Fibroblast surface-associated FGF-2 promotes contact-dependent colorectal cancer cell migration and invasion through FGFR-SRC signaling and integrin αvβ5-mediated adhesion.

Carcinoma-associated fibroblasts were reported to promote colorectal cancer (CRC) invasion by secreting motility factors and extracellular matrix processing enzymes. Less is known whether fibroblasts may induce CRC cancer cell motility by contact-dependent mechanisms. To address this question we characterized the interaction between fibroblasts and SW620 and HT29 colorectal cancer cells in 2D and 3D co-culture models in vitro. Here we show that fibroblasts induce contact-dependent cancer cell elongation, motility and invasiveness independently of deposited matrix or secreted factors. These effects depend on fibroblast cell surface-associated fibroblast growth factor (FGF) -2. Inhibition of FGF-2 or FGF receptors (FGFRs) signaling abolishes these effects. FGFRs activate SRC in cancer cells and inhibition or silencing of SRC in cancer cells, but not in fibroblasts, prevents fibroblasts-mediated effects. Using an RGD-based integrin antagonist and function-blocking antibodies we demonstrate that cancer cell adhesion to fibroblasts requires integrin αvβ5. Taken together, these results demonstrate that fibroblasts induce cell-contact-dependent colorectal cancer cell migration and invasion under 2D and 3D conditions in vitro through fibroblast cell surface-associated FGF-2, FGF receptor-mediated SRC activation and αvβ5 integrin-dependent cancer cell adhesion to fibroblasts. The FGF-2-FGFRs-SRC-αvβ5 integrin loop might be explored as candidate therapeutic target to block colorectal cancer invasion.


INTRODUCTION
Colorectal cancer (CRC) is the third most common cancer in the world. Despite advances in treatments, existing therapies are of limited effectiveness once cancer has become invasive or metastatic [1]. Understanding the molecular mechanisms underlying CRC cell invasion may lead to the identification of patients at high risk for disease progression and to the discovery of novel therapeutic targets and strategies. During tumor progression cancer cells modify the surrounding normal tissue to create a microenvironment supporting tumor growth and progression [2]. The tumor microenvironment consists of a variety of tumor-recruited or locally activated cells, such as bone marrow-derived cells, immune cells and carcinoma-associated fibroblasts (CAF) [3]. Dynamic reciprocal interactions between tumor cells and their microenvironment contribute to tumor cell survival, proliferation, invasion and metastasis and are therefore increasingly considered as potential targets for novel therapeutic strategies [4].
Despite the fact that CAF are among the most abundant and crucial cells of the tumor microenvironment, these cells remains poorly defined due to their inherent plasticity, heterogeneity and different origins, as well as the lack of single, universal specific markers and their pleiotropic functions [5][6][7][8]. Nevertheless, several distinct morphological properties and functions of CAF have been described and characterized. CAF are contractile cells commonly characterized by the expression of multiple activation markers, including α-smooth muscle actin (SMA), vimentin (VIM) or fibroblast-activated protein (FAP) [9,10]. The presence of CAF is associated with aggressive progression and poor prognosis in several cancer types [11][12][13]. Experimental and clinical evidences indicate that CAF regulate a number of tumor-promoting functions, including angiogenesis, invasion and metastasis [9,14]. The higher CAF density is generally found at the invasive front of the tumor [10], and several studies have shown that CAF promote CRC progression though multiple mechanisms [10,11,15]. Firstly, CAF produce soluble factors, such as hepatocyte growth factor/scatter factor (HGF) and platelet-derived growth factor (PDGF), promoting survival, proliferation and motility of cancer cells and stimulating the recruitment and activation of other cells in the tumor microenvironment, in particular inflammatory cells [16,17]. Secondly, CAF induce angiogenesis and enhance vascular permeability through the secretion of vascular endothelial growth factor (VEGF) [18]. Thirdly, CAF promote cancer cells invasion by modifying extracellular matrix (ECM) through the deposition of matrix proteins, including tenascins [11,19], and the secretion of matrix metallo-proteases (MMPs) resulting in altered tensile forces, gradients and release of bioactive fragments promoting cancer cell motility [11,20]. In spite of the fact that CAF can be in contact with cancer cells during stroma invasion [19] only little is known on the putative role of direct CAF-cancer cell contact in promoting cancer cell motility and invasion.
Here we have addressed this question by characterizing the interaction between fibroblasts and CRC-derived cell lines SW620 and HT29 in vitro using 2D and 3D co-culture models. Presented results demonstrate that fibroblasts promote CRC cell migration and invasion through direct cell-cell contact involving fibroblast cell surface associated FGF-2 and FGF receptors (FGFR) -integrin α v β 5 -SRC dependent signaling in cancer cells.

