Platelet-derived growth factor-D and Rho GTPases regulate recruitment of cancer-associated fibroblasts in cholangiocarcinoma


  • Potential conflict of interest: Dr. Colledan advises Novartis.

  • This work was supported by the Associazione Scientifica Gastroenterologica di Treviso (to L.F. and I.F.), the Telethon (grant no.: GGP09189) and Fondazione Amici Dell' Epatologia (to L.F.), the Progetto di Ricerca Ateneo 2008 (grant no.: CPDA083217) and 2011 (grant no.: CPD113799/11) (to L.F. and M.C.), the National Institutes of Health (grant no.: DK34989: Silvio O. Conte Digestive Diseases Research Core Centers – 5P30DK034989), and a grant from PSC partners for a care (to M.S.). R.F is a recipient of a Liver Scholar Award (American Liver Foundation). The support of Fondazione San Martino (Bergamo, Italy) is gratefully acknowledged.

  • See Editorial on Page 853

Address reprint requests to: Luca Fabris, M.D., Ph.D., Department of Surgery, Oncology, and Gastroenterology, University of Padova, School of Medicine, 2 Via Giustiniani, 35128 Padova, Italy. E-mail:; fax: +39 049 827 8906.


Cholangiocarcinoma (CCA) is characterized by an abundant stromal reaction. Cancer-associated fibroblasts (CAFs) are pivotal in tumor growth and invasiveness and represent a potential therapeutic target. To understand the mechanisms leading to CAF recruitment in CCA, we studied (1) expression of epithelial-mesenchymal transition (EMT) in surgical CCA specimens and CCA cells, (2) lineage tracking of an enhanced green fluorescent protein (EGFP)-expressing human male CCA cell line (EGI-1) after xenotransplantation into severe-combined-immunodeficient mice, (3) expression of platelet-derived growth factors (PDGFs) and their receptors in vivo and in vitro, (4) secretion of PDGFs by CCA cells, (5) the role of PDGF-D in fibroblast recruitment in vitro, and (6) downstream effectors of PDGF-D signaling. CCA cells expressed several EMT biomarkers, but not alpha smooth muscle actin (α-SMA). Xenotransplanted CCA masses were surrounded and infiltrated by α-SMA-expressing CAFs, which were negative for EGFP and the human Y-probe, but positive for the murine Y-probe. CCA cells were strongly immunoreactive for PDGF-A and -D, whereas CAFs expressed PDGF receptor (PDGFR)β. PDGF-D, a PDGFRβ agonist, was exclusively secreted by cultured CCA cells. Fibroblast migration was potently induced by PDGF-D and CCA conditioned medium and was significantly inhibited by PDGFRβ blockade with Imatinib and by silencing PDGF-D expression in CCA cells. In fibroblasts, PDGF-D activated the Rac1 and Cdc42 Rho GTPases and c-Jun N-terminal kinase (JNK). Selective inhibition of Rho GTPases (particularly Rac1) and of JNK strongly reduced PDGF-D-induced fibroblast migration. Conclusion: CCA cells express several mesenchymal markers, but do not transdifferentiate into CAFs. Instead, CCA cells recruit CAFs by secreting PDGF-D, which stimulates fibroblast migration through PDGFRβ and Rho GTPase and JNK activation. Targeting tumor or stroma interactions with inhibitors of the PDGF-D pathway may offer a novel therapeutic approach. (Hepatology 2013;53:1042–1053)


alpha smooth muscle actin


cancer-associated fibroblast




discoidin domain receptor tyrosine kinase 2


2-oxoglutarate analogues dimethyloxaloylglycine


enhanced green fluorescent protein


endothelial cells


extracellular matrix


enzyme-linked immunosorbent assay


epithelial-mesenchymal transition


extracellular signal-regulated kinase


fluorescent in situ hybridization


hepatocyte growth factor


hypoxia-inducible factor


hepatic stellate cells








c-Jun N-terminal kinase, MAPK, mitogen-activated protein kinase


methyl tetrazolium salt


platelet-derived growth factor


PDGF receptor


phosphorylated ERK1/2


phosphatidylinositol 3-kinase/protein kinase B


phosphorylated JNK


recombinant human PDGF-D


severe combined immunodeficiency


small interfering RNA


transforming growth factor beta


tumor necrosis factor-related apoptosis-inducing ligand

Y Chr

Y chromosome.

An extensive desmoplastic reaction is a distinctive feature of cholangiocarcinoma (CCA), a highly aggressive cancer originating from the biliary epithelium, characterized by strong invasiveness with limited opportunities of curative treatment.[1] The “tumor reactive stroma” is the site of complex functional interactions between cancer cells and the host microenvironment, and it plays a pivotal role in tumor growth and invasiveness.[2]

Cancer-associated fibroblasts (CAFs) provide tumor cells with proliferative and antiapoptotic signals that ultimately promote cancer growth. On one hand, cancer cells produce a range of signals able to instruct the stromal microenvironment to become permissive and supportive for tumor progression.[3] On the other hand, CAFs communicate with other cell types (endothelial cells [ECs], pericytes, and inflammatory cells) inducing angiogenesis and remodeling of the extracellular matrix (ECM),[3] ultimately favoring tumor invasiveness. In CCA, overexpression of proinflammatory cytokines in the tumor stroma is associated with a more malignant tumor phenotype.[4] Paracrine signals from CAFs protect CCA cells from proapoptotic stimuli.[5]

The origin of CAFs is still uncertain.[6] It has been proposed that CAFs undergo an epithelial-mesenchymal transition (EMT) of carcinoma cells, during which cancer cells lose their epithelial properties and acquire a mesenchymal phenotype that consequently favors increased invasive and migratory capabilities. Alternatively, CAFs may be recruited by cancer cells from resident fibroblasts[6] or from circulating mesenchymal progenitor cells of bone marrow origin.[7]

