Gene expression profiling of cholangiocarcinoma-derived fibroblast reveals alterations related to tumor progression and indicates periostin as a poor prognostic marker

Background Fibroblasts play important roles in several cancers. It was hypothesized that cholangiocarcinoma (CCA)-associated fibroblasts (Cfs) differ from non-tumorigenic liver fibroblasts (Lfs) in their gene expression profiles resulting in the capability to promote cancer. Periostin (PN) is a multi-functional protein and has emerged as a promising marker for tumor progression. The role of PN in CCA, however, has not yet been explored. Results In this study, the gene expression profile of Cfs in comparison to Lfs was performed using oligonucleotide microarrays. The common- and unique-expressed genes in Cfs and the promising roles in cancer promotion and progression were determined. PN was markedly over-expressed in Cfs confirmed by real time RT-PCR and western blot analysis. Immunohistochemistry examination of a number of patients with intrahepatic CCA showed the expression of PN solely in stromal fibroblasts, but was expressed neither in cancer cells nor immune cells. Low to no expression of PN was observed in tissues of benign liver disease and hepatocellular carcinoma. CCA patients with high levels of PN had significantly shorter survival time than those with low levels (P = 0.026). Multivariate analysis revealed high levels of PN (P = 0.045) and presence of lymph node metastasis (P = 0.002) as independent poor prognostic factors. The in vitro study revealed that recombinant PN induced CCA cell proliferation and invasion. Interestingly, interference RNA against integrin α5 significantly reduced the cellular response to PN-stimulated proliferation and invasion. Conclusion The gene expression profile of fibroblasts in CCA is apparently explored for the first time and has determined the genes involving in induction of this cancer progression. High PN can be used to distinguish CCA from other related liver diseases and is proposed as a prognostic factor of poor survival. Regulation of fibroblast-derived PN in CCA proliferation and invasion may be considered as an alternative therapeutic approach.


Background
Cholangiocarcinoma (CCA) originates from biliary epithelial cells and is a unique cancer in northeastern Thailand where the prevalence of a liver fluke, Opisthorchis viverrini infection is higher than elsewhere in the country. A recent study showed a strong positive correlation of CCA incidence and the prevalence of O. viverrini infection [1]. In other countries, CCA has been shown to correlate with Clonorchis sinesis [2,3], and chronic biliary diseases [4]. Even though CCA is caused from the different etiologies, it is well recognized to contain an abundant fibrous stroma that is mainly composed of α-smooth muscle actin (SMA) positive fibroblasts [5,6]. In addition, the degree of α-SMA expression has been shown to correlate with the survival of patients, in part, via the ability of these cancer-associated fibroblasts to induce proliferation of bile duct epithelial and cancer cells [6].
The ability of stromal fibroblasts to generate a favorable microenvironment for cancer cells leading to cancer development, invasion and metastasis has been summarized [7,8]. Mitotic substances have been produced from stromal fibroblasts to promote tumor growth in many cancers [9,10]. In addition, some matrix metalloproteinases which are often mentioned as proteolytic substances of the extracellular matrix (ECM) and have been reported to show increased production from cancer stromal fibroblasts. These proteolytic substances appear to help to promote cancer cell invasion and metastasis [11]. Specifically, in CCA, stromal-derived factor 1 (SDF-1) has been secreted from stromal fibroblasts into the microenvironment in which it was located at the edge of cancer masses and was proposed to play important role in induction of CCA cell invasion and metastasis [12].
To understand the roles of fibroblasts in carcinogenesis, cancer promotion and progression, gene profiling of cancer fibroblasts have been studied in many cancers [13][14][15]. Stromal cancer fibroblasts from breast cancer with invasion were compared with the expression profiles of fibroblasts in benign breast disorders. HYL (Csk-homologous kinase CHK) involving in regulation of Src kinase, GM CSF-1 (granulocyte monocyte colony stimulating factor-1) and osteopontin were up-regulated which may result in induction of tumor growth and metastasis [13]. Among genes encoded for secreted proteins over-expressed in fibroblasts of human basal cell carcinoma [14], genes including CTSK (cathepsin K), SFRP2 (secreted frizzled-related protein 2), PDGFRL (platelet-derived growth factor receptor-like protein), and DCN (decorin) were shown to be up-regulated in non-epithelial cells of breast cancer [16]. In contrast, these genes could not be detected in fibroblasts isolated from cancers of pancreas [17], and liver metastases of colon cancer [18]. Taken together, it is possible to say that differential gene expression profile of cancer fibroblasts is partly similar but actually unique for each cancer type. This supports the importance of specific recognition of the concerted performance between fibroblasts and epithelial cells in carcinogenesis and progression in different organs of origin. So it is of great value to investigate the specific gene expression profile of the CCA-derived fibroblasts to help us better understand the molecular mechanisms that fibroblasts use to promote cancer.
Periostin (PN) is a secreted protein which was first identified in bone and implicated in regulating adhesion and differentiation of osteoblasts. The cancer biology role of PN has been investigated in a wide range of cancers including cell proliferation [19,20], migration [21], invasion/metastasis [22,23], and angiogenesis [23,24]. When not regarding the specific sources, either from cancer cells or fibroblasts, secreted PN has been reported to induce tumorigenic properties of epithelial cells via the activation of integrins (ITGs) receptors [21]. Even though substantial evidence has shown that cancer-associated fibroblasts are involved in tumor promotion and with the evidence that fibroblasts in CCA induce more aggressive tumorigenic properties of cancer cells [6], the role of CCA-derived fibroblasts in this cancer is yet to be determined. In the present study, fibroblasts isolated from CCA tissues or CCA-associated fibroblasts (Cfs) which were already characterized by the present group [6], were explored. The genome wide expressions of these Cfs were determined and compared to non-tumorigenic liver fibroblasts (Lfs). The altered expression of genes focusing on the impact of soluble products from Cfs on the promotion and progression of CCA was investigated. Interestingly, PN, which has never been reported in CCA was found at a high level whereas no-to-low PN was detected in non-tumor liver tissues and cancer of hepatocytes. The overexpression of PN in CCA tissues was detected solely in fibroblasts and associated with poor prognosis and short survival of the patients. The effect of PN to induce cell proliferation and invasion has been examined.

