Transduction of a Mesenchyme-specific Gene Periostin into 293T Cells Induces Cell Invasive Activity through Epithelial-Mesenchymal Transformation*

Tumor metastasis is a multistep pathological process involved in the final phase of tumor development. During this process, epithelium-derived tumor cells undergo fibroblast-like transformation, referred to as epithelial-mesenchymal transition (EMT), which contributes to aggressive behavior of tumors. We identify periostin, a mesenchyme-specific gene product, as a contributor to EMT and metastatic potential. Stable expression of a periostin transgene in tumorigenic but non-metastatic 293T cells caused cells to undergo fibroblast-like transformation accompanied by increased expressions of vimentin, epidermal growth factor receptor (EGFR), and matrix metalloproteinase-9. The cells expressing ectopic periostin increased cell migration, invasion, and adhesion by 2-9-fold. Invasive characteristics required signaling through integrin αvβ5 and EGFR. In addition, periostin-engineered 293T cells formed metastases in immunodeficient mice following either cardiac inoculation or injection into mammary fat pad. These data demonstrate an active role for periostin in EMT and metastasis that requires cross-talk between integrin and EGFR signaling pathways.

Progression of a solid tumor to an invasive tumor is a major prerequisite for metastasis and involves changes in both cell morphology and motility (1,2). Although genetic alterations in cancerous cells may vary in different types of metastatic cancers, they share the outcome that cancerous cells are disseminated to multiple distant organs (3,4). Typically, these aggressive cells detach from the site of origin, move across tissue boundaries to penetrate into lymphatic and blood vessels, and eventually extravasate from these vessels and colonize new sites. Phenotypic conversion of tumor cells from epithelial to mesenchymal phenotype, termed epithelial-mesenchymal transition (EMT), 2 is commonly associated with acquisition of metastatic potential (5)(6)(7). During this transition, tumor cells usually lose epithelial features, including apical-basal polarity, expression on cytokeratin filaments, and membrane-associated adherens or desmosomes junctions. Concurrently, they gain mesenchymal properties, including expression of vimentin filaments, spindle-like morphology, and increased motility allowing cells to invade new sites (8 -10).
The molecular basis underlying the EMT process involves multiple changes in expression, distribution, and/or function of proteins that include vimentin, integrins, matrix metalloproteinases (MMP), and cadherins (6,(11)(12)(13). Acquired expression of vimentin by carcinoma cells, which replaces the cytoskeleton network, often symbolizes mesenchyme-like cell transformation (6,11). Loss of E-cadherin or gain of N-cadherin on tumor cell surface is frequently observed in malignant carcinomas and also correlated with enhanced aggressiveness and dedifferentiation (14). Elevated MMP-9 expression and activity allow tumor cells to degrade and penetrate extracellular matrix (15,16). Integrins, heterodimeric transmembrane receptor complexes, interact with specific regions of extracellular matrix protein and transmit "outside-in" signals to the cells, leading to an array of intracellular signaling events participating in cell proliferation, adhesion, and motility. For example, increased expression of integrin ␣ v ␤ 3 in solid tumor and melanomas was correlated with tumor malignancy, and its function was found to enhance tumor cell growth and invasion (17,18). Likewise, the increased activity of integrin ␣ v ␤ 5 enhanced motility of ovarian cancer cells (19,20).
Because integrins perform important functions in tumor progression, intensive research efforts have focused on the cross-talk between integrins and membrane tyrosine kinase receptors (21)(22)(23). Growth factors that activate receptor tyrosine kinases can alter integrin-mediated activities such as cell adhesion, spreading, and migration via alterations in integrin activation and localization. For instance, breast cancer cells expressing integrin ␣ v ␤ 5 depend on insulin-like growth factor stimulation for integrin ␣ v ␤ 5 -mediated cell migration (24). Conversely, signals from integrins regulate full activation of growth factor signaling. EGFR is activated and required to cooperate with integrin ␣ v ␤ 5 for vitronectin-induced carcinoma cell migration (25). The selective coordination of inputs from different growth factor receptors and integrins largely depends on the cell types that inherently express distinct membrane receptors.
