Extracellular Membrane-proximal Domain of HAb18G/CD147 Binds to Metal Ion-dependent Adhesion Site (MIDAS) Motif of Integrin β1 to Modulate Malignant Properties of Hepatoma Cells*

Background: HAb18G/CD147 interacts with integrin β1 subunit. Results: Extracellular membrane-proximal domain of HAb18G/CD147 (I-type domain) binds at the metal ion-dependent adhesion site (MIDAS) in the βA domain of the integrin β1 subunit. Conclusion: Interaction of HAb18G/CD147 with integrin β1 activates the downstream FAK signaling pathway, enhancing the malignant properties of hepatocellular carcinoma cells. Significance: This is first time binding sites of CD147 and integrin β1 are revealed. Several lines of evidence suggest that HAb18G/CD147 interacts with the integrin variants α3β1 and α6β1. However, the mechanism of the interaction remains largely unknown. In this study, mammalian protein-protein interaction trap (MAPPIT), a mammalian two-hybrid method, was used to study the CD147-integrin β1 subunit interaction. CD147 in human hepatocellular carcinoma (HCC) cells was interfered with by small hairpin RNA. Nude mouse xenograft model and metastatic model of HCC were used to detect the role of CD147 in carcinogenesis and metastasis. We found that the extracellular membrane-proximal domain of HAb18G/CD147 (I-type domain) binds at the metal ion-dependent adhesion site in the βA domain of the integrin β1 subunit, and Asp179 in the I-type domain of HAb18G/CD147 plays an important role in the interaction. The levels of the proteins that act downstream of integrin, including focal adhesion kinase (FAK) and phospho-FAK, were decreased, and the cytoskeletal structures of HCC cells were rearranged bearing the HAb18G/CD147 deletion. Simultaneously, the migration and invasion capacities, secretion of matrix metalloproteinases, colony formation rate in vitro, and tumor growth and metastatic potential in vivo were decreased. These results indicate that the interaction of HAb18G/CD147 extracellular I-type domain with the integrin β1 metal ion-dependent adhesion site motif activates the downstream FAK signaling pathway, subsequently enhancing the malignant properties of HCC cells.

Several lines of evidence suggest that HAb18G/CD147 interacts with the integrin variants ␣3␤1 and ␣6␤1. However, the mechanism of the interaction remains largely unknown. In this study, mammalian protein-protein interaction trap (MAPPIT), a mammalian two-hybrid method, was used to study the CD147integrin ␤1 subunit interaction. CD147 in human hepatocellular carcinoma (HCC) cells was interfered with by small hairpin RNA. Nude mouse xenograft model and metastatic model of HCC were used to detect the role of CD147 in carcinogenesis and metastasis. We found that the extracellular membraneproximal domain of HAb18G/CD147 (I-type domain) binds at the metal ion-dependent adhesion site in the ␤A domain of the integrin ␤1 subunit, and Asp 179 in the I-type domain of HAb18G/CD147 plays an important role in the interaction. The levels of the proteins that act downstream of integrin, including focal adhesion kinase (FAK) and phospho-FAK, were decreased, and the cytoskeletal structures of HCC cells were rearranged bearing the HAb18G/CD147 deletion. Simultaneously, the migration and invasion capacities, secretion of matrix metalloproteinases, colony formation rate in vitro, and tumor growth and metastatic potential in vivo were decreased. These results indicate that the interaction of HAb18G/CD147 extracellular I-type domain with the integrin ␤1 metal ion-dependent adhesion site motif activates the downstream FAK signaling pathway, subsequently enhancing the malignant properties of HCC cells.
CD147 is a transmembrane glycoprotein that is categorized as a member of the immunoglobulin superfamily and is broadly expressed on the surfaces of many kinds of tumor cells. Previous studies in our laboratory have demonstrated that HAb18G/ CD147, a member of the CD147 family that is highly expressed in HCC 4 cells, can promote the invasive and metastatic potentials of HCC cells by stimulating fibroblasts and tumor cells to produce matrix metalloproteinases (MMPs) (1). Specifically, the promotion of invasion and metastasis is mediated by the integrin-mediated FAK-paxillin and FAK-PI3K-Ca 2ϩ signaling pathways (2).
Integrins are composed of two type I transmembrane subunits, ␣ and ␤. Because there are integrin variants that can bind a variety of extracellular matrix molecules and cell-surface receptors (3), integrins can serve as bidirectional transducers of extracellular and intracellular signals in the processes of cell adhesion, cell-cell interactions, proliferation, differentiation, apoptosis, and tumor progression.
CD147 both oligomerizes with itself (4) and interacts with the ␣3␤1 and ␣6␤1 integrin variants at points of cell-cell contact (2,5,6). Published reports support a crucial role for CD147 in the migration and metastatic potential of tumor progression through its interactions with integrin (5,7,8). However, the fundamental mechanism of the interaction between HAb18G/ CD147 and integrin is not well understood; specifically, the binding domains of HAb18G/CD147 and integrin during interaction have not been thoroughly characterized.
Previous results suggest that the RGD motif, which binds to integrin through the metal ion-dependent adhesion site (MIDAS) of the integrin ␤1 subunit, is found in many extracellular matrix proteins, including vitronectin, fibronectin, fibrinogen, laminin, collagen, Von Willebrand's factor, osteoponin, and adenovirus particles (9). Because the RGD motif can inhibit the interaction between HAb18G/CD147 and integrin (8), the question arises of whether MIDAS is also the binding site of HAb18G/CD147 on integrin.
