FAP-overexpressing fibroblasts produce an extracellular matrix that enhances invasive velocity and directionality of pancreatic cancer cells

All cell lines used in this study were originally purchased from ATCC or obtained from the Cell Culture Facility at Fox Chase Cancer Center. Three C.B17/Icr-scid mice were subcutaneously injected with 2 × 10⁶ cells of each pancreatic cell line (Panc-1, Capan-1, AsPC-1, and HPAF-II). After 5 weeks inoculation, tumors were harvested for immunohistochemistry analysis.

Fresh surgical tissue (decoded) samples from a pancreatic Whipple conducted at the Fox Chase Cancer Center was delivered with the assistance of the Protocol Laboratory and the Biosample Repository Facility following protocols approved by the Institutional Review Board. Tissue samples corresponding to pancreatic adenocarcinoma and distant (normal) pancreas were rinsed in cold PBS containing 100 μg/ml streptomycin and 100 U/ml penicillin. The two types of samples were carefully minced and incubated overnight with 0.2% collagenase at 37°C for digestive dissociation. The digested material was subjected to centrifugation at 1200 rpm for 5 min, thus precipitating a fibroblast-enriched fraction. This fraction was filtered through a series of 100 μm followed by 40 μm cell-strainer (BD Bioscience) before culturing in DMEM containing 10% FBS, 100 U/ml penicillin, 100 μg/ml streptomycin and 2 mM L-glutamine at 37°C using a humidified atmosphere and 5% CO₂ for a period of 4 hours before removing the non adherent cells. Cell homogeneity was confirmed by direct microscopic observations, while cells were designated as fibroblastic following confirmation of mesenchymal marker vimentin expression, as well as absence of epithelial marker cytokeratin 19 expression. In addition, normal fibroblasts were characterized as pancreatic stellate cells following confirmation of uptake and storage of vitamin A, by auto-fluorescence stemming from retinyl acetate (Sigma) containing droplets [24]. The resultant (PaSCs) and pancreatic tumor-associated fibroblasts (which were confirmed to have lost the vitamin A droplets) were used for self-derived matrix characterization.

FAP-transfected fibroblasts were harvested at different time points (0, 1, 2, 4, 6, 8, and 10 days) in the presence or absence of Dox. Total protein was extracted using M-Per reagent (Pierce, IL), resolved by 4-12% SDS-PAGE under reducing conditions (Invitrogen, CA), and blotted with rabbit monoclonal anti-mouse FAP antibody (0.08 μg/ml). Immuno positive bands were labeled using goat anti-rabbit-HRP (3000×, Amersham Bioscience, UK) and visualized by ECL reagent (Pierce, IL). Purified murine recombinant FAP protein (92 kD) and parental NIH-3T3 cell lysate were used as positive and negative controls, respectively.

All fibroblasts were seeded at a concentration of 7 × 10⁵ cells per sample onto 35 mm plates pre-coated with 0.2% gelatin. Confluent fibroblastic cultures were treated with media supplemented with 50 μg/ml ascorbic acid (and Dox when necessary) every other day for 8 days to obtain un-extracted 3D cultures. Alkaline detergent treatment (0.5% Triton X-100, 20 mM NH₄OH in PBS) for 5 minutes at 37°C gave rise to cell-free in vivo-like 3D matrices [25, 26]. For control, FAP-transfected fibroblasts were used to make FAP^- matrix in the absence of Dox. To make FAP+inhibitor matrix, FAP-specific small molecule inhibitor naphthalenecarboxy-Gly-boroPro (400 μM, provided by Dr. William Bachovchin, Tufts University, MA) was added to the media during matrix production.

For indirect immunofluorescent labeling of matrix fibers, un-extracted 3D cultures were prepared on glass cover slips as described [21]. Cells were stained with rabbit anti-mouse fibronectin antibody (25 μg/ml, Abcam, UK), a donkey anti-rabbit Cy5 conjugated antibody (15 μg/ml, Jackson ImmunoResearch, PA) and DAPI.