Gene expression analysis of Cf and validation by real time RT-PCR
To reduce the genetic background of different patients, the gene expression profile of Cfs was compared to those of two Lfs namely Lf1 and Lf2. Lf1 was isolated from non-tumorigenic liver tissues of hepatectomized liver from the CCA patient who Cfs were originated from. The Lf2 was isolated from the other CCA patient. Genes with differential expressed levels in Cf compared to Lf1 were 3,560 for 2-fold or more up-regulation and 2,339 for 0.5-fold or less down regulation (Fig 1A and 1B). The comparison of the Cf to Lf2 was 4,579 and 3,348 for upand down-regulation. The common differential genes which are genes altered in their expressions in Cfs when compared to both Lf1 and Lf2 (Cf/Lfs), were 1,466 for up-regulation and 495 for down-regulation. Arylacetamide deacetylase (DAC), procollagen C endopeptidase enhancer 2 (PCPE2), serpin peptidase inhibitor (PAI) and S100 calcium binding protein A4 (S100A4) were predominantly over-expressed at high levels in Cfs whereas bone morphogenic protein 2 (BMP2), matrix-remodeling associated 5 (DKFZp564I1922), bradykinin receptor B1 (BRADYB1), response gene to complement 32 (RGC32) and interleukin 24 (IL-24) were down-regulated with a high array intensity ( Table 1).
The up-regulated levels of these genes in Cfs were verified by relative quantification using real time RT-PCR. In concordance with microarray data, real time RT-PCR results revealed that ADAM12, AREG, ER, JAGL1, PDGF-A, PN and SCG2 had significant up-regulations in Cfs compared to Lfs, but that of NOV was not statistically significantly increased ( Fig 1C). ANG2, LAMA5, and RL, however, showed the opposite direction to the microarray results.

