Cancer cell migration on elongate protrusions of fibroblasts in collagen matrix

Cancer-associated fibroblasts (CAFs) play critical roles in the tumor progression. However, it remains unclear how cancer cells migrate in the three-dimensional (3D) matrix of cancer tissues and how CAFs support the cancer invasion. Here we propose a novel mechanism of fibroblast-dependent cancer cell invasion in the 3D collagen matrix. Human cancer cell lines from the pancreas (Panc-1), lung (A549) and some other organs actively adhered to normal fibroblasts and primary lung CAFs in cultures. To show its significance in tumor invasion, we designed a new invasion assay in which homogeneous microspheroids consisting of cancer cells and fibroblasts were embedded into collagen gel. Time-lapse experiments showed that cancer cells adhered to and quickly migrated on the long protrusions of fibroblasts in the 3D collagen matrix. Fibroblast-free cancer cells poorly invaded the matrix. Experiments with function-blocking antibodies, siRNAs, and immunocytochemistry demonstrated that cancer cells adhered to fibroblasts through integrin α5β1-mediated binding to fibronectin on the surface of fibroblasts. Immunochemical analyses of the co-cultures and lung cancers suggested that cancer cells could acquire the migratory force by the fibronectin/integrin signaling. Our results also revealed that the fibroblast-bound fibronectin was a preferential substrate for cancer cells to migrate in the collagen matrix.

SCientifiC RepORTs | (2019) 9:292 | DOI: 10.1038/s41598-018-36646-z upregulated in cancer stroma 17 . Recent studies have shown that CAF-derived fibronectin matrix guides cancer cell migration [18][19][20] . Although many past studies extensively revealed the roles of various microenvironmental factors in tumor progression, it has not been fully understood yet how cancer cells migrate through the three-dimensional (3D) matrix and how CAFs support the cancer invasion. To address these issues, studies in 3D conditions are critically important. A recent study demonstrated that fibroblasts lead collective cell migration of epidermoid carcinoma cells in collagen matrix by the direct cell adhesion through E-cadherin/N-cadherin heterophilic interaction 21 . Relatively less attention has been paid on the significance of the direct interaction between cancer cells and CAFs in tumor progression [22][23][24][25] . As cancer cells often lose E-cadherin on the cell surface during their malignant progression, the E-/N-cadherin mechanism seems irrelevant to many kinds of invasive cancers [26][27][28] . Our recent study showed that fibroblasts induce EMT-like morphological change of cancer cells on collagen gels 29 . Further attempt to clarify this mechanism has revealed that the heterotypic cell adhesion between cancer cells and fibroblasts facilitates the morphological EMT of the cancer cells and their migration. The present study demonstrates a tumor invasion mechanism in which cancer cells invade the collagen matrix by binding to fibroblasts through the integrin/fibronectin interaction.

Direct interaction of various types of human cancer cells with normal fibroblasts or CAFs in two-dimensional (2D) cultures. To investigate the direct interaction between human cancer cells and
fibroblasts, we mainly used co-culture models of GFP-labeled Panc-1 pancreatic adenocarcinoma cell line or GFP-labeled A549 lung adenocarcinoma cell line with the fetal lung fibroblast line WI-38 or the OUS-11 fibroblast line derived from a non-neoplastic tissue of a lung adenocarcinoma. We also used primary lung CAFs (LuCAFs) and control normal fibroblasts (Ctr-NLFs) from the same adenocarcinoma patient. The expression of the myofibroblast marker αSMA was much higher in WI-38 and OUS-11 than Ctr-NLF and LuCAF as analyzed by immunoblotting, but it was comparable among the four cell lines in immunocytochemistry, indicating that the cytoskeletal αSMA level does not necessarily reflect its total protein level ( Fig. S1a,b). When compared between Ctr-NLF and LuCAF, the latter appeared slightly higher than the former in both analyses. All fibroblast lines expressed fibronectin at comparable levels ( Fig. S1a,c). These results suggested that all these cell line had been activated in culture.
