Comparison of gene expression profiles between Opisthorchis viverrini and non-Opisthorchis viverrini associated human intrahepatic cholangiocarcinoma


  • Potential conflict of interest: Nothing to report.


Intrahepatic cholangiocarcinoma (ICC) is the second most common primary cancer in the liver, and its incidence is highest in the northeastern part of Thailand. ICCs in this region are known to be associated with infection with liver flukes, particularly Opisthorchis viverrini (OV), as well as nitrosamines from food. To clarify molecular mechanisms of ICC associated with or without liver flukes, we analyzed gene expression profiles of OV-associated ICCs from 20 Thai patients and compared their profiles with those of 20 Japanese ICCs that were not associated with OV, by means of laser microbeam microdissection and a cDNA microarray containing 27,648 genes. We identified 77 commonly upregulated genes and 325 commonly downregulated genes in the two ICC groups. Unsupervised hierarchical cluster analysis separated the 40 ICCs into two major branches almost completely according to the fluke status. The putative signature of OV-associated ICC exhibited elevated expression of genes involved in xenobiotic metabolism (UGT2B11, UGT1A10, CHST4, SULT1C1), whereas that of non–OV-associated ICC represented enhanced expression of genes related to growth factor signaling (TGFBI, PGF, IGFBP1, IGFBP3). Additional random permutation tests identified a total of 49 genes whose expression levels were significantly different between the two groups. We also identified genes associated with macroscopic type of ICCs. In conclusion, these data may not only contribute to clarification of common and OV-specific mechanisms underlying ICC, but also may serve as a starting point for the identification of novel diagnostic markers or therapeutic targets for the disease. (HEPATOLOGY 2006;44:1025–1038.)

Intrahepatic cholangiocarcinoma (ICC) is a malignant tumor arising from the biliary epithelium of the intrahepatic biliary duct.1 Its incidence is increasing in Japan as well as in western countries.2 In the United States, the incidence and mortality rates increased by approximately 9% between 1973 and 1997.3 In England and Wales, a marked increase in age standardized mortality rates for this disease has been seen, and it is now ranked as the most common cause of liver tumor–related death since 1993.4 Most ICCs are extremely invasive, develop rapidly without clinical symptoms, and thus are usually diagnosed at an advanced stage when complete removal of the tumor is very difficult. Prognosis of ICC is extremely poor, because neither current anticancer drugs nor radiation therapy is effective. Therefore, to improve the prognosis, identification of novel tumor markers and development of therapeutic strategies are urgently needed.

Significant geographic variation occurs in the incidence of cholangiocarcinoma, and its incidence is significantly high in Southeast Asia.5 Epidemiological studies have uncovered strong associations between ICC and several risk factors including primary sclerosing cholangitis,6 congenital choledochal cyst, chronic hepatolithiasis,7 and infection with liver flukes such as Opisthorchis viverrini or Clonorchis sinensis.8, 9Opisthorchis viverrini (OV) is endemic in Southeast Asia, particularly in the northeastern part of Thailand, where the daily habit of eating raw and salt-fermented freshwater fish repeatedly exposed this local population to both OV and nitrosamine-contaminated food. The prevalence of OV infection is up to 70.8%, and the incidence of cholangiocarcinoma in this region is up to 317.6 per 100,000 person-years.10

Clinical studies11–13 and experimental models14, 15 have uncovered that OV infection plays a crucial role in the promotion of cholangiocarcinoma. The pathogenesis of OV infection includes chronic inflammation, adenomatous hyperplasia of bile duct epithelium, and periductal fibrosis.16 Chronic inflammation induced by infection with liver flukes increases endogenous generation of nitric oxide and production of N-nitroso compounds, which may result in DNA damage leading to neoplastic transformation of biliary epithelial cells.17 Consistently, a recent paper reported that increased nitrative and oxidative DNA damage is associated with elevated inducible nitric oxide synthase expression in the liver of hamsters with OV infection.18 Although liver fluke is proved to be a definite cause of ICCs, molecular mechanisms underlying fluke-associated ICC remain unresolved. In addition, the cause of ICCs in Japan and western countries largely remains unclear, although some are associated with hepatolithiasis, progressive cholangitis, congenital choledochal cyst, or hepatitis C virus infection.19 Clinicopathological studies reported that poorly differentiated adenocarcinoma,20 peribiliary fibrosis, and adenomatous hyperplasia in noncancerous hepatic tissues21 are more frequently observed in Thai ICCs compared with Japanese ICCs. However, a limited number of studies have investigated molecular mechanisms of ICCs of different etiological backgrounds; frequency of K-ras mutation is significantly lower,22 and expression of MIB-1 labeling index is significantly higher21 in fluke-associated tumors than in non–fluke-associated tumors. To disclose mechanisms underlying ICCs of the different etiological backgrounds, it is intriguing to compare global gene expression profiles of ICCs developed in Thai and Japanese populations.