Detection of PN expression in Cf and CCA tissues
Using different biological preparation lots of Cfs from those used in microarray analysis, both real time RT-PCR and western blot analysis confirmed that Cfs had higher expressions of PN than Lfs with statistical significance (Fig 2A and 2B). The expression of PN in KKU-100, KKU-M055, KKU-M156 and KKU-M213 CCA cell lines was detected at a very low level compared to the high expression in Cfs (Fig 2C).
To check whether the increased expression of PN mRNA can be found in CCA tissues, real time PCR was performed using total RNA extracted from pieces of CCA mass. Using b-actin and gapdh as the internal controls, the results showed the median of PN mRNA expression was higher in CCA tissues (4.347 and 2.449 using b-actin and gapdh respectively) than in benign liver tissues (1.064 and 1.625, respectively) ( Fig 2D). This increased up-regulation was not statistically significant. In addition, to achieve the aim to use a rapid method such as real time PCR in place of immunohistochemical detection of PN in CCA tissues, the PN mRNA level was related to the intensity of PN immunoreactivity detected by immunohistochemistry. The results indicated the positive correlation of PN mRNA level and the encoding protein found in CCA tissues with statistical significance (P = 0.045) (Fig 2E).

Expression of PN in CCA tissues and clinicopathological relevance
Immunohistochemistry revealed that the expression of PN was exclusively localized in fibroblasts but not cancer cells (Fig 3). Of all 52 cases, 43 cases or 83% were PN  (Table 4). Among these positive cases, 58% of them showed high expression levels. High expression of PN was observed in well- (Fig 3A), moderately-( Fig 3B) and poorly-differentiated malignant tissues (Fig 3C). For PN-negative CCA tissues, only 17% (9/52) were in this group in which no PN was detected in either fibroblasts or cancer cells (Fig 3D). In contrast, benign liver tissues showed no (2/8) to slight (6/8) PN expression. Similar to benign liver tissues, hepatocellular carcinoma revealed low PN expression in their stromal cells (Fig 3E and 3F). Moreover, double immunofluorescence staining revealed co-localization of α-SMA and PN in the fibroblasts within cancerous area (Fig 3G).
Cumulative survival of CCA patients with low or high PN expression in cancer stromal fibroblasts was analyzed using the Kaplan-Meier method. The patients with  survival time under 14 d were identified as peri-operative deaths (n = 1) and excluded from the analysis. Median survival time was 395 ± 157 d for patients with low and 179 ± 35 d for patients with high PN expression. We found that the patients with high PN positive fibroblasts had statistically significantly shorter survival times than those with low PN positive fibroblasts (P = 0.026) (Fig 4). The prognostic value of PN expression and other clinicopathological factors among CCA patients was analyzed using multivariate Cox Proportional Hazard Regression model. The results revealed that high PN expression (HR = 2.02, P = 0.045), and the presence of lymph node metastasis (HR = 3.13, P = 0.002) were the independent risk factors for the overall survival of CCA patients after hepatectomy (Table 5). However, lymph node metastasis and other clinical data showed no association with PN expression (Table 6).

PN promotes proliferation and invasion of CCA cells
PN could induce proliferation of KKU-M156, KKU-M213 and KKU-M055 CCA cell lines (Fig 5A-C), but not KKU-100 ( Fig 5D). In addition, KKU-M156, KKU-M213 and KKU-M055 responded to the proliferative effect of optimal PN concentration in a time dependent manner with statistical significance at the 24 h-treatment for all cell types ( Fig 5E). To reinforce the proliferation effect of PN on CCA cell lines, colony formation assay with and without soft agar were performed and the result indicated the increased numbers of colonies in the condition of PN treatment in comparison to the negative control without PN stimulation ( Fig 5F). In addition, flow cytometric analysis indicated an increased number of KKU-M213 and KKU-M156 cells distributed in S+G2/M when exposed to PN (Fig 6A and 6B).
To address the invasion effect of PN on CCA cells, the invasion assay of cell lines with high ITGa 5 expression was performed in a Boyden chamber. The results showed that exogenous PN could markedly induce invasion of KKU-M156 and KKU-M213 CCA cell lines up to around 210% and 230% of cells without PN treatment (Fig 6C).