From two independent experiments with duplicate samples, minimum of 5 images per experimental sample were obtained using a z-stack function of the spinning disc microscope (PerkinElmer Life Sciences, PA). Each slice measured 0.5 μm and stacks were reconstituted as a maximum projection using the MetaMorph software as described in detail [21]. Fiber distributions corresponding to each experimental setting were determined by the percentage of fibers that were oriented at 10⁰ variation angles from the identified modes.

FAP-transfected fibroblasts (1.5 × 10⁴ cells/48-well plate) were cultured in the presence or absence of Dox during matrix production. Fibroblasts were fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton X-100 in PBS, and stained with antibodies followed by manufacturer's directions (Li-Cor Bioscience, NE). The antibodies for this assay were tenascin C (50 μg/ml, Abcam, UK), α- SMA (5 μg/ml, Sigma, MO), collagen I (25 μg/ml, Chemicon International, CA), fibronectin (20 μg/ml, Abcam, UK), β-actin (1000×, Cell Signaling Technology, MA), and GAPDH (2 μg/ml, Chemicon International, CA). Fluorescence-labeled secondary antibodies IRDye 680 and IRDye800CW (Li-Cor Bioscience, NE) were used to scan by the Odyssey Infrared Imaging System. The protein levels were normalized by β-actin or GAPDH.

Following 2 days culture on matrices, Panc-1 cells were lysed in extraction buffer (50 mM Tris-HCl, pH8.8, 1% SDS, 5 mM DTT, 13 mM iodoacetamide, 5 mM EDTA, 150 mM NaCl). Proteins were quantified by western blot analysis using ECL reagent (Pierce, IL) or Odyssey infrared imaging system following the manufacturer's directions (Li-Cor Bioscience, NE). The antibodies for this assay were β1-integrin (2500×, BD Transduction Laboratories, NJ), AKT (500×, BD Transduction Laboratories, NJ), pS⁴⁷³-AKT (1000×, Cell Signaling Technology, MA), FAK (1 μg/ml, Upstate, NY), pY³⁹⁷-FAK (1000×, Biosource, CA), and GAPDH (2 μg/ml, Chemicon International, CA). Specific activity of AKT and FAK were calculated as the ratio of the scanned optical density of phosphorylated protein/total protein. Then the ratios were normalized by GAPDH.

As described in detail [26], tumor cells were re-plated onto the matrices and incubated overnight. Cell movements (10~15 cells), recording cells migrating both on and through the matrix, were recorded every 10 minutes during 12 hours using Nikon TE-2000U inverted microscope equipped with cool snap HQ camera. Individual cell dynamics were analyzed using the MetaMorph program following 4 distinctive factors: 1) the net path distance (D, μm) by calculating the number of μm/pixel. 2) The path trajectory of an individual cell during the recording period (T, μm). 3) Average velocity as the motility rate (AV, μm/hr). 4) Directionality calculated by the D/T ratio, which determines random (D/T = significantly smaller than 0.5) versus directional (D/T = close to 1) migration of individual cells [27‐29]. Motility experiments using the assorted matrices were performed a minimum of 3 times using different batches of matrices and following a minimum of 10 individual cells.

In order to assess the effect on motility by integrin inhibitors, motility assays were performed in the presence of functional blocking β₁ integrin antibody mAb13 (50 μg/ml, a kind gift from Dr. K. Yamada at NIH/NIDCR, Bethesda, MD) or the α₅β₁-integrin blocking peptide ATN-161 (Ac-PHSCN-NH2, 50 μg/ml, Attenuon, San Diego, CA). These experiments were performed at least 2 times using two alternative batches of matrices.

Data from 3D matrix fiber distribution were analyzed using "chi-squared test". The normalized western data were analyzed using linear regression. For the motility analysis, the average velocity, net path distance and the path trajectory were analyzed using Gamma regression, and the directionality was analyzed using linear regression.