When Panc-1 cells and WI-38 cells were mixed and plated on plastic dishes in 2D conditions, cancer cells actively bound to fibroblasts by extending pseudopodial protrusions during incubation for 20 h (Fig. 1a). Cancer cells contacted not only the lateral side of fibroblasts but also their apical surface (Fig. 1a, upper right panel). On further incubation, most of Panc-1 cells, like EMT-induced cells, exhibited extremely elongated shapes along the fibroblast structures (Fig. 1b). When Panc-1 cells were plated on a sub-confluent or confluent culture of WI-38 cells which had been pre-cultured for 2 or 3 days, such morphological changes of Panc-1 cells occurred in a few hours ( Fig. S2ab and Video 1). The time-lapse video showed that cancer cells migrated by extending lamellipodia-like protrusions while keeping contact with fibroblasts. A549 lung adenocarcinoma cells also showed the heterotypic cell-cell interaction with WI-38 cells (Fig. 1c) and the normal lung fibroblast line OUS-11 (Fig. 1d). As shown by the time-lapse video, A549 cells more rapidly migrated along the fibroblast structures than Panc-1 cells (Video 2, see also Video 1). In an orthotopic lung cancer model, A549 cells interacted with the lung CAFs (LuCAFs) and the control fibroblasts (Ctr-NLFs) (Fig. 1e,f). There was no significant difference in the interaction with A549 cells between LuCAFs and Ctr-NLFs.
The heterotypic cell adhesion with WI-38 fibroblasts was also found in many types of invasive human cancer cells such as MKN-45 gastric carcinoma cells (Fig. S2c), STKM-1 gastric carcinoma cells (Fig. S2d), MIA-PaCa-2 pancreatic carcinoma cells (Fig. S2e) and MDA-MB-231 breast carcinoma cells (Fig. S2f). Although MKN-45 and STKM-1 cells hardly spread on the usual plastic surface, they could spread in contact with fibroblasts ( Fig. S2c,d). Especially, STKM-1 cells attached to fibroblasts by extending extremely long protrusions. On the other hand, another group of cancer cell lines, such as MCF-7 breast adenocarcinoma ( Fig. S3a,b), A431 vulval epidermoid carcinoma (Fig. S3c,d), CaSki squamous cell carcinoma (Fig. S3e) and MKN-74 gastric carcinoma (Fig. S3f), all of which expressed E-cadherin at high levels, formed tumor cell islands surrounded by fibroblasts in the co-cultures, suggesting stronger homophilic intercellular junctions compared with the heterophilic ones.
Invasion of cancer cells in the presence of fibroblasts in three-dimensional (3D) collagen matrix. The 3D-collagen matrix culture was used to investigate a possible role of the cancer cell/fibroblast interaction in tumor invasion. When Panc-1 cells were embedded into the collagen gel, they kept round morphology or cell aggregates for over 6 days (Fig. 2a, left panel). However, when co-cultured with fibroblasts for a few days, cancer cells adhered to a string structure of fibroblasts (Fig. 2a, center panel). On further incubation, fibroblasts formed a net-work structure of their strings to which many cancer cells attached and extended protrusions (Fig. 2a, right panel). These observations suggested that the direct binding of cancer cells to fibroblasts would support cancer cell invasion in the 3D collagen matrix. It is also noted that fibroblasts extremely elongated, extending invasive protrusions in the 3D matrix.
To facilitate cancer cell invasion in the collagen matrix, we designed a new invasion assay in which homogeneous chimeric spheroids consisting of cancer cells and normal fibroblasts or CAFs were prepared in micro-fabricated 96-well plates and embedded into collagen gel. Panc-1 cells alone formed loose aggregates in the plates (Fig. S4a), but they produced solid chimeric spheroids with WI-38 cells in which cancer cells distributed in both the surface area and inner core of spheroids ( Fig. S4b-d).