Knockdown of ITGa5 attenuates PN-induced proliferation and invasion
Treatment of CCA cells with siITGa 5 and lipofectamine (mock) for 6 h did not affect cell viability (Fig 7A). The  reduction of ITGa 5 expression was observed to be 88% of that expressed in both KKU-M213 and KKU-M156 CCA cells without transient knockdown of this gene (Fig 7B). The knockdown effect could be detected up to 72 h after siITGa 5 treatment (data not shown). Thus, the subsequent investigations of cell proliferation and invasion were done within 72 h after transient knockdown with siITGa 5 . The reduction of ITGa 5 expressions in both KKU-M213 and KKU-M156 CCA cells resulted in a significant decreased response of cells to PN-induced cell proliferation and invasion (Fig 7C and 7D). A 104% of KKU-M213 cell proliferation induction was detected in siITGa 5 -treated cells exposed to PN, whereas cell proliferation could increase up to 130% in cells with intrinsic ITGa 5 expression (Fig 7C). In the same manner, KKU-M156 showed 118% and 98% of cell proliferation induction observed in cells untreated and treated with siITGa 5 . Both ITGa 5 -knockdown CCA cell lines did not respond to PN-activated cell invasion whereas PN dramatically induced invasion of both cell lines having normal intrinsic ITGa 5 expression (168% for KKU-M156 and 172% for KKU-M213) (Fig 7D). Finally, cells with high ITGa 5 expressions were more susceptible to PN stimulation to proliferate and invade than cells without or minimal ITGa 5 expression.