Members of the platelet-derived growth factor (PDGF) family are of interest because of their ability to promote fibroblast and hepatic stellate cell (HSC) migration and proliferation. Furthermore, PDGF expression has been shown to correlate with cancer progression in colon carcinoma as well as to protect CCA cells from apoptosis.[5, 7] The PDGF family encompasses five dimeric ligand isoforms (PDGF-AA, -BB, -AB, -CC, and -DD), which signal through two structurally related tyrosine kinase receptors, PDGF receptor (PDGFR)α and PDGFRβ. Although PDGFRα binds all PDGF isoforms except for PDGF-DD, PDGFRβ has a preferential and high affinity for PDGF-BB and PDGF–DD. The possible role of PDGF-D in tumor development and progression is only starting to be recognized.[8]

To better understand the mechanisms underlying the formation of tumor reactive stroma in CCA, we investigated whether CAFs are generated from cancer cells or are recruited by cancer cells through a PDGF-dependent mechanism. Specifically, this study sought (1) expression of several EMT markers in human CCA specimens and cells (2) the fate of human CCA cells xenotransplanted in severe combined immunodeficient (SCID) mice after transfection with enhanced green fluorescent protein (EGFP), (3) expression of PDGF ligands and receptors in CCA specimens and cells, (4) the ability of cultured human CCA cells to secrete PDGF isoforms, (5) the role of PDGF-D in tumor epithelial/mesenchymal cross-talk in vitro, and (6) the intracellular signaling pathways involved.

Materials and Methods

Tissue Samples and Cells

Fifteen formalin-fixed, paraffin-embedded samples of surgically resected CCA livers, obtained from archival tissue at Bergamo Hospital (Bergamo, Italy) were considered for the immunohistochemical (IHC) study (10 intrahepatic and 5 extrahepatic). For each patient, the matched peritumoral sample was available. We also studied three human CCA cell lines: EGI-1, TFK-1 (Deutsche Sammlung von Mikroorganismen und Zellkulturen, Braunschweig, Germany), and HuCCT-1 (Health Science Research Resource Bank, Osaka, Japan), as well as three primary CCA cell lines isolated from human liver samples derived from surgical resections for intrahepatic CCA at Treviso Regional Hospital (Treviso, Italy) (CCA1, CCA2, and CCA3), as previously described.[9] Human cholangiocytes isolated from liver explants with alcoholic cirrhosis served as controls (n = 3). See the Supporting Materials for further details on human fibroblast isolation. All specimens were reviewed to confirm the histopathological diagnosis of CCA. Informed consent and local regional ethical committee approval were obtained before tissue collection and cell preparations.

Immunophenotyping for EMT Markers and PDGF Family Members of CCA Specimens and Cells

Expression of a panel of phenotypic EMT markers, including E-cadherin, β-catenin, S100A4, the transcription factors, Twist and Snail1, collagen-specific receptor tyrosine kinase discoidin domain receptor tyrosine kinase 2 (DDR2), vimentin, alpha smooth muscle actin (α-SMA) and laminin, and expression of the PDGF family members (PDGF-A,-B,-C, and -D and the cognate receptors, PDGFRα and -β) were evaluated by IHC and immunocytochemistry (ICC) in tissue sections and cultured cells, respectively, and by western blotting (Supporting Table 1). Expression of PDGF ligands and receptors was also studied in tissue sections by dual immunofluorescence (IF) with K7 and α-SMA as markers for neoplastic cholangiocytes and CAFs, respectively. Secretion of PDGF-AA and -BB (RayBiotech, Milan, Italy), and -DD (USCNK, Milan, Italy) was quantified by enzyme-linked immunosorbent assay (ELISA) in culture medium collected from CCA cells and controls. To study whether hypoxia was a stimulus for PDGF-D secretion in CCA cells, PDGF-D secretion was studied in cultured CCA cells treated with 2-oxoglutarate analogs dimethyloxaloylglycine (DMOG; 3 mmol/L for 18 hours) to induce chemical hypoxia.[10] See the Supporting Materials for further details.

Table 1. Immunophenotypic Characterization of EMT and PDGF Signaling Markers in 15 Human CCA Samples Derived From Surgical Resection
MarkerCell TypeNo. of Positive Samples/15
  1. The number of positive samples out of 15 are reported.

  2. a

    Positive samples when down-regulation or delocalization of membrane staining occurred.

  3. b

    Positive samples when uneven immunoreactivity along the bile duct profile occurred.

  4. c

    Only a weak expression (<30% of neoplastic bile ducts) in positive samples.


Morphometric Analysis

Methodological details are given in the Supporting Materials.

Xenotransplantation of CCA Cells in SCID Mice

The well-characterized invasive capabilities of EGI-1 cells in the SCID mouse model meant that this particular CCA cell line was selected for the in vivo experiments.[11] EGI-1 cells were transduced with a lentiviral vector encoding the firefly luciferase and EGFP reporter genes to enable detection of liver engraftment by in vivo bioluminescence imaging and by IF in tissue sections, respectively, as previously described.[11, 12] After transduction, luciferase/EGFP-expressing EGI-1 cells were transplanted by intraportal injection into 10 male SCID mice (6-8 weeks old; Charles River Laboratories Inc., Wilmington, MA). Further details are provided in the Supporting Materials. After the development of liver metastases, mice were sacrificed and tissue samples spanning the liver parenchyma were collected to perform IHC analyses to assess (1) the presence of tumor reactive stroma accompanying CCA liver implants, (2) involvement of EMT in the formation of tumor reactive stroma by EGFP coexpression and fluorescent in situ hybridization (FISH) using both human and murine Y probes,[13] and (3) involvement of a PDGF-mediated cross-talk between CCA cells and CAFs.

