adhesion inhibition assays with anti-integrin ␣9 antibodies (8 -14). Direct integrin ␣9␤1 binding assays have also been employed for tenascin-C, osteopontin, VCAM-1, VEGF, and NGF (13,14,16,18), although no comparative analyses of the affinities of these ligands toward integrin ␣9␤1 have been performed.
In this study, we identified polydom, a putative ECM protein containing an array of complement control protein (CCP) domains along with a von Willebrand factor A domain, a pentraxin domain, and multiple EGF-like domains, as another ligand for integrin ␣9␤1. Polydom was originally identified in murine bone marrow stromal cells as a protein containing EGFlike domains that share strong similarities with those in the Notch family proteins (19). The human orthologue of polydom, named SVEP1/SEL-OB, was reported as a protein expressed on the surface of osteogenic cells (20). We encountered polydom in our in silico screening for functionally unknown ECM proteins (21). Using recombinant expression and in vitro cell adhesion assays, we found that polydom is a secreted protein that mediates cell adhesion in an integrin ␣9␤1-dependent manner. Our findings show that polydom is a hitherto unknown ␣9␤1 ligand with an affinity that far exceeds those of the known ␣9␤1 ligands. Furthermore, polydom colocalizes with integrin ␣9␤1 in mouse tissues, supporting its role as a physiological ligand for integrin ␣9␤1.
cDNA Cloning and Construction of Expression Vectors-DNA segments encoding a FLAG tag and a His 6 tag were PCRamplified and inserted into the NotI/ApaI sites of the pcDNA3.1(ϩ) or pSecTag 2B vector (Invitrogen), yielding pcDNA-FLAG, pSec-FLAG, and pSec-His. For construction of expression vectors for N-terminal FLAG-tagged proteins, the DNA segment encoding the FLAG sequence was inserted into the HindIII/BamHI sites of the pSec-His vector to yield pSec-NFLAG-His. A cDNA encoding mouse polydom was obtained by RT-PCR using RNA extracted from mouse embryos at embryonic day 17 (E17) (Clontech, Palo Alto, CA) and subcloned into pBluescript KS(ϩ) (Stratagene, La Jolla, CA). After sequence verification, error-free cDNA fragments were inserted into pcDNA-FLAG, pSec-FLAG, pSec-His, or pSec-NFLAG-His at the BamHI/NotI sites. Expression vectors for truncated forms of polydom were also constructed by subcloning of PCR-amplified cDNAs into pSec-His at BamHI/NotI or HindIII/NotI sites.
Expression and Purification of Recombinant Polydom Proteins-Recombinant polydom and its fragments were produced using the FreeStyle 293 Expression System (Invitrogen). FreeStyle 293-F cells were transfected with the expression vectors using 293fectin (Invitrogen) and grown in serum-free Free-Style 293 expression medium for 72 h. The conditioned media were collected and clarified by centrifugation. To check the expression, the cells were lysed with TBS containing 1% (w/v) Nonidet P-40, 1 mM PMSF, 5 g/ml aprotinin, 5 g/ml leupeptin, and 5 g/ml pepstatin. The conditioned media and cell lysates were pulled down using Ni-NTA-agarose (Qiagen) or anti-FLAG M2-agarose (Sigma), respectively, and subjected to immunoblotting. For purification of the FLAG-tagged proteins, the conditioned media were applied to an anti-FLAG M2-agarose column, and the bound proteins were eluted with 100 g/ml FLAG peptide in PBS. For purification of the His-tagged proteins, the conditioned media were subjected to affinity chromatography using Ni-NTA-agarose. The columns were washed with PBS, and the bound proteins were eluted with PBS containing 200 mM imidazole. The eluted proteins were dialyzed against PBS. The concentrations of the purified proteins were determined by the Bradford assay using BSA as a standard.
Anti-polydom Antibodies-Antibodies against polydom were raised in rabbits using FLAG-tagged full-length polydom as an immunogen. The antibodies were affinity-purified on columns containing either the N-terminal or C-terminal polydom fragment, designated pol-N (amino acids 18 -789) and pol-C (amino acids 1192-3567), respectively, with a C-terminal His 6 tag. The columns were washed with 10 mM Tris-HCl buffer (pH 7.5) containing 0.5 M NaCl, followed by elution of the bound antibodies with 0.1 M glycine-HCl (pH 2.8). The eluted fractions were immediately neutralized with 1 M Tris-HCl (pH 9.0) and dialyzed against PBS. An Alexa Fluor 555-conjugated anti-polydom (pol-N) antibody was prepared using the APEX Alexa Fluor antibody labeling kit.
SDS-PAGE, Western Blotting, and Amino Acid Sequencing-SDS-PAGE was carried out according to Laemmli (28). The separated proteins were visualized by staining with silver nitrate or Coomassie Brilliant Blue or transferred onto nitrocellulose (for immunoblotting) or polyvinylidene difluoride (for amino acid sequencing) membranes. For immunoblotting, the membranes were probed with antibodies and then visualized with an ECL detection kit (GE Healthcare). For amino acid sequencing, the protein bands on the membrane were visualized by Coomassie Brilliant Blue staining and then dissected out. The N-terminal amino acid sequences were analyzed by the method of Edman (29) using a Procise 491 cLC protein sequencer (Applied Biosystems, Carlsbad, CA).
Preparation of Tissue Extracts-Mouse E15.5 lungs were homogenized in SDS-PAGE sample buffer (2% SDS, 10% glycerol, 125 mM Tris-HCl, pH 6.8), and the protein concentrations of the homogenates were determined using a BCA protein assay reagent kit (Pierce). The homogenates were boiled, and the extracted proteins were subjected to SDS-PAGE and Western blotting.