Discussion
Cancer-associated fibroblasts have been recognized for their impact in the genesis, promotion and progression of many carcinomas and highlighted in several reviews [8,25]. CCA is notoriously associated with dense  desmoplastic stroma with activated fibroblasts [5,6]. Relatively little, however, is known about the contribution of the stromal fibroblasts to CCA. The authors in the present group have recently shown that CCA stromal fibroblasts, with and without direct interaction with cancer cells could induce cancer cell proliferation [6].
Herein, this study focused on the gene expression profile of CCA-derived fibroblasts in order to investigate the molecular mechanism of how fibroblasts induce a favorable microenvironment to promote cancer. Even though the current study is limited to a single cancer fibroblast line isolated from a single CCA patient, the validity of array results was strengthened by comparing gene expression levels in cancer fibroblasts to the two lines of normal fibroblasts; one isolated from the same CCA patient and the other from a second patient. Only genes in cancer fibroblasts altered from both normal fibroblast lines were investigated as the common up-or down-regulated genes. This is to provide evidence that the fibroblasts used in our study are valid representatives of fibroblasts found in CCA. By comparing gene profiles in fibroblasts from CCA with those of other tumor types, it is suggested that CCA fibroblasts display not only common genotypes for activated cells but also unique characteristics. Genes involved in metabolism of cells needed to be up-regulated in order to support the active function of CCA stromal fibroblasts to produce many supporting proteins in the cancer environment. Neuropeptide Y receptor Y1 has been indicated to receive the activation signal to induce neuroproliferation [26] and doublecortin-like kinase 1, a microtubule-associated active protein kinase expressed in growth cones of postmitotic neurons [27] may help facilitate fibroblast proliferation. In similar to human basal cell carcinoma fibroblasts [14], SPARC or osteonectin, was also over-expressed in CCA-derived fibroblasts. SPARC-null mice were recently demonstrated to resist UV-induced squamous cell carcinoma, suggesting a tumor-promoting role of SPARC [28]. In contrast to the cancer-associated fibroblasts in metastatic colon cancer to the liver which showed down-regulation of SDF-1 [18], CCA-derived fibroblasts had upregulated SDF-1 (data not shown).
The highly up-regulated genes in CCA-derived fibroblasts showed several interesting functions involved in cancer progression. Serpin peptidase inhibitor, clade B member 2 (SERPINE2) or plasminogen activator inhibitor type 2 (PAI2) is involved in cancer invasion and metastasis by controlling serine protease urokinase plasminogen activator. In a recent review, several studies led to the suggestion that the significance of PAI2 expression on prognosis of cancers is organ context-dependent [29]. In breast cancer, PAI2 was expressed in both stromal and tumor cells and associated with prolonged disease-free survival [30]. In contrast, high levels of PAI2 in endometrial cancer were reported to correlate with the invasion potential of the cancer [31]. S100 calcium binding protein A4 (S100A4) has been revealed as the metastasis-inducing protein [32]. Genes such as procollagen C-endopeptidase enhancer 2 (PCPE2) were also detected which may involve in collagen synthesis [33]. These results support the function of fibroblasts in CCA to promote a desmoplastic reaction. For down-regulated genes, bone morphogenetic protein 2 (BMP2A), a multi-functional growth factor belonging to the transforming growth factor-β superfamily was decreased in CCA fibroblasts as reported in breast cancer-derived fibroblasts [13]. BMP2A-encoding protein has been elucidated to induce hypophosphorylation of retinoblastoma protein causing cell cycle arrest [34]. Hence, decreased BMP2A in the CCA microenvironment may promote cancer cells to enter the cell cycle. Moreover, a decreased level of interleukin 24 (IL-24), an apoptotic inducible cytokine [35], in cancer tissues, attenuates cancer cells from undergoing apoptosis. The response gene to complement 32 (RGC32), a novel p53inducible gene, and bradykinin receptor B1 (BRADYB1) decreased expression in CCA-derived fibroblasts. Being intracellular protein and membrane receptor, respectively, RGC32 and BRADYB1 have elucidated the function of inhibition of fibroblast cell proliferation [36,37]. It seems possible to conclude that down-regulated genes in fibroblasts encode proteins, if acting in the intracellular region, can inhibit the proliferation of fibroblasts themselves, but if they exist in the extracellular region, they may involve inhibition of cancer cell proliferation. This evidence strengthens the roles of fibroblast-derived pro-  teins released into a tumor environment to induce a high proliferative capability of cancer cells.
Fibroblasts have been proposed the bipolar effects in cancers [38]. In our microarray results, ADAMTS-like 1 (ADAMTSR1) was over-expressed in CCA-derived fibroblasts. The ADAMTS-like proteins have been discussed as the enhancers of ADAMTS proteases [39]. Since some ADAMTS have been proven to be anti-angiogenic factors [40] partly via the trapping of vascular endothelial growth factor by thrombospondin motifs of ADAMTS [41]. So up-regulation of ADAMTSR1 in fibroblasts may inhibit angiogenesis. Moreover, stromelysin-1 or MMP-3 which can degrade ECM and induce cancer invasion and metastasis, showed the decreased expression in fibroblasts. Taken together, the increased expression of ADAMTSR1 and the decreased expression of MMP-3 may highlight fibroblasts in term of suppressing CCA progression.    