Register for Mechanobiology 2016!


Invasion of stromal host cells, such as myofibroblasts, into the epithelial cancer compartment may precede epithelial cancer invasion into the stroma. We investigated how colon cancer-derived myofibroblasts invade extracellular matrices in vitro in the presence of colon cancer cells. Myofibroblast spheroids invade collagen type I in a stellate pattern to form a dendritic network of extensions upon co-culture with HCT-8/E11 colon cancer cells. Single myofibroblasts also invade Matrigel™ when stimulated by HCT-8/E11 colon cancer cells. The confrontation of cancer cells with extracellular matrices and myofibroblasts, showed that cancer-cell-derived transforming growth factor-β (TGF-β) is required and sufficient for invasion of myofibroblasts. In myofibroblasts, N-cadherin expressed at the tips of filopodia is upregulated by TGF-β. Functional N-cadherin activity is implicated in TGF-β stimulated invasion as evidenced by the neutralizing anti-N-cadherin monoclonal antibody (GC-4 mAb), and specific N-cadherin knock-down by short interference RNA (siRNA). TGF-β1 stimulates Jun N-terminal kinase (also known as stress-activated protein kinase) (JNK) activity in myofibroblasts. Pharmacological inhibition of JNK alleviates TGF-β stimulated invasion, N-cadherin expression and wound healing migration. Neutralization of N-cadherin activity by the GC-4 or by a 10-mer N-cadherin peptide or by siRNA reduces directional migration, filopodia formation, polarization and Golgi-complex reorientation during wound healing. Taken together, our study identifies a new mechanism in which cancer cells contribute to the coordination of invasion of stromal myofibroblasts.


Tumor invasion and metastasis is governed by a reciprocal cross-signaling between cancer cells and host cells at the primary tumor site and distant organs. Progression of solid tumors is associated with a remarkable disorganisation of the stromal compartments that normally maintain the mucosal and epithelial cell architecture and functions. In response to signals produced by cancer cells, the tumor stroma is invaded by several cell types. The best characterized mechanism is the hypoxia-dependent angiogenic switch in which host-cell-derived endothelial cells invade into the tumor stroma to form new blood vessels (Carmeliet and Jain, 2000). Other host cells that invade the tumor are fibroblast-like cells (Roni et al., 2003; Yang et al., 2001). Solid tumor implantation in transgenic mice that express GFP under the control of the VEGF promoter leads to induction of VEGF-promoter activity in myofibroblast-like cells. Subsequently, GFP-positive myofibroblast-like cells invade the tumor and can be seen throughout the tumor mass (Fukumura et al., 1998). During avascular growth of developing hepatic metastases, myofibroblast-like cells are already present, before endothelial cell recruitment (Olaso et al., 2003). Thus, myofibroblasts might provide pro-invasive information at the invasion front of cancer cells (De Wever et al., 2004). Does invasion of myofibroblasts into the epithelial compartment precede the invasion of cancer cells into the host compartment? Barriers that myofibroblasts have to traverse are collagen type I-enriched extracellular matrix (ECM) and the basement membrane. Molecular cross-signaling between cancer cells and myofibroblasts might stimulate migration of both cell types towards each other and modify the adjacent ECM and basement membrane. The result is breakdown of normal tissue boundaries (De Wever and Mareel, 2003).

Transforming growth factor (TGF)-β1 is implicated in the signaling from cancer cells to myofibroblasts. TGF-β is a chemotactic agent for fibroblasts (Postlethwaite et al., 1987) and is secreted by colon-cancer cells as a latent complex stored in the ECM. Myofibroblasts release bioactive TGF-β from the latent complex through proteolytic and non-proteolytic mechanisms (Derynck et al., 2001). Bioactive TGF-β1 activates an heterodimeric cell surface complex of two transmembrane receptor serine/threonine kinases, leading to Smad and non-Smad signal transduction (Derynck and Zhang, 2003). Non-Smad signal transduction comprises mitogen activated protein kinase (MAPK) pathways. Functional TGF-β receptors are required for TGF-β-mediated activation of MAPKs (Mulder, 2000). MAPKs are serine/threonine kinases that are activated in response to a wide variety of extracellular stimuli, including TGF-β1. Three distinct groups of MAPKs have been identified in mammalian cells: extracellular signal-regulated kinases (ERKs), Jun N-terminal kinases (JNKs) and p38 MAPK. Activated MAPKs target transcription factors and Smads. The serine/threonine kinase JNK promotes both Smad signal transduction and phosphorylation of Jun. JNK activity is necessary for TGF-β-induced fibronectin expression in HT1080 fibrosarcoma cells (Hocevar et al., 1999) and the phosphorylation of paxillin, an essential signal for migration (Huang et al., 2003).

N-cadherin is a transmembrane glycoprotein composed of extracellular domains that mediate homophilic interactions between neighboring cells, predominantly via a peptide domain containing the His-Ala-Val (HAV) amino acid sequence, which is located near the N-terminus. The cytoplasmic domain of N-cadherin is anchored to the intracellular actin cytoskeleton by interacting with the α-, β- and γ-catenin complexes. N-cadherin has been considered a path-finding molecule involved in invasion, migration and neurite outgrowth (Doherty and Walsh, 1996; Williams et al., 2000; Hazan et al., 2000). Homophilic interaction of N-cadherin induces its rapid and strong anchoring to actin filaments (Lambert et al., 2002). Furthermore, the establishment of cadherin-mediated contacts has the potential to translate adhesive homophilic recognition into changes in actin-mediated cell shape and surface adhesion, and, therefore, cellular motility and invasion (Ehrlich et al., 2002).

The paucity of information on the regulatory circuits that control the interplay between cancer cells and stromal myofibroblasts can be attributed to the difficulty to mimick the tumor environment. In this study, we used combined in vitro approaches to unravel the signal coming from cancer cells affecting the invasive potential of stromal myofibroblasts into collagen and Matrigel. We further investigated how cancer-cell-derived TGF-β1 regulates the invasion of myofibroblasts. Our studies highlight the crucial role of JNK, and N-cadherin as signaling elements involved in TGF-β1- and wound-mediated migration and invasion.