In the present study, we demonstrate that the I-type domain of HAb18G/CD147 can bind at the MIDAS pocket of the ␤A domain of the integrin ␤1 subunit and that Asp 179 in the I-type domain may play an important role in this binding interaction. The FAK signaling pathway and its downstream signals are activated by the interaction of HAb18G/CD147 with the integrin ␤1 subunit. Subsequently, tumor growth potential and the invasive and metastatic potentials of HCC cells are enhanced.

EXPERIMENTAL PROCEDURES
Cell Culture-HCC SMMC-7721 cells and human embryonic kidney (HEK) 293T cells were obtained from the Institute of Cell Biology, Academic Sinica, Shanghai, China. SMMC-7721 cells were cultured in RPMI 1640 medium (Invitrogen), whereas HEK 293T cells were cultured in DMEM (Invitrogen). Both media were supplemented with 10% FBS and 1% penicillin/streptomycin at 37°C in 5% CO 2 .
Construction of Plasmids for Chimeric Proteins in MAPPIT System-MAPPIT, a mammalian two-hybrid method (10,11), was used to study the CD147-integrin ␤1 subunit interaction (Fig. 1). To detect the interaction between the integrin ␤1 subunit and HAb18G/CD147, human extracellular residues of the integrin ␤1 subunit were fused to the bait constructs, and a human extracellular fragment of HAb18G/CD147 was fused to the prey constructs.
All expression plasmid vectors, including LR, pCLL, pMG1, and the pXP2d2-rPAPI-luciferase reporter, were kindly provided by Professor Jan Tavernier, Department of Medical Protein Research, Ghent University, Belgium. Sequences encoding the human extracellular domain residues of the integrin ␤1 subunit (residues 1-441) were amplified from SMMC-7721 cDNA using primer 1 in Table 1. The forward primer contains a SacI site, and the reverse primer contains a NotI site and stop codon. After SacI-NotI digestion, the fragment was cloned into the pCLL1 vector, generating the bait construct pCLL1-ITGB1. The mutants of pCLL-ITGB1 (D150A/ S154A/D279A and D150A/E249K/D279A) were generated step by step by site-directed mutagenesis (Invitrogen) using primers 2-5 in Table 1. The first mutant of pCLL-ITGB1 was named pCLL-ITGB1M 135 , and the second mutant of pCLL-ITGB1 was named pCLL-ITGB1M 145 .
Sequences encoding the human extracellular fragment of HAb18G/CD147 (residues 22-206), which includes the C2and I domains, was amplified with primer 6 listed in Table 1. The forward primer contains an EcoRI site, and the reverse primer contains an XbaI site and stop codon, from the pcDNA3.1-HAb18G/CD147 plasmid (12). The product was digested with EcoRI-XbaI and ligated into the EcoRI-XbaI-digested pMG1 vector, generating the prey construct pMG1-CD147. The pMG1-9 F3-10 F3 construct contains sequences encoding the 9 F3-10 F3 domain of fibronectin, including the RGD motif, which was amplified from SMMC-7721 cDNA using the same strategy by primer 7 listed in Table 1. The same strategy was used for the C2 domain (residues 22-107) and I-type domain (residues 101-206) of HAb18G/CD147 using primers 8 and 9 in Table 1.
Sequences encoding FLAG-CD147 and FLAG-9 F3-10 F3 were created using primers 15 and 16 to amplify the extracellular fragment of HAb18G/CD147 and the 9 F3-10 F3 domain of  10. A, MAPPIT employed the leptin receptor, which signals through the JAK-STAT pathway. B, bait constructs (cloned into pCLL vector) were designed as chimeric receptors consisting of the transmembrane and intracellular regions of a STAT3 recruitmentdeficient leptin receptor and the extracellular portion of the normal leptin receptor with a bait attached to the C terminus. Prey constructs (cloned into pMG1 vector) were composed of a prey polypeptide with a section of the gp130 chain carrying four STAT3 recruitment sites. The ITGB1bait represents the bait construct containing the integrin ␤A propeller domain gene, and the CD147/FNprey represents the prey construct containing the CD147 extracellular domain gene or the fibronectin 9 F3-10 F3 domain gene. If the bait and prey interacted, co-expression would lead to functional complementation of STAT3 activity, which could be measured by luciferase expression driven by the STAT3-responsive rat pancreatitis-associated protein I (rPAPI) promoter.
fibronectin, which contain the N-terminal FLAG tag from the pcDNA3.1-HAb18G/CD147 plasmid and SMMC-7721 cDNA, respectively. We cloned both PCR products into the pMET7 vector using ApaI-XbaI, and the final constructs were named pMET7-FLAG-CD147 and pMET7-FLAG-9 F3-10 F3. All constructs were verified by DNA sequence analysis.
Luciferase Reporter Assays for MAPPIT-Transfection procedures and luciferase assays were performed as described previously (10). HEK 293T cells were seeded in 6-well plates overnight and transfected with the desired constructs together with the luciferase reporter gene. Forty eight hours after transfection, cells were left untreated or stimulated overnight with 100 ng/ml leptin. The luciferase activity of the transfected cells was measured by chemiluminescence.