ICCs are macroscopically classified into three types according to growth characteristics related to the mode of spreading: (1) a mass-forming (MF) type; (2) periductal infiltrating (PDI) type; and (3) intraductal growth (IDG) type. MF type, the most common type of ICC, shows an expansile solid nodule or mass. PDI type exhibits infiltration along the portal pedicle with marked stromal reaction. IDG type is a polypoid tumor that is usually confined within the dilated part of large intrahepatic bile ducts.23 In advanced stages, ICC shows a combined macroscopic appearance. Recent pathological studies have suggested that carcinogenesis process of the IDG type is different from that of the MF and PDI types.24 The IDG type is transformed from adenoma, whereas the MF and PDI type are considered to develop in the hyperplasia-dysplasia-carcinoma sequence.24 However, molecular mechanisms underlying these macroscopic types of ICC have not been clarified so far.

We previously analyzed expression profiles of Japanese ICCs that were not associated with liver flukes.25 To elucidate fluke-specific as well as common mechanisms of ICCs, we analyzed global gene expression profiles of OV-associated ICCs from 20 Thai patients, and compared them with ICCs that were not associated with OV from 20 Japanese patients. In this report, we document different molecular signatures between Thai and Japanese ICCs, which may reflect different causative backgrounds. We also demonstrate genes related to macroscopic appearance of ICCs. These data should contribute to a better understanding of intrahepatic cholangiocarcinogenesis and to development of novel diagnostic and therapeutic targets.


ICC, intrahepatic cholangiocarcinoma; OV, Opisthorchis viverrini; MF, mass-forming; PDI, periductal infiltrating; IDG, intraductal growth; aRNA, twice-amplified RNA; RT-PCR, reverse transcription polymerase chain reaction; VEGF, vascular endothelial epidermal growth factor.

Patients and Methods

Patients and Tissue Samples.

In this study, we analyzed expression profiles of 20 Thai ICC tissues and compared the profiles with those of 20 Japanese ICCs analyzed previously. ICC tissues were obtained with informed consent from Thai and Japanese patients who underwent hepatectomy at Srinagarind hospital, Khon Kaen University, Thailand, and Kyoto University Hospital, Japan, respectively. Noncancerous liver tissues were obtained with informed consent from 10 patients with liver metastasis of colon cancer, who underwent hepatectomy at Kyoto University Hospital, Japan. All tumors were clinically and histologically diagnosed as intrahepatic (or peripheral) cholangiocarcinoma according to WHO classification.1 No perihilar cancers or peripheral cholangiocellular cancers with spindle-shaped cells were included in this study. Patient information was obtained from medical records. Clinical stage was determined according to the IHPBA (International Hepato-Pancreato-Biliary Association) classification.26 No significant difference was observed in terms of gender, age of patients, location, size of tumor, tumor grade, or node involvement between the 20 Thai ICCs and 20 Japanese ICCs. The 20 Thai ICCs included 10 MF-type and 10 MF+IDG–type tumors, and the 20 Japanese ICCs included 8 MF-type, 10 MF+PDI–type and 2 PDI-type tumors. Detailed clinicopathological data of the 40 samples are summarized in Table 1. Infection with OV in Thai patients was diagnosed as positive if the patient fulfilled one of these four criteria: (1) a history of previous positive stool examination for OV or its eggs; (2) identification of OV or its eggs in stool or bile; (3) demonstration of a typical bead-like cholangiogram; and (4) histological evidence of OV in the specimen.27 None of the 20 Japanese ICCs was associated with liver flukes. All samples were immediately frozen and embedded in TissueTek OCT medium (Sakura, Tokyo, Japan) and frozen at −80°C until analysis.

Table 1. Clinicopathological Features of the 40 ICC Clinical Samples Used for cDNA Microarray Analysis
No.AgeSexaLocation of TumorbSize (cm)TcNcMcStagecGross AppearancedHistopathology (WHO, 2000)e
  • a

    M, male; F, female.

  • b

    A, anterior segment; P, posterior segment; M, medial segment; L, lateral segment.

  • c

    International Hepato-Pancreato-Biliary Association (IHPBA) classification.

  • d

    MF, mass forming type; PDI, periductal infiltrating type; IDG, intraductal growth type.

  • e

    World Health Organization, Pathology and genetics of tumors of digestive system, IARC press, Lyon, 2000; WD, well differentiated; MD, moderately differentiated; PD, poorly differentiated.

  • f

    The histopathology of MF + IDGs was determined by the main mass.