Materials and Methods

Cell culture

The human colon cancer cell line HCT-8/E11 (Vermeulen et al., 1995) and the rat myofibroblast cell line DHD-Fib (Dimanche-Boitrel et al., 1994) were cultured as described. Primary myofibroblasts were derived from human colon cancer biopsies (Van Hoorde et al., 1999). Stromal cell lines were cultured in Dulbecco's modified Eagle's medium (DMEM) (Life Technologies, Ghent, Belgium) supplemented with 10% fetal calf serum (FCS), 100 U/ml penicillin and 100 μg/ml streptomycin (Life Technologies, Ghent, Belgium); they were scored immunocytochemically and considered as myofibroblasts when >95% were positive for vimentin, prolyl 4-hydroxylase and α-smooth muscle actin and <5% were positive for smooth muscle myosin and negative for cytokeratin; their identity was routinely controlled during the experiments. Myofibroblast cultures were used until passage 10.

Chemicals and antibodies

The 10-mer HAV-comprizing peptide [NH2-LRAHAVDING-amide; human (hu)N-CAD10] was used, which is homologous to the HAV region in the amino-terminal part of human and rat N-cadherin. A scrambled human N-cadherin 10-mer peptide (NH2-LHDANVGRIA-amide; huN-CAD10scr) served as a control. All decapeptides (Ansynth Service, Roosendaal, The Netherlands) were separated by HPLC to at least 95% purity; they were used at 200 μg/ml, the concentration at which huN-CAD10 or E-CAD10 decapeptides exhibited a maximal effect on invasion and aggregation in our previous experiments (Willems et al., 1995; Noë et al., 1999; Nawrocki-Raby et al., 2003).

The following primary antibodies were used: anti-Golgi (6F4C5, kindly provided by P. Courtoy, ULB, Brussels, Belgium) (Chicheportiche et al., 1984), anti-N-cadherin (recognizing the extracellular domain of N-cadherin, clone GC-4), isotype-matched non-immune IgG1 (MOPC-21) further indicated as control IgG, anti-α-catenin, anti-β-catenin, anti-vimentin, anti-cytokeratin, anti-smooth muscle myosin, anti-α-smooth muscle actin and anti-tubulin (Sigma, St Louis, MO); anti-human prolyl 4-hydroxylase (DAKO, Glostrup, Denmark); anti-phosphotyrosine, anti-ERK1/2, anti-phospho ERK1/2, (Santa Cruz Biotechnology, Santa Cruz, CA); anti-p38 MAPK and anti-phospho p38 MAPK (Biosource, Brussels, Belgium); anti-JNK (Upstate Biotechnology, Campro Scientific, The Netherlands); anti-Jun, anti-phospho Jun (ser63), and anti-phospho Jun (ser73) (Cell Signaling, Westburg, The Netherlands); anti-N-cadherin (recognizing the cytoplasmic domain of N-cadherin; Takara Shuzo, Shiga, Japan); anti-cadherin-11 (Zymed, Sanbio, The Netherlands); anti-TGF-β (R&D systems, Abingdon, UK). Secondary antibodies coupled to horseradish peroxidase and antibodies coupled to fluorescein or biotin were obtained from Amersham Pharmacia Biotechnology (Little Chalfont, UK). An ELISA to determine human TGF-β1 in biological fluids, bioactive recombinant (r)TGF-β1 and latent-rTGF-β1 were from R&D systems. Phalloidin-FITC was purchased from Sigma. SP600125 (used at 0.18-18 μM), a reversible ATP-competitive inhibitor of JNK with 300-fold greater selectivity for JNK than ERK and p38 MAPK (Bennett et al., 2001) was from Calbiochem (La Jolla, CA). The MEK1 inhibitor PD98059 (used at 25 μM) and the p38 MAPK inhibitor SB203580 (used at 5 μM), were from Alexis Corporation (San Diego, CA) and New England Biolabs (Beverly, MA), respectively. The following ECM proteins were used: collagen type I (Upstate Biotechnology, Lake Placid, NY) and Matrigel (Becton Dickinson and Company, Franklin Lakes, NJ).


Human colon myofibroblasts were plated on Petri dishes and, at 70-80% confluency, were trypsinized and collected in an Nucleofector™ certified cuvette (Amaxa GmBH, Cologne, Germany). A mixture of 100 μl Nucleofector solution and 3 μg of short interference RNA (siRNA) was added. The cells were electroporated in the Nucleofector electroporator with the P22 specific Nucleofector program. siRNAs targeting N-cadherin (GenBank/EMBL/DDBJ accession number NM_001792) were designed by Qiagen (Leusden, The Netherlands). Inhibition of N-cadherin expression was achieved by RNA interference using a 1:1 mixture of the following double-stranded oligoribonucleotides: siN-CAD2 5′-AGUGGCAAGUGGCAGUAAA-3′ and siN-CAD3 5′-GGAGUCAGCAGAAGUUGAA-3′ or siN-CAD3 5′-GGAGUCAGCAGAAGUUGAA-3′ and siN-CAD4 5′-CCGUGUCUGUUACAGUUAU-3′. To verify the specificity of the knockdown effect, we used an oligonucleotide sequence with no known mamalian target (con 5′-UUCUCCGAACGUGUCACGU-3′) as a control.

Preparation of colon-cancer-conditioned medium

About 70×106 HCT-8/E11 cells on 175-cm2 tissue flasks were washed three times with 10 ml of serum-free DMEM and incubated for 48 hours at 37°C with 15 ml serum-free DMEM. The medium was harvested, centrifuged at 1250 g for 5 minutes at 4°C and passed through a 0.22 μm filter. Conditioned medium (CM) was stored at -20°C, which did not alter its biological activity for at least 6 months. To prevent the possibility that depletion of nutrients in cancer cell CM is responsible for any observed effect on myofibroblast cultures, we concentrated the medium ten times by using centriprep tubes YM-10 (Amicon, Millipore, Bedford, MA) and diluted the resulting solution with serum-free DMEM until the original volume (14 ml) was reached.