When the Panc-1/WI-38 chimeric spheroids were embedded in the 3D collagen matrix, the cancer cell invasion began in a few hours and became evident in one or two days (Fig. 2b). After fibroblasts sufficiently invaded the collagen matrix in the radial direction from the spheroids, cancer cells attached to and migrated on the  showed invasive morphology compared with those in their presence (Fig. S4e). When the chimeric spheroids of A549 cells and WI-38 cells were embedded in the collagen gel, the lung cancer cells showed fibroblast-dependent invasion more actively than Panc-1 cells (Fig. 3a). More striking invasion of both cancer cells and fibroblasts was observed in a combination of A549 and OUS-11 ( Fig. 3b and Video 4). Cancer cells actively migrated while transferring fibroblasts one by one ( Fig. 3c and Video 4). Some cancer cells that were unable to leave the fibroblast vehicle showed reciprocating movement on a string of fibroblast ( Fig. 3d and Video 4). A strong invasive activity of A549 cells was also found when their chimeric spheroids with the primary lung CAFs (LuCAFs) or the control fibroblasts (Ctr-NLFs) were applied to the 3D invasion assay as the orthotopic lung cancer model ( Fig. 3e and Video 5). Although cancer cells appeared to attach to LuCAFs more than Ctr-NLFs, the difference was not evident. When the four fibroblast lines were compared in the combination with A549 cells, LuCAFs and Ctr-NLFs appeared to have the highest invasive activity in the 3D invasion assay. In the case of OUS-11 fibroblasts, the invasion of fibroblasts was greater in the presence of A549 cells than Panc-1 cells, suggesting that cancer cells also affect invasive activity of fibroblasts (Fig. S4f). All these observations strongly suggest that cancer cells acquire invasive activity by directly binding to fibroblasts in the 3D collagen matrix.
Identification of molecules responsible for heterotypic intercellular adhesion. The molecular mechanism by which cancer cells directly binds to fibroblasts was investigated using the Panc-1/WI-38 co-culture model. Past studies showed the involvement of some integrins in the heterotypic adhesion of gastric cancer cells to CAFs 23 or CAF-guided cancer cell migration 22 . Therefore, we first examined effects of function-blocking antibodies against some integrin molecules and the RGD peptide (integrin α5 inhibitor) on the Panc-1 cell attachment to the confluent culture of WI-38 cells. When Panc-1 cells were pretreated with each of these inhibitors, their attachment was more significantly inhibited by RGD and antibodies against integrin β1 and integrin α5 than the others (Figs 4a and S5a). This suggested that integrin α5β1 in cancer cells might bind to fibronectin on the surface of WI-38 cells. To verify this possibility, function-blocking antibodies were further tested using sub-confluent WI-38 cell cultures. In the assay, we determined the proportion of elongated or extended Panc-1 cells to the total cells in contact with WI-38 cells. The antibodies against integrin α5, β1 and α5β1 strongly ). In addition to the experiments with the function-blocking antibodies, we performed knockdown experiments of integrin α5 and fibronectin genes with specific siRNAs. The treatment of Panc-1 and A549 cells with the specific siRNA pool efficiently blocked the expression of integrin α5 protein in both types of cancer cells as analyzed by immunoblotting (Fig. S6a). The knockdown of integrin α5 gene in Panc-1 and A549 cells significantly inhibited their elongation in contact with WI-38 fibroblasts (Fig. 5a-c). Similarly, the siRNA pool for fibronectin strongly blocked the fibronectin expression in WI-38 cells (Fig. S6b), and this treatment efficiently blocked the spreading of Panc-1 or A549 cancer cells in contact with WI-38 fibroblasts (Fig. 5d-f). The efficient suppression of fibronectin expression in WI-38 cells was confirmed by immunofluorescent staining for cell-associated fibronectin (Fig. S7a). The fibronectin knockdown also caused significant enlargement of WI-38 cells (Fig. S7b). On the other hand, the knockdown of integrin α5 had no significant effect on the morphology of Panc-1 cells (Fig. S7c).