Assessment of Fibroblast Proliferation and Migration

To study the functional effects of PDGF in the cross-talk between cancer cells and fibroblasts, we assessed fibroblast proliferation (methyl tetrazolium salt [MTS] assay) and migration (Boyden chamber) after direct stimulation with recombinant human (rh)PDGF-D (R&D Systems, Milan, Italy) at increasing doses (0.1, 1, 10, and 100 ng/mL), before and after administration of inhibitors of PDGFRβ, to determine a dose-response effect. PDGFRβ antagonism was achieved using the tyrosine kinase inhibitor, imatinib mesylate (1 µM for 24 hours; Cayman, Florence, Italy).[14] Furthermore, fibroblast migration induced by rhPDGF-D (100 ng/mL) was evaluated after selective inhibition of the small Rho GTPases, RhoA, Rac1, and Cdc42, and of c-Jun N-terminal kinase (JNK) (see below). As specific inhibitors, we used NSC23766 (75 nM; Cayman) for Rac1,[15] CASIN (5 µM; Xcess Bioscience, San Diego, CA) for Cdc42,[16] Y-27632 (10 µM; Sigma-Aldrich, Milan, Italy) for RhoA/ROCK,[17] and SP600125 (10 µM; Sigma-Aldrich) for JNK.[18] Proliferation and migration of human fibroblasts were also evaluated after stimulation with conditioned media from CCA cells (EGI-1, TFK-1, and CCA1). Both experiments were run before and after addition of imatinib, and migration experiments also after treatment with small interfering RNA (siRNA) for PDGF-D. siRNA for PDGF-D was performed in EGI-1 cells using RNAiMax and StealthsiRNA (Invitrogen, Milan, Italy). See the Supporting Materials for details.

Assessment of the Downstream Effectors of PDGF-D Signaling in Fibroblasts After Stimulation With rhPDGF-D

Human fibroblasts were exposed to increasing doses of rhPDGF-D (0.1, 1, 10, and 100 ng/mL) for 24 hours. Among the potential effectors of PDGF-D signaling, we assessed extracellular signal-regulated kinase 1/2 (ERK1/2) and JNK by western blotting of total cell lysates, and RhoA, Rac1, and Cdc42 by G-LISA. ERK1/2 (regulating cell proliferation) and the Rho GTPases (regulating cell migration) are downstream effectors of two major signaling pathways activated by PDGF-D, mitogen-activated protein kinase (MAPK) and phosphatidylinositol 3-kinase/protein kinase B (PI3K/Akt), respectively.[8] JNK is a MAPK family member induced in response to a number of growth factors. These signaling molecules were evaluated before and after imatinib treatment. See the Supporting Materials for details.

Statistical Analysis

Results are shown as mean of “x” experiments ± standard deviation. Statistical comparisons were made using Student t tests or Wilcoxon-Mann-Whitney's two-sample rank-sum test. In Wilcoxon-Mann-Whitney's two-sample rank-sum test, the P value was obtained from the exact permutation null distribution. Statistical analysis was performed using SPSS 16.0 software (SPSS, Inc., Bologna, Italy); P values <0.05 were considered as significant.


Partial Expression of EMT Phenotypic Markers in Resected Human CCA Samples

The amount of tumor reactive stroma, measured as the percentage of the α-SMA-positive area present within the boundaries of the neoplastic area, was homogeneously represented in all CCA samples (11.11% ± 4.70%) (Table 1; Supporting Fig. 1). Several phenotypic features of EMT were present in CCA bile ducts, but morphologic criteria supporting a complete transition toward a mesenchymal phenotype (coexpression of K7 and α-SMA) were never met. No EMT phenotype differences were observed between intra- (n = 10) and extrahepatic (n = 5) CCA (Table 1; Supporting Fig. 1).

Figure 1.

Bioluminescence imaging and histological assessment of EGI-1 cells after xenotransplantation into SCID mice. High correspondence between the bioluminescence signal in the liver (A) and the macroscopic detection of liver tumors at autopsy (arrows, B) after xenotransplantation of EGI-1 cells into SCID mice by intraportal injection. Before transplantation, EGI-1 cells were transduced with lentiviral vectors encoding the firefly luciferase gene. Mice were sacrificed once the bioluminescence signal intensity in the liver reached a value >105 p/sec/cm2/sr. Histological analysis of liver metastases showed that EGI-1 cells laid embedded in a rich stroma (hematoxylin and eosin staining, C). By dual IF for EGFP (green) and α-SMA (red), we showed that CAFs were strictly adjacent to EGI-1-derived tumors, but coincident labeling between EGFP and α-SMA was never observed (D). In liver tumors formed by EGI-1 cells, FISH showed that α-SMA-positive CAFs (green, E and F) coexpressed mouse (red, white arrows, E), but not human (red, F) Y Chr, which was instead expressed by tumoral EGI-1 cells (F). High specificity of both Y-probes was confirmed in preliminary experiments. Original magnification: C-F: ×200; insets in E and F: ×400.

Lack of Evidence of Complete EMT in Liver Tumors Derived From Xenografted EGI-1 Cells