Preparation of spheroid-myofibroblast aggregates

Six ml of a suspension containing 4×105 myofibroblasts/ml were incubated in a 50-ml Erlenmeyer flask on a Gyrotory shaker at 37°C for 2 days. The cultures were viewed under a macroscope equipped with a calibrated ocular grid and spheroid cells with a diameter between 100 and 300 μm were collected by removing the supernatant (Bracke et al., 2001).

Spheroid invasion into collagen type I

A collagen solution was prepared for three wells of a 6-well plate by mixing the following pre-cooled (at 4°C) components in a sterile 50-ml Erlenmeyer flask on melting ice: 2.1 ml of collagen type I (3.98 mg/ml) (Upstate Biotechnology) 0.8 ml of MEM (10×), 4.6 ml of calcium- and magnesium-free Hanks' balanced salt solution pH 7.4, 0.8 ml of NaHCO3 (0.25 M). We added 0.15 ml of NaOH (1 M) to make the solution alkaline. The solution was mixed gently by pipetting, avoiding the introduction of air bubbles. The final solution looks purple, the phenol red pH indicator shows a pH >9. Of this solution, 1.25 ml was poured in three wells to cover the bottom. This bottom layer prevents contact of spheroid cells with the plastic of the well. We let the collagen set on a flat surface at 37°C in air that was water-saturated and contained 10% CO2 for at least 2 hours. The spheroid cells were suspended in the remaining collagen type I solution, and 1.25 ml of this spheroid-cell-suspension was poured in the three wells. After collagen gelation, 1 ml of the following media (serum-free CMHCT-8/E11, serum-free DMEM with testing products in desired concentrations, serum-free DMEM containing 106 HCT-8/E11 cells) was carefully added. We refreshed the medium every 2 days, and scored invasion every day, up to day 7. Invasion was scored by two independent observers (O.D.W. and A.V.).

Matrigel invasion assay

Transwell chambers with polycarbonate membrane filters (6.5 mm diameter, 8 μm pore size) were overlaid with a Matrigel gel. The filter was placed in a 6-well plate which had as a bottom layer a collagen gel containing either 106 HCT-8/E11 colon cancer cells or rTGF-β1 (0.1-10 ng/ml) as a chemo-attractant. Collagen type I is used as a substrate to mix cells in a 3-D gel instead of seeding them on 2-D substrate; this was necessary, because we wanted to have more cells in the small space of the lower compartment. Notice, that there is no correlation with the physiologic situation in which cancer cells rest on the basement membrane and myofibroblasts reside in the collagen type I. To the upper compartment of the Transwell chamber, 4×105 myofibroblasts were added. After 48 hours, the myofibroblasts that had invaded to the membrane's undersurface were stained with 0.4 mg/ml 4′,6-diamino-2-phenylindole (DAPI) (Sigma) and were counted on 50 fields/filter (Hall and Brooks, 2001).

Wound healing migration assay

Human colon myofibroblasts were grown in 6-well tissue culture dishes until confluent. Cultures were incubated for 3 minutes with Ca2+- and Mg2+-free phosphate-buffered saline (PBS) pH 7.4. After removal of PBS, confluent monolayers were wounded with a sterilized razor blade to remove part of the cell monolayer sheet (André et al., 1999). Wounded monolayers were washed three times with serum-free DMEM to remove dead cells. Myofibroblast migration occurred in the presence of 1 ml DMEM with or without monoclonal antibody (mAb) or peptides or pharmacological inhibitors and was assessed after 24 hours by counting the number of intersections at different distances. Starting point of migration was revealed by a small incision in the culture dish that was made with the razor blade.

Polarization and Golgi-complex reorientation

Confluent DHD-Fib myofibroblasts grown on glass coverslips were wounded with a plastic pipet tip. After 8 hours of migration, cells were fixed in 4% paraformaldehyde diluted in PBS, blocked in PBS containing 50 mM NH4Cl, permeabilized in 0.2% Triton X-100 in PBS, and stained with a rat specific anti-Golgi mAb. The orientation of the Golgi complex was assessed as described previously (Nobes and Hall, 1999). The significance of the inhibition of the Golgi-complex reorientation was determined by the Student's t-test (P<0.05). Samples were viewed by fluorescence microscopy (Dialux 20) (Leitz, Wetzlar, Germany) and photographed with the Leitz Orthomat E camera system.

F-actin staining and numerical evaluation of filopodia

Confluent human colon myofibroblasts grown on glass coverslips were wounded with a plastic pipet tip. After 8 hours of migration, cells were fixed in 4% paraformaldehyde with 1% glutaraldehyde in PBS (to preserve fine actin structures such as filopodia), blocked in 50 mM NH4Cl in PBS, permeabilized in 0.2% Triton X-100 in PBS, and stained with phalloidin-FITC. The number of filopodia were counted and averaged from 50 myofibroblasts in each condition. Samples were viewed by the Axiovert 200 microscope (Carl Zeiss, Göttingen, Germany).

Protein analysis

Total cell lysis, immunoprecipitation and western blot analyses were performed as previously described (De Wever et al., 2004). Scanning densitometry was carried out with the Quantity One program (Bio-Rad). Quantitative determination of human TGF-β1 in CM was performed with ELISAs following the manufacturer's instructions (R&D systems).

Statistical analysis

All functional assays were performed at least in triplicate unless otherwise indicated. All values are give the means±s.d. or 95% confidence interval. Significance of Matrigel invasion was determined with Statview (Brainpower; Calabasas, CA) on an Apple Macintosh system. Comparisons were performed using an unpaired Student's t-test. P<0.01 was considered to indicate a significant difference (indicated by *).