Cell Adhesion Assays-Cell adhesion assays were performed as described (30). For the assays involving blocking antibodies, RD cells were preincubated with the specified mAbs (10 g/ml) for 15 min at room temperature and plated on 96-well plates coated with recombinant polydom proteins or fibronectin. The cells were incubated for 30 min, washed, fixed with 3.7% formaldehyde, and stained with 0.5% toluidine blue.
Flow Cytometry-The integrin ␣9␤1 expression levels on A549, HT1080, and RD cells were verified by flow cytometry using anti-integrin ␣9␤1 mAb Y9A2. Mouse normal IgG was used as a control.
RNA Interference-RD cells were grown on 6-well plates in DMEM containing 10% FBS. The cells were transfected with 30 pmol of siRNA using Lipofectamine TM RNAiMAX (Invitrogen). After 5 h, the cells were passaged into 10-cm dishes. At 3 days after transfection, the cells were subjected to flow cytometry or cell adhesion assays.
Expression and Purification of Recombinant Integrins-A cDNA encoding the extracellular region of human integrin ␣9 was amplified by RT-PCR using total RNA extracted from RD cells and cloned into pBluescript KS(ϩ). After sequence verification, an error-free cDNA fragment was inserted into pcDNA-ACID-FLAG (31). An expression vector for the extracellular regions of integrin ␤1 was provided by Dr. Junichi Takagi (Osaka University, Osaka, Japan) (25). The recombinant integrins were produced using the FreeStyle 293 expression system and purified as described (31).
Integrin Binding Assay-Integrin binding assays were performed as described previously (31). In some experiments, integrins were preincubated with synthetic peptides at various concentrations to evaluate their inhibitory activities. The apparent dissociation constants were determined as described (32).
Expression and Purification of GST-fused Polydom, Tenascin-C, and Osteopontin Fragments-cDNAs encoding the 21st CCP domain or its deletion mutants were amplified by PCR and subcloned into pGEX4T-1 (GE Healthcare) at the EcoRI/SalI sites. The 21st CCP domain and the same domain lacking the Glu 2628 -Ser 2664 segment, designated ⌬D2628-S2664, were His 6 -tagged at their C termini. GST fusion proteins were induced in Escherichia coli BL21 by incubation with 0.1 mM isopropyl ␤-D-thiogalactopyranoside at 25°C for 2 h. The cells were lysed by sonication, and the supernatants were passed over a glutathione-Sepharose 4B (GE Healthcare) column. The bound proteins were eluted with 50 mM Tris-HCl (pH 8.0) containing 10 mM glutathione. The GST fusion proteins of the intact and mutant 21st CCP domains were further purified on an Ni-NTA-agarose column. The purified proteins were dialyzed against 20 mM HEPES buffer (pH 8.0) containing 130 mM NaCl and quantified by the Bradford assay. A cDNA encoding the third fibronectin type III domain of human tenascin-C (TNfn3) was obtained by RT-PCR using RNA extracted from HT1080 cells and subcloned into pBluescript KS(ϩ). A cDNA encoding human osteopontin was purchased from the National Institutes of Health Mammalian Gene Collection (Invitrogen), and a cDNA encoding the osteopontin N-terminal fragment (OPN-Nhalf) was amplified by PCR. The RGD motifs within TNfn3 and OPN-Nhalf were then mutated to RAA for specific binding to integrin ␣9␤1 (referred to as TNfn3-RAA and OPN-Nhalf-RAA, respectively) (8,33). After sequence verification, the cDNA fragments were inserted into pGEX4T-1 at the EcoRI/SalI sites. Both proteins were expressed and purified as described above.
In Situ Integrin Binding Assay-Frozen sections (8 -10-m thickness) of mouse embryos were fixed with ice-cold acetone for 15 min, washed with TBS, and blocked with 1% BSA in TBS. After three washes with TBS, the sections were incubated with recombinant integrin ␣9␤1 (3 g/ml) in TBS containing 1% BSA and 1 mM MnCl 2 at 4°C overnight. The sections were then washed with TBS containing 1 mM MnCl 2 (TBS/Mn) and incubated with the anti-Velcro antibody (0.5 g/ml) in TBS/Mn containing 1% BSA (TBS/Mn/BSA) for 2 h at room temperature. After washing with TBS/Mn, the sections were incubated with Alexa Fluor 488-conjugated goat anti-rabbit IgG in TBS/ Mn/BSA for 2 h to visualize the bound anti-Velcro antibody. For double immunofluorescence staining with the anti-polydom or anti-tenascin-C antibody, sections were incubated with 100 g/ml normal rabbit IgG (Dako) in TBS/Mn/BSA for 1 h to saturate the antigen-binding sites of the bound Alexa Fluor 488-conjugated goat anti-rabbit IgG. After washing with 20 mM HEPES buffer (pH 7.5) containing 130 mM NaCl and 1 mM MnCl 2 , the sections were refixed with 3.7% formaldehyde in 20 mM HEPES buffer (pH 7.5) containing 130 mM NaCl and 1 mM MnCl 2 for 10 min. The sections were then washed with TBS and incubated with Alexa Fluor 555-conjugated anti-pol-N antibody or the anti-tenascin-C antibody in TBS containing 1% BSA at 4°C overnight. After washing in TBS, the bound antitenascin-C antibody was detected with Alexa Fluor 546-conjugated goat anti-rat IgG in TBS containing 1% BSA for 2 h. The sections were washed with TBS and mounted with fluorescent mounting medium (Dako).