Quick snap frozen tumor tissues sections (10 μm thick) from pancreatic cancer patients and xenografted mouse tumor tissues were incubated with rat monoclonal anti-human FAP antibody D8 (20×, a kind gift from Dr. Wen-Tien Chen, State University of New York, NY) and rabbit polyclonal anti-mouse FAP antibody (2 μg/ml) [16], respectively. Biotin-streptavidin detection with horseradish peroxidase (BioGenex, CA) was applied for the amplification and visualization of signals following the manufacturer's directions. Note that results stemming from this method are depicted in supplemental Figure 1.

Some reports suggest that FAP expression can occur in both the stroma and epithelial compartments of cancers [30‐33]. However, the selectivity of FAP for stromal fibroblasts, but not epithelial tumor cells, has been confirmed either by immunohistochemical studies in pancreas, colorectal and breast cancer patients [12, 34, 35], or by RT-PCR of pancreas, lung, and renal cell xenografts [36]. To confirm the putative significance of studying FAP-expression effects in stromal matrices, we first examined FAP expression in the human pancreatic cancer patient tissues and mouse xenografts by immunohistochemistry. We observed that FAP is highly expressed on the stromal fibroblasts of human pancreatic cancer, while it is undetectable both in the epithelial cancer cells (Additional file 1, Fig. S1A) and normal tissue (not shown). In xenograft mouse models inoculating the human pancreatic cancer cell lines (HPAF-II, Capan-1, AsPC-1, and Panc-1), murine FAP expression was also found up-regulated specifically at the tumor stroma (Additional file 1, Fig. S1B). This observation confirmed the existence of a selectivity of FAP expression in tumor-associated fibroblasts and prompted us to believe that perhaps FAP is an attractive protein to study stromal effects imparted upon tumor behaviors.

In vivo, FAP is highly expressed on pancreatic tumor stromal fibroblasts, but its expression ex vivo on cultured primary cells is not maintained. To establish stable FAP-expressing fibroblasts, mouse fap gene was cloned under the Tet-inducible system. Induction of FAP expression was achieved as early as 24 hours following Dox treatment, and its expression was clearly maintained for at least 10 days (Figure 1A). Parental NIH-3T3 cells and FAP⁺ fibroblasts cultured in the absence of Dox showed no detectable FAP induction, allowing the use of these fibroblasts as negative controls.

During the 8 days required to produce a fibroblast-derived 3D matrix, clear morphological differences on fibroblasts were observed between the parental NIH-3T3 and FAP^- vs. FAP⁺ fibroblasts. FAP⁺ cells were elongated into an enhanced spindled shape, and they organized in a parallel pattern (data not shown). Since tumor-associated ECMs are organized in parallel patterns [21, 22, 37], we tested whether FAP expression affects the topography of fibroblasts-derived ECMs by indirect immunofluorescence using anti-fibronectin (Figure 1) or anti-collagen I (Additional file 2, Fig. S2) antibodies on un-extracted matrices containing their original matrix-synthesizing fibroblasts [21]. Indeed, FAP⁺ fibroblasts produced ECM fibers oriented in parallel patterns when compared to FAP^- matrices (Figure 1B). These patterns are reminiscent of previously observed tumor-associated patterns in vitro [21]. In addition, the level of organization of the assorted fibroblast-produced fibronectin fibers was quantified by measuring the relative orientation angles of fibers [21]. The average percent of parallel fibers that were oriented within ±10⁰ of the mode angle was determined using the MetaMorph software. The measured percentages were 27% and 45% for FAP^- and FAP⁺ 3D ECMs, respectively (p < 0.001) (Figure 1B). Furthermore, inhibition of FAP enzymatic activity by the FAP inhibitor during the matrix production effectively reversed the FAP-induced parallel matrix orientation (29%). Moreover, the observed architectural patterns were similar to the ones seen in desmoplastic reactions in human pancreatic cancers and in xenografted tumors formed in mice using human pancreatic cancer cells in vivo (see asterisks in Additional file 1, Fig. S1). Importantly, when assessing the architectural patterns of collagen I fibers using the FAP series of un-extracted matrices in vitro, our results showed that, although the collagen I fibrillogenesis levels during matrix production (8 days) were not as substantial as to allow the analysis of the fiber orientation, simple microscopy observations clearly demonstrated that FAP⁺ fibroblasts indeed affect collagen I fiber organization (Additional file 2, Fig. S2). These results suggest that the FAP enzymatic activity during matrix production is important for the topographical organization of the ECM fibers. Because of this topographical similarity to that of tumor permissive 3D ECMs [25, 37], the FAP⁺ 3D matrix system was used to study the mechanism of matrix-supported pancreatic cancer cell invasion.