20 Thai ICCs (liver fluke–associated)
 300168FA5.3 × 4.5300IIIMFWD tubular adenocarcinoma
 300243MAP9.5 × 8.5401IV-BMFPD tubular adenocarcinoma
 300359MAP8.0 × 6.5300IIIMF + IDGMucinous carcinomaf
 300853MAP3.5 × 3.0210IV-AMFWD tubular adenocarcinoma
 301061FP8.0 × 5.5201IV-BMFWD tubular adenocarcinoma
 301468MP7.5 × 6.0201IV-BMF + IDGPD adenocarcinomaf
 301874MML5.5 × 5.0411IV-BMF + IDGAdenosquamous carcinomaf
 301948FAP14.0 × 10.0411IV-BMFMD tubular adenocarcinoma
 302057FP5.5 × 4.3411IV-BMFWD tubular adenocarcinoma
 302156FML7.0 × 6.5310IV-AMFMD tubular adenocarcinoma
 302242FAP7.7 × 6.3411IV-BMFMD tubular adenocarcinoma
 302444MAP8.5 × 6.3310IV-AMF + IDGWD adenocarcinomaf
 302548MP8.0 × 6.2310IV-AMF + IDGWD adenocarcinomaf
 302761MP11.0 × 7.0211IV-AMF + IDGPap-tubular adenocarcinomaf
 400164MML7.8 × 6.0411IV-BMFMD tubular adenocarcinoma
 400272MP4.5 × 4.2310IV-AMF + IDGWD adenocarcinomaf
 400358MA3.0 × 3.0200IIMF + IDGMucinous carcinomaf
 401053MML12.7 × 11.3311IV-BMF + IDGMucinous carcinomaf
 401142FA3.7 × 3.0300IIIMF + IDGWD adenocarcinomaf
 401862FP7.2 × 5.8401IV-BMFWD tubular adenocarcinoma
20 Japanese ICCs (non-liver fluke–associated)
154MML3.0 × 3.0211IV-BPDIWD tubular adenocarcinoma
554FAML4.8 × 4.3310IV-BMF + PDIPD tubular adenocarcinoma
1073MPA14.0 × 10.5410IV-BMF + PDIMD tubular adenocarcinoma
1326FMA4.5 × 4.0310IV-BMF + PDIMD tubular adenocarcinoma
1571MA5.6 × 5.5200IIMF + PDIPD tubular adenocarcinoma
1661MMA7.5 × 6.5300IIIMF + PDIWD tubular adenocarcinoma
1776FA2.5 × 3.0211IV-BMF + PDIMD tubular adenocarcinoma
1864FPAM5.0 × 5.0310IV-BMF + PDIWD tubular adenocarcinoma
2663FML6.8 × 4.8410IV-BMFMD tubular adenocarcinoma
4857FL9.0 × 6.0211IV-BMFPD tubular adenocarcinoma
4962MAM6.0 × 3.5211IV-BPDIMD tubular adenocarcinoma
5170FLM6.0 × 6.0311IV-BMFMD tubular adenocarcinoma
5462MAM4.0 × 2.0311IV-BMF + PDIMD tubular adenocarcinoma
5648ML5.6 × 5.5410IV-BMFMD tubular adenocarcinoma
5776ML4.5 × 3.5310IV-BMF + PDIMD tubular adenocarcinoma
6067MLM6.5 × 6.0311IV-BMFMD tubular adenocarcinoma
6167FM3.4 × 2.8400IV-AMFWD tubular adenocarcinoma
6246MML7.2 × 5.0410IV-BMFMD tubular adenocarcinoma
6565FAM8.5 × 7.0300IIIMF + PDIPD tubular adenocarcinoma
6663FP2.4 × 2.2300IIIMFPD tubular adenocarcinoma

Laser Microbeam Microdissection and Extraction of RNA.

Preparation of sections, laser microbeam microdissection, extraction of total RNA, T7-based amplification, and labeling of probes were performed as described previously.25 Approximately 2 to 3 × 104 cells from each tissue were collected selectively using the EZ cut system (SL Microtest GmbH, Germany) according to the manufacturer's protocol. A mixture of normal intrahepatic biliary epithelial cells in liver tissues from 10 patients without ICC served as a universal control. 2.5-μg aliquots of twice-amplified RNA (aRNA) from each cancerous and noncancerous tissue were then labeled respectively with Cy3-dCTP or Cy5-dCTP (Amersham Biosciences, Buckinghamshire, UK).

cDNA Microarray and Analysis of Data.

We fabricated a genome-wide cDNA microarray containing 27,648 cDNAs selected from the UniGene Database (build #131) of the National Center for Biotechnology Information. The procedures for hybridization, washing, detection of signals, and normalization of data were carried out using the same method as described previously.25 Because data from spots with low signal intensity are less reliable, we assigned a cutoff value to each slide according to the intensity of spots with S/N (signal to noise) ratios of 3. If both Cy3 and Cy5 dyes gave signal intensities lower than the cutoff value, we excluded those genes from further analysis. After leaving out the unreliable spots, we selected upregulated or downregulated genes based on their Cy3/Cy5 ratios (r): upregulated (r > 5.0) and downregulated (r < 0.2). Genes with Cy3/Cy5 ratios greater than 5.0 or less than 0.2 in more than 50% of both 20 Thai and 20 Japanese cases were defined as commonly upregulated or downregulated genes, respectively.