RESULTS
Polydom Is an ECM Protein-Previously, we established a bioinformatics-based protocol for screening ECM proteins from transcriptome databases, and identified 16 new ECM proteins from Ͼ65,000 mouse cDNAs available in the RIKEN FANTOM cDNA collection (21). We extended this protocol to the Ensembl mouse and human genome databases to identify more ECM proteins by focusing on transcripts encoding proteins with Ͼ1500 amino acid residues. One of the obtained candidates was polydom (also named SVEP1), a putative secreted protein harboring EGF domains with strong similarities to those in the Notch family proteins (19,20). Polydom comprises multiple domains characteristic of ECM proteins (i.e. a von Willebrand factor A domain, a pentraxin domain, 10 EGF-like domains, and 34 CCP (also called "sushi") domains) (19,20) (Fig. 1A). To explore the biological functions of polydom, we amplified the cDNA encoding full-length mouse polydom by RT-PCR and expressed it in 293-F cells with a C-terminal FLAG tag. The recombinant polydom was detected in the culture medium and enriched by immunoprecipitation with an anti-FLAG antibody (Fig. 1B, lane 1), confirming that polydom is a secreted protein. However, the amount of the recombinant protein secreted into the medium was very low, yielding only a faint band on immunoblotting detection. To increase the expression level, recombinant polydom was expressed with a mouse immunoglobulin leader sequence and a C-terminal FLAG tag (Fig. 1B, lane 2) and purified from the spent medium by immunoaffinity chromatography using an anti-FLAG mAb. The purified recombinant polydom gave two major bands migrating at 250 and 100 kDa on SDS-PAGE under non-reducing conditions (Fig. 1C). Similar banding patterns were obtained under reducing conditions except that the two major bands migrated more slowly (270 and 115 kDa), suggesting that both were enriched in intrapeptide disulfide bonds. Determination of the N-terminal amino acid sequences of the two protein bands revealed that the 115 kDa band started with DAAQ, the putative N-terminal sequence after removal of the mouse immunoglobulin-derived leader sequence, whereas the 270 kDa band started with VAPG, which is identical to amino acids 1177-1180 located N-terminal to the first EGF domain (see Fig.  1A). These findings indicated that polydom was proteolytically processed, before or after secretion, into two parts (i.e. the N-terminal 115-kDa fragment and C-terminal 270-kDa fragment), which remained associated after the proteolytic cleavage.
To corroborate this conclusion, we expressed recombinant polydom with two tags, a FLAG tag at the N terminus and a His 6 tag at the C terminus, in 293-F cells and analyzed both the spent media and cell lysates by precipitation with anti-FLAG or Ni-NTA beads. Both the 270-kDa and 115-kDa fragments were recovered in the precipitates from the spent medium irrespective of the type of beads used (Fig. 1D), thereby confirming that these fragments remain associated. Notably, a high molecular weight band migrating in the Ͼ300 kDa region was detected in the precipitates from cell lysates with both the anti-FLAG and Ni-NTA beads. It therefore seems likely that the proteolytic processing into the N-terminal 115-kDa and C-terminal 270-kDa fragments occurs after secretion.
To examine whether polydom is an ECM protein, we produced antibodies against polydom and used them as immunohistochemical probes. Because polydom was cleaved into the N-terminal and C-terminal parts after secretion, we raised antibodies against recombinant full-length polydom in rabbits, followed by affinity purification of the antibodies using columns containing recombinant fragments for either amino acids 18 -789 (designated pol-N) or amino acids 1192-3567 (designated pol-C) (see Fig. 1A). The specificities of these antibodies were verified by Western blotting using recombinant pol-N and pol-C fragments (Fig. 1E) as well as tissue extracts from E15.5 mouse embryo lungs (Fig. 1F). It should be noted that a higher molecular weight band migrating at the Ͼ300 kDa region was also detected by both the anti-pol-N and anti-pol-C antibodies in the tissue extracts, together with the N-terminally derived 90-kDa and C-terminally derived 250-kDa fragments (Fig. 1F,  arrowheads), supporting the conclusion that polydom is proteolytically processed into two fragments both in vivo and in vitro. We used these antibodies for immunohistochemical staining of frozen sections of mouse embryos at E16.5 and obtained essentially the same results ( Fig. 1G and supplemental Fig. S1). Polydom was detected in a variety of tissues, including the stomach, intestine, and lungs, where it was predominantly detected in the mesenchymal ECM. Polydom was also detected in the kidneys, liver, nerve fiber bundles, and choroid plexus (data not shown). Given the same staining patterns with the pol-N and pol-C antibodies, the N-terminal and C-terminal fragments appear to remain associated even after deposition in the ECM.
