Background
Cholangiocarcinoma (CC) is a highly lethal adenocarcinoma arising from bile duct epithelial cells. CC accounts for approximately 15% of the total liver cancer cases worldwide, and its incidence is rising [
1,
2]. The prognosis for CC is quite poor because of difficulties in early diagnosis, and relative resistance of the tumors to chemotherapy [
3,
4]. At the time of diagnosis, approximately 70% of CC patients have an occult metastasis or advanced local disease that precludes curative resection. Of candidates for curative resection, 30% develop recurrent disease at the anastomotic site or within the intrahepatic biliary tree, and succumb to disease progression or cholangitis [
5]. Established risk factors for ductal cholangiocarcinomas include primary sclerosing cholangitis, infection with
Clonorchis sinensis or
Opisthorchis viverrini (liver flukes), Calori's disease, congenital choledochal cysts, and chronic intrahepatic lithiasis [
6]. However, for most CCs, the cause is unknown.
Recently, molecular investigations have provided evidence that CC carcinogenesis involves a number of genetic alterations, including activating point mutations in the
K-ras oncogene, and in
p53 and
BRAF [
7‐
9]. The deregulated expression of a number of other genes has also been reported, and cyclooxygenase-2 and c-erbB-2 are frequently overexpressed in CCs, suggesting an involvement in early biliary carcinogenesis [
10]. In addition, increased expression of interleukin-6 is frequently observed in CC [
11]. CC also develops after the liver-specific targeted disruption of the tumor suppressors
SMAD4 and
PTEN [
12]. The incidence of sarcomatoid changes in CC is estimated to be approximately 5% [
13], and sarcomatoid cells are thought to result from de-differentiation of ordinary carcinomatous CC cells. Sarcomatoid neoplasms are highly aggressive and have a poorer survival rate than ordinary CCs [
14], but the underlying molecular alterations, which may be related to the epithelial-mesenchymal transition (EMT), remain unclear. Little extensive genome-wide information about altered gene expression in CCs is available, and only a few published studies have reported a comprehensive analysis of gene expression among biliary tract cancers in general [
15,
16]. The advancement of microarray technology now enables us to analyze genome-wide gene expression in a single experiment, opening avenues for the molecular classification of tumors, detection of the biological nature of tumors, and prediction of prognosis and sensitivity to treatments.
In this study, we generated genome-wide gene expression profiles of 10 cell lines (9 CC cell lines and 1 immortalized cholangiocyte line), and 19 CC tissues using a BeadChip oligonucleotide technology containing 48,000 genes. This procedure allowed us to observe a comprehensive pattern of gene expression in CC compared to cultured normal biliary epithelia (NBE). In addition, we identified a set of genes associated with sarcomatoid transdifferentiation. These data are useful not only because they provide a more profound understanding of cholangiocarcinogenesis and transdifferentiation, but also because they may help to develop diagnostic tools and improve the accuracy of CC prognosis.
Methods
Cell lines and cultures
Tumor tissues were obtained from surgical specimens and biopsy specimens in Korean cholangiocarcinoma patients. Tumor tissues were washed three times in Opti-MEM I (Gibco, Grand Island, NY) containing antibiotics. Washed tissue was transferred to a sterile Petri dish and finely minced with scalpels into 1- to 2-mm
3 fragments. Tissue fragments in culture medium were seeded in T25 culture flasks (Corning, Medfield, MA) in Opti-MEM supplemented with 10% fetal bovine serum (FBS, Gibco), 30-mM sodium bicarbonate and antibiotics. Tumor cells were cultured undisturbed and passaged as described [
17]. Near the 20th passage, the medium was changed from Opti-MEM I to DMEM supplemented with 10% FBS and antibiotics. NBE cells were isolated from mucosal slices of normal bile ducts, with informed consent from liver transplantation donors, and
ex-vivo cultured in T25 culture flasks in Opti-MEM supplemented with 10% FBS, 30 mM sodium bicarbonate and antibiotics at 37°C with 5% CO
2 in air. Near-confluent NBE cells were harvested and stored at -80°C until use. Cells were routinely tested for mycoplasma and found to be negative using a Gen Probe kit (San Diego, CA). CC cell lines are in Table
1.