The knockdown effects of the integrin α5 and fibronectin genes were evaluated for the Panc-1/WI-38 interaction in the EZSPHERE micro-fabricated vessels. The integrin knockdown in Panc-1 cells tended to reduce the number of the chimeric spheroids produced with WI-38 cells (p = 0.056) (Fig. 6a, left panel), while the fibronectin knockdown in WI-38 cells greatly inhibited the spheroid formation (p < 0.001) (Fig. 6a, right panel). In addition, the sizes of the chimeric spheroids were weakly but significantly reduced by both the knockdown of integrin α5 and fibronectin (Fig. S7d,e). We also attempted to quantify the knockdown effects on the cancer cell invasion in the 3D collagen matrix by counting cancer cells attached to invading fibroblasts on fluorescent images for one cross section per spheroid. The Panc-1 cell adhesion to the fibroblast protrusions in the collagen matrix was significantly blocked by both the integrin α5 and the fibronectin knockdown (p < 0.05 and p < 0.01, respectively) (Fig. 6b,c and Video 7). These results indicate that the integrin α5β1/fibronectin interaction is required not only for the heterotypic cell adhesion in both 2D and 3D conditions but also for the chimeric spheroid formation in the micro-fabricated vessels. Moreover, poor interaction between cancer cells and fibroblasts in the 3D collagen matrix was observed when the Panc-1/WI-38 chimeric spheroids without siRNA treatment were treated with the anti-integrin-α5β1 or the anti-fibronectin antibody (p < 0.05 and p = 0.066, respectively) (Fig. 6d,e). These antibody treatments also appeared to suppress the invasion of fibroblast protrusions. Localization of cell adhesion molecules. The mechanism for the heterotypic cell-cell adhesion was further verified by immunocytochemistry. Immunofluorescent staining of 2D co-cultures for fibronectin revealed that abundant fibronectin fibrils, to which Panc-1 cells appeared to adhere, were assembled on WI-38 cells (Fig. 7a,b). Similar or rather stronger signals for the deposited fibronectin fibrils were found on LuCAFs, Ctr-NLFs and OUS-11 cells in 2D cultures (Fig. S1c). The assembly of fibronectin fibrils on fibroblasts was further confirmed in the 3D co-cultures of the Panc-1/WI-38 (Fig. 7c), A549/OUS-11 (Fig. 7d) and A549/LuCAF (Fig. 7e). Together with the inhibition data shown in Figs 4-6, these immunostaining data indicate that cancer cells attach to and migrate on fibronectin fibrils deposited on fibroblasts. Notably, A549 cells extended long invasive protrusions along the fibronectin-rich fibroblast strings (Fig. 7d,e). We also analyzed the localization of integrin molecules in the 2D Panc-1/WI-38 co-cultures. Integrin α5β1 was co-localized with fibronectin fibrils at the Panc-1/WI-38 contact sites (Fig. 7f). It is known that the polymerized fibronectin has stronger cell adhesion activity than the monomer 30 . Therefore, it was expected that the cancer cells bound to the fibrillar fibronectin might have stronger intracellular signaling than those bound to the monomer fibronectin or unbound cells. Immunofluorescent staining for phospho-Erk1/2 (p-Erk) (Fig. 7g, left panel) and phospho-Akt (p-Akt) (Fig. 7g, right panel) indicated that these signaling mediators were indeed activated in the cancer cells bound to fibroblasts.
All the results described above support the mechanism that Panc-1 and A549 cancer cells bind to fibroblasts through the integrin α5β1/fibronectin interaction and migrate on the fibroblast protrusions in the collagen matrix. To examine the possible involvement of the E-cadherin/N-cadherin interaction in our heterotypic cell-cell adhesion models, we analyzed the expression of E-cadherin and N-cadherin in the 2D co-cultures of Panc-1 or A549 cancer cells with WI-38 fibroblasts. In immunoblotting analysis, E-cadherin expression was negligible in Panc-1 cells (<0.4%) and far lower in A549 cells (<17%) than that in MCF-7 cells, but the N-cadherin expression levels were comparable among the three cancer cell lines and WI-38 fibroblast (Fig. S8a). Immunofluorescent staining was unable to detect any clear E-cadherin localization in the A549/WI-38 contact sites (Fig. S8b). Even in the co-cultures of the high E-cadherin-expressing cell lines (MCF-7 and A431) with WI-38 cells, localization of E-cadherin at their heterotypic contact sites was not evident in our experimental conditions (Fig. S3a-d). In the case of the MCF-7/WI-38 combination, localization of integrin α5β1 in MCF-7 cells was detected at their contact sites with fibronectin-binding WI-38 cells but less evidently compared with the case of Panc-1 cells (Fig. S8c, see also Fig. 7f). This suggested that the integrin α5β1/fibronectin interaction might also be involved in the weak adhesion of MCF-7 cells to WI-38 cells (Fig. S3a).