Polydom Mediates Integrin ␣9␤1-dependent Cell Adhesion-To explore the biological functions of polydom, we examined whether it was capable of promoting cell adhesion using three human cell lines: A549 lung adenocarcinoma, HT1080 fibrosarcoma, and RD rhabdomyosarcoma. Although A549 and HT1080 cells failed adhere to polydom (data not shown), RD cells adhered to polydom in a coating concentration-dependent manner, attaining the maximum adhesion at 3 g/ml ( Fig. 2A). When the substrates were coated with either the recombinant pol-N or pol-C fragments, only the pol-C fragment was capable of promoting cell adhesion, with a potency that was almost equivalent to that of full-length polydom. The cells adhering to polydom and the pol-C fragment assumed spread morphology, although the extent of the spreading was less pronounced than that of cells adhering to fibronectin (Fig. 2B). These findings indicated that polydom is capable of mediating cell adhesion was expressed in 293-F cells with a C-terminal FLAG tag. The conditioned media (med), cell lysates (cell), and anti-FLAG immunoprecipitates from conditioned media (IP) were subjected to 6% SDS-PAGE under non-reducing conditions and subsequent immunoblotting with an HRP-conjugated anti-FLAG mAb. Untransfected 293-F cells were similarly processed in parallel (lane 3). C, affinity-purified FLAG-tagged polydom was subjected to 6% SDS-PAGE under non-reducing (NR) or reducing (R) conditions and subsequent silver staining. The N-terminal amino acid sequences determined by peptide sequencing are indicated on the right. D, recombinant polydom with both an N-terminal FLAG tag and a C-terminal His 6 tag was transfected into 293-F cells and then pulled down from culture media (med) and cell lysates (cell) using Ni-NTA (His) or anti-FLAG mAb-conjugated (FLAG) agarose. The precipitates were subjected to 5% SDS-PAGE under reducing conditions and analyzed by immunoblotting with anti-His (left) and anti-FLAG (right) mAbs. Untransfected (lane 1) and transfected (lane 2) cells were analyzed in parallel. Note that uncleaved polydom is detected in the cell lysates (arrowheads). E, recombinant pol-N or pol-C fragments were subjected to 6% SDS-PAGE under non-reducing conditions, followed by Coomassie Brilliant Blue staining (left) or immunoblotting with anti-pol-N (middle) and anti-pol-C (right) antibodies. Fragments at the 90 and 250 kDa regions were detected by the anti-pol-N and anti-pol-C antibodies, respectively (arrows). Oligomeric forms of the N-terminally derived fragment were also detected by the anti-pol-N antibody (arrowheads). Degradation products of the C-terminally derived fragment were also detected by the anti-pol-C antibody (asterisk). F, mouse E15.5 lung extracts (ϳ100 g of protein for the anti-pol-N antibody and ϳ35 g of protein for the anti-pol-C antibody) were subjected to 5% SDS-PAGE under non-reducing conditions, followed by immunoblotting with the anti-pol-N (left) and anti-pol-C (right) antibodies. Both antibodies detected the Ͼ300-kDa full-length protein (arrowheads) as well as the 90-kDa (anti-pol-N) or 250-kDa (anti-pol-C) fragment (arrows). G, cryosections of mouse E16.5 embryos were stained with an anti-pol-N antibody. Representative images of the stomach, intestine, and lung are shown. The bottom panels show magnified views of the boxed areas in the top panels. Bars, 100 m. and subsequent cell spreading and that the cell adhesion-promoting activity resides within the C-terminal region composed of an array of CCP domains.
Next, we explored whether the adhesion of RD cells to polydom was dependent on integrins. The spread morphology of the cells adhering to polydom was indicative of the involvement of integrin-dependent cytoskeletal reorganization. In support of this possibility, the mAb against integrin ␤1 strongly inhibited the adhesion of RD cells to polydom. Among the mAbs against various integrin ␣ subunits, the mAbs against integrins ␣2, ␣3, ␣5, and ␣6 did not inhibit cell adhesion to polydom, whereas the mAb against integrin ␣9␤1 was strongly inhibitory toward the RD cell adhesion to polydom (Fig. 2C). These findings are consistent with the failure of A549 and HT1080 cells to adhere to polydom because integrin ␣9␤1 was expressed on RD cells (34) but not on A549 or HT1080 cells (supplemental Fig.  S2A). Thus, it seems likely that the adhesion to polydom is primarily mediated by integrin ␣9␤1.
To further investigate the role of integrin ␣9␤1 in cell adhesion to polydom, we knocked down the expression of integrin ␣9 in RD cells by RNA interference. A significant reduction in the level of integrin ␣9 expression was verified by flow cytometric analysis (supplemental Fig. S2B). The siRNA-mediated integrin ␣9 knockdown resulted in an ϳ70% reduction in the number of cells adhering to pol-C, whereas cells treated with a control siRNA did not exhibit any significant decrease (Fig. 2D). The integrin ␣9 knockdown did not cause any reduction in cell adhesion to fibronectin, confirming that polydom is a specific ligand for integrin ␣9␤1.
Because tenascin-C and osteopontin are well known ligands for integrin ␣9␤1, we compared the cell adhesive activity of polydom with those of tenascin-C and osteopontin. Integrin ␣9␤1 has been shown to bind to TNfn3 and the OPN-Nhalf (8,9). We recombinantly expressed and purified these ␣9␤1 ligands as GST fusion proteins, in which the RGD cell-adhesive motifs were replaced with RAA to nullify their abilities to interact with RGD-binding integrins, such as those containing the ␣V subunit (8,33). RD cells adhered only poorly to GST-Tnfn3-RAA and GST-OPN-Nhalf-RAA, even at the highest coating concentration (i.e. Ͼ30 g/ml), whereas the cells were fully competent in adhering to the recombinant pol-C fragment, attaining the maximum cell-adhesive activity at 7 g/ml (Fig.  2E). These results indicate that polydom is more potent than tenascin-C and osteopontin as an integrin ␣9␤1 ligand.
Polydom Is a Preferred Ligand for Integrin ␣9␤1-To corroborate the role of integrin ␣9␤1 as an adhesion receptor for polydom, we performed direct integrin binding assays using recombinant integrin ␣9␤1, which was expressed and purified as a truncated disulfide-linked heterodimer with C-terminal FLAG (␣9) and His 6 (␤1) tags (31). The authenticity of the resulting integrin ␣9␤1 was verified by SDS-PAGE and immunoblotting (Fig. 3, A and B). Solid-phase binding assays showed that the recombinant integrin ␣9␤1 bound to full-length polydom and the pol-C fragment but not to the pol-N fragment (Fig. 3C). The binding of integrin ␣9␤1 to polydom was completely abrogated in the presence of EDTA (Fig. 3C), thereby confirming the specificity of the integrin binding assays. Although cellular fibronectin containing the EIIIA domain is a putative ligand for integrin ␣9␤1, recombinant integrin ␣9␤1 exhibited only marginal binding activity toward recombinant cellular fibronectin, which was fully active in binding to integrin ␣5␤1 (data not shown). These findings, together with those of the cell adhesion assays, corroborate the notion that integrin ␣9␤1 binds directly to polydom, which has its integrin-binding site in the C-terminal 270-kDa region.