Table 1
Clinicopathological features of nine patients with intrahepatic cholangiocarcinomas used to generate CC cells lines.
1 | CK-Choi (Choi-CK) | M/68 | IVB | 184 | WD | + | |
2 | CK-Cho (Cho-CK) | M/82 | IVA | 500 | MD | + | |
3 | CK-J (JCK) | M/72 | IVA | 125 | PD | + |
C. sinensis
|
4 | CK-S (SCK) | M/68 | IVA | 235.6 | PD | + | Sarcomatoid |
5 | CK-L1 | M/46 | IVA | 0.01 | PD | + | Combined with HCC |
6 | CK-L2 | M/65 | III | 2050.1 | MD | + | |
7 | CK-P1 | M/66 | IVA | 23.7 | MD | - | |
8 | CK-P2 | F/66 | IVA | 121.4 | MD | + | |
9 | CK-Y1 | M/52 | IVA | 0.01 | PD | + | Combined with HCC |
5-Aza-2'-deoxycytidine (Aza) treatment
Choi-CK, Cho-CK, and JCK cells were seeded at 1 × 106 cells/ml. After overnight culture, cells were treated with 5 μM of the DNA methylating agent Aza (Sigma-Aldrich, St. Louis, MO) for 4 days, and then harvested.
Patients and tissue samples
CC tissues were obtained with informed consent from Korean patients who underwent hepatectomy and common bile duct exploration at Chonbuk National University Hospital. All tumors were clinically and histologically diagnosed as cholangiocarcinoma. Detailed clinocopathological data of the 19 samples are in Table
2. All samples were immediately frozen in nitrogen tanks. Patient information was obtained from medical records. Clinical stage was determined according to the International Hepato-Pancreato-Biliary Association (IHPBA) classification [
18].
Table 2
Clinicopathological features of 19 CC samples used for microarray analysis.
1 (CC-GHS) | 68/F | L | 8.7 × 5.4 | 3 | 1 | 0 | IVA | MF | PD | A | |
2 (CC-CYS) | 57/M | L | NA | 3 | 1 | 0 | IVA | MF +PDI | MD | A | |
3 (CC-LJS) | 42/M | A | NA | 1 | 0 | 0 | I | ID | WD | B | Intraductal papillary |
4 (CC-BJP) | 62/M | P | 7.8 × 5.6 | 1 | 0 | 0 | I | ID | WD | B | Intraductal papillary |
5 (CC-HSR) | 66/M | AP | 7.3 × 6 | 2 | 0 | 0 | II | MF | MD | B | |
6 (CC-HSW) | 59/M | AP | 9 × 6.8 | 2 | 0 | 0 | II | MF | MD | B | |
7 (CC-CSB) | 60/M | L | 4 × 4.5 | 3 | 0 | 0 | III | MF | MD | B | |
8 (CC-SJS) | 71/M | A | 2.1 × 1.9 | 1 | 0 | 0 | II | MF | WD | B | |
9 (CC-HDS) | 63/M | CBD | 1.1 × 0.9 | 1 | 1 | 1 | IVB | ID | MD | B | |
10 (CC-KHC) | 47/M | L | 14 × 10 | 4 | 0 | 0 | IVA | MF | PD | C | |
11 (CC-LHG) | 42/M | L | 5.6 × 3.9 | 3 | 1 | 0 | IVA | MF | PD | C | Combined with HCC |
12 (CC-LSH) | 40/F | P | 8.6 × 4 | 2 | 0 | 1 | IVB | MF | PD | C | Combined with HCC |
13 (CC-KHS) | 70/F | L | 8.5 × 4.8 | 3 | 0 | 1 | IVB | MF +PDI | PD | C | Combined with HCC |
14 (CC-LMS) | 38/F | AP | 5 × 3.7 | 4 | 0 | 1 | IVB | MF | PD | C | Combined with HCC |
15 (CC-KJA) | 39/F | LP | 6 × 5 | 4 | 0 | 0 | IVA | MF | MD | C | Combined with HCC |
16 (CC-JSJ) | 64/M | L | 0.5 × 0.5 | 2 | 0 | 1 | IVB | ID | PD | C | |
17 (CC-YCU) | 53/M | L | 4.3 × 2.1 | 3 | 1 | 0 | IVA | MF | MD | C | |
18 (CC-GMG) | 50/M | L | 7 × 3.5 | 2 | 0 | 0 | II | MF | MD | C | |
19 (CC-BSD) | 67/M | L | 2.9 × 2.7 | 4 | 0 | 0 | IVA | MF+P야 | PD | C | |
Primer labeling and Illumina Beadchip array hybridization