Interaction between integrin α5β1 and fibronectin in invasive adenocarcinomas. The possible role
of the integrin α5β1/fibronectin interaction in human cancer invasion was investigated by immunohistochemistry of human lung adenocarcinoma tissues. In at least 10 cases of invasive lung adenocarcinomas tested, invasion fronts of cancer cells were surrounded by the stroma rich in fibronectin fibrils. Double immunofluorescent staining revealed co-localization of integrin α5β1 on cancer cells with the stromal fibronectin at the invasion fronts (Fig. 8a, insets). In another case of adenocarcinoma, the stromal tissue directly surrounding a cluster of integrin-α5β1-positive cancer cells was strongly stained for α-tubulin (Fig. 8b). This confocal image also shows a few cancer cells which have singly invaded into the adjacent α-tubulin-rich stroma (Fig. 8b, insets). These invading cells were also positive for integrin α5β1. At the invasion fronts of the same tissue, the strong α-tubulin signals were co-localized with strong fibronectin signals (Fig. 8c). Because microtubules construct various cell structures including string-like protrusions of fibroblasts, the dense fibronectin fibrils in the tumor stroma are likely to be mainly cell-associated fibronectin fibrils rather than the fibronectin deposits in the ECM. This possibility was also supported by close localization of the immunofluorescent signals for fibronectin and the mesenchymal cell marker vimentin in the lung cancer tissues (Fig. S8d,e). Some cancer cells appeared to directly attach to the vimentin-positive CAF-like cells.
The cancer cells collectively or singly invading into dense fibronectin fibrils showed stronger positive signals for the activated Erk (p-Erk) (Fig. 8d) and the activated Akt (p-Akt) compared with inner cancer cells in fibronectin-less environments (Fig. 8e). These data also support the mechanism that the binding of cancer cells to fibroblasts via the integrin α5β1/fibronectin interaction promotes cancer cell invasion.

Discussion
How do cancer cells migrate through the 3D matrix in vivo? This has long been investigated as one of the principle subjects in cancer biology. In the 2D culture conditions, cancer cells attach to a suitable substrate via cell-surface adhesion molecules and spread by constructing a new actin cytoskeleton. These cellular changes provide the contractile force for the cells to migrate. How is this scenario attained in the 3D matrix? Although the cell migration mechanism is much more complex in the 3D conditions than the 2D ones, an appropriate cell adhesion substrate is essential for the cells to migrate in both 2D and 3D conditions. The present study using the 2D and 3D co-culture models demonstrates a novel tumor invasion mechanism by which invasive cancer cells migrate on the long strings of fibroblast protrusions in the 3D collagen matrix as if trains run on a railroad track (Fig. 8f). The cancer cell migration was mediated by their adhesion to the fibrillary fibronectin assembled on the surface of fibroblasts. The integrin-mediated binding to the fibronectin is expected to supply a sufficient migratory force via the integrin signaling to the cancer cells. This was supported by the results from the immunostaining experiments of the 2D co-cultures and lung cancer tissues for p-Erk and p-Akt. The extended or elongate morphology of cancer cells in contact with fibroblasts in the 3D matrix also suggested that they acquired enough tension for their migration from the heterotypic cell-cell adhesion.