It is known that FAP exhibits very restricted expression in normal adult tissue [38] and that pancreatic stellate cells are the 'normal' fibroblastic cells found in the exocrine pancreas. Activated pancreatic stellate cells (i.e., myofibroblastic cells) express α-SMA and FAP (among other markers) and correspond to the cells believed to be responsible for the fibrosis in chronic pancreatitis and the desmoplastic reactions in pancreatic adenocarcinomas [39]. To compare our observations of the specific matrix architectures stemming from the FAP matrix series to matrices derived from normal pancreas human stellate cells and activated adenocarcinoma-associated fibroblasts, we proceeded to isolate these two types of human pancreatic fibroblastic cells from matched fresh surgical normal and pancreatic tumor (i.e., adenocarcinoma) samples (see Materials and Methods). Our results shown in Figure 2 present the standard bundle pattern of ECM fiber structures observed in FAP^- matrix. Moreover, human pancreatic adenocarcinoma-associated fibroblasts produced ECMs oriented in parallel patterns, which are similar to the patterns formed by FAP⁺ matrices. This result supports the notion suggesting that FAP⁺ fibroblast-derived matrices effectively recapitulate tumor stromal ECMs and that perhaps these matrices also have the capacity of producing matrices permissive/inductive of pancreatic tumorigenesis.

The ECM proteins and endogenous desmoplastic proteins (α-SMA) have been known to be up-regulated in pancreatic stromal reactions (desmoplasia) and play an active role in tumorigenesis [40]. The levels of ECM proteins (tenascin C, collagen I and fibronectin) and cellular α-SMA potentially underlying the matrix differences mediated by FAP were measured on the assorted un-extracted matrices (Figure 3). The expressions of α-SMA, fibronectin, and collagen I were significantly up-regulated in FAP⁺ matrix compared to FAP^- matrix (p = 0.009, p = 0.002, p = 0.001, respectively), while tenascin C was significantly down-regulated in FAP⁺ matrix (p = 0.001). Thus FAP expression in fibroblasts during matrix production alters ECMs composition, as well as its topography (e.g., fibronectin-fiber orientation). Interestingly, compared with FAP^-, inhibition of FAP enzymatic activity during FAP⁺ matrix production enhanced fibronectin and α-SMA expression (p = 0.04 and p = 0.04, respectively) to levels comparable to the ones observed in FAP⁺ cultures in the absence of the inhibitor. In fact, the P values for FAP⁺ vs. FAP⁺inhibitor were 0.32 and 0.61 for fibronectin and α-SMA, respectively. On the other hand, blocking FAP activity during FAP⁺ matrix production significantly down-regulated collagen I (p = 0.007), bringing the levels of this ECM protein back to levels similar to the ones observed in FAP^- matrix. These results suggest that collagen I production may be FAP activity dependent (Figure 3).