Total RNA from CC samples was isolated using TRIzol reagent (Invitrogen, CA) according to the manufacturer's instructions. RNA quality was determined by gel electrophoresis, and concentrations were determined using an Ultrospec 3100 pro spectrophotometer (Amersham Bioscience, Buckinghamshire, UK). Biotin-labeled cRNA samples for hybridization were prepared according to Illumina's recommended sample-labeling procedure: 500 ng of total RNA was used for cDNA synthesis, followed by an amplification/labeling step (in vitro transcription) to synthesize biotin-labeled cRNA using the Illumina TotalPrep RNA Amplification kit (Ambion Inc., Austin, TX). cRNA concentrations were measured by the RiboGreen method (Quant-iT RiboGreen RNA assay kit; Invitrogen-Molecular Probes, ON, Canada) using a Victor3 spectrophotometer (PerkinElmer, CT), and cRNA quality was determined on a 1% agarose gel. Labeled, amplified material (1500 ng per array) was hybridized to Illumina Human-6 BeadChips v2 containing 48,701 probes for 24,498 genes, according to the manufacturer's instructions (Illumina, San Diego, CA). Array signals were developed by Amersham fluorolink streptavidin-Cy3 (GE Healthcare Bio-Sciences, Little Chalfont, UK) following the BeadChip manual. Arrays were scanned with an Illumina Bead-array Reader confocal scanner (BeadStation 500GXDW; Illumina) according to the manufacturer's instructions. Array data processing and analysis were performed using Illumina BeadStudio software. The BeadStudio Gene Expression Module is a tool for analyzing gene expression data from scanned microarray images generated by the Illumina BeadArray Reader.
Data analysis
Normalization algorithms were used to adjust sample signals to minimize the effects of variation from non-biological factors. To reduce variation between microarrays, the intensity values for samples in each microarray were rescaled using a quartile normalization method in the BeadStudio module. Measured gene expression values were log2-transformed and median-centered across genes and samples for further analysis. To generate an overview of the gene expression profile and to identify major relationships in cell lines, we used unsupervised hierarchical clustering analysis. Genes with an expression ratio of at least a two-fold difference relative to the median gene expression level across all samples in at least 10% of samples were selected for clustering analysis. Average linkage hierarchical cluster analysis was carried out using a Pearson correlation as the similarity metric, using the GeneCluster/TreeView program (
http://rana.lbl.gov/EisenSoftware.htm). Expression profiles for the differentially expressed genes were selected by
t-test with false discovery rate (FDR) and q-values as gene significance measures, using R software (version 2.5). Because of varying significance in the analyzed comparisons, using a fixed FDR (or q-value) cut-off value was not practical. Therefore, we used
t-test
P = 0.01. To ascertain biological relevance, a fold-change cut-off value of 2 or 4 from the mean was chosen. The gene ontology (GO) program (
http://david.abcc.ncifcrf.gov/) was used to categorize genes in subgroups based on biological function. Values for each GO group were calculated as a percentage of total mRNA change. For example, the Fisher exact test was used to determine whether the proportions of genes in each category differed by group. The microarray data were registered with the Gene Expression Omnibus (GEO) database (Accession No. GSE22633)
Immunoblotting
Extracted protein (30 μg) from cell lysates was resolved by SDS-PAGE and transferred to a nitrocellulose membrane. Membranes were incubated for 1 h at room temperature with primary antibody at 1:1000 dilution. After incubation, blots were washed three times in TBS/0.1% Tween 20. Immunoreactivity was detected using alkaline phosphatase-conjugated goat anti-rabbit IgG or a commercial chemiluminescence detection kit (Amersham), according to the manufacturer's instructions.