The heterogeneous cell-cell interaction has been reported by a relatively small number of studies. Gaggioli et al. 22   junction. The intercellular junction via E-cadherin/N-cadherin heterophilic interaction has been reported by other groups 31 . In the present study, we also showed fibroblast-dependent cancer cell invasion in the 3D matrix. In our model, however, fibroblasts function as a track for running vehicles rather than the traction vehicles themselves. Moreover, our results with function-blocking antibodies and siRNAs for gene silencing clearly indicated that the principal molecules mediating the heterotypic cell-cell adhesion were integrin α5β1 on cancer cells and fibronectin assembled on the surface of fibroblasts. However, our results do not exclude the possible involvement of other additional cell adhesion molecules in the intercellular adhesion. E-cadherin is down-regulated at later stages of cancer progression, while N-cadherin is reciprocally up-regulated 27,28 . N-cadherin is thought to be an EMT marker, and its expression is associated with the invasive activity of cancer cells 28 . In the present study, Panc-1 and A549 cell lines, as well as fibroblast cell lines, expressed N-cadherin at high levels. N-cadherin and other cell adhesion molecules such as integrins α9β1, α4β1 and αvβ3 for fibronectin 16,32 , cadherin-23 24 , nectins/ afadin 21,33 , and integrin α4β1/V-CAM 34 may also support the heterotypic cell-cell adhesion. However, the stable cell adhesion mechanisms such as the adherence and tight junctions seem unsuitable for our cancer invasion model. It should be emphasized that cancer cells which have lost E-cadherin still have a strong capacity to bind to fibroblasts. It is expected that cancer cells having low affinity for fibroblasts would have poor invasive activity in the 3D co-culture model. In preliminary experiments, A431 cells also showed the fibroblast-dependent invasion in the 3D invasion assay but at a very low speed compared with Panc-1 and A549 cells.
Fibronectin is a major ECM component produced by stromal fibroblasts 16 and essential for the mesodermal cell migration and mesoderm formation in the development of mouse 35 . However, it has not been thought to be a molecule that mediates the intercellular adhesion. Soluble fibronectin is assembled on cell surface to form polymerized fibronectin fibrils 16 . The major cell receptor for fibronectin and its assembly is integrin α5β1 36,37 . The binding of soluble fibronectin to the integrin causes its conformational change from the compact form to the extended form, thus exposing the cell-binding sites and fibronectin-binding sites 16 . Polymerized fibronectin exerts much higher cell adhesion activity than the soluble one 30 . In addition to integrin α5β1, integrin αvβ3, α9β1, α4β1 and syndecan-4 are thought to interact with fibronectin 16 . Among them, integrin α9β1 specifically recognizes an EIIIA-containing isoform of fibronectin 32 . There are numerous studies showing that the integrin/fibronectin interaction is involved in cancer progression 17 . Fibronectin is over-expressed in primary and metastatic cancer tissues, and it is associated with poor survival of the cancer patients 18,38 . Similarly, high expression of integrin α5β1 correlates with poor prognosis in cancer patients 39,40 . The integrin α5β1-mediated adhesion to fibrillary fibronectin facilitates cancer cell invasion by activating FAK, Erk, Akt and other signal mediators and hence remodeling cytoskeleton [41][42][43] .
Very recently, three groups have reported important roles of fibronectin matrices in CAF-guided cancer cell migration in 2D cultures. CAFs lead invasion of CT26 mouse intestinal cancer cells by assembling fibronectin via integrin αvβ3 19 . GAL33 squamous carcinoma cells collectively migrate via integrins αvβ6 and α9β1, rather than α5β1, on 2D cell-free fibronectin-rich matrices 18 . Moreover, Erdogan et al. 20 found that CAFs promote directional migration of DU145 prostate carcinoma cells by organizing aligned fibronectin matrix. In this case, CAFs utilized integrin α5β1 to organize the fibronectin matrix, whereas cancer cells did αv integrins as the fibronectin receptors. Consistent with these studies, we showed important functions of CAF-derived fibronectin in cancer cell invasion. In our study, however, cancer cells utilized integrin α5β1 at least as a major receptor to bind to fibronectin fibrils on fibroblasts. Fibronectin receptors may vary depending on the type of cancer cells.