Western Blot Analysis and Co-immunoprecipitation-The peptides GRGDS, containing the RGD motif, and GRGES, containing the RGE motif, were synthesized by CL Xi'an Bio-Scientific Co., Ltd. Cells with different treatments were lysed with RIPA cell lysis buffer (Beyotime Institute of Biotechnology, China), and the resulting lysates were cleared by centrifugation. Proteins were resolved by 10% SDS-PAGE and transferred to a polyvinylidene fluoride (PVDF) microporous membrane (Millipore). After being blocked with 5% fat-free dry milk, the membrane was probed with primary antibodies, including HAb18, against HAb18G/CD147 (prepared by our laboratory), anti-FAK (BD Biosciences), anti-p-FAK (BD Biosciences), anti-integrin ␤1 (Santa Cruz Biotechnology, Santa Cruz, CA), anti-p-Akt (Cell Signaling Technology), and anti-␣-tubulin antibodies (Santa Cruz Biotechnology). Following incubation with horseradish peroxidase-conjugated goat anti-mouse or rabbit antigoat IgG (Pierce), protein bands on the membrane were visualized using a chemiluminescence kit (Beyotime, Shanghai, China) according to the manufacturer's instructions. Co-immunoprecipitation was then carried out to detect the interaction of HAb18G/CD147 with the integrin ␤1 subunit. 7721 cells were seeded into culture dishes and incubated with the peptides HAb18G/CD147 I-type domain (residues 101-206) Forward GRGDS (100 g/ml) and GRGES (100 g/ml), respectively. Cells were lysed with the M-PER Reagent from the mammalian co-immunoprecipitation kit (Pierce) after adding or omitting the designated peptides (100 g/ml). The lysates were then collected onto a coupling gel that was pre-bound with HAb18 overnight at 4°C. To detect interaction of the integrin ␤1 subunit with Asp-mutated CD147 prey, lysates of HEK 293T cells co-transfected with the plasmids encoding ITGB1 and Aspmutated CD147 prey were immunoprecipitated with anti-FLAG antibody. Immunocomplexes were washed four times with the co-immunoprecipitation buffer, analyzed by immunoblotting with the indicated antibodies. Enzyme Linked Immunosorbent Assay (ELISA)-The extracellular portion of HAb18G/CD147, produced and purified as described previously (13), was added to the wells of a 96-well ELISA plate at a concentration of 50 g/100 l. Plates were incubated overnight at 4°C, and wells were blocked for 1 h with 200 l of 5% fat-free milk. The pellet of 7721-shCD147 cells (1 ϫ 10 9 ) was extracted at 4°C with 10 ml of 150 mM NaCl, 25 mM Tris-HCl, pH 7.4, 2% (w/v) Triton X-100, 1 mM phenylmethylsulfonyl fluoride, 10 mg/ml leupeptin, 2 mg/ml bovine serum albumin and protease inhibitor mixture, EDTA-free tablet (Roche Applied Science). The extract was centrifuged at 12,000 ϫ g for 10 min. and the supernatant was concentrated by passing it through an Amicon ultracentrifugal filter (50 kDa, Millipore). The concentrated extract was diluted as indicated and added to wells (100 l/well). After incubation overnight at 4°C, wells were washed with 200 l of PBST three times and aliquots (100 l each) of integrin ␤1 pAb (Santa Cruz Biotechnology, 1:50) with different concentrations of cations or EDTA. To analyze the effect of a range of Mn 2ϩ , Mg 2ϩ , and Ca 2ϩ concentrations on the binding of CD147 to integrin ␤1, Mn 2ϩ , Mg 2ϩ , or Ca 2ϩ was added to a final concentration of 0.25-4 mM. The plate was then incubated at 37°C for 2 h, and the wells were washed three times with PBST. After incubation with horseradish peroxidase-conjugated goat anti-mouse IgG for 20 min at room temperature, wells were then washed, and color was developed using 3,3Ј,5,5Ј-tetramethylbenzidine solution. After incubation for 10 min, an equal volume of stopping solution (2 M H 2 SO 4 ) was added, and the absorbance was read at 450 nm.
Immunofluorescence-7721-shCD147 cells and 7721-snc cells were grown on coverslips and incubated with GRGDS or GRGES peptides (100 g/ml) for 24 h and fixed with 4% paraformaldehyde. The human HCC tissues for frozen sections were supplied by the General Hospital of the People's Liberation Army. The coverslips or frozen sections were first blocked with 10% bovine serum albumin in phosphate-buffered saline (PBS), pH 7.0, and then incubated with primary antibodies, including HAb18 against HAb18G/CD147 (prepared by our laboratory), anti-integrin ␤1 (Santa Cruz Biotechnology), and anti-␣-tubulin antibodies (Santa Cruz Biotechnology). Alexa Fluor 594 goat anti-mouse IgG (HϩL) (Molecular Probes, Eugene, OR), fluorescein isothiocyanate (FITC)-conjugated goat anti-mouse IgG (Pierce), and FITC-conjugated donkey anti-goat IgG (Santa Cruz Biotechnology) were used as secondary antibodies. To detect F-actin, the cells were probed with Alexa Fluor 488-phalloidin (Molecular Probes) at 1:40 for 20 min. Cell nuclei were stained with DAPI. Images were obtained with an FV1000 laser scanning confocal microscope (Olympus, Japan).
Gelatin Zymography-7721-shCD147 cells and 7721-snc cells were incubated with serum-free medium in the presence of the GRGDS or GRGES peptides (100 g/ml). The conditioned medium was collected, and the volume from each cell line was adjusted to provide a predetermined quantity of protein. Media were separated by 10% acrylamide gels containing 0.1% gelatin. The gels were incubated overnight in a reaction buffer (0.05 mol/liter Tris-HCl, pH 7.5, 0.2 mol/liter NaCl, and 0.01 mol/liter CaCl 2 ) at 37°C after the SDS was removed by washing with buffer containing 2.5% Triton X-100 with gentle agitation. After the reaction, the gels were stained with Coomassie Brilliant Blue R-250 for 6 h and destained for ϳ0.5 h. The zones of gelatinolytic activity were indicated by negative staining.