Immunohistochemistry
Immunohistochemical staining was performed on formalin-fixed, paraffin-embedded 4-μM tissue sections, as described preciously [
19]. Briefly, a deparaffinized section was pretreated by microwave epitope retrieval (750 W during 15 min in citrate buffer 10 mmol; pH 6.0) after rehydration. Before applying primary antibodies, the endogenous peroxidase activity was inhibited with 3% hydrogen peroxide, and a blocking step with biotin and bovine albumin was performed. Primary monoclonal or polyclonal antibodies were detected using a secondary biotinylated antibody and a streptavidin-horseradish peroxidase conjugate according to the manufacturer's instructions (DAKO, Glostrup, Denmark). Counterstaining was performed using Meyer's hematoxylin. Tumors were evaluated for the percentage of positive cells and the staining intensity. Negative controls were samples incubated with either PBS or mouse IgG
1 instead of primary antibody.
Real-time RT-PCR
RNA prepared from dissected tissues was precipitated with isopropanol and dissolved in DEPC-treated distilled water. Reverse transcription (RT) was performed using 2 μg total RNA, 50 μM decamer and 1 μl (200 units) and RT-PCR Superscript II (Invitrogen) at 37°C for 50 min, as previously described. Specific primers for each gene were designed using the Primerdepot website (
http://primerdepot.nci.nih.gov/) and are in Additional file
1. The control
18S ribosomal RNA primer was from Applied Biosystems (Foster City, CA) and was used as the invariant control. The real-time RT-PCR reaction mixture consisted of 10 ng reverse-transcribed total RNA, 167 nM forward and reverse primers, and 2 × PCR master mixture in a final volume of 10 μl PCR, was in 384-well plates using the ABI Prism 7900HT Sequence Detection System (Applied Biosystems).
Animal model of cholangiocarcinoma
The hamster CC model was modified from a previous study [
20]. On the first day of the experiment, hamsters in the experimental group were infected with 15 metacercariae of the liver fluke,
C. sinensis. One day after parasite infestation, hamsters received 15 ppm of dimethylnitrosamine (DMN; Kasei, Japan) in the drinking water for 4 weeks with a normal diet. Thereafter, hamsters were given tap water with a normal diet for the rest of the study. An interim stage of cholangiocarcinogenesis was confirmed at 8 weeks after experiment initiation. Control and CC model hamsters were maintained for a total of 27 weeks for CC to develop.
Discussion
In this study, our experimental design primarily investigated the gene expression profiles of 10 cell lines and 19 CC tissues, and compared these profiles to those from four cultured NBE cell line using genome-wide BeadChip microarray analysis. Transdifferentiation-related genes were analyzed by same method. Using unsupervised hierarchical clustering analysis, we found that the
SPP1,
EFNB2, and
E2F genes were commonly upregulated in both cell and tissue samples.
IRX3,
PTTG1, and
PPARγ were upregulated in the cell samples, and were immunohistochemically verified in human and hamster CC tissues. SPP1 (osteopontin), a secretory adhesive alycoprotein, was identified as a highly overexpressed gene in CC lines and tissues. SPP1 is a ligand of CD44 that binds to αV-containing integrins and is important in malignant cell attachment and tumor invasion [
43]. It was a highly overexpressed gene in HCC, and its expression correlated with earlier recurrence, poorer prognosis, and metastasis [
44]. Consistent with our findings, a recent oligonucleotide microarray study reported that SPP1 was the most highly expressed gene in intrahepatic cholangiocarcinoma [
45]. EFNB2 was identified as a preferentially expressed genes in CC. EFNB2 overexpression is reported to be significantly correlated with the number of lymph node metastases and clinical stage in esophageal cancer [
46]. Several reports have examined concomitant expression of the ligand EFNB2 and its receptor EphB4 in leukemia-lymphoma cell lines [
47], and in endometrial cancer [
48]. E2Fs 1-3 are characterized as "activator E2Fs" since their binding to promoters results in increased transcription, while E2Fs 4 and 5 are "repressor E2Fs" since they form complexes with p130, HDACs, and other factors to block transcription [
49]. During hepatocarcinogenesis in c-myc/TGFalpha double-transgenic mice, expression of E2F-1 and E2F-2 increases, and putative E2F target genes are induced [
50].