Next, we compared the integrin ␣9␤1 binding activity of polydom with those of other known integrin ␣9␤1 ligands (i.e. tenascin-C (TNfn3-RAA) and osteopontin (OPN-Nhalf-RAA)). Although TNfn3-RAA and OPN-Nhalf-RAA were capable of binding to integrin ␣9␤1, their affinities for integrin ␣9␤1 were significantly lower than that of pol-C (Fig. 3D). We were unable to determine the apparent dissociation constants for these ␣9␤1 ligands due to incomplete saturation of binding at the highest integrin concentration, whereas the dissociation constant between integrin ␣9␤1 and pol-C was estimated from three independent determinations to be 32.4 Ϯ 2.7 nM. These results were consistent with those obtained by the cell adhesion assays and support the conclusion that polydom is a more potent ligand for integrin ␣9␤1 than tenascin-C and osteopontin. Integrin ␣9␤1 Binds to the 21st CCP Domain-To locate the integrin ␣9␤1-binding site within polydom, we constructed a series of N-terminal deletion mutants of pol-C (Fig. 4A) and examined their binding activities toward integrin ␣9␤1. Although deletion up to the 20th CCP domain (designated ⌬CCP20) did not compromise the integrin binding activity of pol-C, deletion of the 21st CCP domain (designated ⌬CCP21) resulted in a dramatic loss of the activity (Fig. 4B), thereby underscoring the critical role of the 21st CCP domain (hereafter referred to as CCP21). To explore whether CCP21 harbors the integrin ␣9␤1-binding site, we produced recombinant CCP21 as a GST fusion protein and examined its integrin ␣9␤1 binding activity. CCP21 alone was fully active in binding to integrin ␣9␤1 (Fig. 4B), confirming the critical role of CCP21 in the binding of polydom to integrin ␣9␤1.
Among the 34 CCP domains within polydom, CCP21 is unique because it contains extra segment of ϳ40 amino acids compared with the other CCP domains (Fig. 4C and supplemental Fig. S3) (19). To examine whether the extra segment within CCP21 is involved in ␣9␤1 binding, we deleted this segment (i.e. 37 amino acids encompassing Asp 2628 -Ser 2664 ) from the GST-CCP21 fusion protein (designated ⌬D2628-S2664; see   4C). The resulting CCP21 mutant was unable to bind to integrin ␣9␤1 (Fig. 4D), indicating the involvement of this extra segment in integrin ␣9␤1 binding to CCP21. We also produced a GST fusion containing only the extra segment (Asp 2628 -Ser 2664 ) and found that this extra segment retained the integrin binding activity of CCP21. These findings support the conclusion that the binding site for integrin ␣9␤1 is located within the extra 37-amino acid segment of CCP21.
Integrin ␣9␤1 Recognizes the Sequence EDDMMEVPY-To further narrow down the region responsible for ␣9␤1 binding, we divided the extra 37-amino acid segment of CCP21 into N-terminal and C-terminal halves and determined their integrin binding activities (Fig. 5A). Integrin ␣9␤1 only bound to the N-terminal segment Asp 2628 -Leu 2645 . We then divided the N-terminal segment into two partially overlapping segments, Asp 2628 -Asp 2638 and Asp 2634 -Leu 2645 . Only the C-terminally derived segment Asp 2634 -Leu 2645 was capable of binding to integrin ␣9␤1 (Fig. 5A, D2634-L2645), showing that the binding site for integrin ␣9␤1 could be mapped to the 12-amino acid segment DQEDDMMEVPYL within the extra segment of CCP21.
To identify the residues involved in binding to integrin ␣9␤1, we performed alanine-scanning mutagenesis of the DQEDDM-MEVPYL segment. Alanine substitution of Glu 2641 almost completely abrogated the integrin binding activity, whereas mutations at the residues from Glu 2636 to Tyr 2644 caused partial reductions in the integrin binding activity to variable extents (Fig. 5B). These findings indicate that integrin ␣9␤1 recognizes the EDDMMEVPY sequence, within which Glu 2641 is the critical acidic residue involved in the polydom recognition by integrin ␣9␤1.
To corroborate this conclusion, we produced two series of deletion mutants of the DQEDDMMEVPYL segment (i.e. those with N-terminal deletions and those with C-terminal deletions) (Fig. 5C). Integrin binding assays of the N-terminal deletion mutants revealed that deletion of Glu 2636 and Asp 2637 caused a small stepwise reduction in the integrin binding activity, whereas deletion of Asp 2638 resulted in a dramatic loss of the activity. Similarly, deletion of Pro 2643 and Val 2642 from the C terminus of the DQEDDMMEVPYL segment caused a stepwise decrease in the integrin binding activity, although deletion of Tyr 2644 did not. These findings were consistent with the conclusion that EDDMMEVPY is the recognition sequence for integrin ␣9␤1, except that the involvement of Tyr 2644 was not evident with the C-terminal deletion mutants.