EGI-1 cells were xenotransplanted in SCID male mice after transduction with lentiviral vectors encoding firefly luciferase and EGFP to detect tumor engraftment in the liver in vivo (Fig. 1). Nine of ten xenotransplanted SCID mice developed a luminescent signal over the liver area 30-150 days postxenotransplantation. One animal died at day 55 before developing a detectable luciferase signal. Once the bioluminescent signal intensity in the liver reached a value >1 × 105 p/sec/cm2/sr, tumor-bearing mice were sacrificed at a median of 71 days after xenotransplantation (range, 50-155). Fig. 1A,B shows the correspondence between the bioluminescent signal and the macroscopic presence of liver tumors. Liver tumors were analyzed by dual IF for EGFP (expressed by transplanted EGI-1 cells) and α-SMA (myofibroblast/CAF marker). Xenotransplanted cancer cells that underwent a complete EMT would be expected to coexpress EGFP and α-SMA. EGFP-positive, EGI-1-derived tumors were found embedded in abundant stroma, rich in α-SMA-positive cells strictly adjacent to tumor cells (Fig. 1C,D). However, coincident labeling between EGFP and α-SMA was never observed (Fig. 1D). In selected mice, a FISH analysis was performed using both human and mouse Y-probes for their coexpression with CAFs to confirm the above-mentioned results. Preliminary studies in mouse (n = 2) and human liver specimens (n = 2) indicated that both Y-probes were highly specific and did not cross-react between the two species. Consistent with the EGFP data, α-SMA-positive cells expressed the mouse, but not the human, Y-probe, which was instead normally expressed by infiltrating EGI-1 cells (Fig. 1E,F). These data demonstrate that CAF-infiltrating liver metastases are not generated through an EMT of xenografted EGI-1 cells.

In Human CCA Samples, CCA Cells Express PDGF-A, -D, and PDGFRα, Whereas CAFs Express PDGFRβ

IHC analysis of CCA specimens showed that neoplastic bile ducts were strongly positive (>70% of ducts) for PDGF-A (Fig. 2A), PDGF–D (Fig. 2D), and for PDGFRα (Fig. 2E), but only weakly positive (<30%) for PDGF-B (Fig. 2B), and negative for PDGF-C (Fig. 2C) and PDGFRβ (Fig. 2F). CAFs, identified as α-SMA-positive cells localized in close vicinity to neoplastic ducts outside of vascular structures, were instead extensively positive for PDGFRβ (Fig. 2F), whereas their expression of PDGFRα was patchy (Fig. 2E). In extratumoral liver samples, bile ducts were consistently negative for PDGF ligands and receptors (not shown). This reciprocal expression of the members of the PDGF family between neoplastic bile ducts and CAFs suggests a role for PDGF-mediated cross-talk in CAF recruitment. In addition to CCA cells, IF studies showed that PDGF-D was also expressed by a fraction of CD45-positive inflammatory cells, scattered within the tumor reactive stroma, whereas it was negative in ECs (Supporting Fig. 2A,B). This finding indicates that inflammatory cells populating the stromal microenvironment behave as an additional paracrine source of PDGF-D (Table 1; Fig. 2 and Supporting Fig. 2).

Figure 2.

Expression of PDGF ligands and receptors in human CCA samples. Neoplastic bile ducts (K7, green, A-E) were strongly positive for PDGF-A (red, A) and -D (red, D), weakly positive for PDGF-B (red, B), and negative for PDGF-C (red, C). In addition, neoplastic bile ducts expressed the PDGFRα, though extra-CCA PDGFRα staining was also found in some scattered surrounding cells, including CAFs (red, E). On the other hand, CAFs (α-SMA, green, F) were strongly decorated by PDGFRβ (red, F), which was negative in CCA cholangiocytes. Coincident staining appears in yellow. Original magnification: A-F: ×200; insets in D-F: ×400.

Cultured CCA Cell Lines Secrete PDGF-A and -D

The immunophenotype of cultured CCA cells assessed by ICC reproduced the expression pattern of the neoplastic bile ducts observed in CCA histological sections (Supporting Table 2; Supporting Figs. 3 and 4). Expression of PDGFRs was confirmed by western blotting. ELISA was used to assess the secretory functions of the different PDGF isoforms. PDGF-B secretion was undetectable in both CCA cells and controls; secretion of PDGF-A was similar between CCA cells and controls, whereas PDGF-D was variably secreted only by CCA cells (from 65.88 to 420.52 pg/mL), and was undetectable in controls (Table 2).

Table 2. Levels of PDGF-A, -B, and -D Secretion by Established (EGI-1, TFK-1, and HuCCT-1) and Primary (CCA1, CCA2, and CCA3) CCA Cell Lines and Control Cholangiocytes (Control) Measured by ELISA (pg/mL)
  1. Abbreviation: ND, not detectable.

  2. a

    P < 0.05 versus control.

EGI-1 (n = 8)484.71 ± 128.51ND420.52 ± 120.23a
TFK-1 (n = 5)484.71 ± 125.516.81 ± 5.76108.75 ± 29.02a
HuCCT-1 (n = 5)50.72 ± 84.66a4.23 ± 2.8296.89 ± 24.94a
CCA1 (n = 5)556.30 ± 160.51ND138.16 ± 26.18a
CCA2 (n = 5)680 ± 160.515.47 ± 5.9265.88 ± 22.96a
CCA3 (n = 5)249.38 ± 77.40a2.36 ± 2.0187.17 ± 40.60a
Control (n = 5)380 ± 51NDND
Figure 3.