For immunohistochemical verification, the representative genes were selected from the list of top 25 commonly upregulated genes, according to antibody available for immunohistochemistry. In addition, other genes were selected from only cell-based microarray database. The same immunohistochemical staining in hamster CC tissues induced by
Clonorchiasis infestation was compared with control staining in normal hamster livers. IRX3 is involved in dorsal-ventral patterning in spinal cord development and coordination with other homeobox genes [
51]. IRX3 is preferentially expressed in the examined CC tissues and localized to the nucleus of human and hamster malignant biliary epithelial cells, independent of cell differentiation. A methylated CpG island was detected in exon 2 of the IRX3 locus, rather than in the promoter, and is responsible for IRX3 overexpression in brain tumor cells and tissues [
52]. PTTG1, a critical mitotic checkpoint protein, is a known proto-oncogene that is highly expressed in HCC [
53]. Our data showed that PTTG1 was preferentially expressed in the cytoplasm of the human and hamster CC cells. PPAR-
γ, a member of the nuclear receptor superfamily, functions as a ligand-activated transcription factor [
54]. It is overexpressed in a variety of cancers, including HCC and pancreatic cancer [
55,
56]. Positive immunostaining was localized in the cytoplasm and nuclei of human CC cells. However, positive immunostaining was exclusively detected in the nuclei of the hamster CC cells. Our data also immunohistochemically validated the downregulation of proteins KRT17, UCHL1, IGFBP7, and SPARC. Our hamster model showed the similar expression patterns of human CC related genes and therefore might be a relevant model to study human CC.
Analysis of genes involved in the transdifferentiation of CC cells showed two clusters in the gene axis, with genes that were upregulated (cluster II), and downregulated (cluster I) in the SC group as compared to the AC group. The mesenchymal antigen VIM and the transcriptional factor TWIST1 were upregulated in JCK and SCK cells by tumor dedifferentiation. The overexpression of these proteins is reported to be associated with the EMT [
57,
58]. Intriguingly, genes silenced by promoter hypermethylation during CC development were restored at the point of sarcomatous transdifferentation, which implied that the demethylation may be involved in the EMT progression of CC.
In addition to tumor-related genes known to be overexprssed in intrahepatic CC, we identified other strongly and consistently dysregulated genes in CC that are known to be involved in other human cancers. Our data support a correlation between the expression of these genes and CC tumor differentiation, and the gene expression patterns found in this study are consistent with those associated with a poor clinical prognosis for this cancer. gene expression profiling appears to be a useful diagnostic tool, especially for differentiating CC from other liver masses, as well as for the subclassification of intrahepatic CC compared to histopathological findings.
Acknowledgements
We gratefully thank Prof. Yoon B-I (Kangwon National University, South Korea) for providing the hamster CC tissues. This study was supported by grants from the Korean Association of Internal Medicine (Chungram Research Fund, 1997), a Korea Research Foundation grant from the Korean government (Basic Research Promotion Fund, KRF-2008-313-E00434), and the National R&D Program for Cancer Control (0620220) and the Korean Health Technology R&D Project (A101834), Ministry for Health, Welfare and Family affairs (0620220), Republic of Korea.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
MS and IC performed most of the experiments and drafted the manuscript. ML and GY carried out the tissue collection and the establishment of cell lines. XC participated in the immunohistochemical analysis. BC and IK participated in the design and coordination of the study and helped to draft the manuscript. EA and SL participated in the array data processing and analysis. DK conceived of the study, and participated in its design and coordination. All authors read and approved the final manuscript.