Colon-cancer-cell-derived TGF-β is necessary and sufficient to stimulate invasion of myofibroblasts into collagen type I and Matrigel in vitro

After 48 hours, spheroid myofibroblasts that were suspended in collagen type I gel, a major constituent of the stromal ECM, were still non-invasive. By contrast, when HCT-8/E11 colon cancer cells were co-cultured on top of the collagen or when treated with CM of HCT-8/E11 (CMHCT-8/E11), spheroid myofibroblasts invade collagen type I in a stellate pattern radiating from the spheroid (Fig. 1A). Cell density at the surface of the spheroid appeared to decrease, suggesting that a resident population of myofibroblast spheroid cells invades the gel, rather than only new cells generated by cell proliferation. When myofibroblasts invaded the collagen, a reorganization of the collagen into thick collagen fibers (straps) occurred that aligned parallel to the axes between spheroids (data not shown). This phenomenon is explained by the traction exerted by the invasive myofibroblasts that combine to realign collagen fibers into a ligament-like `strap' on the axis between the spheroids (Sawhney and Howard, 2002). Invasion and formation of the strap was independent of the distance between the spheroids (distances less than 5 mm). Since HCT-8/E11 cells produced between 200 and 500 pg/ml TGF-β1 in the CM, we next examined the role of this cytokine in the invasion of the myofibroblasts. The addition of neutralizing TGF-β mAb but not control IgG1 inhibits the invasion-stimulating effect of HCT-8/E11 colon cancer cells and their CM. To confirm that colon-cancer-cell-derived TGF-β1 plays a role in myofibroblast invasion we used recombinant (r)TGF-β1 in either its latent or its bioactive form. After 48 hours, both latent and bioactive rTGF-β1 stimulated myofibroblast spheroid invasion, as observed for HCT-8/E11 cells and CMHCT-8/E11 (data not shown). We cultured these myofibroblasts further by exchanging the medium on day 3 and 5. After 7 days, untreated spheroid myofibroblasts were invasive (Fig. 1B). This phenotype did not depend on autocrine TGF-β1 secretion by myofibroblasts as evidenced by using TGF-β neutralizing mAb. Long-term invasion of myofibroblast spheroids was further increased by both latent and bioactive rTGF-β1, resulting in a remarkable stellate pattern of invasion into collagen type I. The larger magnification in Fig. 1B shows that myofibroblasts originating from two different myofibroblast spheroids establish cell-cell contacts via dendritic-like extensions.

Fig. 1.

Colon-cancer-cell-derived TGF-β is necessary and sufficient to stimulate invasion of myofibroblasts into collagen type I and Matrigel. (A) Myofibroblast spheroids were embedded in a collagen type I gel on top of which serum-free DMEM was added without (untreated), with 106 HCT-8/E11 cells or with CMHCT-8/E11 cells, in the presence of control IgG1 or of neutralizing TGF-β mAb. (B) Myofibroblast spheroids in collagen type I were cultured for 7 days with the indicated treatments. The medium was refreshed on days 3 and 5. A representative phase-contrast micrograph is shown for each condition, taken after 48 hours (panel A) or after 7 days (panel B). Experiments in A and B were repeated at least three times and had similar results. Scale bars, 100 μm. (C) Single-cell myofibroblasts at a density of 4×105 were seeded upon a Matrigel coated filter. HCT-8/E11 cells at a density of106 were seeded or rTGF-β1 was placed in the lower compartment in the presence of control IgG1 or neutralizing TGF-β mAb. Bars indicate means of three results ± s.d. Asterisks show statistically significant difference from untreated control.

Next, we examined the behavior of single myofibroblasts seeded on top of Matrigel-coated filters, to mimic invasion through the basement membrane. The lower compartment contained a collagen gel mixed with 106 HCT-8/E11 cells alone, combined with a neutralizing TGF-β mAb, or with control IgG1. As shown in Fig. 1C, HCT-8/E11 colon cancer cells significantly stimulate the invasion of myofibroblasts into Matrigel after 72 hours in a TGF-β-dependent way. The role of TGF-β in invasion was confirmed by embedding rTGF-β1 in the collagen gels of the lower compartment. At concentrations ranging from 0.1 to 10 ng/ml, rTGF-β1 stimulated the invasion of myofibroblasts into the Matrigel in a dose-dependent manner.

N-cadherin is essential for TGF-β1 stimulated invasion of myofibroblasts

We have previously demonstrated that N-cadherin is expressed by human colon myofibroblasts at the tips of contacting filopodia (Van Hoorde et al., 1999). In the Matrigel invasion assay, a single treatment by the N-cadherin-neutralizing GC-4 mAb was tested at concentrations ranging from 20 to 160 μg/ml (Fig. 2A). Invasion of the myofibroblasts was reduced by 70% in the presence of 80-160 μg/ml GC-4 mAb. Since control IgG1 showed no effect, this experiment strongly suggests a specific role for N-cadherin in rTGF-β1-stimulated myofibroblast invasion. With the GC-4 mAb (80 μg/ml), a similar level of inhibition was seen with HCT-8/E11 colon cancer cells as stimulators of invasion. Fourty-eight-hour invasion of myofibroblasts in collagen type I stimulated by CMHCT-8/E11 was also inhibited by a single dose of GC-4 mAb. Inhibition was also achieved in a 7-day assay with rTGF-β1 as a stimulant, provided the GC-4 mAb was added at day 1, 3 and 5. Here, also the control IgG1 had no effect (Fig. 2B) and spontaneous invasion of myofibroblasts was also sensitive to inhibition with the N-cadherin neutralizing GC-4 mAb.

Fig. 2.