Proliferation (A) and migration (B) of human fibroblasts stimulated by conditioned media from CCA cholangiocytes. Proliferation of human fibroblasts was evaluated by MTS assay and is expressed as absorbance at 490 nm (A), whereas recruitment was evaluated by Matrigel-coated transwell chambers and is expressed as number of transwell-invaded nuclei (B). (A) Conditioned media obtained from EGI-1, TFK-1, and CCA1 cholangiocytes induced only a slight proliferative response with respect to starved fibroblasts and to control cholangiocytes (ctrl) (black columns: EGI-1, 132.33 ± 5.05; TFK-1, 144.33 ± 2.94; CCA1, 140.33 ± 7.09, versus control, 123.67 ± 6.19 and starved, 114.33 ± 6.15; n = 6 experiments). This effect, although small, was significantly reduced after treatment of cultured cells with imatinib mesylate (1 µM) (A) (gray columns: EGI-1, 119.83 ± 7.22; TFK-1, 135.33 ± 6.31; CCA1, 129.17 ± 10.17; control, 117.83 ± 6.74; n = 6 experiments). (B) Conditioned media from the same cell lines induced a potent chemotactic response on human fibroblasts with respect to ctrl (black columns: EGI-1, 45.9 ± 4.01; TFK-1, 44.17 ± 3.59; CCA1, 43.82 ± 3.99 versus control, 11.76 ± 9.83; n = 3 experiments), an effect significantly reduced by imatinib mesylate (1 µM) (gray columns: EGI-1, 35.08 ± 3.33; TFK-1, 37.83 ± 1.55; CCA1, 36.82 ± 0.41, control, 9.16 ± 7.62; n = 3 experiments). A similar significant reduction in fibroblast invasion was achieved using conditioned media from EGI-1 cells treated with PDGF-D siRNA, as compared with EGI-1 scramble (siRNA1, 37.45 ± 2.08; siRNA2, 35.45 ± 4.70 versus EGI-1 scramble, 44.98 ± 2.30; n = 3 experiments). *P < 0.05 treated versus control; **P < 0.01 treated versus control; #P < 0.05 treated versus IM; ##P < 0.01 treated versus IM; ^^P < 0.01 treated versus starved; §P < 0.05 siRNA versus scramble. IM, imatinib mesylate.

Figure 4.

Activation of ERK1/2 and JNK in human fibroblasts stimulated with rhPDGF-D. (A) By western blotting, stimulation of human fibroblasts with increasing doses of rhPDGF-D resulted in a mild increase of p-ERK1/2/ERK1/2, which reached significance only at the highest doses (black columns; 0.1 ng/mL, 0.47 ± 0.05; 1 ng/mL, 0.57 ± 0.08; 10 ng/mL, 0.67 ± 0.03; 100 ng/mL, 0.94 ± 0.09 versus untreated, 0.40 ± 0.14; n = 5 experiments), and was inhibited after the addition of imatinib (1 µM) (gray columns; untreated, 0.33 ± 0.29; 0.1 ng/mL, 0.30 ± 0.26; 1 ng/mL, 0.29 ± 0.25; 10 ng/mL, 0.51 ± 0.09; 100 ng/mL, 0.58 ± 0.04). (B) In contrast with ERK1/2, JNK was activated by all doses of rhPDGF-D (black columns; 0.1 ng/mL, 0.66 ± 0.21; 1 ng/mL, 0.93 ± 0.17; 10 ng/mL, 1.07 ± 0.34; 100 ng/mL, 0.91 ± 0.33 versus untreated, ND; n = 4 experiments), an effect that was significantly reduced by imatinib (1 µM) (gray columns; untreated, 0.27 ± 0.20; 0.1 ng/mL, 0.18 ± 0.23; 1 ng/mL, 0.32 ± 0.27; 10 ng/mL, 0.27 ± 0.26; 100 ng/mL, 0.36 ± 0.25). *P < 0.05 treated versus untreated; **P < 0.01 treated versus untreated; #P < 0.05 treated versus imatinib (1 µM); ##P < 0.01 treated versus imatinib (1 µM). IM, imatinib mesylate. Columns of bands in the western blots below are respective of each of the five conditions displayed in the graph. ND, not detectable.

PDGF-D Secretion Is Further and Significantly Increased in Cultured PDGF-D- Secreting CCA Cells After Chemically Induced Hypoxic Stimulus

In CCA cells with high PDGF-D secretion (EGI-1, TFK-1, and CCA1), PDGF-D secretion was also measured in conditions of chemical hypoxia after treatment with DMOG. In all three CCA cell lines, DMOG induced a further and significant increase in PDGF-D secretion (Supporting Fig. 5A), greater than 3 times, and was associated with a significant up-regulation of hypoxia-inducible factor (HIF)−1α of the same degree (Supporting Fig. 5B).

Figure 5.

Dose-response activation of RhoA, Rac1, and Cdc42 in human fibroblasts stimulated with rhPDGF-D. Human fibroblasts were stimulated for 1 minute with increasing doses of rhPDGF-D (0.1, 1, 10, and 100 ng/mL) to assess a dose-response effect. Levels of activation are expressed as normalization on untreated cells. A linear dose-dependent increase was observed for Rac1 (B, black columns; 0.1 ng/mL, 2.41 ± 0.86; 1 ng/mL, 3.14 ± 1.70; 10 ng/mL, 3.39 ± 1.40; 100 ng/mL, 3.93 ± 1.24), and for Cdc42 (C, black columns; 0.1 ng/mL, 1.73 ± 0.28; 1 ng/mL, 1.77 ± 0.22; 10 ng/mL, 2.26 ± 0.61; 100 ng/mL, 2.56 ± 0.49) that was significant from the lowest dose. In contrast, activation of RhoA was observed only at the highest doses (A, black columns; 0.1 ng/mL, 1.29 ± 0.32; 1 ng/mL, 1.39 ± 0.49; 10 ng/mL, 2.08 ± 0.50; 100 ng/mL, 2.32 ± 0.47). Imatinib (1 µM) blunted the activating effects of rhPDGF-D in all cases (gray columns; RhoA, 0.1 ng/mL, 1.10 ± 0.28; 1 ng/mL, 0.97 ± 0.25; 10 ng/mL, 1.02 ± 0.51; 100 ng/mL, 1.08 ± 0.28; Rac1, 0.1 ng/mL, 1.28 ± 0.37; 1 ng/mL, 1.49 ± 0.48; 10 ng/mL, 1.30 ± 0.26; 100 ng/mL, 1.43 ± 0.66. Cdc42, 0.1 ng/mL, 1.11 ± 0.07; 1 ng/mL, 1.04 ± 0.37; 10 ng/mL, 1.31 ± 0.11; 100 ng/mL, 1.05 ± 0.16) (n = 4 experiments). *P < 0.05 treated versus untreated; **P < 0.01 treated versus untreated; #P < 0.05 treated versus imatinib (1 µM); ##P < 0.01 treated versus imatinib (1 µM). IM, imatinib mesylate.