Integrins also play important roles in the activation of endothelial cells as well as tumor cells (26,27). Engagement of acti-vated vascular endothelial growth factor (VEGF) receptor 2 (Flk-1/KDR) with integrin ␣ v ␤ 3 in endothelial cells is required for cell migration and adhesion in response to VEGF, whereas integrin ␣ v ␤ 5 is required for cancer cell adhesion and motility (20,28,29). Therefore, the convergence and cascades of signaling from integrins and tyrosine kinase receptors appear to be essential in both endothelial cells and tumor cells for the induction of tumor angiogenesis, invasion, and metastasis.
The secretory protein periostin, a mesenchyme-specific gene, is normally expressed in osteoblasts. Gillan et al. (30) have found that periostin promotes adhesion and potentiates cancer cell motility through binding to integrins ␣ v ␤ 3 and ␣ v ␤ 5 . More importantly, clinical studies of periostin expression in human cancers have demonstrated that increased expression of periostin is correlated with tumor angiogenesis and metastasis (31)(32)(33). In the present study, we tested the hypothesis that periostin acquired by tumorigenic cells may facilitate EMT and induce metastatic behavior. In our studies the epithelium-derived, tumorigenic 293T cells were used to determine the effects of periostin both in vitro and in vivo. We found that 293T cells overproducing periostin induced EMT and metastasis and that EMT required the coordination of integrin ␣ v ␤ 5 and EGFR signals.

EXPERIMENTAL PROCEDUERS
Generation of Periostin-producing Cells-Full-length human periostin cDNA (MluI-XhoI) was subcloned into a retroviral pCMV-neo-vector. 293T retroviral packaging cells were transfected with the periostin construct or vector control in the presence of pCL 10A1 vector using FuGENE 6 as the delivery vehicle. 48 h after transfection, the supernatant was harvested and filtered through a 0.45-m pore size filter, and the virus-containing medium was used to infect parental 293T cells. Selection with 800 g/ml of G418 was started 48 h after infection, and the drug-resistant cell populations were used for subsequent studies.
Periostin siRNA Gene Knock Down-Specific oligos (21 bp) targeting the periostin gene were selected, and an oligo template (64 oligo nucleotides) containing the 21-bp oligos was subcloned into a retroviral pSUPER-puro-vector. Retroviral medium was generated as described above, and consequently the medium infected 293T cells expressing periostin to establish a stable line containing periostin siRNA.
Gelatin Zymography-Cell-conditioned serum-free medium was collected for zymographic analysis. Gelatinolytic activities were assessed under non-reducing conditions using a 6% SDSpolyacrylamide gel with 1 mg/ml of gelatin. After being washed with 2.5% Triton X-100 twice for 30 min, the gel was incubated in zymography buffer containing 150 mM NaCl, 5 mM CaCl 2 , 50 mM Tris-HCl, pH 7.5, for overnight at 37°C. The gel was then Left, cell lysates from parental, control, and periostin-producing cells were employed to examine vimentin, fibronectin, and cytokeratin expression. Medium was collected to detect secreted protein periostin. Right, the serum-free medium from 24-h cultured cells was used to determine the active forms of MMP-2 and MMP-9. C, periostin increases MMP-9 proteolytic activity. The serum-free medium was collected from parental, control, and periostin-expressing cells to determine MMP-9 activity in a zymographic analysis.  Recombinant Periostin-Periostin was isolated and purified from baculoviral system as described previously (31).
Migration Assay-Cells (2 ϫ 10 5 ) were preincubated with serum-free medium for 24 h and transferred onto transwells (24-well plates) precoated with fibronectin (50 g/ml). The lower chamber of transwells included EGF, periostin, tyrphostin 25, or anti-integrin ␣ v ␤ 3, ␣ v ␤ 5 antibody. After 4 h of incubation, top cells on the transwell membrane were removed using Q-tips. The cells trapped by membrane were fixed and stained with hematoxylin. Average cell numbers from five different areas in each example were counted.