Fibroblasts promote SW620 and HT29 CRC cell elongation and motility
To monitor the effect of fibroblasts on CRC cell we cultured the CRC-derived cell lines SW620 and HT29 in the absence or presence of skin-derived fibroblasts. When cultured alone, SW620 and HT29 have a rather rounded morphology, while after 48 hours culture in the presence of fibroblasts they acquire an elongated morphology (Fig. 1A). Time lapse imaging revealed that only cancer cells establishing contacts with fibroblasts develop pseudopodia at the attachment site and progressively acquire an elongated morphology over time (about 70% of SW620 and 50% of HT29 compared to less than 10% in the absence of fibroblasts) ( Fig. 1B and 1C). Concomitant to elongation, cancer cells cultured with fibroblasts massively increased their motility, as monitored by tracking the distance travelled by individual cells (Fig. 1D).
These results demonstrate that fibroblasts induce colon cancer cell elongation and motility.

Cultured dermal, normal colon or colon cancer fibroblasts have equivalent gene expression and activation profiles and induce comparable cancer cell elongation and motility
Next we tested whether fibroblasts isolated from normal colon (CFB) or colon cancer (CAF) tissues were also able to induce cancer cell elongation and motility. Indeed, CFB and CAF induced SW620 and HT29 elongation and motility to extents comparable to those exerted by dermal fibroblasts ( Fig. 2A-2C). The fact that dermal fibroblasts and CFB were able to induce these effects on CRC cells was unexpected, as previous studies demonstrated that only freshly isolated CAF, but not normal fibroblasts, induced cancer progression in vivo [21,22].
To explain these similar properties, we hypothesized that fibroblasts cultured and expanded in vitro might acquire common functional capabilities regardless of their in vivo origin. To substantiate this hypothesis we performed gene expression profiling analyses on CFB, CAF and dermal fibroblasts (FB). Self-organizing heatmaps of the top 100 differentially expressed genes revealed that all fibroblasts display a very similar expression profile (Fig. 2D). As comparison, umbilical cord endothelial cells (HUVEC) have a clearly different gene expression profile. Moreover Principal Component Analysis (PCA) confirmed that all tree fibroblasts populations cluster together and clearly segregate from HUVEC (Fig. 2E). In addition volcano plot analysis confirms the results (data not shown).
To further strengthen these observations we monitored transcripts profiles for fibroblasts activation markers typically observed in CAF [10,15]: α-SMA, FAP, stroma-derived factor (SDF)-1, interleukin-6 (IL-6), VIM and fibroblasts specific protein (FSP)-1. Transcripts for all these markers were similarly expressed across all fibroblasts populations, thereby indicating equivalent activation states (Fig. 2F). FSP-1 and VIM were also expressed in cancer cells, consistent with previous reports [23,24]. www.impactjournals.com/oncotarget To collect further evidence supporting the notion that ex vivo culture alters gene expression profile in fibroblasts, we performed gene expression profiling analyses on CAF and CFB and compared them to expression profiles of laser-capture micro-dissected CRC stroma and normal colon stroma. PCA demonstrate that laser micro-dissected normal stroma and reactive stroma have different expression profile, while cultured CAF and CAB have similar expression profiles (Fig. 3A). Normal colonic epithelial cells and cancer segregate separately. Self-organizing maps of genomics profiles further confirmed that expression profiles of cultured CAF and CFB were indistinguishable (Fig. 3B), while genomics profiles of laser-capture micro-dissected tumor stroma Taken together these results indicate that cultured dermal fibroblasts, CFB and CAF are in a similar state of activation and are equally effective in inducing colon cancer cells elongation and motility in our model. Since CFB and CAF have reduced proliferative capacity in vitro compared to dermal fibroblasts (data not shown), we performed most of the experiments using dermal fibroblasts whereas CFB and CAF were used in selected validation experiments.