Hierarchical Cluster Analysis and Permutation Tests.

To reveal difference in gene expression profiles of Thai and Japanese ICCs, we applied an unsupervised two-way hierarchical cluster analysis across the 40 ICCs using the data of 27,648 genes. We selected data of 126 genes that had given valid values in more than 50% of all samples and standard deviations (SD) of the values greater than 1.8. Then average linkage hierarchical clustering was performed using the “Cluster3.0” software (∼mdehoon/software/cluster/), and the results were visualized using Java TreeView 1.0.8 software (∼alok/TreeView/) packages. We further carried out random permutation tests25 to identify genes whose expression was significantly different between the 20 Thai and 20 Japanese ICCs. The analysis identified a total of 49 genes with P values less than .01, and |MedT-MedJ| > 1, where MedT and MedJ indicate the median of log2-transformed relative expression ratios in Thai or Japanese ICCs, respectively. For the identification of genes expressed differentially between MF-type Thai ICCs and MF-type Japanese ICCs, MF-type and MF+IDG–type Thai ICCs, or MF+PDI–type and MF-type Japanese ICCs, we performed random permutation tests, and selected genes with P values less than .01, and |Med1–Med2| > 1.5, 1, and 0.8, respectively, where Med1 and Med2 indicate the medians derived from log2-transformed relative expression ratios in the former type or the latter type of ICCs. We subsequently carried out hierarchical cluster analyses using the data of genes that were identified by random permutation tests.

Quantitative RT-PCR.

Preparation of cDNA was carried out as described previously,25 and real-time polymerase chain reaction (PCR) experiments (TaqMan PCR; Applied Biosystems, Foster City, CA) were performed according to the manufacturer's protocol with the same aRNAs that had been used for microarray analysis. The sequences of primers and probes are summarized in Supplementary Table 1 (Supplementary material for this article can be found on the HEPATOLOGY website: FDFT1 served as an internal control because it showed the smallest Cy3/Cy5 fluctuation in our microarray experiments for both Thai and Japanese ICCs.

Tissue Microarrays and Immunohistochemical Staining.

Tissue microarrays were provided by Department of Pathology, Faculty of Medicine, Khon Kaen University, Thailand.28 The arrays contained a total of 273 OV-associated Thai ICCs, including the 20 Thai tumors used in our microarray analysis in quadruplicate. Paraffin-embedded tissue sections of the 20 Japanese ICCs were used for immunohistochemical analysis. Paraffin-embedded tissue sections were subjected to the Envision Plus Detection kit (Dako, Carpenteria, CA) for P-cadherin, ADAM8, IGFBP3, EphA4, TFF1, and MUC1, or the SAB-PO peroxidase immunostaining system (Nichirei, Tokyo, Japan) for survivin and PGF, according to the manufacturers' recommendations. Antigens were retrieved from deparaffinized and re-hydrated tissues by pre-treating the slides in citrate buffer (pH 6.0) for 10 minutes at 108°C by autoclave. Immunohistochemical staining was performed using anti-P-cadherin (Calbiochem, San Diego, CA), anti-ADAM8, anti-IGFBP3 (R&D Systems, Minneapolis, MN), anti-EphA4 (Santa Cruz Biotechnology, Santa Cruz, CA), anti-TFF1, anti-MUC1 (Zymed Laboratories, South San Francisco, CA), anti-survivin (NOVUS Biologicals, Littleton, CO), or anti-PGF (RELIATech, Braunschweig, Germany) antibodies. Scoring was assessed semi-quantitatively as negative (no detectable staining or positive staining in <10% of tumor cells); weak-positive (positive staining between 10% to 50% of tumor cells); strong-positive (positive staining in >50% of tumor cells), by three independent investigators. Cases were considered as positive if at least one of four tissue cores showed positive staining, and as negative if none of four tissue cores showed positive staining.


Identification of Genes Commonly Up- or Downregulated in the 40 ICCs.