Role of N-cadherin in rTGF-β1-stimulated invasion of myofibroblasts. (A) 4×105 single myofibroblasts were seeded upon a Matrigel coated filter with rTGF-β1 or HCT-8/E11 cells in the lower compartment to stimulate invasion. N-cadherin was neutralized by the GC-4 mAb which was added together with the myofibroblasts. Bars indicate means of three results ± s.d. Asterisks show statistically significant inhibition of invasion. (B) Myofibroblast spheroids cultured in collagen type I for 48 hours in the presence of HCT-8/E11 cells, and cultured in collagen type I for 7 days untreated or treated with rTGF-β1, were challenged with control IgG1 (80 μg/ml) or GC-4 mAb (80 μg/ml). In the 7-day experiment, medium was exchanged on day 3 and 5. A representative phase-contrast micrograph is shown for each condition. Scale bar, 100 μm. (C) Western blot, showing the effect of combinations of siN-CAD oligonucleotides on N-cadherin expression and cadherin-11 expression in myofibroblasts. Tubulin was used as loading control. (D) Spheroid myofibroblasts of cells that had been electroporated with siN-CAD oligonucleotides and cultured in collagen type I for 2 or 4 days in the presence of rTGF-β1. con, control oligonucleotide. Scale bar, 100 μm.

To further strengthen the idea that N-cadherin expression is essential in TGF-β-stimulated invasion, we knocked down N-cadherin expression in human colon myofibroblasts using siRNA. For this, myofibroblasts were electroporated with two double-stranded oligonucleotides derived from two distinct regions of N-cadherin cDNA. Myofibroblasts electroporated either with control oligonucleotide or without were used as controls. As revealed by western blot analysis 2 and 4 days after electroporation, two paired combinations of siN-cadherin oligonucleotides (siN-CAD2 with siN-CAD3 and siN-CAD3 with siN-CAD4) efficiently reduced N-cadherin (up to 97%) but not cadherin-11 expression (Fig. 2C). Tubulin expression was used as a control for equal protein loading. Electroporated spheroid myofibroblasts were subjected to the spheroid-cell-collagen-invasion-assay in the presence of rTGF-β1. As shown in Fig. 2D, inhibition of N-cadherin expression by siRNA substantially inhibited TGF-β-stimulated invasion. We thus conclude that, N-cadherin expression is necessary to observe the pro-invasive effect of TGF-β, at least in the spheroid-cell-invasion-assay and in the cell system analyzed.

We next investigated the relationship between TGF-β1 and the expression of N-cadherin in human colon myofibroblasts. As shown in Fig. 3A, the N-cadherin protein is upregulated by rTGF-β1 in a dose-dependent manner, whereas the expression levels of the molecular partners of N-cadherin, such as β-catenin and α-SMA, remain unchanged upon rTGF-β1 treatment. The stoechiometry of the molecular component of the N-cadherin-β-catenin-α-SMA complex was then analyzed in control and rTGF-β1-treated myofibroblasts by means of immunoprecipitation (Fig. 3B). Approximately 2.5-fold higher amounts of β-catenin and α-SMA were coprecipitated with antibodies recognizing N-cadherin in rTGF-β1-treated myofibroblast cultures, as compared with controls.

Fig. 3.

Effect of rTGF-β1 on N-cadherin expression and the organization of N-cadherin-catenin-α-SMA complex in myofibroblasts. (A) Treatment of myofibroblasts with rTGF-β1 for 7 days stimulates N-cadherin protein expression in a dose-dependent manner. Expression levels of β-ctn and α-SMA proteins are not significantly changed. Tubulin protein expression is used as loading control. (B) Immunoprecipitation (IP) on lysates of myofibroblasts treated or not for 7 days with rTGF-β1 (1 ng/ml) were performed to investigate the association of N-cadherin with β-catenin and α-SMA. Total lysates were stained for N-cadherin and tubulin (loading control).

Role of MAPK in N-cadherin-dependent myofibroblast invasion stimulated by colon-cancer-cell-derived TGF-β1

TGF-β1 has been reported to signal through the MAPK pathways (Mulder, 2000). We therefore, examined the possible contribution of the JNK, p38 and MEK1 as signaling components in TGF-β stimulated invasion of myofibroblasts into Matrigel. As shown in Fig. 4A, the JNK inhibitor SP600125 inhibited rTGF-β1-stimulated invasion of myofibroblasts in a dose-dependent manner, whereas the pharmacological inhibitors of MEK1 (PD98059 and UO126), and p38 (SB203580) were ineffective during the 72-hour assay. Similarly, only the JNK inhibitor SP600125 was effective to antagonize rTGF-β-stimulated myofibroblast invasion in the long-term spheroid-invasion-assay performed for 7 days, using collagen type I as an ECM substrate (Fig. 4B). Interestingly, none of the MAPK inhibitors affected the baseline levels of spontaneous invasion in untreated myofibroblasts (Fig. 4B). We performed experiments to determine whether JNK is involved in N-cadherin upregulation stimulated by rTGF-β1 in myofibroblasts. As shown in Fig. 4C, only SP600125 reversed rTGF-β1 stimulated N-cadherin expression at concentrations ranging from 0.18 to 18 μM. The other MAPK inhibitors SB203580, PD98059 and UO126 were ineffective at concentrations that prevented p38 and MEK1 activity in other studies. None of the MAPK inhibitors had an effect on N-cadherin expression in untreated myofibroblasts (data not shown).

Fig. 4.

Effect of MAPK inhibitors on rTGF-β1-stimulated N-cadherin expression and N-cadherin dependent myofibroblast invasion. (A) 4×105 single myofibroblasts were seeded on a Matrigel coated filter, with rTGF-β1 in the lower compartment to stimulate invasion. MAPK inhibitors were added in the upper compartment together with the myofibroblasts. Incubation was for 72 hours. Bars indicate the mean number of cells at the lower side of the filter (means of three results ± s.d.). The asterisk shows statistically significant inhibition of invasion. (B) Myofibroblast spheroids were cultured for 7 days in collagen gels in presence or not of rTGF-β1 and MAPK inhibitors. Medium was exchanged on day 3 and 5. A representative micrograph for each condition is shown. Scale bars, 100 μm. (C) Western blot analysis of lysates from myofibroblasts treated for 7 days with rTGF-β1 in presence or not of MAPK inhibitors. The upper part of the blot was probed for N-cadherin, the lower part for tubulin.