Cancer cell elongation and migration require direct contact with living fibroblasts
As elongated cancer cells were closely associated with fibroblasts we asked the question whether direct contact was required or whether soluble factors or extracellular matrix were sufficient in mediating such effects. To this purpose SW620 and HT29 were exposed to conditioned medium from fibroblast cultures or fibroblast-cancer cell co-cultures. Both conditions were ineffective in inducing cancer cell elongation or motility ( Fig. 4A and Supplementary Fig. 1A). In addition, cancer cells cultured on an ECM deposited by fibroblasts, as demonstrated by the detection of fibronectin (Supplementary Fig. 2A), or on fixed fibroblasts, did not acquire elongation or motility ( Fig. 4B and Supplementary Fig. 1B). Elongated cancer cells isolated from an established co-culture and re-plated in the absence of fibroblasts reverted to a round morphology ( Fig. 4C and Supplementary Fig. 1C). The purity of cancer cell fraction isolated from co-culture with fibroblasts was over 95% (Supplementary Fig. 2B and 2C). Importantly cancer cell co-cultured with HUVECs did not result in any effect on elongation or migration ( Fig. 4D and Supplementary Fig. 1D). Exposure of SW620 or HT29 to exogenous FGF-2, EGF and TGFβ, three cytokines known to induce cancer cell motility, failed to induce elongation and migration (Supplementary Fig. 3A and 3B).
From these experiments we concluded that fibroblast-induced SW620 and HT29 cell elongation and motility require continuous contact with living fibroblasts.

Fibroblasts promote contact-dependent cancer cell invasion
To test whether fibroblasts can also promote cancer cell invasion, we used 2D and 3D spheroid assays. For the 2D assay, SW620 spheroids were cultured on a fibroblasts monolayer for 4 days. Fibroblasts induced SW620 cell scattering and migration out of the spheroids (Fig. 5A-5B). SW620 migrating along fibroblasts acquired an elongated morphology (Fig. 5C). Culturing tumor spheroids alone, with co-culture conditioned medium, on a gelatin substrate or on a HUVEC monolayer did not induce cancer cell invasion ( Fig. 5D and Supplementary Fig. 3C), thereby confirming that the effect was fibroblast specific and contact dependent. We then tested for fibroblasts-induced cancer cell invasion in 3D conditions by embedding SW620 cells spheroids in fibroblasts-supplemented Matrigel. Fibroblasts migrated toward the spheroids, established contacts with cancer cells and induced SW620 invasion into the surrounding matrix. SW620 spheroids embedded in Matrigel without fibroblasts did not develop invasive properties (Fig. 5E).
From these experiments we concluded that fibroblasts induce contact-dependent cancer cell invasion.

Figure 4: Fibroblast-induced elongation and motility of cancer cells require direct contact with living fibroblasts.
A. Representative images of SW620-GFP cultured in the presence or absence of fibroblasts, fibroblasts and co-culture (cc) conditioned media. Bar graphs represent quantification of SW620 elongation and motility for 48 h. B. Representative images of SW620-GFP cultured on ECM deposited by fibroblasts and on a fixed fibroblast layer. Bar graphs represent quantification of SW620 cell elongation and motility for 48 h. C. Representative image and quantification of elongation of SW620-GFP returned to culture after separation from established co-culture. D. Representative images and quantification of elongation and motility of SW620-GFP cultured with HUVECs and fibroblasts for 48 h. All data are represented as mean +/− SD.

Cancer cell elongation, migration and invasion require FGFR signaling
In order to define the molecular mechanism involved in contact-dependent fibroblast-induced cancer cell elongation, migration and invasion, we applied pharmacological inhibitors of several known invasionpromoting kinases (i.e. PI3K/AKT, c-MET, MAPK/ERK, EGFR or FGFR) to co-cultures. Of all the tested inhibitors, only the FGFR inhibitors PD-161570 and PD-173074 prevented fibroblasts-induced cancer cell elongation  Fig. 4A and 4B). PD-161570 and PD-173074 also inhibited cancer cell migration ( Fig. 6A and Supplementary Fig. 4C) and invasion in the 2D (Fig. 6B) and 3D (Fig. 6C) assays.
Exogenous soluble FGF-2 did not induce cancer cell elongation ( Supplementary Fig. 3A and 3B) and FGF-2 protein was readily detected by Western blotting in fibroblast but not in cancer cell lysates (Fig. 6D). mRNA level of FGF-2 was also much higher in fibroblasts compared to cancer cells (Fig. 6E), however FGF-2 protein levels were identical in culture supernatants of fibroblasts, cancer cells and co-cultures (Fig. 6F). Co-culture with  Fig. 4D).