In addition to our previous expression analysis of Japanese ICCs, we analyzed 20 OV-associated ICCs from Thai patients using the same microarray system and the same pooled RNA from noncancerous biliary epithelial cells as a control. We searched for commonly upregulated or downregulated genes with Cy3/Cy5 ratios greater than 5.0 or less than 0.2, respectively, in more than half of both Thai and Japanese cases to ensure that the elevated or reduced expression of the selected genes was not only confined to one group. This filter identified a total of 77 commonly upregulated genes, and 325 commonly downregulated genes, compared with normal intrahepatic bile duct epithelia (Table 2 and Supplementary Table 2). The 77 commonly upregulated genes contained those associated with cell cycle regulation and cytokinesis (TTK, STK6, CKS2, BUB1, BUB1B, CDC20, PRC1, and KIF2C), signal transduction (GNAZ, RAI3, and PLCL4), transcription factors and growth factors (FOXM1, HOXB7, IGFBP2, IGFBP3, and MDK), regulation of apoptosis (BIRC5 and BCL2L14), xenobiotic metabolisms (CYP2S1 and UGT1A10), and cell adhesion (CDH3, ADAM8, and ITGA2). The 325 commonly downregulated genes encode a number of proteins involved in immune response (IGHG1, APOH, TPSB1, CD53, CD1C, CCL19, CXCL12, IL6R, and IL1R1), and small-molecule transport (SLC4A4, SLC28A2, SLCO4C1, FXYD2, TTR, and RBP4). Other members of the downregulated genes consist of those associated with a variety of functions such as cell adhesion or cytoskeleton (CLDN5, CLDN10, DOC1, and CDH6), and Wnt signaling pathway (SFRP5 and APCDD1). These commonly up- or downregulated genes may play a crucial role in the carcinogenesis of both OV- and non-OV–associated ICCs.

Table 2. List of 77 commonly upregulated genes in the 40 ICCs
No.Accession no.aSymbolTitleFunctionbThai ICCscJapanese ICCscAll ICCs
  • a

    GenBank accession number.

  • b

    Gene functions were summarized from literature sources or according to Entrez Gene in NCBI (

  • c

    Thai group: 20 liver fluke-associated ICCs, Japanese group: 20 non-liver fluke-associated ICCs.

  • d

    Percent of cases with valid data.

  • e

    Median value of log2-transformed expression ratios of tumor to control.