To validate our data obtained with the pharmacological inhibitors, we next examined the direct action of rTGF-β1 on the activation status of the MAPKs in myofibroblasts. JNK activity was determined by investigating the phosphorylation level of its direct and specific substrate Jun. We found that the addition of rTGF-β1 led to phosphorylation of Jun at serine residues 63 and 73, as shown in the western analysis with phosphospecific antibodies (Fig. 5). These activated (phosphorylated) forms of Jun, were increased 1.6-fold and 2-fold after rTGF-β1 treatment for 1 hour and 2 hours, respectively. By contrast, phosphorylation levels of ERK1/2 (substrate of MEK-1) and p38 were either not affected or even decreased following treatment with rTGF-β1.

Fig. 5.

Kinetics of rTGF-β1 activation of p38, ERK1/2 and JNK MAPK in myofibroblasts. Western analysis of lysates from myofibroblasts that had been treated with rTGF-β1 for different times (h=hours) and probed for the phosphorylated forms of p38MAPK (p38MAPK-P), ERK1/2 (ERK-P), and the JNK substrate Jun, mainly phosphorylated ser63 or ser73 (Jun-P63 and Jun-P73). Blots were stripped and re-probed for total amounts of p38MAPK, ERK1/2 and Jun and tubulin. The relative intensity of total ERK1/2, Jun and p38MAPK was normalized to tubulin levels. Consequently, the relative intensity of phosphorylated forms of ERK1/2, Jun and p38MAPK was normalized to total values of ERK, Jun and p38MAPK.

Role of N-cadherin in the migratory phenotype of myofibroblasts

Since N-cadherin plays a pivotal role in neurite outgrowth and in migration and invasion (Doherty and Walsh, 1996; Hazan et al., 2000; Williams et al., 2000), we investigated its role in the migratory phenotype of myofibroblasts via a wound healing assay. Confluent myofibroblast cultures were serum-starved for a minimum of 24 hours to establish quiescence, such that the presence of cells in the wounded area would owing to cell motility rather than cell proliferation. Myofibroblasts moved perpendicularly to the wound, in an irregular shaped front and not as a compact front like epithelial cells do (Suriano et al., 2003). Counting the number of intersections of myofibroblasts - with marks at various distances from the original wound margin - showed that the N-cadherin-neutralizing GC-4 mAb and the decapeptide huNCAD10, but not the control IgG1 or the control huN-CAD10scr peptide, significantly inhibit the migration of myofibroblasts (Fig. 6A). Previous studies have shown that HAV peptides can interfere with several cadherin-dependent processes, including neurite outgrowth (Doherty and Walsh, 1996), hippocampal long-term potentiation (Tang et al., 1998), osteoclast formation (Mbalaviele et al., 1995), myoblast fusion (Mege et al., 1992) and cancer invasion (Noë et al., 1999; Nawrocki-Raby et al., 2003). To confirm the role of N-cadherin in migration, we used the siRNA knock-down method. As shown in Fig. 6B, silencing of N-cadherin expression significantly inhibits myofibroblast migration.

Fig. 6.

Role of N-cadherin in wound-healing migration of myofibroblasts. (A-B) Confluent monolayers of serum-starved myofibroblasts were wounded with a razor blade. Migration of myofibroblasts was quantified as number of intersections/microscopic field at different distances after a 24-hour treatment with (A) N-cadherin neutralizing mAb or peptide and (B) electroporation in the presence of N-cadherin siRNA. Of the different conditions tested in A, representative pictures were taken. The arrowhead indicates the scratch (time 0) made by the razor blade. Bar, 100 μm.

F-actin staining showed filopodia-like cellular extensions and actin stress fibers that were oriented towards the wound in control IgG1-treated myofibroblast cultures (Fig. 7A,B). The GC-4 mAb neutralizing N-cadherin, significantly reduced the number of filopodia (Fig. 7A,B). A comparable decrease in filopodia formation was observed in myofibroblasts in which N-cadherin expression was knocked-down by siRNA (Fig. 7A,C). N-cadherin is extensively expressed at contacting filopodia in control electroporated myofibroblasts. By contrast, myofibroblasts electroporated with siN-CAD oligonucleotides do not show N-cadherin staining at all (Fig. 7C).

Fig. 7.

Role of N-cadherin in filopodia formation of myofibroblasts. (A) Bars show mean numbers of filopodia revealed by F-actin staining and counted on 50 myofibroblasts at the migration front (means of three results ± s.d.); asterisks indicated statistically significant difference. (B) Representative pictures of F-actin-staining of myofibroblasts at the migration front of cultures treated with control IgG1 or the GC-4 mAb. Bar, 25 μm (C) Representative pictures of F-actin and N-cadherin-staining at the migration front of myofibroblasts electroporated with control (con), or siN-CAD3 and siN-CAD4 oligonucleotides. Bar, 25 μm.

We determined myofibroblast polarization by phase-contrast microscopy and analyzed the Golgi-complex orientation during wound healing, using a specific and efficient mAb that recognized Golgi proteins originating from rat (Chicheportiche et al., 1984). Therefore, experiments were performed in N-cadherin-positive rat DHD-Fib myofibroblasts which migrate unidirectionally and develop cell protrusions perpendicular to the wound that become visible after 4-6 hours after wounding and are prominent after 8 hours (Fig. 8A). The huN-CAD10, bearing the same HAV-flanking amino acids as the rat N-cadherin but not the same as its scrambled control huN-CAD10scr, caused loss of polarity. This was evidenced by the loss of unidirectional migration and the formation of cell protrusions in a random direction. The GC-4 mAb, not recognizing rat N-cadherin in western blots or immunocytochemistry (data not shown), had no effect. Cells along the edge of monolayer wounds were considered polarized when the Golgi-complex was localized at the wound-side of the nucleus, (Nobes and Hall, 1999). We found that significantly less huN-CAD10-treated cells (relative to huNCAD10scr-treated cells) were polarized. In fact, polarization of huN-CAD10-treated cells at the edge of wounds was not significantly different from the random distribution observed in a confluent monolayer (Fig. 8B).