Cancer cells adhesion on fibroblast requires fibroblast surface-associated FGF-2, FGFR signaling and α V β 5 integrin ligation
These observations suggested that immobilized, rather than soluble FGF-2 might be responsible for these effects. To demonstrate that FGF-2 was localized at the cell surface, cell surface proteins were labeled with biotin, isolated by avidin precipitation and analyzed by Western blotting for FGF-2. Indeed FGF-2, but not GAPDH (a typical cytoplasmic protein) was detected in the biotinylated membrane protein fraction of fibroblasts cultures (Fig. 7A). These results demonstrate the presence of FGF-2 at the cell surface. Consistent with this observation, SW620 and HT29 adhesion on fibroblasts was strongly reduced in the presence of an anti-FGF-2 antibody and small molecular FGFR kinase inhibitors ( Fig. 7B and Supplementary  Fig. 5A). Furthermore, an α V β 3 /α V β 5 integrin blocking cyclic RGD peptide, EMD-121974, and an anti-α V β 5 blocking antibody, but not anti-α V β 6 or anti-β 1 blocking antibodies, strongly reduced SW620 and HT29 adhesion on fibroblasts ( Fig. 7C and Supplementary Fig. 5B). Expression of α V β 5 and α V β 6 , but not α V β 3 on cancer cells, was demonstrated by flow cytometry analysis and was not altered by co-culture with cancer cells (Supplementary Fig. 5C). From these experiments we concluded that SW620 and HT29 CRC cell adhesion to fibroblasts depends on fibroblasts cell surface associated FGF-2, FGFR signaling and α V β 5 integrin ligation.

Fibroblasts-induced cancer cells adhesion, elongation, migration and invasion require SRC activation in cancer cells
Next we looked for an intracellular signaling pathway mediating FGF-2/FGFR-induced cancer cell elongation, migration and invasion. As inhibition of PI3K/AKT, MAPK/ERK, ROCK and FAK did not prevent these effects ( Supplementary Fig. 4A), we focused on SRC, a kinase downstream of FGFR and integrins known to modulate cell adhesion, migration and invasion [25,26]. Basal levels of membraneassociated phospho-SRC were observed in SW620 and HT29 cultured alone and more frequently in elongated cancer cells, in particular at sites of contact with fibroblasts. Nuclear phospho-SRC was observed in cancer cells in contact with fibroblasts and in fibroblasts themselves ( Fig. 8A and Supplementary  Fig. 6A). Inhibition of FGFR kinase activity fully or partially inhibited SRC phosphorylation ( Supplementary  Fig. 6B). Inhibition of SRC activity with PP-2 or CGP-77675 ( Supplementary Fig. 6C) effectively inhibited SW620 and HT29 adhesion on fibroblasts ( Fig. 8B and Supplementary Fig. 6D), fibroblasts-induced SW620 and HT29 elongation and motility during co-cultures (Fig. 8C-8E and Supplementary Fig. 6E and 6F). The related inactive compound PP-3 was ineffective to induce any effect. SRC inhibition also prevented scattering and invasion of cancer cells in the 2D and 3D spheroid models ( Fig. 8F and Fig. 9). Importantly, under these experimental conditions, SRC and FGFR inhibitors were not cytotoxic ( Supplementary  Fig. 6G).
The use of pharmacological inhibitors in these experiments has two main limitations: inhibitors are not fully specific and may affect other SRC family members (i.e. Yes, Fyn, Fgr, Lck or Hck) and also target SRC in fibroblasts. To circumvent these limitations we silenced SRC through lentiviral-mediated expression of SRC-specific shRNA in cancer cells or in fibroblasts ( Supplementary Fig. 7A). SW620 and HT29 cells with silenced SRC failed to elongate and migrate in the presence of fibroblasts ( Supplementary Fig. 7B-7D). In contrast, fibroblasts with silenced SRC expression were still able to induce cancer cell elongation ( Supplementary Fig. 7E and 7F).
From these experiments we concluded that fibroblast-induced cancer cell elongation, motility and invasion depend on SRC activation in cancer cells.