1M16937HOXB7homeo box B7Transcription factor9015.867516.598316.35
2NM_202002FOXM1forkhead box M1Transcription factor9515.518515.359015.43
3AA709155FLJ10134hypothetical protein FLJ10134Unknown5015.406513.355814.93
4CR624652TTKTTK protein kinaseCell cycle regulation5013.185014.975014.75
5NM_018685ANLNanillin, actin binding protein (scraps homolog, Drosophila)Cell division906.787014.428014.39
6D88308SLC27A2solute carrier family 27 (fatty acid transporter), member 2Fatty acid transporter5014.375513.645314.22
7BC008947C10orf3chromosome 10 open reading frame 3Unknown8014.656013.377013.73
8AF095288PTTG2pituitary tumor-transforming 2Cell cycle8014.59755.627812.81
9CR610173ALDOCaldolase C, fructose-bisphosphateFructose metabolism6013.07506.335511.79
10BC041846CDH3cadherin 3, type 1, P-cadherin (placental)Cell adhesion756.957513.01759.53
11BM971909 CDNA FLJ31668 fis, clone NT2RI2004916Unknown657.715014.13589.40
12BX640908EVI1ecotropic viral integration site 1Oncogene7513.60505.48637.33
13AY358603CYP2S1cytochrome P450, family 2, subfamily S, polypeptide 1Drug metabolism757.27557.37657.27
14AF037335CA12carbonic anhydrase XIICarbon dioxide metabolism808.01706.55757.12
15AF159456DMBT1deleted in malignant brain tumors 1Immune response706.78504.97606.44
16NM_198433STK6serine/threonine kinase 6Cell cycle regulation6011.39506.40556.40
17BC008718BIRC5baculoviral IAP repeat-containing 5 (survivin)Apoptosis inhibitor955.21856.53906.06
18AF195765RAMPRA-regulated nuclear matrix-associated proteinNeuronal differentiation955.291006.27985.93
20D00244PLAUplasminogen activator, urokinaseProteolysis, Chemotaxis555.26655.42605.35
21BC011409UGT1A10UDP glycosyltransferase 1 family, polypeptide A10Xenobiotic metabolism854.78954.73904.77
22NM_178229IQGAP3IQ motif containing GTPase activating protein 3Cytoskeleton regulation1005.741004.191004.65
23AF017790KNTC2kinetochore associated 2Cell division9013.33804.19854.60
24AF053306BUB1BBUB1 budding uninhibited by benzimidazoles 1 homolog beta (yeast)Cell cycle954.31955.80954.57
25XM_290629C14orf78chromosome 14 open reading frame 78Unknown853.51905.61884.39
26NM_182964NAV2neuron navigator 2Neuronal development753.607011.61734.35
27BC010044CDC20CDC20 cell division cycle 20 homolog (S. cerevisiae)Cell cycle regulation804.26505.39654.33
28BM726315 Transcribed sequence with weak similarity to protein sp:P39194 (H. sapiens) ALU7_HUMANUnknown854.25605.03734.33
29BC004312IGFBP2Insulin-like growth factor binding protein 2, 36kDaGrowth factor653.65554.87604.33
30NM_006845KIF2Ckinesin family member 2CCell division1004.55953.10984.28
31CR625760TOP2Atopoisomerase (DNA) II alpha 170kDaDNA transcription and replication503.87504.92504.27
32BM554470UBE2Cubiquitin-conjugating enzyme E2CUbiquitination905.09753.94834.26
33BE538546FLJ20641hypothetical protein FLJ20641Unknown854.08754.33804.11
35BM912233CKS2CDC28 protein kinase regulatory subunit 2Cell cycle854.01754.12804.07
36AF044588PRC1protein regulator of cytokinesis 1Cytokinesis904.10903.40904.05
37AK095136RASGEF1ARasGEF domain family, member 1ASignal transduction503.86554.33533.92
38AY376439ECT2epithelial cell transforming sequence 2 oncogeneCytokinesis953.90803.82883.84
39AK122672RAI3retinoic acid induced 3Signal transduction953.981003.80983.83
40BQ230791TNN13troponin I, cardiacMuscle contraction603.67504.17553.81
41AB032261SCDstearoyl-CoA desaturase (delta-9-desaturase)Fatty acid synthesis853.59853.54853.57
42AK023744EPS15L1epidermal growth factor receptor pathway substrate 15-like 1Endocytosis903.18753.72833.57
43AF027153 Transcribed sequenceUnknown503.33653.66583.49
44BC019679PLCL4phospholipase C-like 4Lipid metabolism1003.58753.22883.46
45NM_004523KIF11kinesin family member 11Cell mitosis553.23553.66553.46
46D26579ADAM8a disintegrin and metalloproteinase domain 8Cell adhesion and proteolysis503.51553.26533.44
47BC000013IGFBP3insulin-like growth factor binding protein 3Growth factor502.84703.60603.35
48AL834247MYPNmyopalladinMuscle contraction953.33753.12853.32
49BU736378COL1A1collagen, type I, alpha 1ECM703.00804.01753.30
50BC015050OIP5Opa-interacting protein 5Cell communication553.67502.64533.24
51X92518HMGA2high mobility group AT-hook 2Chromatin architecture902.98703.67803.20
53AK126185PPFIA4protein tyrosine phosphatase, receptor type, f polypeptide, interacting protein, alpha 4Cell communication902.97753.98833.14
54BC069193HIST2H2BEhistone 2, H2beNucleosome structure1002.981003.321003.12
55AB006000LECT1leukocyte cell derived chemotaxin 1Skeletal development703.56903.08803.12
56AK092758LOC58489hypothetical protein from EUROIMAGE 588495Unknown852.94753.75803.11
57AF281255BCL2L14BCL2-like 14 (apoptosis facilitator)Apoptosis regulator802.94603.30703.05
58NM_002391MDKmidkine (neurite growth-promoting factor 2)Growth factor902.77953.64933.03
59AA777954 Transcribed sequenceUnknown652.50503.56582.98
60BQ021339KIAA1582KIAA1582 proteinUnknown752.70502.89632.88
61BC067289CTSL2cathepsin L2Proteolysis752.44603.95682.81
62NM_002073GNAZguanine nucleotide binding protein (G protein), alpha z polypeptideSignal transduction1002.961002.691002.79
63NM_003558PIP5K1Bphosphatidylinositol-4-phosphate 5-kinase, type I, betaPhosphorylation702.84602.50652.73
64X17033ITGA2integrin, alpha 2 (CD49B, alpha 2 subunit of VLA-2 receptor)Cell-ECM interaction752.66554.11652.68
65NM_182798FLJ39155hypothetical protein FLJ39155Unknown902.72702.41802.65
66NM_005915MCM6MCM6 minichromosome maintenance deficient 6 (MIS5 homolog, S. pombe) (S. cerevisiae)DNA replication902.38603.04752.65
67BC028905 CDNA clone IMAGE:4792407, partial cdsUnknown902.55603.03752.64
68AI791563TNFRSF5tumor necrosis factor receptor superfamily, member 5B-cell proliferation1002.621002.701002.64
69BX109199 Transcribed sequenceUnknown952.93752.60852.63
70CR598555KIF20Akinesin family member 20AVesicle-mediated transport1002.62702.86852.62
71U14658PMS2PMS2 postmeiotic segregation increased 2 (S. cerevisiae)DNA mismatch repair502.51802.57652.57
72BC069736LAP1Blamina-associated polypeptide 1BMembrane integrity752.46653.01702.50
73AF053305BUB1BUB1 budding uninhibited by benzimidazoles 1 homolog (yeast)Cell cycle regulation952.49902.35932.49
74AA429665 Transcribed sequenceUnknown902.48802.34852.44
75NM_198270NHSNance-Horan syndrome (congenital cataracts and dental anomalies)Eye, tooth and brain development902.37952.44932.42
76BG209407 Transcribed sequenceUnknown952.371002.49982.41
77CF595688 Full-length cDNA clone C50DC011YB24 of Neuroblastoma Cot 25-normalized of Homo sapiensUnknown1002.361002.501002.40

Unsupervised Hierarchical Clustering of Liver Fluke– and Non-Liver Fluke–Associated ICCs.