Theoretically, proteins secreted from fibroblasts having interplay with cancer cells could be detected in the extracellular region and be involved in ECM organization and biosynthesis. Within these 2 groups of genes, we focused our interest on genes encoded secreted proteins and their products have been previously reported of their tumorigenic effects. ADAM12, AREG, AGN2, ER, JAGL1, LAMA5, NOV, PDGF-A, PN, RL, and SCG2 were selected to explore. AREG, ER, JAGL1, and LAMA5 are predominantly reported for proliferation induction in cancer cells [42][43][44][45]. PDGF-A, NOV, AGN2, and SCG2 are involved in angiogenesis [46][47][48][49], whereas ADAM12 and RL play an important role in cell motility, invasion and metastasis [50,51]. For PN, many carcinogenic functions including cell proliferation, invasion, metastasis and angiogenesis have been demonstrated [20][21][22][23][24]. This study employed real time PCR to verify the up-regulation of these genes, and found that only ADAM12, AREG, ER, JAGL1, PDGF-A, PN and SCG2 were significantly increased in their expression levels in CCA-derived fibroblasts and may promote CCA progression through activation of cancer growth, invasion and angiogenesis.
Herein PN was chosen to deeply explore since its well accepted multifunction in cancer as mentioned above. Moreover, the result from our group about the expressions of ADAM12, AREG, ER, JAGL1, PDGF-A, PN and SCG2 in whole CCA tissues (n = 20) showed that only AREG, PDGF-A and PN had higher level in cancer than those in benign liver tissues with statistical significance (data not shown). AREG and PDGF-A could be detected in not only fibroblasts but also in cancer and endothelial cells, however PN expressed exclusively in CCA fibroblasts. In order to demonstrate role of fibroblast-derived proteins in CCA, we determined that PN should be the first target to explore.
The strong evidence using different biological preparations of CCA-derived fibroblasts and CCA tissues confirmed the increased levels of PN at both mRNA and protein. Most of CCA tissues of all differentiated types had high levels of PN and expressed exclusively in α-SMA positive fibroblasts. In the same direction, the findings showed no expression of PN in CCA cell lines when compared to the high level expressed in the fibroblasts. This may strengthen the results of the absence of PN in cancer cells in CCA tissues. From these results taken together, it can be concluded that PN detected in CCA tissues is only of fibroblast origin as reported in some cancers [52][53][54]. In cancers of head and neck, ovary, and colon, PN was found in cancer cells and has been proposed to induce tumorigenic properties of cancer cells via an autocrine mechanism [21,22]. Hence results from the present study allow the speculation to propose a phenomenon that fibroblast-derived PN in CCA may affect cancer cells by a paracrine mode and has a promising role in cancer promotion. These results revealed that a high PN level in fibroblasts was an independent risk factor in CCA patients and those having high PN had significantly low cumulative survival time after surgery. PN might therefore be used as a poor prognostic marker in patients suffering from CCA. Detections of PN at both mRNA and encoding protein in CCA tissues are in the same direction to distinguish CCA from non-cancer syndromes of bile ducts. In addition, most benign liver tissues and hepatocellular carcinoma showed no to only a minimal expression of PN when compared to the high level detected in CCA tissues. Hence, serum PN may help to distinguish CCA from benign conditions and closely-related liver cancer and may use as the prognostic or predictive marker as previously reported [52,55].
To show the tumorigenic impacts of PN on CCA cells, recombinant PN was employed as extracellular PN to mimic the paracrine effect of PN produced from cancer stromal fibroblasts to induce CCA cell proliferation and invasion. Though receptors ITGα v β 3 and ITGα v β 5 have been shown to be the receptors for PN in several cancer cells [21], PN promoted invasiveness of pancreatic cancer cells via the β 4 integrin [56]. This suggests the cell type dependent on a specific ITG responded to PN. The study herein reveals that PN-induced cell proliferation and invasion could be inhibited by RNAi against ITGa 5. Hence, ITGα 5 is a potentially promising receptor for PN in CCA cells. As the well known receptor for fibronectin, the apparent reason for ITGα 5 production in CCA cells is to support the abundance of fibronectin found in CCA [57]. In addition, ITGα 5 can only be from dimerization with the β 1 subunit and activation of ITGα 5 β 1 has been revealed to support cell survival [58] and induce invasion and angiogenesis [59,60]. Though further studies need to be performed before such a conclusion is valid in CCA, this work highlights the PNinduced-ITGα 5 pathway as one of the activated pathways to induce an aggressive CCA.
TGF-β has been proposed to induce the expression of PN [61]. O. viverrini excretory/secretory product has also recently been shown to be the stimulator of fibroblast proliferation via the TGF-β-mediated signal transduction pathway [62] and this pathway seems likely to be the cause of PN expression in CCA-derived fibroblasts. The authors' laboratory has checked the effect of parasitic product-treated fibroblasts and found that these fibroblasts increased PN expression compared to the normal liver fibroblasts without exposure (unpublished data). It is interesting to propose that in CCA cells; the expression of fibroblast-derived PN could be induced by TGF-β produced from infected parasites since the early stage of carcinogenesis and may be in concert with TGF-β produced from CCA cells in a late stage of cancer [12]. Though in vivo experiments are needed to confirm, fibroblast-derived PN may influence O. viverrini-associated CCA at the early stage of cancer as well as to promote cancer progression in the later time. With this information, targeting the stroma in CCA may not only be effective in treatment of primary, invasive and metastatic tumors, but may also play role in prevention of tumor development.