Invasion Assay-Cells (2 ϫ 10 5 ) were preincubated with serum-free medium for 24 h and transferred onto transwells (24-well plates) preloaded with 50 l of matrigel (1 g/ml). After 18 h of incubation, the gel including non-migrated cells was removed and cells invading into the membrane were fixed and stained.
Adhesion Assay-Cells (5 ϫ 010 4 ) were preincubated with serum-free medium for 24 h and transferred to 48-well plates precoated with periostin or vitronectin (10 g/ml). After 1 h of incubation, cells were gently washed and attached cells were fixed, stained, and counted.
Immunocytochemistry-Cells were transferred onto glass slides to culture for 2 days. After being fixed with 4% paraformaldehyde, cells were blocked with phosphate-buffered saline containing 5% goat serum, 1% bovine serum albumin, and 0.05% Nonidet P-40 followed by incubation with anti-vimentin antibody for 1 h. Fluorescein-conjugated secondary antibody was then added for 1 h, and fluorescence was examined under a microscope.
Induction of Tumor Xenografts and Metastasis in Mice-4-week-old female SCID-Beige/NOD mice (Charles River, Wilm-

. Periostin increases EGFR expression, and tyrosine-phosphorylated EGFR is detectable in the presence of periostin and EGF.
A, 293T cells mainly express integrin subunits ␣ v and ␤ 5 , and periostin induces EGFR expression. Cell lysates from control or periostin-producing cells were used to determine integrin subunits ␣ v, ␤ 3, ␤ 5 , and EGFR expression by immunoblotting. B, integrin ␣ v ␤ 5 is associated with EGFR in the periostin-producing cells. Cell lysates from control or periostin-producing cells were immunoprecipitated with anti-integrin ␤ 3 or ␤ 5 antibody followed by immunoblotting with anti-EGFR antibody. No specific interaction between integrin ␤ 3 and EGFR was found in either control or periostin-expressing cells (data not shown). C, integrin ␣ v ␤ 5 is associated with tyrosine-phosphorylated EGFR in the presence of periostin and EGF . Left, following starvation by the removal of serum for 24 h, control or periostin-producing cells were stimulated with EGF (100 ng/ml) for 5-10 min and cell lysates were immunoprecipitated with anti-integrin ␤ 3 or ␤ 5 antibody. The immunocomplex was then blotted with antibody PY-20, an antibody against phosphotyrosine protein.
Right, periostin-producing 293T cells were pretreated with serum-free medium in the absence or presence of ␣ v ␤ 5 antibody (10 g/ml) or tyrphostin 25 (50 M) for 1 h, followed by stimulation with EGF (100 ng/ml). The lysates were immunoprecipitated with anti-integrin ␤ 5 antibody prior to immunoblotting with PY-20. D, reciprocal immunoprecipitation and immunoblot confirming the interaction between integrin ␣ v ␤ 5 and tyrosinephosphorylated EGFR. Periostin-producing cells were stimulated with EGF (100 ng/ml) for 5-10 min and immunoprecipitated with anti-EGFR or p-EGFR (Tyr-1086) antibody, followed by immunoblotting with anti-integrin ␤ 5 or ␤ 3 antibody. No specific bands were observed in the control cells with either anti-integrin ␤ 3 or ␤ 5 antibody (data not shown).
ington, MA) were injected at mammary fat pad tissue with control or periostin-producing 293T cells (1 ϫ 10 7 ) in 0.2 ml of Hanks' balanced buffer without calcium and magnesium. The growth of solid tumors from the injected cells was monitored daily for up to 50 days when animals became moribund. The tumors were measured and calculated as follows: volume ϭ length ϫ width 2 ϫ 0.52. The removed lung and liver were fixed in 10% neutral-buffered formalin and embedded in paraffin. Tissue sections were cut to 6 -10-m thickness and stained with hematoxylin and eosin by a standard method. For cardiac inoculation, cell suspension (2 ϫ 10 6 cells/0.15 ml) in phosphate-buffered saline was injected into the left ventricle of pentobarbiturate-anesthetized mice. Mice were monitored daily for up to 90 days. Necropsies were performed to examine tumor formation throughout the body.