DISCUSSION
CAF promote cancer invasion and metastasis through the combined promotion of cancer cell growth, survival and motility, induction of angiogenesis and ECM modification [10, 15 , 19]. These effects are largely considered mediated by released factors (i.e. cytokines, chemokines, angiogenic factors and matrix proteases). Considering the close physical proximity of cancer cells and CAF at CRC invading fronts [9], it appears reasonable that direct CAF-cancer cell contact may also play a role. To date, however, the role of direct CAF contact in CRC cell invasion and the putative involved mechanisms remain largely elusive. Here we addressed this question by using in vitro 2D and 3D co-cultures systems SW620 and HT29 CRC cells and cultured dermal, normal colon and colon cancer fibroblasts. We found that fibroblasts induce cancer cell motility and invasion in a cell-contact dependent manner and we provide a mechanism: Fibroblasts surface associated FGF-2 activates FGFR on cancer cells, which in turn activates SRC. Activated SRC induces integrin α v β 5dependent tumor cell adhesion to fibroblasts and tumor cell motility (Fig. 10). In 2D models, cancer cells migrate along fibroblasts while in the 3D models fibroblasts "pull" cancer cells to initiate and guide migration into the gel.
The relevance of our results relies on the following implications. Firstly, they demonstrate the functional importance of direct fibroblast-cancer cell contact. In vivo this may be especially important under conditions of migration along precise paths, such as matrix fibers [27], or when cancer cells have not yet acquired full invasive capacities, for instance during invasion in the absence of EMT [28]. To corroborate this hypothesis additional experiments using several cancer cell lines with different degree of malignancy and in vivo experiments are necessary. Secondly, they provide a new role for fibroblasts cell surface-associated FGF-2. The role of FGF-2 in cancer progression is well recognized [29,30]. While it is well known that FGF can associate to the cell surface through its high affinity for glycosaminoglycans [31], the biological significance of cell surface FGF has not been well characterized. Rather, its biological activities have been attributed to the soluble and matrix-associated fraction [29]. Our observations suggest that FGF-2 may contribute to distinct phases of cell invasion: cell surface associated FGF-2 may initiate migration of poorly invasive cells requiring additional cell-derived stimuli, while matrix-associated FGF-2 might enhance invasion of already invasive cells. The exact process of FGF-2 presentation at the cell surface leading to these effects remains unknown at this point and its elucidation will require additional work. Thirdly, we identify FGFR, SRC and α v β 5 as the molecules mediating cancer cell adhesion on fibroblasts, migration and invasion. To our surprise, α v β 6 , an integrin up-regulated in CRC (and present on CRC cells) and contributing to CRC invasiveness through the activation of TGF-β and production of MMPs [32], was not involved in this effect. How FGFR, SRC and α v β 5 interact in this context, in particular whether activated FGFR associate with α v β 5 integrin, remains still unknown. As β 5 integrin and FGFR1 are reported to promote CRC progression [33,34], we performed co-immunoprecipitation experiments to test whether these receptors may associate upon FGF-2 presentation. However, we could not demonstrate co-precipitation of β 5 integrin and FGFR1 under mild detergent conditions and complex stabilization by surface protein-protein crosslinking ( Supplementary Fig. 8). Although preliminary, these results suggest that either FGFR1 is not the receptor of interest for this effect, among the FGFR family members, or that indeed FGFR1 and α v β 5 do not associate. Additional experiments are necessary to clarify this point. The exact role of SRC signaling and nuclear translocation remains also open. Preliminary observations indicate that SRC is not regulating α v β 5 activity through direct phosphorylation of the β 5 cytoplasmic domain (data not shown). A further important open question concerns the α v β 5 ligand on the fibroblast surface. As FGF-2 was reported to bind to α v β 3 [35], we tested whether SW620 and HT29 cells adhered to immobilized FGF in a α v β 5 -dependent manner. While cancer cells adhered to immobilized FGF-2 this adhesion was not integrin dependent (data not shown). We also tested the potential role of heparin, which is a necessary bindingpartner for FGF-2 in order to stabilize receptor binding and dimerization. However, results suggest that heparin is not playing a detectable role in promoting/perturbing cancer cell adhesion to fibroblasts (data not shown). A role of fibroblast surface fibronectin, as a possible α v β 5 ligand for cancer cell adhesion, was also excluded   using anti-fibronectin-adhesion blocking antibodies (data not shown). Further, N-cadherin, which was reported to promote cancer cell invasion and metastasis in association with FGFR1 [36] and fibroblasts, was also excluded as not expressed in these cancer cells (data not shown). We are now considering additional candidate ligands. Fourthly, as inhibitors of FGFs, FGFRs, SRC and α v integrins are in preclinical development or already available for clinical testing [37][38][39][40], these results provide a rationale for in vivo preclinical testing and focusing on their clinical assessment in invasive CRC. To assess the clinical relevance of these in vitro observations, we performed preliminary immunohistochemical stainings for FGF-2 and FGFR1 on human colorectal cancer samples. Representative results, shown in Supplementary  Fig. 9, confirm the presence of FGF-2 in both cancer cells and stromal cells. FGFR1 staining is present on more differentiate areas of the cancer, but lost in more transformed, and invasive regions. This is consistent with concurrent observations suggesting that FGFR1 is not the major receptor involved in this process. Membrane localization on fibroblasts cannot be confirmed in these stainings. Fifthly, they raise a fundamental question on the role of fibroblasts in cancer cell invasion: which cues do cancer cells gain from adhering on fibroblasts to initiate elongation and migration, that they cannot acquire through fibroblasts-released factors and deposited matrix? Initial experiments addressing this question indicate that cell contact induces important changes in gene expression in both fibroblasts and cancer cells.
A corollary of this study is the observation that fibroblasts cultured and expanded in vitro have similar gene expression profiles and activation markers independently on the in vivo origin, including for those originally from the tumor stroma. These observations, in combination with the results from laser micro-dissected stroma, suggest that comparative studies of CAF vs. normal fibroblasts are best performed using freshly isolated, non-cultured, cells.
In conclusion, we show that fibroblasts induce cellcontact-dependent CRC cell migration and invasion under 2D and 3D conditions in vitro through surface associated FGF-2, FGFR-mediated activation of the SRC in cancer cells and α v β 5 -mediated adhesion. Further characterization of this effect may reveal novel aspects on the cancer invasion-promoting role of fibroblasts.