To further corroborate the role of EDDMMEVPY as the integrin ␣9␤1 recognition sequence, we performed the integrin inhibition assay with a synthetic EDDMMEVPY peptide. The EDDMMEVPY peptide inhibited the interaction of the pol-C fragment with integrin ␣9␤1 in a dose-dependent manner, with an IC 50 of 0.18 M (Fig. 5D). A smaller peptide, DMMEVPY, was nearly as potent as EDDMMEVPY in inhibiting integrin ␣9␤1 binding, consistent with the relatively small contribution of the N-terminal Glu-Asp residues to the interaction with integrin ␣9␤1. Further deletion of the C-terminal Tyr residue from DMMEVPY resulted in a significant reduction in the inhibitory potency, supporting the involvement of Tyr 2644 in the CCP21 recognition by integrin ␣9␤1. Alanine substitution for Glu 2641 in the DMMEVPY peptide resulted in a marked reduction of its inhibitory potency, consistent with the importance of Glu 2641 in the recognition of polydom by integrin ␣9␤1.
We also compared the inhibitory activity of the EDDM-MEVPY peptide with those of the synthetic peptides AEIDG-IEL and TYSSPEDGIHE, which represent the integrin ␣9␤1 recognition sequences in tenascin-C and the fibronectin EIIIA domain, respectively (17,35). AEIDGIEL was capable of inhibiting the integrin ␣9␤1 binding to pol-C, although its potency was more than 1 order of magnitude less than that of EDDM-MEVPY, with an IC 50 of 7.8 M (Fig. 5D). TYSSPEDGIHE was barely inhibitory toward the interaction between integrin ␣9␤1 and pol-C. These findings support the conclusion that integrin ␣9␤1 binds to polydom with an affinity that is significantly higher than those to tenascin-C and that EDDMMEVPY is a preferred sequence for ligand recognition by integrin ␣9␤1.
Polydom Partially Colocalizes with Integrin ␣9 in Tissues-The relatively high binding affinity of polydom toward integrin ␣9␤1 suggests that polydom functions as a physiological ligand of integrin ␣9␤1. To address this possibility, we performed immunofluorescence staining of mouse embryonic tissues with antibodies against polydom and integrin ␣9. Polydom was colocalized with integrin ␣9 at the submucosal mesenchyme in the stomach and intestine (Fig. 6, A-F). Integrin ␣9 was also expressed in the smooth muscle layer in these tissues, where polydom was barely expressed. Polydom and integrin ␣9 were also colocalized at the sinusoids in the liver (Fig. 6, G-I) and Bowman's capsules (arrowheads) and the mesenchyme between renal tubules (arrows) in the kidney (Fig. 6, J-L). In the lung, polydom was detected in the mesenchyme, where integrin ␣9 was only partially colocalized (Fig. 6, M-O). Integrin ␣9 was strongly expressed in the smooth muscle layer of the lung, where polydom was barely detectable. These findings demonstrated that polydom was colocalized with integrin ␣9 in the embryonic mesenchyme of various organs, except for the smooth muscle layers where integrin ␣9 was highly expressed, consistent with the possibility that polydom serves as one of the physiological ligands of integrin ␣9␤1.
To further corroborate the role of polydom as a physiological ligand for integrin ␣9␤1, we performed in situ integrin overlay assays to visualize the integrin ␣9␤1 ligands in tissues as a whole. Incubation of frozen tissue sections of mouse embryos with recombinant integrin ␣9␤1 detected its ligands in the mesenchyme and smooth muscle layer of the stomach, intestine, and lung when the incubation was carried out in the presence of Mn 2ϩ (Fig. 7, A, E, and I). No signals were detected when the assay was performed in the presence of EDTA (Fig. 7, D, H, and  L), confirming the specificity of the in situ integrin overlay assay. Double immunofluorescence detection of polydom using an Alexa Fluor 555-conjugated anti-polydom antibody demonstrated that the signals for polydom overlapped with those for bound integrin ␣9␤1 in the mesenchymal region of the stomach, intestine, and lung (Fig. 7, C, G, and K). We also performed double immunofluorescence detection using an anti-tenascin-C antibody (supplemental Fig. S4). The signals for tenascin-C overlapped with those for bound integrin ␣9␤1 in the smooth muscle layer of the stomach and part of the mesen-

Polydom Is a Ligand for Integrin ␣9␤1
JULY 20, 2012 • VOLUME 287 • NUMBER 30 chyme in the lung, although tenascin-C was not detected in the submucosal mesenchyme in the stomach, where signals for polydom and bound integrin ␣9␤1 were colocalized (Fig. 7C). Given the potent binding affinity of polydom toward integrin ␣9␤1, these results support the possibility that polydom functions as a physiological ligand for integrin ␣9␤1, complementing tenascin-C and other known integrin ␣9␤1 ligands. It should also be noted that strong signals for bound integrin ␣9␤1 were detected in the smooth muscle layer surrounding the trachea, where polydom and tenascin-C were barely expressed (arrows in Fig. 7K and supplemental Fig. S4F), indi-cating the presence of one or more other ligands for integrin ␣9␤1 in these regions.

DISCUSSION
In the present study, we obtained evidence that polydom is an ECM protein that functions as a ligand for integrin ␣9␤1. Polydom was secreted into the culture media when transfected into 293-F cells. Deposition to the ECM in vivo was demonstrated by immunohistochemistry. Polydom was capable of mediating cell-to-substrate adhesion of RD cells that express integrin ␣9 but not of A549 or HT1080 cells that do not express integrin ␣9. The adhesion of RD cells to polydom was specifically inhibited by treatment of the cells with anti-integrin ␣9␤1 antibodies and by siRNA-mediated knockdown of integrin ␣9, indicating that the cell adhesion to polydom was integrin ␣9␤1dependent. In support of the conclusion, recombinant integrin ␣9␤1 bound to polydom in a divalent metal ion-dependent manner with an affinity that was significantly greater than those to tenascin-C and osteopontin, which are two well known ECM ligands for integrin ␣9␤1. Polydom was partially colocalized with integrin ␣9 in mouse embryonic tissues. Furthermore, the in situ integrin overlay assay, which visualizes the entire ligands available for a given integrin overlaid on tissue sections, demonstrated that the immunohistochemical signals for polydom significantly overlapped with those obtained by the integrin ␣9␤1 overlay. Taken together, these findings led us to conclude that polydom is an ECM protein that functions as a ligand for integrin ␣9␤1 in vivo.