RESULTS
We employed tumorigenic 293T cells, a typical non-metastatic, epithelium-derived tumor line, to investigate the effects of periostin on metastasis. First, we tested whether periostin is capable of promoting EMT, a process tightly associated with cell malignant behavior. We used a retroviral infection system to introduce the periostin cDNA into 293T cells and to stably produce secreted periostin. Transduction of periostin gene resulted in extensive cell spreading and elongation, displaying a spindle-like morphology, whereas this alteration was not observed in control cells (Fig. 1A, top panel). 293T cells ectopically expressing periostin dramatically increased vimentin expression, a hallmark for mesenchymal cells, by 7-8-fold (Fig.  1, A and B). The level of vimentin in vector control cells and parental 293T cells was barely detectable. The elevation of vimentin expression was validated by immunocytochemistry staining. Fibronectin expression was also induced by the periostin transgene in the cells. Expression of the epithelial marker cytokeratin was not significantly altered in periostin-producing 293T cells. The active form of MMP-9 was strikingly increased in periostin-producing cells and accumulated in the medium ϳ5-8-fold higher than the level produced from control cells, accompanied by high proteolytic activity in the cells engineered with periostin (Fig. 1C ). In contrast, neither the active form of MMP-2 (Fig. 1B) nor MMP-3 (not shown) was changed in the conditioned medium. Expression of E-or N-cadherin was also unaltered (data not shown). In aggregate, the data demonstrate that periostin expressed by 293T cells drives the cells to undergo EMT.
To explore membrane receptors that may mediate periostininduced cell EMT, we characterized its interaction with integrins on the cell surface. As periostin was shown to be a ligand of integrins ␣ v ␤ 3 and ␣ v ␤ 5 on ovarian cancer cells (30), its interactions with integrins on 293T cells were examined (Fig. 2). Periostin-coated plates enhanced cell adhesion by 3.5-fold compared with controls. This interaction was abolished by incubation with anti-integrin ␣ v ␤ 5 antibody. This effect was specific for integrin ␣ v ␤ 5 , as anti-integrin ␣ v ␤ 3 antibody failed to block periostin-induced cell adhesion.
293T cells predominantly expressed subunits integrin ␣ v and ␤ 5 , but not ␤ 3 , implicating that a heterodimer is mainly composed of the integrin ␣ v ␤ 5 form in the cells (Fig. 3A). A similar expression pattern was observed in both control cells and periostin-expressing cells. To determine whether the cross-talk between integrins and growth factor receptors is required for

. Both integrin ␣ v ␤ 5 and EGFR activation are required for periostin-induced cell migration and invasion behavior.
A, control or periostin-producing cells were pretreated with serum-free medium for 24 h and transferred onto transwells precoated with fibronectin in the presence of tyrphostin 25 (50 M) or anti-integrin ␣ v ␤ 3 or ␣ v ␤ 5 antibody (10 g/ml) for 4 h. Migrated cells were fixed and stained, followed by quantitative analysis. For cell invasion, cells were pretreated with serum-free medium for 24 h in the absence or presence of tyrphostin 25 (50 M). The reason for pretreatment with tyrphostin 25 was that Me 2 SO dissolved for tyrphostin 25 can disrupt the matrigel. The cells were then washed and loaded to the transwells preloaded with matrigel. Following incubation in the presence of anti-integrin ␣ v ␤ 3 or ␣ v ␤ 5 antibody (10 g/ml) for 18 h, cells invading through the matrigel were fixed and stained. The data were quantitatively analyzed. *, p Ͻ0.05 compared with control group; ϩ, p Ͻ0.05 compared with the periostintreated group alone. B, 24 h following treatment with serum-free medium in the presence of anti-integrin ␣ v ␤ 3, ␣ v ␤ 5 antibody (10 g/ml) or tyrphostin 25 (50 M), the conditioned medium was collected for testing the active form of MMP-9. Cell lysates were used to examine actin expression by immunoblotting.