PDGF-A, -D, and PDGFRα Are Expressed by Xenografted EGI-1 Cells, Whereas PDGFRβ Is Expressed by Infiltrating CAFs

Specimens from xenotransplanted CCA were analyzed by dual and triple IF to assess expression of PDGF ligands and receptors. PDGF-A (Supporting Fig. 6A) and -D (Supporting Fig. 6B) were expressed by EGI-1 cells, together with PDGFRα (Supporting Fig. 6C), but not PDGFRβ (Supporting Fig. 6D). Conversely, CAFs localized in close vicinity to EGFP-positive cells were diffusely and intensely decorated by the anti-PDGFRβ antibody (Supporting Fig. 6D), but unevenly by anti-PDGFRα. These findings confirmed that the reciprocal expression of ligands and receptors between cholangiocytes and CAFs observed in native CCA was maintained in our experimental model of CCA.

Figure 6.

Effects of RhoA, Rac1, Cdc42, and JNK inhibitors on migration of human fibroblasts stimulated by rhPDGF-D. Human fibroblasts treated with rhPDGF-D (100 ng/mL) showed a significant reduction in migration after treatment with chemical inhibitors of small GTPases. (A) Treatment with Y-27632 (10 µM) (RhoA inhibitor), NSC23766 (75 nM) (Rac1 inhibitor), CASIN (5 µm) (Cdc42 inhibitor), and SP600125 (10 µM) (JNK inhibitor) induced a significant reduction in migration of different degrees (43.79 ± 3.75 for RhoA antagonism, 22.53 ± 3.81 for Rac1 antagonism, 33.78 ± 3.15 for Cdc42 antagonism, and 20.09 ± 3.08 for JNK antagonism versus 52.01 ± 2.21 for PDGF-D treatment; P < 0.01 in all cases). Combined treatment with the three inhibitors of small GTPases (mix) completely abolished the PDGF-D-stimulated fibroblast migration (8.78 ± 3.97 versus untreated, 7.52 ± 3.42; P = not significant) (n = 4 experiments). It is worth noting that the spindle-shaped morphology of fibroblasts induced by PDGF-D (B) is lost after treatment with NSC23766 (C). **P < 0.01 PDGF-D treated versus untreated; ##P < 0.01 PDGF-D treated versus inhibitors. Mix, Y-27632 10 µM + NSC23766 75 nM + CASIN 5 µm.

Immunophenotyping of Human Fibroblasts

The phenotype of cultured fibroblasts isolated from alcoholic cirrhosis was similar to CAF isolated from CCA, characterized by the expression of vimentin and α-SMA, in conjunction with that of PDGFRβ (Supporting Fig. 7A-C). See the Results section of the Supporting Materials for further details.

PDGF-D Is Responsible for CAF Recruitment, Increasing Motility Without Affecting Proliferation

EGI-1, TFK-1, and CCA1 were selected for experiments with human fibroblasts (Fig. 3 and Supporting Figs. 8 and 9).[8] Effects of conditioned media from CCA cells on fibroblast proliferation (MTS assay) and migration (Boyden chamber) were studied before and after addition of imatinib, a PDGFRβ antagonist. Effects of EGI-1 cells on fibroblast recruitment were also tested after treatment with siRNA for PDGF-D, resulting in a significant down-regulation of PDGF-D secretion (of approximately 35%-40%, as compared with scramble, P < 0.01 with siRNA1, P < 0.05 with siRNA2) (Supporting Fig. 8).

Conditioned Media From CCA Cells Induce Only a Minor Increase in Fibroblast Proliferation

As compared with starved fibroblasts, or with fibroblasts exposed to conditioned medium derived from control cholangiocytes, human fibroblasts showed only a mild increase in proliferative activity after exposure to conditioned media from the different CCA cells (from 7% to 15%, as compared to control cholangiocytes) (Fig. 3A). PDGFRβ blockade induced a significant reduction in the rate of proliferating cells in fibroblasts stimulated by EGI-1 and TFK-1 (P < 0.01 and P < 0.05, respectively).

CCA Cells Secreting PDGF-D Strongly Stimulate Fibroblast Recruitment, an Effect That Is Significantly Reduced by PDGFRβ Antagonism and by PDGF-D siRNA

As compared to control cholangiocytes, all conditioned media from the different CCA cells induced a potent migration of human fibroblasts (increase of approximately 73%-74%), which reduced significantly after PDGFRβ blockade (P < 0.05 for all CCA cells) (Fig. 3B). Notably, in EGI-1 cells, both PDGF-D siRNA showed a significant reduction in fibroblast recruitment of an extent similar to PDGFRβ blocker (P < 0.05, as compared with scramble).

Human fibroblasts exposed to rhPDGF-D exhibit a similar behavior. Effects of rhPDGF-D on migration were significantly reduced when fibroblasts were exposed to imatinib. These results are detailed in the Supporting Materials and shown in Supporting Fig. 9.

The Small Rho GTPases Are Activated in Fibroblasts After Stimulation With rhPDGF-D and Are Inhibited by Imatinib

To study the signaling pathways activated by PDGFRβ in response to PDGF-D, we stimulated human fibroblasts with rhPDGF-D at increasing doses (0.1, 1, 10, and 100 ng/mL), and then modulation of phosphorylated ERK1/2 (p-ERK1/2) and phosphorylated JNK (p-JNK) expression (by western blotting) and activation of RhoA, Rac1, and Cdc42 (by G-LISA) were evaluated in the presence or absence of imatinib treatment (Figs. 4 and 5 and Supporting Fig. 10). To determine the kinetics of activation of RhoA, Rac1, and Cdc42, preliminary G-LISA experiments were run at 1, 10, 20, 30, and 60 minutes after stimulation with rhPDGF-D (100 ng/mL). PDGF-D induced a significant increase of p-ERK1/2 only at the highest doses (P < 0.05 at 10 and 100 ng/mL), those able to stimulate also fibroblast proliferation, and this effect was abrogated by imatinib (Fig. 4A). In contrast, increase of p-JNK was significant starting from the lowest doses of rhPDGF-D (0.1 ng/mL; P < 0.01) and was abolished by imatinib (P < 0.01) (Fig. 4B).