In Vitro Wound Healing Assay-7721-shCD147 cells and 7721-snc cells were seeded in 6-well plates and cultured to 100% confluence. After a cell-free area was created by scraping the monolayer with a pipette tip, the cells were incubated with medium supplemented with 1% FBS or medium supplemented with 1% FBS and 100 g/ml GRGDS or GRGES peptides. Digital images were taken with an inverted phase-contrast microscope (BX60, Olympus) at time points up to 20 h.
Invasion Assay-An invasion assay was performed in 24-well Transwell units with an 8-mm pore size polycarbonate (Millipore), according to the manufacturer's instructions. Briefly, filters were coated with Matrigel to form a continuous thin layer. The cells were then seeded in 0.5% fetal calf serum in RPMI 1640 medium in the upper chamber, and the lower chamber was filled with 10% fetal calf serum in RPMI 1640 medium. GRGDS or GRGES (100 g/ml) peptides were added to the upper chambers at the same time. Following 24 h of incubation at 37°C, cells remaining in the upper compartment were completely removed using cotton swabs. The cells that invaded through the filter into the lower compartment were fixed with 4% paraformaldehyde/HBS-Ca 2ϩ , stained with crystal violet (0.5% in 20% methanol), and counted.
Colony Formation Assay-7721-shCD147 cells and 7721-snc cells were plated on a 0.6% agarose base in a 24-well plate (1.0 ϫ 10 3 cells per well) in 1 ml of DMEM containing 10% FBS, 0.3% agarose, GRGDS, or GRGES peptides (100 g/ml). At day 15 after plating, colonies of Ͼ50 cells were counted. The colony formation rate was calculated using the following formula: colony formation rate ϭ number of colonies/10 3 cells.
Nude Mouse Xenograft Model of HCC-Four-week-old nude mice were divided randomly into three groups (six mice per group). An identical number of 7721-shCD147 cells and 7721snc cells were subcutaneously injected into the right and left flanks of every nude mouse. From the 6th day after inoculation, GRGDS (100 g), GRGES (100 g) or 1ϫ PBS were injected into the growing tumors of groups 1-3 mice, respectively. The tumors were monitored with a caliper every 2 days. The tumor volume was determined for each mouse (in cubic millimeters) by measuring the tumor in two directions and was calculated as tumor volume ϭ length ϫ (width) 2 /2. All animal procedures were performed in accordance with Laboratory Animal Ethics Committee of Fourth Military Medical University.
Immunohistochemistry-After the tumor model experiment, half of the human tumor xenografts in the nude mice were fixed with 10% formalin and embedded in paraffin. Sections were deparaffinized and incubated with primary antibodies, including HAb18 against HAb18G/CD147, anti-MMP-2 (Santa Cruz Biotechnology), and anti-integrin ␤1 (Santa Cruz Biotechnology), followed by visualization with the Histostain-Plus kit (Invitrogen).
Real Time PCR Amplification-Total RNA was isolated from the tumor tissues of the nude mouse xenograft model of HCC using TRIzol reagents (Invitrogen), according to the manufacturer's instructions. After RNA isolation, total RNA was reverse-transcribed into cDNA with the ReverTra Ace-a kit (Toyobo, Japan). SYBR Green real time RT-PCR was performed as described in the product instructions using SYBR Premix EX TaqII (2ϫ) (Takara, Japan) with the sequence detection system Stratagene Mx3005P (Agilent Technologies, Germany). In all PCRs, a negative control corresponding to an RT reaction without the reverse transcriptase enzyme and a blank sample were carried out; they exhibited no PCR product amplification. Amplification of GAPDH cDNA was used as an internal control to quantify the expression of a given gene. For screening expression and quantification studies, PCR analysis was performed using specific primers made by Shanghai Sangon Biological Engineering Technology & Services Co., Ltd., listed in Table 2.

Metastatic Model of Human Hepatocellular Carcinoma in Nude Mice via Orthotopic Implantation of HCC Cells-Four-
week-old nude mice were divided randomly into four groups (7721-snc group, 7721-shCD147 group, 7721-snc-RGE group and 7721-snc-RGD group). Viable HCC cells (2.5 ϫ 10 6 ) mixed with Matrigel (BD Biosciences) were injected into left lobe of liver in nude mice. From the 7th day after the injection, RGE peptide (GRGES) and RGD peptide (GRGDS) (1 mg/kg body weight) were injected into 7721-snc-RGE group or 7721-snc-RGD group intraperitoneally every 2 days for 2 weeks. At 3 weeks after the injection, the mice were sacrificed under anesthesia. Tumor formation in the liver, lung, and spleen was macroscopically counted.
Statistical Analysis-Student's t test was used to compare the colony formation rates in vitro. Fisher's exact test was used to evaluate the effect of HAb18G/CD147 and GRGDS on the formation of metastases. For other experiments, a one-way analysis of variance t test was performed for the comparison of mean values that represent data obtained from three independent experiments (p Ͻ 0.05 was considered statistically significant).