periostin activity, we first examined EGFR expression in 293T cells. As shown in Fig. 3A, EGFR expression was augmented in the cells expressing periostin ϳ2-fold higher than that in control. We examined the physical interaction between integrins and EGFR by immunoprecipitation and immunoblotting after the cells were exposed to EGF. Following stimulation of the cells with EGF, integrin ␤ 5 -associated proteins were co-immunoprecipitated. EGFR was associated with integrin ␤ 5 in the periostin-expressing cells but not the cells with vector alone (Fig. 3B). Activated EGFR was examined by immunoblotting with an antibody against tyrosine-phosphorylated molecules. Once cells were exposed to EGF, tyrosine-phosphorylated EGFR was detected in integrin ␤ 5 -associated complex from periostin-engineered 293T cells but not control cells (Fig. 3C). No tyrosine-phosphorylated EGFR was observed in integrin ␤ 3 -associated complex from either control or periostin-engineered cells. This specific association between integrin ␤ 5 and activated EGFR was inhibited by the preincubation of periostinproducing cells with anti-integrin ␣ v ␤ 5 antibody or tyrphostin 25, an EGFR kinase inhibitor (Fig. 3C). The specific interaction was also confirmed by the reciprocal immunoprecipitation and immunoblot. There was no detectable interaction between EGFR and integrin ␤ 3 , whereas phosphorylated EGFR showed the association with integrin ␤ 5 (Fig. 3D). The data suggest that integrin ␣ v ␤ 5 and EGFR form a complex but require both periostin and EGF as ligands.
Next, to test whether periostin expression can induce aggressive activities in the absence of exogenous EGF, we monitored functional alterations in cell migration and invasion. 293T cells expressing periostin exhibited a significant increase in cell motility (Fig. 4A). More than two times as many periostin-producing cells migrated into the membrane relative to control cells. When either anti-integrin ␣ v ␤ 5 antibody or tyrphostin 25 was added, the increased motility in periostin-producing cells was blocked, whereas treatment with anti-integrin ␣ v ␤ 3 antibody failed to inhibit the cell migration. We also monitored cell invasive behavior using a modified migration assay in which a layer of matrigel was preloaded on transwells. As shown in Fig.  4A, ectopic expression of periostin in the cells resulted in a 9-fold increase in cell invasive activity relative to the control cells. Treatment of these aggressive cells with either anti-integrin ␣ v ␤ 5 antibody or tyrphostin 25 substantially impaired cell invasion as the activity was reduced to the basal level. As expected, inclusion of anti-integrin ␣ v ␤ 3 antibody did not block invasion. Likewise, periostin enhanced cell adhesive activity, but it was abrogated by anti-integrin ␣ v ␤ 5 antibody or tyrphostin 25 (data not shown). To determine the regulation of intracellular molecules that may play an important role in cell aggressive behavior, we examined MMP-9 gene expression.
Consistent with the previous result, periostin-expressing cells contained a higher level of MMP-9 than in control cells. However, the increased MMP-9 level in periostin-producing 293T cells was fully inhibited to the control level by an addition of either anti-integrin ␣ v ␤ 5 antibody or tyrphostin 25, whereas blockage of integrin ␣ v ␤ 3 failed to attenuate MMP-9 production (Fig. 4B). The data suggest that co-activation of integrin ␣ v ␤ 5 and EGFR is required for periostin-induced MMP-9 production and cell invasive function. The effectiveness of inhibition by anti-integrin ␣ v ␤ 5 antibody on periostin-induced cell functions demonstrates that periostin produced from the cells acts through its secreted protein that interacts with integrin ␣ v ␤ 5 but not via intracellular pro-periostin.