A strong PDGF-D-dependent modulation of Rho GTPases was also documented. Time-course studies (Supporting Fig. 10) showed that PDGF-D induced a strong and early activation of Rac1 (nearly 5-fold increase) at 1 minute, followed by a rapid return to basal values (Supporting Fig. 10B). RhoA kinetics also showed an early, but smaller, increase (2-fold), then fluctuated (Supporting Fig. 10A). In contrast with Rac1 and RhoA, Cdc42 remained persistently activated up to 60 minutes (nearly 4-fold increase; Supporting Fig. 10C).

We next performed a dose-response curve with increasing doses of rhPDGF-D (Fig. 5). Rac1 and Cdc42 activity gave a clear dose-dependent linear increase that was significant from the lowest dose (Fig. 5B,C), whereas RhoA was activated only at the highest doses (Fig. 5A). In all cases, GTPase activation was inhibited by imatinib (P < 0.05; Fig. 5A-C). These data strongly suggest that PDGF-D secreted by CCA cells, by interacting with PDGFRβ expressed by mesenchymal cells, induces migratory effects resulting in CAF recruitment through activation of Rho GTPases, in particular, Rac1 and Cdc42.

Treatment With Small GTPases and JNK Inhibitors Abrogated PDGF-D-Induced Fibroblast Migration

To further confirm this hypothesis, we next tested the effects of selective inhibitors of RhoA/ROCK (Y-27632), Rac1 (NSC23766), Cdc42 (CASIN), and JNK (SP600125) on fibroblast migration stimulated by PDGF-D (Fig. 6). The inhibitors induced a significant reduction in fibroblast migration of approximately 15% for Y-27632, 35% for CASIN, and up to 60% for NSC23766 and SP600125 (P < 0.001 in all cases; Fig. 6A). Notably, the combined treatment with all the small GTPases inhibitors (mix) completely abrogated the migratory effects of PDGF-D, thereby indicating a synergic effect of Rho GTPases (Fig. 6A). In addition, when Rac-1 was inhibited, PDGF-D-stimulated fibroblasts showed relevant morphological changes, characterized by the loss of the spindle-shape morphology and by the presence of short surface protrusions, consistent with a motile-halting phenotype (Fig. 6B,C).


The incidence of CCA is increasing in Western countries and accounts for 10%-20% of deaths from primary hepatobiliary malignancies. CCA is characterized by the presence of an abundant tumor reactive stroma, a feature common to other aggressive malignancies of ductal origin, such as pancreatic and breast carcinomas.

The tumor reactive stroma is the microanatomical site of multiple functional interactions between cancer cells and several kinds of host cells and thus it behaves as an important determinant of cancer invasiveness. CAFs, the main cellular component of the tumoral stroma, produce tumoral matrix and release a variety of growth factors and chemokines, which modulate tumor cell survival, migration, and invasion.[4] For example, it has been shown that CAF-derived PDGF protects CCA cells from death induced by tumor necrosis factor-related apoptosis-inducing ligand (TRAIL) in a Hedgehog-signaling-dependent manner.[5] CAFs are also an important source of matrix metalloproteinases, cathepsins, and plasminogen activators that enable cancer cells to escape from the primary site of growth.[19] Some researchers propose that factors originating from the stroma (transforming growth factor beta [TGF-β] and hepatocyte growth factor [HGF]) signal to cancer cells to undergo an EMT and to become endowed with functional properties that favor the metastatic process, such as the ability to detach from the neoplastic cluster and to migrate to and invade lymphatic or blood vessels.[20] The role of EMT in liver diseases and tumors remains unclear and controversial.[21]

In this study, CCA cells expressed several phenotypic features, known to correlate with increased motility and invasiveness, including down-regulation of E-cadherin and β-catenin and up-regulation of Snail1, Twist, and S100A4. However, there was no evidence of EMT. This conclusion is based on the lack of coexpression of K7 and α-SMA in CCA tissue sections as well as on the lack of coincidence between CCA cholangiocyte lineage markers (EGFP and human Y chromosome [Y Chr]) and an activated myofibroblast marker (α-SMA) after intraportal injection of the highly invasive EGI-1 cells into SCID mice. EGFP-positive CCA cholangiocytes expressed the human Y-probe, but did not express α-SMA, whereas α-SMA-positive CAFs expressed the murine Y-probe, rather than the human Y-probe (Fig. 1). After xenotransplantation, in spite of the immunotolerant environment, an abundant stroma formed around the CCA cholangiocytes, suggesting a direct effect of factors secreted by tumoral cells.

Several factors can regulate epithelial-mesenchymal cross-talk, including Hedgehog, Wnt, and PDGF. We present IHC and in vitro evidences suggesting that PDGF secreted by tumoral cells plays a key role on migratory properties of CAFs.

We demonstrate that PDGF-D is secreted by neoplastic, but not by control, cholangiocytes. PDGF-D is one of the players responsible for the increased migration of fibroblasts when exposed to CCA conditioned medium. In contrast with the other members of the PDGF family, PDGF-D binds only to the PDGFRβ.[22] Mechanisms leading to the up-regulation of PDGF-D in neoplastic cholangiocytes are uncertain. However, our data suggest that hypoxia may behave as a critical inducer of PDGF-D secretion, as shown by the potent stimulation exerted on CCA cells by DMOG, an agent that prevents HIF-1α degradation. This effect is in line with the typical hypovascularization featured in CCA. Our IF studies show that a subset of inflammatory cells may represent an additional source of PDGF-D released in the tumor microenvironment, albeit their PDGF-D expression is less relevant than CCA cells.