RESULTS
Extracellular Portion of HAb18G/CD147 Interacts with the ␤A Domain of Integrin ␤1 Subunit-We fused a leptin receptor with a bait protein, the fragment of the integrin ␤1 subunit's ovoid head containing residues 1-387 (wild type or two different mutants of MIDAS). To determine the interaction, the integrin ␤1 fusion protein (pCLL-ITGB1) was transiently coexpressed with the extracellular fragment of the HAb18G/ CD147 prey. As shown in Fig. 2A, the two bait protein mutants of MIDAS (pCLL-ITGB1M 135 and pCLL-ITGB1M 145 ) of the ␤1 subunit were sufficient to abrogate MAPPIT signals. GRGDS also displayed a strong interaction with the integrin ␤1 subunit, although the interactions of mutations of the ␤1 subunit with GRGDS were significantly attenuated. The expression of the FLAG-tagged HAb18G/CD147 extracellular fragment prey was revealed by immunoblotting using an anti-FLAG antibody (Fig. 2B). Identical expression levels of various preys confirmed that the elevated MAPPIT signal was not due to unequal prey levels.
Fibronectin 9 F3-10 F3 Domain Inhibits the Interaction of HAb18G/CD147 and Integrin ␤ 1 Subunit-The Arg-Gly-Asp (RGD) motif in the central cell-binding domain of many ligands was an important site of cell recognition for integrins. The fibronectin (FN) central cell-binding 9 F3-10 F3 domain also contained the RGD sequence. The ␣5␤1 integrin variant binds to the RGD motif of FN at the interface between the ␣ and ␤ subunits of integrin, and the aspartic acid residues of the RGD motif contacts MIDAS at the ␤A directly, with the mediation of a divalent cation, forming the Mn 2ϩ -bound integrin-RGD complex (14,15). In this study, the sequences encoding 9 F3-10 F3 domain of FN was inserted into a prey construct. In HEK 293T cells, the co-expression of the FN prey and integrin bait elevated the MAPPIT signal, and integrin mutants abolished the MAPPIT signal ( Fig. 2A). To gain further insight into whether the MIDAS of the integrin ␤ 1 subunit is the CD147 integrin-binding site, the CD147 prey was co-expressed with the wild type fibronectin 9 F3-10 F3 domain. Fibronectin 9 F3-10 F3 domain expression markedly reduced the CD147 prey signal (Fig. 2C). Co-expression of the extracellular CD147 fragment also repressed the fibronectin 9 F3-10 F3 prey signal (Fig.  2C). The expression levels of the FLAG-tagged proteins were confirmed by immunoblotting using an anti-FLAG antibody (Fig. 2D).
HAb18G/CD147 I Domain, but Not the C2 Domain, Interacts with the Integrin ␤1 Subunit-We fused a leptin receptor with a bait protein fragment containing the C2 domain of CD147 (residues 22-107) and a fragment containing the I domain of

CD147 Binds to MIDAS Motif of Integrin ␤1
FEBRUARY 10, 2012 • VOLUME 287 • NUMBER 7 CD147 (residues 101-206). The prey proteins, including a fragment of the integrin ␤1 subunit's ovoid head that contained residues 1-387, and two MIDAS-targeted mutants or the fibronectin 9 F3-10 F3 domain, which contains the RGD sequence, were linked to the gp130 chain. Preys containing the C2 domain of CD147 (residues 22-107) or the I domain of CD147 (residues 101-206) were co-transfected with the integrin ␤ 1 fusion protein. As shown in Fig. 2A, the fragment containing the I domain exhibited a novel interaction with the integrin ␤1 subunit, unlike the fragment containing the C2 domain of CD147. Mutants of the ␤1 subunit showed a weak interaction with the fragment containing the I domain. These results indicated that the I domain of CD147, rather than the C2 domain, might play a central role in the interaction with the integrin ␤1 subunit. Similar prey expression levels confirmed that the reduced signal was not caused by variations in prey levels (Fig. 2B).
Asp 179 -targeted Mutation in HAb18G/CD147 I Domain Inhibits the CD147-Integrin Interaction-To further indicate whether aspartic acid-based sequences such as RGD, LDV, KGD, RTD, and KQAGD containing Asp (16) in HAb18G/CD147 might play an important role in the interaction between integrin and CD147, each Asp of the I domain was mutated to Ala by sitedirected mutagenesis, represented as pMG1-CD147-A 136 , pMG1-CD147-A 144 , pMG1-CD147-A 147 , pMG1-CD147-A 179 , and pMG1-CD147-A 194 , and each construct was co-transfected with the integrin ␤ 1 fusion protein. As shown in Fig. 2E, MAPPIT signals were reduced markedly in an experiment using pMG1-CD147-A 179 as the prey, compared with the signals obtained using a wild type HAb18G/CD147 extracellular fragment or other mutants of HAb18G/CD147 extracellular fragment. Equal prey expression levels confirmed that the reduced signal was not caused by differences in prey levels (Fig. 2F). As shown in Fig. 2G, Asp 179 mutation inhibited binding of CD147 to integrin ␤1, although other mutations did not inhibit the interaction significantly.
Co-immunoprecipitation between HAb18G/CD147 and the integrin ␤1 subunit is decreased by GRGDS Treatment-A previous study indicated that HAb18G/CD147 co-immunoprecipitates with integrin ␤1 in HCC cells (2). The decreased coimmunoprecipitation (by 71.1%) in GRGDS-treated 7721 cells compared with control 7721 cells demonstrates that incubation with the GRGDS peptide could inhibit the interaction of HAb18G/CD147 and the integrin ␤1 subunit (Fig. 3A).