The in vitro data that periostin induces EMT and invasive activity encouraged us to test whether periostin promotes tumor invasion and metastasis in vivo. For this, we injected control and periostin-producing 293T cells into mouse mammary fat pad tissue and monitored growth of tumors at the injection site as well as in distant organs. The mice receiving periostin-producing 293T cells grew local tumors significantly faster than did the mice receiving control cells during the 35-day observation (Fig. 5A). During the next 2 weeks, the mice injected with periostin-producing cells began to die, resulting in a rapid decrease in survival rate. But no death occurred in the control group within the same time period of observation (Fig.  5B). Strikingly, we found that the tumors bearing periostinproducing cells metastasized to the lung and liver ( Fig. 5C and Table 1), in which the rate of tumor formation was 83 and 17%, respectively. No tumor metastasis was identified in either organ in mice injected with control cells. New blood vessels adjacent to the tumors developed in some lung metastatic cases, indicating that periostin promotes tumor angiogenesis in the secondary solid tumor formation during metastasis. To confirm that periostin has ability to promote tumor cell invasion and metastasis, we utilized the siRNA gene knockdown approach in periostin-expressing 293T cells. As shown in Fig. 5D, the periostin  6). B, decreased survival rate in mice carrying periostin tumors. Survival of mice bearing control and periostin-producing tumors was observed daily for 50 days following injection. C, periostin-producing tumors metastasize to the lung and liver. Following sacrifice of moribund mice during day 40 -50, the lung and liver were fixed in formalin, embedded in paraffin, and processed for hematoxylin and eosin staining. Arrows indicate tumor mass, and arrowheads indicate blood vessel formation (40ϫ). D, periostin gene knock down inhibits tumor invasive behavior. Periostin-expressing 293T cells were engineered with control vector or vector plus siRNA oligos that specifically target the periostin gene. Inhibition of periostin was evaluated by immunoblotting as shown in the top panel. Accordingly, blockage of cell invasive function was determined by the invasion assay (middle panel). In addition, control or periostin siRNA cells were injected into the mice as described in panel A. Lung metastases were found in mice injected with siRNA control cells but not in mice injected with YKL-40 siRNA cells (n ϭ 5).

TABLE 1 Administration of periostin-producing 293T cells in the mammary fat pad tissue results in lung and liver tumor formation, but periostin gene knock down inhibits metastasis
Cells expressing control vector or periostin were introduced into the mice through mammary fat pad tissue injection. The organs with metastasized tumors were examined and verified by hematoxylin and eosin staining in tissue sections (10 -20 sections/organ). Periostin Induces Cell Invasive Activity through EMT JULY 14, 2006 • VOLUME 281 • NUMBER 28 gene in siRNA cells was suppressed to ϳ30% of the level in periostin-producing cells. This decreased expression of periostin was sufficient to abrogate cell invasion in vitro and metastasis in vivo ( Fig. 5D and Table 1), although the local tumor growth was similar between the siRNA vector control and periostin siRNA tumor (data not shown).
To further evaluate the metastatic activity, we injected control or periostin-producing 293T cells directly into the blood system via left ventricle inoculation. In agreement with the above results, the survival rate in the mice bearing periostinproducing cells was significantly lower than that of the mice carrying control cells (Fig. 6A). A variety of visible large second-ary tumors in the liver, lung, and abdomen were identified in two of six mice receiving periostin-producing cells (Fig. 6B). Moreover, an extensive arbor of ramified blood vessels was also found in those tumors, consistent with enhanced tumor angiogenesis. No tumors were detected in organs from the control mice. The data strengthen our hypothesis that expression of periostin by tumor cells prompts tumor invasion and metastasis.

DISCUSSION
A wealth of evidence has demonstrated that tumors of epithelial origin usually undergo phenotypic transformation dur-FIGURE 6. Survival rate is shortened in mice bearing periostin-producing cells, and a variety of tumors in multiple organs are observed after a cardiac inoculation with periostin-producing cells. Control or periostin-producing 293T cells (2 ϫ 10 6 cells) were introduced directly into the blood system through left ventricle injection as described under "Experimental Procedures." A, survival of mice was observed daily for up to 90 days (n ϭ 6). p Ͻ0.05 compared with the control group. B, two of six mice injected with periostin-producing 293T cells were found to form different sizes of tumors in the lung, liver, and abdomen. Arrows indicate tumors; arrowheads indicate blood vessels.