Fig. 8.

Role of N-cadherin in the polarity of myofibroblasts. (A) Representative phase-contrast micrographs of DHD-Fib myofibroblasts, taken 8 hours after wounding. Cells were treated with huN-CAD10scr or huN-CAD10 peptide. Loss of polarity is indicated by loss of unidirectional migration. Bars, 50 μm. (B) (upper panel) Golgi-complex immunostaining of DHD-Fib myofibroblasts in confluent conditions or 8 hours after wounding. Direction of movement is shown by arrows; (lower panel) bars show percentage of cells with their Golgi complex facing the wound for 50 cells of the first 2 front rows or at random in wounded or confluent monolayers. Bar, 25 μm.

Role of MAPKs in migratory phenotype of myofibroblasts

MAPKs are implicated in TGF-β stimulated invasion of myofibroblasts and we were therefore interested whether MAPKs are also implicated in wound-healing-induced migration of myofibroblasts (Fig. 9). The JNK inhibitor SP600125 inhibited wound-healing-induced migration of myofibroblasts in a dose-dependent manner. By contrast, the pharmacological inhibitors of MEK1 (PD98059) and p38 (SB203580) were ineffective during the 24-hour assay, as also observed in the invasion studies (Fig. 4A).

Fig. 9.

Effect of MAPK inhibitors on wound healing-induced migration of myofibroblasts. Serum-starved confluent myofibroblast monolayers were wounded with a razor blade. Migration of myofibroblasts was quantified as number of intersections/microscopic field at different distances after 24 hours in cultures and treated as indicated.


There is accumulating evidence that non-cancer cells in the stroma of solid tumors provide critical information for cancer cell growth, ectopic survival and invasion (De Wever and Mareel, 2003). The interaction between the epithelial cancer cells and the stromal compartment creates a local heterotypic invasion field in which it is not always clear `who is invading whom' (Liotta and Kohn, 2001). The importance of such cancer-stroma interactions are becoming more and more recognized, and this article describes a potential new mechanism how stromal myofibroblasts invade in vitro upon stimulation by colon-cancer-cell-derived factors. One can argue that both cancer cells and stromal myofibroblasts are interdependent for their migratory behavior, and can form coordinated invasive layers during local invasion and tumor spreading. The 3-D spheroid-myofibroblast-collagen-type-I-assay enabled us to demonstrate myofibroblast invasion in a typical stellate pattern, radiating from the spheroid cells upon contact with factors derived from cancer cells. Cancer cells stimulate invasion of myofibroblasts and attract them to pass Matrigel, which is a mimic of the basement membrane. Thus, myofibroblasts might also perform counter-current invasion, from the stroma into the epithelial cancer compartment, providing roads for cancer cell invasion, as has been described for inflammatory cells and immunocytes (Opdenakker and Van Damme, 1992). Alternatively, myofibroblasts might dock close to accumulations of cancer cells and locally provide them with pro-invasive signals (De Wever et al., 2004).

Our in vitro ecosystem, consisting of colon cancer cells, ECM barriers such as collagen type I or Matrigel, and host cells such as myofibroblasts, does not perfectly mimic the natural tumor tissue. It might, however, provide information about the reciprocal effects of one cell type upon another. In vitro methods widely make use of the 3-D collagen type I and Matrigel invasion assays to measure the ability of cells to attach to the matrix and migrate into it. Wound-healing-migration is a 2-D motility assay where myofibroblasts migrate as an irregularly shaped front towards the wound, a process that differs from invasion into matrices. We used the wound-healing-migration assay as a tool to study the migration response of myofibroblasts independently of cancer-cell-derived stimuli. Furthermore, 2-D analysis is an easy way to get information about defects in migratory cues such as filopodia formation, Golgi-complex reorientation and polarity.

We have shown that TGF-β1 derived from colon cancer cells was a necessary and sufficient signal to stimulate myofibroblast invasion in collagen type I and Matrigel. Cancer cells of different origins produce TGF-β and are resistant to its growth-inhibitory effects, in contrast to their benign precursor cells, which are sensitive to these effects (Derynck et al., 2001). TGF-β accumulates in the ECM where it is sequestered and can be processed into a bioactive ligand by myofibroblast-derived MMP-2 and MMP-9 (Yu et al., 2000). Accordingly, active TGF-β stimulates myofibroblasts to form a dendritic network of extensions in the spheroid myofibroblast invasion assay.

We have shown by using three independent approaches that, N-cadherin plays a crucial role in TGF-β-stimulated invasion and wound-induced migration. Also, N-cadherin is implicated in invasion of untreated myofibroblasts. This suggests that TGF-β aggravates the invasion-properties of a spontaneously invasive cell type by upregulating N-cadherin protein expression. Precipitation studies revealed that N-cadherin is complexed with β-ctn and the α-SMA cytoskeleton, suggesting that N-cadherin might stimulate invasion by forming adherens junctions linked to the cytoskeleton (Aberle et al., 1996). The expression of cadherin-11 in myofibroblasts is not affected by silencing N-cadherin expression. This suggests that the siRNA approach is very specific and that a reduction in N-cadherin protein expression does not necessarily lead to changes in cadherin-11 expression. Downregulation of one cadherin might lead to the upregulation of another cadherin that functionally substitutes or differs from the one that is downregulated. The latter was shown by Islam et al. in invasive squamous carcinoma cells, where downregulation of N-cadherin expression by an antisense construct increases the expression of E-cadherin, leading to reduced invasion (Islam et al., 1996).