To determine the effects of the ECM on the invasive behavior of cancer cells, time-lapse acquisition assays were performed using pancreatic cancer cell lines that are known for their distinct invasive and metastatic potentials in vivo [41]. The more metastatic cell lines (Capan-1, Panc-1) and the less aggressive cells (AsPC-1, HPAF-II) were plated onto matrices, and their behaviors were measured for average velocity (AV), net path distance (D), path trajectory (T), and directionality (D/T ratio) (Figure 4). The invasive behaviors of AsPC-1 (AV = 5.1 ± 0.7, D = 9.5 ± 2.9, T = 54.8 ± 5.0) and HPAF-II cells (AV = 5.5 ± 1.7, D = 16.6 ± 4.8, T = 58.1 ± 18.1) were found to be less effective than the behaviors of Capan-1 (AV = 11.3 ± 0.7, D = 32.2 ± 0.7, T = 120.8 ± 7.9) and Panc-1 cells (AV = 12.0 ± 1.4, D = 94.1 ± 13.1, T = 128.2 ± 15.0) within the FAP⁺ matrix. The greater average velocity of Capan-1 and Panc-1 correlated with their known enhanced ability to metastasize in vivo [41]. Although Capan-1 and Panc-1 cells presented similar velocities, they showed significant differences in their movement patterns as Panc-1 cells invaded using greater directional motilities (D/T = 0.72 ± 0.05) compared to the more random paths taken by the Capan-1 cells (D/T = 0.28 ± 0.02, p < 0.001). However, no significant differences in velocity or directionality were observed using these cells cultured within the FAP^- matrices (not shown).

In order to question whether FAP⁺ matrix effects promoting cell motility of invasive pancreatic cancer cells (i.e., Panc-1) are specific for pancreatic cancers, we repeated this series of experiments using three well characterized breast cell lines. Additional file 3, Fig. S3 shows that compared to immortalized normal MCF-10A (AV = 7.59±1.3, D/T = 0.348±0.09) and tumorigenic MCF-7 (AV = 4.88±0.65, D/T = 0.17±0.042), highly invasive MDA-MB-231 cells moved faster (AV = 12.0±1.0, p = 0.04, p < 0.001, respectively) and more direct (D/T = 0.68±0.04, p = 0.001, p < 0.001, respectively) on FAP⁺ matrix. The results demonstrate that FAP⁺ matrix effects may be important in neoplasias other than just pancreatic cancers.

In summary, FAP⁺ 3D matrices represent a permissive environment for pancreatic (and breast cancer) invasion, and perhaps these matrices play an equally important role along with the epithelial cell component. Given their matrix-induced velocity and persistent directionality, Panc-1 (and in some instances MDA-MB-231) cells were selected for the further studies.

To test the contribution of FAP in the matrix-induced permissive tumor behavior, Panc-1 cells were seeded onto the three distinct matrices produced from the FAP^- , FAP⁺, or FAP+inhibitor fibroblasts (Figure 5). Compared to cells in FAP^- matrix (AV = 7.8 ± 0.8, D/T = 0.36 ± 0.04), Panc-1 cells presented greater velocity and directional motility within FAP⁺ matrix (AV = 12.0 ± 1.4 and D/T = 0.72 ± 0.06) (p = 0.05 and p < 0.001, respectively). In addition, directionality was significantly attenuated on FAP+inhibitor matrix (D/T = 0.52 ± 0.06, p = 0.01), with a minor inhibition in the velocity (AV = 9.1 ± 1.6, p = 0.19). These results demonstrate that velocity and directionality of cells depend on the type of ECM, which can differ if the matrix-producing fibroblasts express FAP or not. In addition, these results suggest that while velocity is FAP dependent regardless of its activity, directionality is strictly dependent on FAP enzymatic activity.