Cell culture
The human colorectal carcinoma cell lines SW620 and HT29, and HEK-293T cells were purchased from ATCC (LGC Standards). CAF, dermal and colon fibroblasts and Human Umbilical Vein Endothelial Cells (HUVEC) were obtained from human samples. SW620 were cultured in RPMI GlutaMAX™, HT29, all fibroblasts and HEK-293T in DMEM GlutaMAX™, all supplemented with 10% FCS, 100 U/ml Penicillin and 100 μg/ml Streptomycin. HUVECs were cultured as described previously [41]. All cell culture reagents were purchased from Life Technologies.
For co-culture experiments, fibroblasts and colon cancer cells were grown for 48 hours at a 1:1 ratio in DMEM GlutaMAX™. Conditioned media were harvested after 72 hours of culture and filter 0.45 μM before use. Coatings were performed over-night using porcine Gelatin 0.5% (Sigma), Poly-L-Lysine 0.1% (Life Technologies), human Fibronectin 0.1% (Sigma) and 1% Rat Tail Collagen Type I (BD Bioscience) in PBS. TGF-β (Sigma) was used at 20 ng/ml, EGF and FGF-2 (Cell Signaling) at 100 pg/ml. Cell counting and viability determination was performed by trypan blue exclusion using Neubauer Counting Chamber.
Inhibitors were used at the indicated concentrations and blocking antibodies at 10 μg/ml, all added to the serum free medium and incubated with cancer cells and fibroblasts for 48 hours.

Co-culture separation
Cancer cells were separated from fibroblasts using MACS ® separation technique, following manufacturer's instructions. MS columns and anti-human Fibroblast micro-beads were used. All reagents were from Miltenyi Biotec.