Time-lapse experiments in the present study demonstrated that cancer cells preferentially adhered to the elongate protrusions of fibroblasts through the integrin α5β1/fibronectin interaction and efficiently migrated through the collagen matrix. The fact that cancer cells which left the fibroblast protrusions greatly lost their invasive potential suggested that the cancer cells bound to fibroblasts acquired the migratory force by the integrin-mediated signaling. The collagen gel co-cultures seem to contain soluble fibronectin and its polymers released from fibroblasts. Indeed, the immunofluorescent staining detected fibronectin-rich membrane fragments around the chimeric spheroids ( Fig. 7c-e). However, these cell-free fibronectins appeared unsuitable as the substrate that sufficiently supports cancer cell spreading and migration. The fibronectin fibrils anchored to fibroblasts seem to function as the most favorable substrate for cancer cells to migrate through the collagen matrix. Fibronectin is known to interact with collagen and other ECM components 16 . However, it is unclear how fibronectin exists in the tumor stroma. As reported in other types of cancers 18,20 , fibrillary fibronectin was densely detected at the invasion fronts of cancer cells in lung cancer tissues. Such accumulation of fibronectin fibrils could be produced by CAFs, at least in parts, as a result from the stimulation by invading cancer cells. The co-localization of α-tubulin with the fibronectin matrix suggests that the fibrillary fibronectin is associated with the filamentous protrusion of CAFs. It is supposed that such fibronectin fibrils serve for a good substrate of cancer cell growth and invasion in vivo.
It is thought that there are differences between resting and activated fibroblasts, or between normal fibroblasts and CAFs, in many aspects, e.g. activities of proliferation, migration and ECM production 4,8,20 . In our experimental model, the fibroblast-dependent cancer cell migration was similarly found in all of four types of fibroblasts though their invasion-supporting activity differed among the cell types. The difference in the activity between the lung CAFs (LuCAFs) and their normal counterparts (Ctr-NLFs) was, if any, not significant. Considering the diversity in the origins and phenotypes of fibroblasts in cancer tissues [3][4][5] , the functional difference should be investigated in more detail using many different pairs of fibroblasts in future studies. We previously found that TGF-β, which is expressed in Panc-1 cells 29 , enhances the expression of fibronectin, angiomodulin and α-SMA in WI-38 cells and human dermal fibroblasts 44 . In Panc-1 and A549 cells, TGF-β down-regulates E-cadherin, up-regulates EMT markers such as vimentin, laminin-γ2 and MMPs, and enhances the activities of these cells to adhere to fibronectin and to migrate 45 . Thus, the direct interaction between fibroblasts and cancer cells as well as the indirect one via TGF-β and other soluble factors are expected to affect the fibronectin assembly in fibroblasts and the invasive activities of both types of cells.
In conclusion, we propose a new tumor invasion mechanism by which cancer cells bind to fibroblast-associated fibrillary fibronectin via integrin α5β1, leading to the activation of the integrin signaling and to the enhanced cancer cell migration along the network of fibroblast protrusions. It is expected that our results provide a clue to clarify the central mechanism by which cancer cells invade human tissues.

Materials and Methods
Antibodies and reagents. The sources and types of primary antibodies used in this study are listed in Table S1. In addition, anti-mouse-IgG and anti-rabbit-IgG goat polyclonal antibodies labeled with Cy3 or FITC were purchased from Vector (Burlingame, CA). For double immunofluorescent staining with mouse antibodies, the anti-fibronectin monoclonal antibody FN12-8 was labeled with FITC using Dojindo Ab-10 Rapid Fluorescein Labeling Kit (Kumamoto, Japan). The RGD peptide (Gly-Arg-Gly-Asp-Ser-Pro) and the control RGE peptide (Gly-Arg-Gly-Glu-Ser-Pro) were purchased from Takara (Shiga, Japan). Cell lines and culture condition. All human cancer cell lines used in this study are listed in Table S2 46 . GFP-labeled Panc-1 and A549 cell lines were previously established by introducing