Conclusions
To the authors' knowledge, this study is the first to describe the gene expression profile of CCA-derived fibroblasts. Molecular understanding of fibroblasts in CCA by the functions of certain up-and down-regulated genes has been revealed and has suggested certain groups of genes in controlling cancer cell proliferation, invasion, metastasis and angiogenesis (Fig 8). These findings provide evidence that fibroblasts are important sources of tumorigenic substances, particularly PN, when produced into the microenvironment of CCA. High levels of PN are found in most CCA patients and can be used as a poor prognostic marker. In addition, the level of PN can be used to distinguish CCA from other benign liver conditions and hepatocellular carcinoma. The interaction of fibroblast-derived PN and CCA cells helps to promote cell proliferation and invasion probably via ITGα 5 . Though further investigations are needed, this study suggests promising evidence of the value of using serum PN as a prognostic marker of poor survival in CCA patients. Moreover, targeting fibroblasts or fibroblast-derived-PN-stimulated pathways in cancer cells to attenuate the tumorigenic induction of PN is a further challenge to inhibit CCA progression in the patients.

Cells and culture condition
Cfs and Lfs were established and characterized in this lab as previously reported [6]. CCA cell lines including KKU-M213; KKU-M156; KKU-M055; and KKU-100 were received as a kind gift from Associate Professor Dr. Banchob Sripa (Khon Kaen University). Cfs and Lfs were grown in the complete media which is 10% FBS containing DMEM with 20% epidermal growth factor (CytoLab Ltd., Rehovot, Israel). CCA cells were cultured in 10% FBS containing Ham-F12 (Invitrogen, Carlsbad, CA) supplemented with antibiotics and an anti-fungal agent at 37°C and in 5% CO 2 incubator.

Gene expression profiling study and data analysis
Total RNA was extracted using RNeasy Micro Kit (Qiagen, Valencia, CA) using the manufacturer's instructions. The quality of the RNA was assessed by an Agilent RNA 6000 Nano Kit (Agilent Technologies, Waldbronn, Germany). Affymetrix GeneChip Human Genome U133 plus 2.0 Array containing 38,500 human genes (Affymetrix, Santa Clara, CA) was used following Affymetrix's instruction. The array data were scanned by a GeneChip Scanner 3000 (Affymetrix) and analysed by Affymetrix microarray suite, version 5.0. Raw data from the GeneChips were used to analyze expression levels and expressed as fold changes and gene ontology was categorized by Gene Spring G.X.7.3 software (Agilent Technologies). Average fold change of gene expression was determined by intensity comparison between Cf and Lf1 and Lf2.

Semi-quantitative real time PCR
Complementary DNA was synthesized from 1 μg of total RNA using the first strand cDNA synthesis kit (AMV) (Roche Molecular Biochemicals, Mannheim, Germany) according to the instructions. Relative expression levels in genes of Cfs and Lfs were determined by SYBR Green-based real time PCR using ABI 7500 (Applied Biosystem, Foster City, CA) and calculated by the 2 -ΔΔC T equation. In this case, ΔC T = C T (Cf)-C T (Lf). b-actin served as an internal control to adjust the amount of starting cDNA. The sequences of genes tested in this study were retrieved from PubMed http:// www.ncbi.nlm.nih.gov and the primers were designed by Primer 3 (Table 7).

Human CCA tissues and immunohistochemistry
Fifty-two cases of CCA tissues were obtained from patients who had undergone hepatectomy using the protocol approved by the Human Research Ethics Committee, Khon Kaen University (HE490143). The age, sex, tumor size, histological type and staging data were derived from the medical charts and pathological records. Benign liver tissues were characterized as chronic inflammation by other causes rather than CCA.
Paraffin-embedded tissues were used and antigens were retrieved in 10 mM citrate buffer pH 6.0 at 95°C for 40 min and endogenous peroxidase was blocked in 3% H 2 O 2 for 5 min. After blocking non-specific binding with 2% bovine serum albumin for 20 min, 1:10,000 rabbit anti-human PN (Biovendor, Heidelberg, Germany) was applied to the sections at room temperature overnight, followed by anti-rabbit Envision + System-HRP labeled polymer (Dako, Carpinteria, CA) for 30 min at room temperature. The immunoreactive signal was developed by diaminobenzidine (DAB; Sigma, St Louis, MO) and counterstained with hematoxylin. The signal was checked under light microscope.
PN expression of intratumoral fibroblasts on the histologic sections was semi-quantitatively scored on the basis of PN-positive fibroblasts percentage and the immunostaining intensity. The number of positive fibroblast cells were classified as < 10% (negative); 10-25% (+1); 26-50% (+2); and >50% (+3). The intensity of PN expression in fibroblasts was scored no staining, 0; weak staining, 1; intermediate or focal weak and focal intense staining, 2; intense staining, 3. The interpretation of PN expression was performed by summarization the scores of the percent positive cell (0-3) and the scores of staining intensity (1-3) to reach the total final score of 0-6. The results were then categorized as follows; low expression, score ≤ 4; and high expression, score > 4. All samples were anonymized and independently scored by one pathologist (KC) and 2 investigators (PT and CT). In case of disagreement, the slides were reexamined and a consensus was reached by at least 2 observers.