Isolation of fibroblasts and HUVEC
Dermal fibroblasts were isolated from human neonatal foreskin. Excised skin samples were was placed into Ca 2+ , Mg 2+ -free Hank's balanced salt solution (Life Technologies) with 100 U/ml Penicillin-Streptomycin (Life Technologies) and 5 μg/ml Gentamicin (Sigma). After removal of subcutaneous fat, skin samples were incubated 2 hours at 37°C with 2-4 ml of collagenase type I 1 mg/ml (Roche). Cell suspension was harvested on ice, quenched by adding DMEM and filtered 70 μM. 2'000'000 cells per dish were plated with DMEM. The medium was replaced twice a week and cells maintained up to 80% confluence. Dermal fibroblasts were used up to 15 passages.
Colon fibroblasts and CAF were isolated according to the protocol by Orimo et al. [21]. Tissue sections, obtained following colectomy for colon cancer, of about 1 cm 3 were collected in 15 ml DMEM, transported on ice and washed about 5 times with DMEM, before being cut into small pieces, washed again and incubated for 12 h at 37°C with agitation in 2-4 ml DMEM containing collagenase type I 1 mg/ml and 120 units hyaluronidase (Sigma). The dissociated tissues were incubated without shaking for 5 minutes at room temperature, followed by the separation of stromal cell-enriched supernatant to a new tube. The stromal fraction was centrifuged and cells were cultured on tissue culture plates in DMEM for 15-20 cell passages.
HUVEC were isolated as previously described [42] from fresh umbilical cords dissociation.
Protocols for collection and use of human samples were approved by the Ethic committees of Cantons Vaud, Berne and Ticino, Switzerland.

Spheroids assay
Cancer cells spheroids were prepared as previously described [43], using the hanging drop technic in a Terazaki plate with 500 cells per well in 20 μl medium for 72 h. 3D spheroids assay were performed according to the 3D-On-Top method [44]. 7 mg/ml Matrigel growth factors reduced (Corning) was used as matrix, mixed or not with 75'000 fibroblasts/ml. For 2D spheroids assay, cancer cell spheroids were placed on top of a confluent fibroblasts layer or on coated wells.

Vectors and infections
LifeAct lentiviral vectors in GFP and mCherry were kindly provided by Dr. Olivier Pertz. pSD44 lentivector expressing GFP and DsRED were pRRLSIN.cPPT. PGK/GFP.WPRE lentiviral vector under the control of the phospho glycerate kinase promoter. SRC pLKO.1 shRNA were from Thermo Fisher Scientific Inc. and used following manufacturer's instructions.
Lentiviral particles were generated in HEK-293T cells by transducing the vector of interest with pMD2G (pSD11) and pMDLgpRRE (pSD16) plasmids using calcium phosphate transfection [45], followed, depending of the vector used, by an antibiotic-based selection of the infected cells.
Cell Surface Protein Isolation Kit (Pierce) was used for isolation of membrane proteins fraction following manufacturer's instructions.

Co-Immunoprecipitation
Cells were lysed using RIPA non-denaturant lysis buffer (Cell Signaling) supplemented with protease Cocktail inhibitor, 2 mM PMSF, 2 mM BGE and 0.2 mM Orthovanadate on ice. After sonication and centrifugation the supernatant was incubated for 12 hours at 4°C under agitation with primary antibody at 1/50 dilution. The mix was then incubate for 30 minutes at 4°C with protein A magnetic beads (Cell Signaling), followed by beads isolation using magnetic separation rack (Cell Signaling). Samples was heated for 5 minutes at 95°C, mixed with SDS buffer and processed by SDS-PAGE and Western blotting.
For protein cross-linking experiments, cells were treated for 30 minutes at room temperature with 2 mM DSP (Thermo Fisher Scientific Inc.) followed by 15 minutes quenching using 1 M Tris pH 7.5. Cells were lysed and processed as above.

ELISA
For FGF-2 quantification, ELISA assays (Biolegend) was performed following manufacturer's instructions. For fibronectin detection, fibroblasts were removed from cell culture plates as previously described, wells were blocked using 5% BSA and incubated with anti-fibronectin antibody (Sigma), followed by a goat-anti-mouse-HRP antibody and a chromogenic substrate (Biolegend). Absorption was measured at 630 nm and normalized comparing fibronectin coated-wells.

Flow cytometry
Annexin V/PI staining was performed following manufacturer's instructions (Life Technologies). Integrin staining were performed using directly labeled antiintegrin primary antibodies 1:100 dilution 30 minutes on ice or unlabeled primary antibodies followed by 30 minutes secondary antibody anti-rabbit or mouse-PE (BD Bioscience) incubation on ice (1:100 dilution). PBS with 3% FCS and 5 mM EDTA was used as buffer. BD FACSCalibur (Becton Dickinson) instrument was used to perform experiments and FlowJo 9.7.4 (Treestar Inc.) software was used to analyze all data.

Adhesion assay
8'000 fibroblasts were cultured in a 96-wells plate to reach confluence. 15'000 GFP-cancer cells were added on fibroblasts layer and incubated in serum free medium for 3 hours. Non-adherent cells were removed using ice cold PBS and adherent cells were observed and quantified using fluorescence microscopy.