ing tumor metastasis (34,35). A variety of classic molecules involved in cell motility, cell-cell contacts, and cell-extracellular matrix interaction have been well characterized in metastasis such as vimentin, E/N-cadherin expression, and MMP production. In addition, a number of growth factors including TGF-␤, FGF1, EGF, and SF/HGF can induce tumor cell EMT process under culture conditions (36). In the present study, we have observed the direct effects of periostin on aggressive cell behavior by expressing periostin in 293T cells. We found that exogenous expression of periostin in 293T cells promoted the cells to undergo EMT. They gained vimentin expression, a hallmark of EMT, and transformed to fibroblast-like phenotype including spreading of cells accompanied with increased MMP-9 production. Consequently, these cells increased cell adhesion, migration, and invasion in vitro and metastatic potential in vivo.
Cadherins, the cell-cell adhesion molecules, have been shown to exert important functions in the regulation of cell-cell communication. E-cadherin usually maintains epithelial properties, whereas N-cadherin mainly contributes to the mesenchymal transformation. In addition, recent evidence has demonstrated that N-cadherin up-regulates MMP-9 expression in tumor progression (37)(38)(39). Thus, either gain of N-cadherin or loss of E-cadherin expression is frequently associated with tumor metastasis (14,40,41). In the present study, no alteration was detected in either E-cadherin or N-cadherin expression after introduction of periostin into the cells. Although it is possible that N-and E-cadherin may participate in the actions of periostin, they were not rate-limiting for the responses in gene expression, migration, or invasion.
Given that periostin directly interacts with integrin ␣ v ␤ 5 and the activation of basal EGFR level is necessary for aggressiveness in response to periostin, we have revealed a model of the cross-talk for periostin that requires coordination of integrin ␣ v ␤ 5 and EGFR pathways. Periostin induced features of metastasis including cell adhesion, migration, and invasion. Tyrosine-phosphorylated EGFR was found to physically associate with integrin ␣ v ␤ 5 exclusively in the cells when they were exposed to both receptor stimuli, suggesting that signal transduction is enhanced once both receptor ligands are present. The cytoplasmic tail of integrin ␤ subunit participates in the physical association with the cytoplasmic domain of growth factor receptors favoring the coordination of both membrane receptor-mediated signaling pathways (42)(43)(44). A similar active role of the specific interaction between integrins and tyrosine kinase receptors has been established in a variety of cell types (45). For example, in endothelial cells, the interaction between integrin ␣ v ␤ 3 and vascular endothelial growth factor receptor 2 is essential for vascular endothelial growth factorinduced angiogenesis (46,47). Likewise, activation of EGFR is required for integrin ␣ v ␤ 5 -directed motility in FG human pancreatic carcinoma cells, whereas the interaction between integrin ␣ v ␤ 3 and EGFR mediates cell survival and proliferation in smooth muscle cells (25,44). Our results, in context with other reports, have underscored the paradigm that the coordination of inputs from integrins and growth factor receptors is essential for periostin-induced cell invasive activity.
Clinical studies found higher serum levels of periostin in cancer patients with malignant diseases compared with patients with local cancers (32,33,48). The high concentration of periostin in the blood was directly correlated with poor prognosis and short survival in multiple cancers, including breast, lung, head, and neck cancers. Therefore, periostin may be a valuable biomarker for tumor metastasis. The elevated production of periostin in metastatic diseases may directly cause cancerous cells to induce malignant transformation of mesenchymal phenotype. To this end, our data have identified important functional and molecular mechanisms for periostin in tumor metastasis. Periostin triggers the co-activation of integrin ␣ v ␤ 5 and EGFR signaling and induces expression of multiple genes such as MMP-9, vimentin, and fibronectin, leading to EMT and metastasis (Fig. 7). Elevated serum levels of periostin in cancer patients provide a promising tool for early detection of cancer and for monitoring cancer patients. More importantly, the results underscore possible applications of EGFR kinase inhibitors for control of metastasis in multiple human cancers in which both EGFR and periostin are overproduced (33,49,50).