To compare the gene expression profiles across the 40 ICCs, we further carried out a two-way unsupervised hierarchical cluster analysis using data of the filtered 126 genes that had given valid values in more than 50% of all samples and standard deviations (SD) of the values greater than 1.8. This analysis classified the 40 samples into two major groups (Fig. 1); one group contained 17 Thai tumors and 3 Japanese tumors, and the other contained 17 Japanese and 3 Thai tumors. The unsupervised cluster analysis correctly separated 8 of 10 MF-type Thai tumors and six of eight Japanese MF-type tumors into two groups according to the different causes, although the 18 MF-type tumors were widely distributed across the horizontal axis. These data suggest that OV-associated and non–OV-associated ICCs have distinct expression signatures. Although we investigated the relationship between the clustered groups and other clinicopathological data, including age and sex of the patients, location, macroscopic types, node involvement, and stage of the tumors, we did not find any other factors that were significantly associated with the two groups. In addition, no significant difference in clinicopathological features was observed between the three misclassified tumors and the 17 correctly classified tumors in either Japanese or Thai group. The cluster analysis also identified two major clusters in the gene axis, one containing genes that were relatively upregulated in Japanese tumors, and the other containing genes that were relatively upregulated in Thai tumors. In the former cluster containing 64 genes, we found that a number of genes associated with signal transduction, especially growth-factor signaling (IGFBP1, IGFBP3, TGFBI, PGF, and EPHA4), and those associated with inflammatory and immune response (SERPINA1, SERPINA3, C1S, C3, and C5) were uniquely expressed at relatively higher levels in Japanese tumors than in Thai tumors. Among the 62 genes in the latter cluster, we found that genes associated with xenobiotic or estrogen metabolism (UGT1A10, UGT2B11, and CHST4), other metabolisms (SULT1C1, ALDH1A1, and ATP8A1), or other functions (TFF1 and MUC1) were specifically expressed higher in the Thai tumors compared with Japanese tumors. No genes associated with metabolism were included in the former cluster. Notably, clones corresponding to the same gene that had been spotted at different positions on our microarray slides, such as SERPINA1 and TFF1, were present in the same branch of the gene-cluster dendrogram, indicating a high degree of reliability of our microarray data.

Figure 1.

An unsupervised two-way hierarchical clustering analysis using 126 genes across 40 intrahepatic cholangiocarcinoma (ICC) samples. In the horizontal axis, the 40 tumors were classified into two major trunks, whereas the 126 genes were clustered according to similarities in relative expression ratios in the vertical axis. The color of each cell in the matrix represents the relative expression level of each gene. Red and green indicate expression levels above and below the median derived from log-transformed relative expression ratios for that gene across all samples, respectively. Black represents unchanged expression, gray indicates no or insignificant expression (intensities of both Cy3 and Cy5 under the cutoff value).

Identification of Genes Expressed Differently Between Thai and Japanese ICCs.

To identify genes whose expression levels were significantly different between Thai and Japanese ICCs, we further performed random permutation tests. We selected genes that had given valid values in more than 90% of all samples and showed significant difference in their expression (P < .01) between the two groups. This analysis identified a total of 49 genes that included 29 genes relatively upregulated in Japanese tumors and 20 in Thai tumors (Table 3). Consistent with our unsupervised cluster analysis, the 29 genes included those associated with growth-factor signaling (TGFBI, PGF, and EPHA4). We also found that genes encoding actin cytoskeleton (RND3, TMSB4X, GSN, and TAGLN) showed significantly higher expression levels in Japanese tumors than Thai tumors. The 20 genes that showed significantly higher expression levels in Thai tumors than Japanese tumors, contained genes related with a variety of functions such as regulation of cell growth (TOB1, FOS), mitochondrial energy transfer (CKMT1A, ACSSI), ion channel and transport (HCN3, ATP1B1, KCNH2), and genes implicated in metabolisms (ALDH1A1). In line with our unsupervised cluster analysis, ALDH1A1 showed relatively abundant expression in Thai tumors compared with Japanese tumors. The genes differentially expressed between Thai and Japanese ICCs may reflect their different nature or cause. Subsequent hierarchical clustering using the expression profiles of the 49 genes correctly separated the 20 Thai from 20 Japanese tumors, as we expected (Fig. 2A). Another random permutation test using the data of 10 MF-type Thai and eight MF-type Japanese tumors identified 38 genes differentially expressed between the two groups. Cluster analysis using the data of these genes correctly separated the 18 MF-type tumors into Thai and Japanese groups, suggesting that the OV- and non–OV-associated tumors do have different expression profiles (Fig. 2B). The 38 genes contained EPHA4, PGF, TAGLN, ANXA4, and HCN3, which were also in the list of the 49 genes expressed differently between the 20 Thai and 20 Japanese ICCs.