Bivalent Cations Modulate the Apparent Affinity of Binding of HAb18G/CD147 to Integrin ␤1-Although it is clear that bivalent-cation occupancy and integrin-ligand binding are intimately linked, the precise role of bivalent cations in CD147integrin interaction remains uncertain. We tested the effect of a range of cation concentrations on the binding of CD147 to integrin ␤1. As shown in Fig. 3B, CD147 binding was increased by Mn 2ϩ in a dose-dependent manner and to a less extent by Mg 2ϩ and Ca 2ϩ . Different concentrations of cell extract influenced the CD147-binding activity of integrin ␤1 in the presence of Mn 2ϩ . EDTA inhibited CD147 binding to integrin ␤1 in a dose-dependent manner (Fig. 3C). Replacing Mn 2ϩ with EDTA led to reduced binding of CD147 to integrin ␤1 (Fig. 3D).
HAb18G/CD147 Regulates Expression and Phosphorylation of FAK and Mediates Cytoskeletal Rearrangements-FAK is a cytoplasmic tyrosine kinase that plays a major role in integrin signaling. To identify whether the FAK pathway is involved in the HAb18G/CD147-mediated integrin signaling pathway, we tested the levels of HAb18G/CD147, integrin, FAK, and p-FAK in 7721-shCD147 cells or snc-transfected 7721 cells, with or without GRGDS or GRGES treatment. The expression levels of both FAK and p-FAK were significantly diminished, by 65.5 and 71.4%, respectively, in 7721-shCD147 cells as compared with that in 7721-snc cells (Fig. 4B). However, in 7721-shCD147 cells, the levels of FAK and p-FAK were not further diminished after treatment with the GRGDS peptide (Fig. 4B). As shown in Fig. 4C, the expression level of phosphorylation of Akt (p-Akt) was decreased in 7721-shCD147 cells and GRGDS-treated 7721-snc cells, indicating the diminished activation of PI3K/ Akt signaling. The level of p-Akt was not decreased further in 7721-shCD147 cells after treatment with the GRGDS peptide. To explore the mechanism responsible for FAK pathway activation, 7721 cells were transfected with si-RNAs and collected at the indicated times. mRNA levels of CD147 and FAK and protein levels of CD147, FAK, p-FAK, and p-Akt decreased in a time-dependent manner (Fig. 4, D and E). As shown in Fig. 4F, CD147 extracellular domain increased protein levels of p-Akt  ␤1 subunit, analyzed by co-immunoprecipitation and ELISA. A, CD147 immunoprecipitation with the integrin ␤1 subunit. Top, lysates of 7721 cells incubated with the GRGDS or GRGES peptides (100 g/ml) were subjected to immunoprecipitation, with the GRGDS or GRGES peptides (100 g/ml) present in the co-immunoprecipitation process. The integrin ␤1 subunit was detected by Western blot analysis with anti-HAb18G/CD147 antibody-prebound coupling gel. Mouse IgG was used as a negative control. Bottom, CD147 expression was examined by Western blot analysis. N indicates negative control group without treatment of peptides. B, effects of bivalent cations Mn 2ϩ , Mg 2ϩ , and Ca 2ϩ on the binding of HAb18G/ CD147 to integrin ␤1. The concentrated extract of cells was diluted to 30% before the incubation. C, binding of HAb18G/CD147 to integrin ␤1 was measured for varying concentrations of EDTA. The concentrated extract of cells was diluted to 30% before the incubation. D, binding of HAb18G/CD147 to integrin ␤1, measured in the presence of 1 mM Mn 2ϩ or in the presence of 1 mM EDTA. The concentrated extract of cells was diluted as labeled on the x axis. B, C, and D, data are representative of the absorbance read at 450 nm in three duplicated wells and are expressed as the mean Ϯ S.D. and p-FAK in 7721 cells in a dose-dependent manner. However, FAK expression levels were not affected. Both RGD peptides and integrin ␤1 attenuated activation of Akt and FAK pathways, although expression of FAK was not significantly changed (Fig. 4G).
There are less co-localizations of HAb18G/CD147 with integrin ␤1 on the cell membrane of 7721-shCD147 cells compared with 7721-snc cells (Fig. 5A). Compared with 7721-snc cells, cells treated with GRGDS acquired a round-shaped morphology and had decreased co-localizations of HAb18G/ CD147 with integrin ␤1 at cell-cell junctions. As shown in Fig.  5B, HAb18G/CD147 co-localized with integrin ␤1 on HCC cells but not on mesenchymal cells (white arrows) of human HCC tissues.
The integrin downstream signaling pathways involve many cytoskeletal proteins and enzymes. Activation of integrin signaling pathways ordinarily results in cytoskeletal reorganization (17). We tried to determine whether GRGDS and HAb18G/CD147 deletion could cause a cytoskeletal rearrangement in HCC cells. To quantify the changes in cytoskeletal arrangements, we labeled F-actin stress fibers with rhodaminephalloidin, with or without GRGDS treatment, in 7721-snc cells or 7721-shCD147 cells. F-actin stress fibers were reorganized in 7721-shCD147 cells compared with those observed in 7721-snc cells and were reorganized in GRGDS-treated 7721-snc cells compared with those observed in GRGES-treated 7721-snc cells (Fig. 5C). F-actin stress fibers in 7721-shCD147 cells and GRGDS-treated 7721-snc cells indicated changes from polarization throughout the cytoplasm toward polarization under the cell membrane. Meanwhile, the treatment of 7721-shCD147 cells with GRGDS did not result in further changes in F-actin stress fiber polarization.