To date, a number of proteins have been shown to bind to integrin ␣9␤1. These include tenascin-C, osteopontin, the EIIIA domain of fibronectin, ADAMs, and VEGF (8 -10, 13, 16, 17, 35-39). Among these, tenascin-C, and particularly its third fibronectin type III domain (TNfn3), has been used as a representative for integrin ␣9␤1 ligands. Our results showed that polydom bound to integrin ␣9␤1 with higher affinity than TNfn3, with a dissociation constant of 32.4 Ϯ 2.7 nM. This dissociation constant is comparable with those for the interactions of integrin ␣6␤4 with laminin-332 (12 Ϯ 3 nM) and laminin-511/521 (25 Ϯ 1 nM) (31), suggesting that polydom is a more preferred physiological ligand for integrin ␣9␤1 than tenascin-C. Indeed, polydom can explain, at least in part, the integrin ␣9␤1 ligands detected by the in situ integrin ␣9␤1 overlay assay in the submucosal mesenchyme of the developing gastrointestinal tract. It should be noted, however, that polydom cannot explain the integrin ␣9␤1 ligands in the smooth muscle layers in the stomach, intestine, and lung, where polydom was barely expressed. Recently, EMILIN-1 has been shown to be a ligand for integrin ␣9␤1 (40). Because EMILIN-1 is highly expressed in smooth muscle cells (41), it may account for the major integrin ␣9␤1 ligands in these smooth muscle layers. It remains to be determined whether EMILIN-1 binds to integrin ␣9␤1 with an affinity that is comparable with that of polydom.
The peptide sequences recognized by integrin ␣9␤1 have been reported for a number of integrin ␣9␤1 ligands. These sequences include AEIDGIEL in TNfn3 (35) and TYS-SPEDGIHE in the EIIIA domain of fibronectin (17). In the present study, we found that EDDMMEVPY is the polydom sequence recognized by integrin ␣9␤1. This conclusion was drawn from the integrin ␣9␤1 binding assays using a series of alanine substitution as well as N-terminal and C-terminal deletion mutants of CCP21, which harbors the integrin ␣9␤1 binding activity of polydom. The importance of the EDDMMEVPY sequence for binding to integrin ␣9␤1 was further confirmed by integrin binding inhibition assays with a synthetic EDDM-MEVPY peptide, which exhibited a potent inhibitory activity with an IC 50 of 0.18 M, whereas AEIDGIEL and TYS-SPEDGIHE were about Ͼ10-fold and Ͼ100-fold less potent than EDDMMEVPY.
In the EDDMMEVPY sequence, the sixth Glu residue (Glu 2641 ) is critical for integrin ␣9␤1 binding, because alanine substitution of this residue completely abolished the integrin ␣9␤1 binding activity of CCP21. Consistent with this conclusion, the DMMAVPY peptide, in which alanine was substituted for Glu 2641 , was Ͼ50-fold less potent than the control DMMEVPY peptide in inhibiting the interaction between integrin ␣9␤1 and polydom. It is known that integrin ligands contain a critical acidic residue for their interactions with integrins (42)(43)(44). Accumulating evidence indicates that the carboxyl group of the acidic residue coordinates the divalent metal ion in the so-called metal ion-dependent adhesion site (MIDAS), thereby securing the binding to integrins (45). It therefore seems likely that Glu 2641 is the acidic residue that coordinates the metal ion in the MIDAS of the integrin ␤1 chain. However, one may argue against this conclusion because deletion of this Glu residue from the DQEDDMME peptide did not cause a further reduction in the integrin binding activity (Fig. 5C). A possible explanation for the apparent discrepancy would be that the ␣-carboxyl group of the C-terminal methionine residue of the resulting DQEDDMM peptide may play a role that is functionally equivalent to the side chain of Glu 2641 in coordinating the metal ion in the MIDAS. In support of this possibility, the crystal structure of integrin ␣IIb␤3 complexed with the peptide derived from the C-terminal region of the ␥ subunit of fibrinogen, referred to as the ␥C peptide, revealed that the free ␣-carboxyl group of the C-terminal residue of the ␥C peptide coordinates the divalent metal cation at the "adjacent to MIDAS" (ADMIDAS) region of integrin ␤3 (46). Thus, the free ␣-carboxyl group of the C-terminal methionine residue of the DQEDDMM peptide may coordinate the divalent metal ion at the MIDAS or ADMIDAS of the ␤1 subunit of integrin ␣9␤1 and thereby partially substitute for the side chain of Glu 2641 to stabilize the binding of integrin ␣9␤1 to the truncated peptide.