Double immunofluorescence staining of a-SMA and PN
In order to localize the expression of PN and α-SMA in CCA tissues, double immunofluorescence staining was performed. The 1:200 mouse anti-human α-SMA antibody (Sigma) and 1:500 rabbit anti-human PN antibody (Biovendor) were used as primary antibodies. Anti-mouse IgG-Alexa 488 and anti-rabbit IgG-Cy3 (Invitrogen) were used as the second antibodies. Nucleus was stained with Hoechst (Invitrogen). The signal was observed under the LSM 510 Meta laser scanning confocal microscope (Carl Zeiss, Jena, Germany) at the Division of Medical Molecular Biology, Office for Research and Development, Faculty of Medicine Siriraj Hospital, Mahidol University.

Protein extraction and western blot analysis
Twenty μgs of total protein from the cell lysate were separated in 10% SDS-PAGE and transferred onto a PVDF membrane (Millipore, Billerica, MA). For PN detection, 1:500 rabbit anti-human PN (Biovendor) and 1:1,000 goat anti-rabbit conjugated HRP (Abcam, Cambridge, MA) were used. The signal was visualized by ECL (Pierce, Rockford, IL). The expression of β-actin was used as an internal control to determine an equal amount of loading proteins.

Cell proliferation assay
CCA cells with or without treatment with siITGa 5 cells were arrested in HAM-F12 without serum supplement for 12 h. Different concentrations of recombinant PN (Biovendor) prepared in 1% FBS containing HAM-F12 were incubated with cells for 6, 12 and 24 h. The viable cells in each condition were determined using an MTS assay (Promega, Madison, WI) according to the manufacturer's instruction.

Cell cycle analysis by flow cytometry
Cell cycle distribution analysis used cells stained with propidium iodide (Invitrogen) as previously described [63]. The distribution of cells in each stage of the cell cycle was quantitated in a flow cytometer and CellQuest software (Becton Dickinson, Franklin Lakes, NJ). Numbers of CCA cell lines in S+G2/M phases of the cell cycle were measured and compared between conditions with and without recombinant PN treatment. These experiments were repeated two times using replicate culture dishes in the same experiment.
Colony formation assay CCA cell lines were cultured in 6-well plate. After 24-h culture, recombinant PN diluted in 1% FBS containing medium was added and the plate was incubated in CO 2 incubator. Soft agar colony formation assay was also performed using 0.5% and 0.35% MetaPhor® agarose (Cambrex Bio Science, Rockland, ME) as lower and upper layers, respectively. After 12 d, cells were fixed with 5% v/v glutaraldehyde and stained with 0.5% w/v crystal violet in 40% v/v methanol. Cell growth was estimated by counting numbers of colonies with more than 30 cells under inverted microscope and compared between those of treat and untreated with PN. The experiment was performed in duplicate.

Invasion assay
KKU-M213 and KKU-M156 CCA cells were seeded in PN (100 ng/ml) containing medium into the Matrigel invasion chamber (BD Biosciences, San Jose, CA) and incubated for 24 h. Invaded cells were fixed with 5% v/v glutaraldehyde and stained with 0.5% w/v crystal violet in 40% v/v methanol for 30 min each. The number of invaded cells was counted under a microscope by two independent investigators using 100× magnification fields. The assays were done in replicate and three independent experiments were performed.

Statistical analysis
Statistical analyses were performed using SPSS version 16.0 (SPSS Inc., Chicago, IL). The correlation of PN expression and pathological parameters of CCA patients was analyzed by the χ 2 -test and binary logistic multivariate analysis. Patient survival was calculated from the time of surgical resection to death and the survival curves were constructed according to Kaplan-Meier, with a Log-Rank test. A multivariate analysis was performed by the Cox proportional hazard regression model. The significance of the different data was determined by the Student's t-test. A P value of equal to or less than 0.05 was defined as statistically significant.