GRGDS Attenuates HAb18G/CD147-mediated Invasion, Migration, Growth, and Metastasis Potentials-To confirm the involvement of an HAb18G/CD147-integrin interaction in the invasion, migration, and metastasis potentials of HCC cells, 7721-shCD147 cells and 7721-snc cells were subjected to wound healing assays, invasion assays, and gelatin zymography, with or without treatment with GRGDS or GRGES peptides. As shown in Fig. 6, A-C, both sh-CD147 and GRGDS treatments inhibited cell migration, MMP secretion, and the invasion potentials of 7721 cells. The levels of pro-MMP-9, MMP-9, pro-MMP-2, and MMP-2 examined by gelatin zymography were decreased by 60.0, 53.5, 77.6, and 46.8%, respectively, in 7721-shCD147 cells, and they were diminished by 56.8, 59.4, 84.7, and 43.2%, respectively, in GRGDS-treated 7721-snc cells, compared in both cases with the levels observed in 7721-snc cells (p Ͻ 0.01, Fig. 6B). As shown in Fig. 6D, the colony formation ability was significantly decreased in 7721-shCD147 cells and GRGDS-treated 7721-snc cells (p Ͻ 0.01). However, the co-treatment of sh-CD147 and GRGDS did not result in further inhibition of malignant properties compared with the single treatments (p Ͼ 0.05, Fig. 6, A-D).
Next, we inoculated the same number of 7721-shCD147 cells and 7721-snc cells subcutaneously into the flanks of nude mice to determine the role of HAb18G/CD147 in tumor formation in vivo. 7721-shCD147 cells showed decreased capacity of tumor growth compared with 7721snc cells, and GRGDS treatment to tumors generated by 7721-snc cells decelerated tumor growth compared with the effects of PBS treatment. However, the treatment of 7721-shCD147 cells with GRGDS did not further decelerate tumor growth (Fig. 7A). The mRNA levels of CD147, ITGB1, MMP-2, and MMP-9 and the protein expression levels of HAb18G/CD147, integrin ␤1, FAK, and p-FAK in tumor tissues examined using real time RT-PCR (Fig. 7B), Western blot analysis (Fig. 7C), and immunohistochemistry (Fig. 7D) were consistent with the results obtained by the Western blot analysis and the immunofluorescent double staining in vitro (Figs. 4B and 5, A and C). Tumors generated by 7721-shCD147 cells or treated with GRGDS showed decreased levels of MMP-2 in mesenchymal cells adjacent to tumor cells, compared with the level in tumors generated by 7721snc cells (Fig. 7D). We further tested the potential of HAb18G/CD147 and integrin ␤1 in tumor metastasis. The incidence of metastasis after intrahepatic injection of HCC cells into nude mice was shown in Table 3 and Fig. 7E. Eight of eight mice in the 7721-snc cell group and six of seven mice in the RGE-treated 7721-snc cell group developed visible intrahepatic metastasis, whereas only three of seven mice in the 7721-shCD147 cell group and one of seven mice in the RGD-treated 7721-snc cell group developed visible intrahepatic metastasis. These results showed that both shCD147 and RGD treatments significantly decreased incidences of experimental intrahepatic metastasis of 7721 cells.

DISCUSSION
As an HCC-associated antigen, HAb18G/CD147 has been shown to play crucial roles in the intercellular interactions involved in tumor metastasis (18). Previous studies have found that CD147 interacts with the integrins ␣3␤1 and ␣6␤1 in HCC cells and activates the downstream FAK-PI3K-Ca 2ϩ pathway, thus contributing to the enhanced invasion and metastatic potentials of HCC cells (2,6). Although the intracellular molecules interacting with HAb18G/CD147 and  their signal transduction pathways have been partially revealed, the fundamental mechanism of the interaction between HAb18G/CD147 and integrin was not well understood. This study attempted to accurately reveal the binding sites of HAb18G/CD147 and the integrin ␤1 subunit and the mechanism regulating the CD147-integrin ␤1 interaction and its downstream pathways.
Composed of ␣ and ␤ subunits, the integrin ␣V␤3 heterodimer has a jellyfish-like appearance, with a globular "head" and two "legs," with the headpiece containing the ligand-binding sites (19,20). In ␣A-lacking integrins, such as ␣3␤1, a predicted ␣A-like domain (␤A) from the ␤ subunit is essential for ligand binding. Many studies have suggested that ␤A might bind ligands through a putative MIDAS-like motif (14,21,22). In this study, we found that the extracellular head of the ␤1 subunit (the ␤A domain) interacts with the extracellular portion of HAb18G/CD147. The mutated MIDAS motif of the ␤A domain in the ␤1 subunit abrogates the interaction of CD147 and the integrin ␤1 subunit, indicating a crucial role for MIDAS in the CD147-integrin ␤1 interaction.
As an important cell recognition site with the integrin, the Arg-Gly-Asp (RGD) sequence exists in various ligands, including FN. The Asp of the RGD motif has previously been proven to bind to the MIDAS motif in the ␤3 subunit (14), and ␤1 also contains a MIDAS motif (21). In this study, FN interacted with the ␤1 subunit and competitively repressed the CD147-integrin ␤1 interaction. The results of the co-immunoprecipitation experiments showed that the CD147-integrin ␤1 interaction is mediated by the MIDAS motif. It has been reported that Mn 2ϩ is effective at inducing the conformational changes associated with integrin MIDAS activation (14). Analysis of the binding of CD147 to integrin ␤1 over a range of Mn 2ϩ , Mg 2ϩ , or Ca 2ϩ concentrations demonstrated the ability of cations (especially Mn 2ϩ ) to elicit conformational changes and to modulate the CD147-binding potential of integrin ␤1 MIDAS. These results imply two possibilities. First, the CD147-binding site on the integrin ␤1 subunit is a MIDAS motif. Second, occupation of the MIDAS site might help to expose/reorient the synergy site to create an optimal and stable complementary interface between the integrin and CD147.