It is interesting to note that most, if not all, of the sequences reported as integrin ␣9␤1 recognition sites contain a hydrophobic residue that precedes the acidic residue critical for binding to the integrin (Fig. 8). Furthermore, another acidic residue is often situated two residues N-terminal to the hydrophobic residue, as has been shown for the proposed ␣9␤1 recognition sequences of tenascin-C (AEIDGIEL) and ADAM-15 (DLPEF) (11,35). The DXXE motif is well conserved among the ϳ30 members of the ADAM family proteases, except for ADAM-10 and ADAM-17, which both lack integrin ␣9␤1-dependent celladhesive activity (11). Given that the EDDMMEVPY sequence in polydom also agrees with this rule, the DX⌽E motif (where ⌽ represents a hydrophobic residue) seems to be the preferred sequence recognized by integrin ␣9␤1 (Fig. 8). Interestingly, a motif similar to DX⌽E is also found in the ligands recognized by integrin ␣4␤1 (44, 47) (i.e. fibronectin (PEILDVPST) and EMI-LIN-1 (PEGLENKP, the latter of which was recently shown to bind to integrin ␣9␤1 as well) (40). Consistent with this similarity between integrins ␣4␤1 and ␣9␤1 in their recognition sequences, integrin ␣4 exhibits the highest sequence homology with integrin ␣9 among the 18 integrin ␣ subunits (2). Other integrin ␣4␤1 ligands contain a ⌽(E/D) motif in their integrin recognition sequences, including IDSP in VCAM-1 (47) and IDAP in the C-terminal heparin-binding domain of fibronectin (48), underscoring the importance of the hydrophobic residue preceding the critical acidic residue in ligand recognition by integrins ␣4␤1 and ␣9␤1 (Fig. 8).
Although the coordination of the carboxyl group of an acidic residue with the divalent metal ion in the MIDAS is critical for ligand binding to integrins, additional interactions between residues in the integrin headpiece and those in the integrin recognition sequence of the ligand are required to secure the ligand binding to integrins. The crystal structure of integrin ␣V␤3 complexed with an RGD-containing ligand revealed that the side chain of the Arg residue in the RGD motif formed salt bridges to two Asp residues in the ␤-propeller domain of the integrin ␣V subunit (49). Interestingly, the two Asp residues in integrin ␣V are replaced by Lys or Arg residues in integrins ␣4 and ␣9, and these basic residues are well conserved within vertebrates (supplemental Fig. S5). By analogy with the RGD recognition by integrin ␣v␤3, it is tempting to speculate that these basic residues in integrins ␣4 and ␣9 may form salt bridges with the most N-terminal acidic residue in the (E/D)X⌽(E/D) motif, whereas the critical acidic residue coordinates the divalent metal ion in the MIDAS. Thus, the N-terminal three acidic residues of the EDDMMEVPY motif in polydom may well be engaged in forming salt bridges with the basic residues in the ␤-propeller domain of integrin ␣9.
Polydom is a large modular protein containing a von Willebrand factor A domain, a pentraxin domain, EGF-like domains and an array of CCP domains. Genes encoding its homologues with similar domain compositions and alignments have been found in both vertebrates and invertebrates, including honey bees, mosquitoes, sea urchins, fish, mice, rats, dogs, and monkeys (50). Thus, polydom is an evolutionarily conserved protein with function(s) shared among metazoans. It should be noted, however, that long arrays of CCP domains, including the one containing the DX⌽E motif, are absent from invertebrates. Furthermore, the EDDMMEVPY sequence in the extra region of CCP21 is only conserved in mammals (supplemental Fig. S6). It seems likely that the ability to interact with integrin ␣9␤1 was acquired in a later stage of evolution and was not the prototypical function of polydom. Consistent with this possibility, integrin ␣9 homologues are found only in vertebrates (51). Given that the N-terminal half comprising a von Willebrand factor A domain, a pentraxin domain, EGF-like domains, and a stretch of a few CCP domains is conserved in both vertebrates and invertebrates, this region should have another function, other than binding to integrin ␣9␤1, that remains to be explored to understand the physiological roles of polydom in metazoans.
Integrins transduce signals into cells upon binding with their ligands, thereby regulating cell proliferation, morphology, and migration (52,53). Despite a plethora of information on the signaling events triggered by RGD-binding or laminin-binding integrins (54,55), there is only limited information regarding those triggered by integrin ␣9␤1, partly because of the limited availability of cells expressing high levels of the integrin and also because of the difficulties in analyzing integrin-mediated signals on low affinity ligands. Gupta and Vlahakis (34) reported that integrin ␣9␤1-dependent cell adhesion to TNfn3-RAA not only activates c-Src kinase with concomitant tyrosine phosphorylation of p130 Cas and activation of Rac-1, both of which are well known signaling events triggered by a variety of integrins, including integrins ␣3␤1 and ␣5␤1 (55, 56), but also enhances nitric oxide production through activation of inducible nitricoxide synthase, which is so far unique to the signaling events triggered by integrin ␣9␤1. Furthermore, Sheppard et al. (57) demonstrated that integrin ␣9␤1 binds to spermidine/spermine-N 1 -acetyltransferase via the cytoplasmic domain of integrin ␣9 and enhances cell migration by modulating the local concentration of spermidine/spermine, which in turn regulates the K ϩ ion efflux by blocking inward rectifier K ϩ channels. Thus, integrin ␣9␤1 may trigger both signaling pathways, one common to other integrins and the other unique to integrin ␣9␤1. In this respect, cells cultured in the presence of an EMILIN-1 fragment exhibited reduced cell proliferation in an integrin ␣9␤1-dependent manner (40), raising the possibility that, unlike other integrins, integrin ␣9␤1 elicits signals that suppress, but do not accelerate, cell cycle progression. Currently, the mechanism by which integrin ␣9␤1 negatively regulates cell proliferation remains only poorly understood. Given its high binding affinity toward integrin ␣9␤1, polydom represents an excellent adhesive ligand for analyzing the signaling pathways activated by ligation of integrin ␣9␤1 and should contribute to uncovering the hitherto unknown physiological functions of integrin ␣9␤1 in diverse biological processes, including cell migration, cell proliferation, wound healing, lymphangiogenesis, and hematopoietic homeostasis, in which integrin ␣9␤1 has been implicated.