Integrins are the major receptors that link the ECM tumor microenvironment with the migrating cancer cell. Given the enhanced expression of ECM proteins and rearranged fibronectin fiber orientation in FAP⁺ matrix, the contribution of integrins responsible for the invasiveness of Panc-1 on matrix was investigated by time-lapse acquisition assays performed in the presence of functional blocking agents, including the β₁-integrin antibody mAb13 and the α₅β₁-integrin blocking peptide ATN-161 (Figure 6). Compared to the control experiment using rabbit serum (AV = 13.55 ± 2.25, D/T = 0.64 ± 0.07), inhibition of β₁-integrins significantly attenuated both velocity (AV = 4.46 ± 0.49, 67% inhibition, p < 0.001) and directionality (D = 0.19 ± 0.03, 70% inhibition, p < 0.001) of Panc-1 cells on FAP⁺ matrix. However, these significant inhibitions were not observed on cells using FAP^- matrix (p = 0.07 and p = 0.06, respectively). This result suggests that β₁-integrin is an important regulator of tumor cell motility in FAP⁺ but not FAP^- matrices. Similar to results above, questioning the matrix specific effect on only pancreas tumorigenesis, here we decided to test FAP⁺ matrix motility effects imparted upon MDA-MB-231 cells in the presence or absence of mAb-13. Results suggest that the observed FAP⁺ matrix induced increase in MDA-MB-231 motility is also dependent on β₁-integrin activity (Additional file 4, Fig. S4).

Given the enhanced expression and spatial orientation of fibronectin in FAP⁺ matrix, the importance of the fibronectin receptor α₅β₁-integrin was also tested. Inhibition of α₅β₁-integrin using ATN-161 showed significant decrease in the FAP⁺ matrix-induced invasive behavior of Panc-1 cells (Figure 6). Compared with controls, inhibition of α₅β₁-integrin decreased both velocity (AV = 7.73 ± 0.84, 43% inhibition, p = 0.002) and directionality (D = 0.34 ± 0.03, 53% inhibition, p < 0.001) of cells on FAP⁺ matrix, although not to the same extent as that of more general integrin inhibitors. Given the incomplete abrogation of invasive behavior of Panc-1 cells in the presence of ATN-161, it is possible that additional β₁-integrins could also contribute to pancreatic cancer invasion facilitated by the FAP⁺ 3D matrices. In summary we concluded that, since the enhanced motility of Panc-1 cells on FAP⁺ 3D matrices can be significantly reversed by blocking the function of β₁-integrins, a potential mechanism of matrix mediated tumor invasion is an attractive possibility.

Integrin-mediated signaling pathways are often associated with AKT and/or FAK effectors, resulting in proliferation, survival, and invasion of tumor cells [42]. To determine the downstream components of β₁-integrins, Panc-1 cells were cultured within assorted 3D matrices and the constitutive activity levels of both AKT and FAK were analyzed by the ratios of pS⁴⁷³-AKT/total AKT and pY³⁹⁷-FAK/total FAK, respectively (Figure 7). As shown in Figure 7A, constitutive FAK activity levels of Panc-1 cells cultured within both FAP⁺ and FAP+inhibitor matrices were significantly up-regulated ~1.5 fold compared to its activity in FAP^- matrix (p = 0.007 and p = 0.03, respectively). When activity levels were compared for Panc-1 cells cultured within the assorted matrices in the presence or absence of mAb13, a significant down-regulation of constitutive FAK activity levels under β₁-integrins inhibition was observed in all three matrices (p = 0.02, p = 0.002, p = 0.02 for FAP^-, FAP⁺, FAP+inhibitor matrices, respectively). However, constitutive levels of AKT activity in Panc-1 cells cultured within the various matrices remained unchanged (all p > 0.18, Figure 7B). In addition, inhibition of α₅β₁-integrin with ATN-161 did not show any changes in the activities of FAK or AKT (all p > 0.2), suggesting that the specific integrin contributions to the total AKT and FAK activity levels are undetectable by this method, nevertheless, ATN-161 effects were shown in the motility assay (Figure 6). In summary, up-regulation of FAK activity in Panc-1 cells on FAP⁺ matrix and down-regulation of its activity by function blocking β₁-integrin suggest that the FAP⁺-dependent enhanced invasion behavior of Panc-1 is regulated by the β₁-integrin/FAK signaling pathway, and although α₅β₁-integrin seems to play a role, additional β₁-integrins may also participate in this regulatory process.