Our laboratory has found that the crystal structure of the extracellular portion of HAb18G/CD147 comprises a membrane-distal N-terminal Ig C2 domain and a membrane-proximal C-terminal Ig I domain, which are connected by a fiveresidue flexible linker (13). Here, we showed that the I domain, but not the C2 domain, could bind the ␤A domain of integrin ␤1. Because the interactions of the MIDAS motif with the ␤A domain and various ligands are mediated by aspartic acid-based sequences such as RGD, LDV, KGD, RTD, and KQAGD in the ligands, we wondered whether aspartic acid-based sequences in HAb18G/CD147 also play an important role in the interaction of the ␤A domain and CD147. Asps in the I domain of CD147 were mutated by site-directed mutagenesis, and of five targeted Asp mutations, only the fourth Asp (Asp 179 ) mutation partially abolished the interaction with the ␤A domain compared with the other mutations.
The MIDAS motif in the ␤A domain of the integrin ␤1 subunit interacts with the I domain of CD147; moreover, the exposed Asp 179 residue in CD147 is critical for recognition by the integrin ␤1 subunit. A structural model of the integrin ␣3␤1  in a complex with CD147 is shown in Fig. 8. The ovoid head of the integrin ␣v␤3 containing the ␤A domain cannot approach a position adjacent to the cell membrane in the activated state, similar to the condition of ␤1 (15). Because Asp 179 of the I domain is adjacent to the cell membrane, we presume that the interaction of CD147 and the integrin ␤1 subunit does not occur in the same HCC cell, but instead it occurs between neighboring HCC cells (Fig. 8). The resulting interactions between HCC cells also provide a possible explanation for enhanced cell junction and adhesion capacity.
The integrin signaling pathways consist of many cytoskeletal proteins and enzymes, including FAK, paxillin, and PI3K. We previously discovered that the activated FAK-paxillin and FAK-PI3K-Ca 2ϩ signaling pathways, which enhance the metastatic potentials of HCC cells, closely correlate with HAb18G/CD147 expression (2,6). The interaction of CD147 with the integrin ␤1 subunit could be competitively blocked by both the GRGDS peptide and the transfection of a CD147-specific shRNA (sh-CD147). Because of an attenuated interaction, the downstream FAK pathway failed to be activated, which resulted in a rearranged actin cytoskeleton. Simultaneously, HCC cells exhibited attenuated malignant properties, including MMP release, invasion, and migration potential and tumor formation capability. Co-treatment of sh-CD147 and GRGDS did not result in further inhibition to FAK pathway activation and malignant properties of the cells, indicating that HAb18G/CD147 interacts with integrin at the RGD-binding site (MIDAS pocket) to activate the FAK pathway.
A previous study in our laboratory has found down-regulation of FAK in 7721 cells transfected with si-RNA targeting CD147 mRNA (23). Our study showed that RNA interference (either by si-RNA or sh-RNA) resulted in decreased levels of both FAK and phosphorylated FAK. However, treatment with CD147 extracellular domain did not alter the FAK protein levels, and neither did treatments with GRGDS peptides or integrin ␤1 antibody. These results confirmed that attenuated interaction of HAb18G/CD147 with integrin decreased the phosphorylation level of FAK. Meanwhile, HAb18G/CD147 seemed to be capable of affecting the protein level of FAK by some other pathways. Besides CD147-integrin interaction, new mechanisms downstream of CD147 are being explored in our laboratory. As a transmembrane glycoprotein, CD147 localizes to both the cell membrane and the endomembrane system in HCC cells. A recent study in our laboratory has proved that CD147 localizing to endomembrane system inhibits the RhoA/ ROCK signaling pathway and amoeboid movement via depression of annexin II phosphorylation (24). Although a previous study claimed that the ERK pathway was involved in FAK regulation by CD147 (23), further investigation is needed to study the responsible mechanism.
These data demonstrate that the CD147-integrin interaction executes an accelerative function of CD147 in the metastatic process of HCC cells. When the interaction was disrupted, CD147 failed to ensure the transduction of outside-in signals across integrins and proper intracellular functions.
RGD peptides have been proven to competitively interact with integrin, thus blocking the interaction of ligands with integrin in various tumor types. Additional RGD peptides could induce apoptosis of tumor cells through the activation of procaspase 3 (25). The radiolabeled RGD peptides could be used as a potent tumor imaging reagent, and recombinant adenoviral vectors containing RGD peptides are very effective in targeted cancer therapies, including those of breast cancer, melanoma, osteosarcoma, glioma, ovarian cancer, and pancreatic carcinoma (26 -32).
It is significant to consider CD147 as a target for the development of diagnostic and treatment methods for cancers. Licartin (a 131 I-labeled mAb specific for HAb18G/CD147) was developed in our laboratory and has been used safely and effectively in treating hepatocellular carcinoma patients (33,34). Our results show that the RGD motif inhibits the CD147-integrin ␤1 interaction, blocking the outside-in signals of CD147 and attenuating the malignant properties of tumor cells that are induced by CD147. The regulatory binding site of the CD147integrin ␤1 interaction might be a novel potential target for tumor therapy.