Immunostaining
Cells were grown on a glass chamber-slide previously coated with Poly-L-lysine followed by 10 minutes fixation with 4% PFA, 5 minutes permeabilisation with 0.1% Triton X-100 (Sigma) in PBS and blocking for 30 minutes with PBS containing 0.05% BSA, 5% Donkey serum (Fitzgerald) and 0.1% Triton X-100. Primary antibody was incubated over-night at 4°C at 1/100 dilution. Secondary and third antibodies were incubated for 1 hour at room temperature in blocking at 1/100 dilution. DAPI staining was included in the mounting medium.

RT-PCR
mRNA was extracted using RNeasy Mini kit (Qiagen) and total RNA was retro-transcribed using Superscript following manufacturer's instructions (Promega AG). cDNA was subjected to PCR amplifications using the following primer pairs (Eurofins MWG Operon) at the indicated hybridization temperatures. Real-time qPCR was performed using the StepOne SYBR System (Life Technologies).

Gene expression analysis by microarray hybridization
Laser capture micro-dissections were performed as described in the paper of Farmer et al. [47]. RNA extraction, micro-array and data analysis were performed as described in the paper of Christensen et al. [48]. The complete data set is publicly available at GEO (http:// www.ncbi.nlm.nih.gov/geo/) through the accession number GSE30292 and GSE23583.
Robust multi-array averaging (RMA) and quantile normalization were used to quantify gene expression, for non-supervised hierarchical clustering Pearson's correlation and average linkage clustering, respectively, were used for similarity measurement and clustering using www.impactjournals.com/oncotarget R (R 2.12.0, Bioconductor 2.7). Principle component analysis was performed using R 3.1.

Gene expression analysis by RNASeq
All fibroblasts and HUVEC cells were grown at the same time in triplicate from independent patients at passage 6. Total RNA was extracted using the total RNA extraction Nucleospin II kit (Machery-Nagel). The quality and quantity of all RNA samples were examined by Agilent 2100 Bioanalyzer (Agilent Biotechnologies) and NanoDrop (Witec AG).
RNA-seq libraries were prepared using 15 ng of total RNA and the Ovation Nugen RNA-Seq System v2 kit (Nugen). 100 ng of the ds-cDNAs obtained were fragmented to 350 pb using Covaris S2 (Covaris). The Illumina TruSeq nano reagents (Illumina) were used to generate the final libraries. Cluster generation was performed with the resulting libraries using the Illumina HiSeq PE Cluster Kit v4 reagents and sequenced on the Illumina HiSeq 2500 using HiSeq SBS Kit v4 reagents. Sequencing data were processed using the Illumina Pipeline Software version 1.82.

Imaging and data analysis
For co-culture experiments images were taken every hour at different magnifications with an inverted fluorescence microscope (Leica AF6000). Cancer cell elongation was quantified manually by scoring individual cells using Image J "Cell Counter" plugin on taken images at 48 h co-culture. Elongated cell were defined as single cell with a minimal length-width rapport of 2:1, as well as at least one sharp-ending extremity. Cancer cells motility was quantified based on single cell track follow on 24 h microscope movies using Image J "Manual Tracking" and "Chemotaxis Tool" plugins.
Imaging of 2D and 3D spheroids assay were performed with 5× and 40× objectives with Leica AF6000 inverted fluorescence microscope. Invasion was quantified using Photoshop CS6 (Adobe) by calculating the invaded area on the 4 days image, normalized on initial spheroid size.
Western Blot images were quantified for protein level using Image J "Gel Analyze Tool" plugin.
For immunostaining images were taken using a DeltaVision Elite Microscope (GE Healthcare) with 63× and 100× objectives, followed by deconvolution treatment.
For immunohistochemistry images, slides were scanned using a NanoZoomer 2.0 HT (Hamamatsu) and images were extracted using NDP.view2 analysis software (Hamamatsu).

Statistical analysis
Each experiment was repeated independently a minimum of 3 times in triplicate conditions. Acquired data were analyzed using Prism Software (GraphPad). Statistical comparisons were performed by un-paired two-tailed Student's t test or by two-way ANOVA with Bonferroni post-test. Results were considered to be significantly from p < 0.05. In the figures the various p values thresholds are presented as follow: ≤0.05 = *, ≤0.01 = **, ≤0.001 = ***, ≤0.0001 = ****.