The importance of PDGF-D in cancer biology is just beginning to be understood.[23, 24] Our findings strongly suggest that PDGF-D plays a major role in promoting CAF recruitment in CCA. In fact, siRNA of PDGF-D significantly impaired the ability of CCA cholangiocytes to promote fibroblast migration. In addition, rhPDGF-D induced a clear dose-dependent effect on fibroblast migration, whereas the effect on proliferation was milder and evident only at the highest dosages. PDGFRα, which binds all isoforms except for PDGF-D, may theoretically contribute to CAF recruitment in CCA, because PDGFRα was also expressed by CAF, and EGI-1 cells were able to secrete PDGF-A. However, administration of conditioned medium from control cholangiocytes, which contained amounts of PDGF-A comparable to those produced by CCA cells, exerted only a weak effect on fibroblast transwell migration. Interestingly, whereas PDGFRα signaling plays a pivotal role in embryonic development and in fibrosis of nonhepatic conditions, PDGFRβ seems to be more relevant in activating HSCs[25] and in stimulating the production of profibrogenic growth factors and ECM components by liver myofibroblasts.

By interacting with its cognate receptor, PDGFRβ, PDGF-D can activate several signaling cascades to regulate cell survival, cell growth, cell differentiation, cell invasion, and angiogenesis.[8] Because MAPK and PI3K/Akt are two major signal transduction pathways known to be activated by PDGF-D,[8] we studied ERK1/2, JNK, and the small Rho GTPases as downstream effectors, respectively, of MAPK and PI3K/Akt, which are able to control cell proliferation (ERK1/2)[10] and migration (JNK and Rho GTPases).[18, 26] The ability of PDGF-B to induce cytoskeletal remodeling by Rac1 and JNK has recently been reported in NIH3T3 cells,[26, 27] but the effects of PDGF-D on these molecular effectors are hitherto largely unknown. Our findings show that exposure of fibroblasts even to low doses of PDGF-D strongly activates Rho GTPases and JNK, whereas expression levels of p-ERK increased only at the highest doses. These results strongly correlate with the different functional effects on fibroblast migration and proliferation of PDGF-D (as shown in Figs. 3, 5 and Supporting Fig. 9). By regulating the cytoskeleton and adhesion dynamics, the Rho GTPases are key drivers of cell migration. The time-course study of Rho GTPase activation further enforces the role of PDGF-D as a fundamental mediator of CAF recruitment. Rac1 and Cdc42 are two of the members of the family that are most activated by PDGF-D; however, they show different kinetics of activation. Rac1, which induces the assembly of actin-rich surface protrusions (lamellipodia) enabling the start of the mesenchymal cell movement (“random” migration),[27] shows a brisk, but transient, activation by PDGF-D. In contrast, Cdc42, which promotes the formation of actin-rich, finger-like membrane extensions (filopodia) regulating chemotaxis,[28] shows a significantly sustained activation. These data indicate that by activating Rac1 and Cdc42 with different time-dependent patterns, PDGF-D may potentially regulate distinct steps of CAF recruitment, including chemotaxis toward tumoral cells, a critical function in the generation of the tumor stroma. The capability of the small GTPases to orchestrate fibroblast recruitment driven by PDGF-D is confirmed by the observation that fibroblast transwell migration elicited by PDGF-D was completely inhibited by a mix of selective inhibitors (Y-27632, NSC23766, and CASIN). Although the regulation of cell motility by the Rho GTPases has been well documented in cancer cells,[29] their involvement as fundamental molecular determinants of the tumor stromal reaction has not been reported yet. In addition to Rho GTPases, fibroblast migration in response to PDGF-D is also modulated by JNK, as previously shown in murine HSCs and portal myofibroblasts.[30] Notably, our data show that specific JNK inhibition halts fibroblast migration to an extent similar to Rac1, likely indicating that both pathways act in concert to orchestrate the PDGF-D-mediated paracrine fibroblast recruitment by CCA cells.

In addition to Rho GTPase and JNK inhibitors, we found that tyrosine kinase inhibitors were also highly effective in halting fibroblast migration and proliferation induced by PDGF-D. The potential clinical usefulness of tyrosine kinase inhibitors in CCA has recently been outlined by Andersen et al.,[4] particularly in those patients where overexpression of inflammatory functions in the microenvironment is a critical signature related to a worse prognosis. Data in this study show that selective blockade of PDGFRβ with imatinib mesylate, a tyrosine kinase inhibitor already in clinical use for other indications, significantly reduces fibroblast recruitment by CCA cholangiocytes in Boyden chambers. The therapeutic relevance of specifically targeting PDGFRβ in CCA is a topic of growing interest.[5] Recently, the ability of PDGFRβ inhibitors to interfere with CAF-to-CCA paracrine signaling mediated by PDGF-BB has been reported on. In fact, PDGFRβ activation promotes Hedgehog survival signaling in CCA cholangiocytes through protection from TRAIL cytotoxicity.[5] Our study further extends the role of PDGFRβ molecular targeting in CCA because it can prevent CAF recruitment induced by CCA cholangiocyte-derived PDGF-D. Notably, overexpression of PDGFRβ in the stromal compartment of CCA was related to the most significant “network connectivity” with the tumoral compartment.[4] Pharmacological targeting of tumor/stroma interactions using PDGF inhibitors may represent a novel molecularly targeted therapeutic approach in CCA.[31, 32]


The authors wish to thank Dr. Scott Swenson (Section of Digestive Disease, Yale University School of Medicine, New Haven, CT) for assistance with FISH experiments.