Intercellular contacts between fibroblasts, involving interactions of cadherins, actin and myosin, generate the forces required for wound closure (Adams and Nelson, 1998). Therefore, N-cadherin has been implicated in intercellular mechanotransduction by activating Ca2+-permeable channels, consequently triggering the induction of actin assembly (Ko et al., 2001). The SK-N-SH neuroblastoma cell line, which expresses N-cadherin at the adherens junctions, is invasive into collagen type I (Van Aken et al., 2003). Also, N-cadherin is expressed in a punctuate pattern over the entire surface of axonal growth cones during neurite outgrowth (Riehl et al., 1996) and in Y79 retinoblastoma cells was shown to be invasive into collagen type I (Van Aken et al., 2002). In all cases, invasion and neurite outgrowth depended on N-cadherin as evidenced by using the GC-4 mAb. These findings suggest that the pro-invasive activity of N-cadherin cannot solely be attributed to its localization in the membrane and/or association with the cytoskeleton. As well the extracellular domain, the juxtamembrane and the β-catenin binding domain of N-cadherin are involved in pro-invasive signaling pathways. First, the N-cadherin extracellular repeat 4 interacts with the fibroblast growth factor receptor and mediates increased motility in epithelial breast cancer cell lines (Kim et al., 2000). Second, the juxtamembrane domain which binds p120-catenin, is implicated in N-cadherin-dependent neurite outgrowth (Riehl et al., 1996). Third, the tyrosine phosphatase 1B might function as a regulatory switch, controlling cadherin function by dephosphorylating β-catenin and thereby maintaining cells in an adhesion-competent state (Balsamo et al., 1998). We have shown that neutralization of N-cadherin by the GC-4 mAb or the huN-CAD10 peptide inhibits wound healing migration. Also, specific N-cadherin knock-down decreases wound-healing migration. A scratch through the monolayer initiates an increase in Cdc42-activity leading to directional movement of the myofibroblasts (O.D.W., W.W., A.V., N.B., M.B., C.G., E.B. and M.M., unpublished data). Cell migration in control conditions is initiated by a change in morphology of myofibroblasts at the wound edge; they extend a long protrusion into the space, generating an elongated shape with filopodia sensing the environment. Movement became disoriented and leading edge cells lost their filopodia after N-cadherin was neutralized following treatment with antibody or peptide or after silencing. These findings suggest that homophilic binding of N-cadherin at the trailing edge of cells might provide cues in defining the direction of invasion and migration. N-cadherin may also promote invasion and migration by forming labile cellular adhesions that facilitate dynamic cell-cell interactions. Labile contacts would allow the release of N-cadherin at some adherens complexes while propulsive forces are being generated at others. The engagement of cadherins probably has complex consequences for Rho-GTPase signaling that might be influenced by the cellular and extracellular context and type of cadherin (Goodwin et al., 2003; Kovacs et al., 2002; Noren et al., 2003). Furthermore, cell polarity is characterized by (1) a perpendicular orientation of the protrusion and the cell migration to the wound, and, (2) a reorientation of the Golgi complex towards direction of migration. Although the relatively long time course suggests that Golgi-complex realignment is not essential for movement, it might facilitate movement and it clearly indicates the development of a polarized morphology. Cdc42 was shown to regulate filopodia formation and polarity in fibroblasts in an in vitro wound assay (Nobes and Hall, 1999).

It is currently not exactly known, how N-cadherin expression is regulated by TGF-β1. In mouse mammary epithelial cells (NMuMG), TGF-β1 stimulates epithelial-tomesenchymal differentiation and N-cadherin expression through a RhoA-dependent pathway (Bhowmick et al., 2001). We have shown that upregulation of N-cadherin expression by rTGF-β1 occurs via JNK activation with coincident myofibroblast invasion into collagen type I and Matrigel. Both MEK-1 or p38 MAPK blockade are inefficient in inhibiting TGF-β1-stimulated invasion into collagen type I and Matrigel. Accordingly, rTGF-β1 stimulates JNK activity causing JNK to phosphorylate the JNK substrate Jun at serine residues 63 and 73. Jun is a member of the AP1 family of transcription factors. Interestingly, analysis of the chicken N-cadherin promoter revealed several consensus Sp1 and AP2 elements, and an AP1 binding site (Li et al., 1997). Alternatively, the stimulation of JNK by TGF-β might regulate Smad activation and therefore Smad dependent transcriptional activity (Engel et al., 1999). In that way, TGF-β-dependent N-cadherin expression might be the result of a convergence between Smad and JNK signaling pathways. Basal JNK, MEK1 or p38MAPK activity is not involved in spontaneous long-term spheroid-cell-invasion of myofibroblasts. However, JNK activity but not MEK1 or p38MAPK activity is necessary for wound-healing-induced migration of myofibroblasts. Consistent with this are data showing that, JNK activity correlates with cell migration in several systems (Abassi and Vuori, 2002; Huynh-Do et al., 2002; Huang et al., 2003) and JNK is a common downstream target of multiple signaling pathways that control cell movement, such as Rac and Src (Minden et al., 1995; Oktay et al., 1999).

In malignant tumors, interactive cross signaling between cancer cells and stromal host cells influences the invasive behavior of both cell populations, leading to concerted invasion. Neutralization of N-cadherin activity by several independent approaches inhibited the invasion response of myofibroblasts. Furthermore, TGF-β stimulated N-cadherin protein expression via a JNK-dependent pathway in myofibroblasts. By using an in vitro wound-healing assay we demonstrated that neutralization of N-cadherin or the specific N-cadherin knock-down inhibit Golgi-complex reorientation, filopodia formation, polarization and finally wound-healing-migration. From a therapeutical point of view, it is interesting that N-cadherin which is expressed by invasive cancer cells and myofibroblasts might be targeted by one drug; and stable cyclic-peptide-antagonists of N-cadherin may be used to inhibit N-cadherin activity (Williams et al., 2000).


D. Vandekerkhove and G. De Bruyne are gratefully acknowledged for technical assistance, and J. Roels for preparation of the illustrations. This work was funded by `Kom Op Tegen Kanker, FWOVlaanderen' (Brussels, Belgium). O.D.W. was supported by a fellowship from the `Stichting Emmanuel van der Schueren'. We thank Jean Willems (Kortrijk, Belgium) and P. Courtoy (Brussels, Belgium) for providing reagents.

  • Accepted May 20, 2004.


View Abstract