the pAcGFP-N3 vector from Clontech (Mountain view, CA) into these cell lines 29 . Human fetal lung fibroblast line WI-38 and adult lung fibroblast line OUS-11 (JCRB1034), which had been established from a non-neoplastic tissue of a lung cancer by Dr. M. Namba (Okayama University, Japan), were provided from Japanese Collection of Research and Bioresources (JCRB, Tokyo, Japan). Primary cultures of human lung cancer-associated fibroblasts (Lu-CAFs) and control normal fibroblasts (Ctr-NLFs) were prepared by the method of Holz et al. 47  Cancer cell invasion assay in three-dimensional (3D) collagen gel. EZSPHERE micro-fabricated 96-well plate (AGC Techno Glass) has about 90 small pores in each well and hence allows to easily make many homogeneous spheroids of cells 48 . Using this plate, chimeric spheroids with 120-150 μm in diameter were prepared by incubating 1.5 × 10 4 cells/well each of GFP-labeled cancer cells and fibroblasts for 20 h. The spheroids were recovered into a 1.5-ml tube from each well and precipitated by centrifugation at 800 rpm for 2 min. For collagen gel culture, cold 1.9 mg/ml collagen solution containing × 1 DMEM/F12 and 10% FBS was prepared from 0.3% bovine dermis native collagen (Koken Atelo Cell IAC-30, Tokyo). Seventy μl of the unsolidified collagen/ medium solution was placed on the 12-mm glass bottom of 35-mm culture dishes, spread and solidified at 37 °C for at least 30 min. The spheroid precipitate was suspended in 100 μl of the ice-cold collagen/medium solution, and its 50-μl aliquot was spread on the lower collagen gel and solidified. Two ml of the complete medium was gently added to each dish, and the 3D collagen gel culture was incubated at 37 °C in a CO 2 incubation chamber of the Keyence digital fluorescence microscope BZ-9000. Time-lapse images were taken every 30 min for one representative spheroid after pre-incubation for 1 h unless otherwise noted. To quantitate the fibroblast-dependent cancer cell invasion, cancer cells bound to invading fibroblasts were counted on representative fluorescent images with one cross section per spheroid from triplicated dishes.

Inhibition of cell adhesion by function-blocking antibodies. To block cell adhesion proteins in 2D
co-cultures, GFP-labeled Panc-1 or A549 cells were pre-incubated with 10 μg/ml of normal mouse IgG, 1 μg/ml of anti-E-cadherin antibody, or anti-integrin antibodies at a 100-fold dilution in the standard medium at room temperature (RT) for 30 min. After the treatment, the cells were transferred to the WI-38 cell culture and incubated for 2 h, followed by the cell adhesion assay as described above. The concentrations of IgG and antibodies were diluted 1.5-fold in the final assay medium. In the case of the anti-fibronectin antibody FN12-8, WI-38 cells were pre-treated with 10 μg/ml of the control mouse IgG or FN12-8 at RT for 2 h. GFP-labeled cancer cells were then applied to the WI-38 cell culture. The concentrations of the test samples were diluted two-fold in the final assay medium. In the case of 3D collagen co-cultures, control IgG or FN12-8 was included at 10 μg/ml in the upper collagen gel and the culture medium.
siRNAs and transfection. Knockdown experiments of integrin α5 and fibronectin were performed using pools of 4 siRNAs and a negative control RNA (Dhamacon SMARTpool; GE Healthcare) (Table S3). These RNAs were transfected to Panc-1 or A549 cancer cells or WI-38 fibroblasts using Lipofectamine RNAiMAX (Invitrogen). Two or three days after the transfection, these cells were used to see the knockdown effects.
Immunochemical analyses. Immunofluorescent staining was performed using the specific antibodies listed in Table S1, according to the standard protocol. Briefly, cultured cells or 5-μm frozen sections of human cancer tissues were fixed in 10% formalin/PBS and blocked with 3% (w/v) bovine serum albumin (BSA)/PBS. These samples were treated with the primary antibody, such as FN12-8 (fibronectin) at x 200 dilution and #1969 (integrin α5β1) at x100 dilution at RT for 2 h or at 4 °C overnight (Table S1) and then with the secondary antibody conjugated with Cy3 (red), Alexa Fluor 488 (green) or FITC (green) (x200 dilution) at RT for 1 h. For intracellular proteins, fixed samples were permeabilized with 0.1% (v/v) Triton X-100 in PBS for 15 min. In the case of collagen