Table 3. List of genes expressed differently between the 20 Thai and 20 Japanese ICCs
No.Accession no.aSymbolTitleP valueb|MedT-MedJ|c
  • a

    GenBank accession number.

  • b

    Permutation P-values calculated as described in Materials and Methods.

  • c

    MedT and MedJ indicate the median value of log-transformed relative expression ratios in Thai and Japanese ICCs, respectively.

Genes significantly expressed higher in Japanese ICCs than Thai ICCs
1AB011173AOF2amine oxidase (flavin containing) domain 25.97 × 10−41.05
2L10678PFN2profilin 29.10 × 10−31.20
3BC022332UBE2E2ubiquitin-conjugating enzyme E2E 2 (UBC4/5 homolog, yeast)9.64 × 10−31.31
4NM_000311PRNPprion protein (p27–30) (Creutzfeld-Jakob disease, Gerstmann-Strausler-Scheinker syndrome)3.13 × 10−31.50
5CA425779PRDX4peroxiredoxin 44.44 × 10−31.25
6X63187WFDC2WAP four-disulfide core domain 24.63 × 10−31.26
7BU616881RND3Rho family GTPase 31.01 × 10−52.13
8NM_001649APXLapical protein-like (Xenopus laevis)1.22 × 10−73.51
9NM_020947 KIAA1609 protein3.50 × 10−41.01
10AB191261FN1fibronectin 16.84 × 10−32.16
11M77349TGFBItransforming growth factor, beta-induced, 68kDa3.05 × 10−32.36
12BC016815DCBLD2discoidin, CUB and LCCL domain containing 28.28 × 10−31.49
13NM_000177GSNgelsolin (amyloidosis, Finnish type)1.01 × 10−31.63
14BQ067165TMSB4Xthymosin, beta 4, X-linked5.82 × 10−41.11
15BX537913CTSCcathepsin C2.81 × 10−31.15
16CR624122TUSC3tumor suppressor candidate 32.18 × 10−32.04
17AK023319CLUAP1,Clusterin associated protein 15.44 × 10−81.75
18NM_001304CPDcarboxypeptidase D3.23 × 10−31.17
19AY048757ABCG1ATP-binding cassette, sub-family G (WHITE), member 14.83 × 10−31.13
20L20688ARHGDIBRho GDP dissociation inhibitor (GDI) beta6.56 × 10−31.67
21NM_004438EPHA4EPH receptor 43.43 × 10−53.07
22AF000381CLEC12AC-type lectin domain family 12, member A9.86 × 10−246.49
23X54936PGFplacental growth factor, vascular endothelial growth factor-related protein1.12 × 10−112.57
24AA581409 Transcribed sequence2.63 × 10−152.38
25AA620862 Transcribed sequence3.79 × 10−92.06
26M95787TAGLNtransgelin5.78 × 10−41.53
27AW976457MBNL1Muscleblind-like (Drosophila)5.63 × 10−31.74
28N38751KLHL22kelch-like 22 (Drosophila)8.21 × 10−31.20
29NM_003617RGS5regulator of G-protein signalling 59.02 × 10−31.06
Genes significantly expressed higher in Thai ICCs than Japanese ICCs
1K03000ALDH1A1aldehyde dehydrogenase 1 family, member A14.30 × 10−32.71
2AY358336NPNTNephronectin5.14 × 10−31.80
3BC000407SYNGR2synaptogyrin 24.79 × 10−41.54
4D38305TOB1transducer of ERBB2, 16.22 × 10−31.59
5NM_001677ATP1B1ATPase, Na+/K+ transporting, beta 1 polypeptide5.15 × 10−31.13
6M19383ANXA4annexin A46.28 × 10−31.18
7BC053563TMEM49Transmembrane protein 497.11 × 10−31.29
8AF035594PRKCAprotein kinase C, alpha5.42 × 10−31.14
9NM_005252FOSv-fos FBJ murine osteosarcoma viral oncogene homolog8.56 × 10−31.33
10BC071580SEC31L2SEC31-like 2 (S. cerevisiae)1.84 × 10−31.50
11NM_003047SLC9A1solute carrier family 9 (sodium/hydrogen exchanger), isoform 1 (antiporter, Na+/H+, amiloride sensitive)9.78 × 10−31.12
12BC035812PCDH1protocadherin 1 (cadherin-like 1)1.15 × 10−41.43
13AA639753 Transcribed sequences1.42 × 10−52.11
14AK125058ACSS1acetyl-Coenzyme A synthetase short-chain family member 13.11 × 10−31.62
15U11862KCNH2Potassium voltage-gated channel, subfamily H (eag-related), member 22.58 × 10−31.70
16AF289554PERQ1PERQ amino acid rich, with GYF domain 15.03 × 10−41.42
17AB018335TMEM63ATransmembrane protein 63A3.37 × 10−41.43
18NM-024761MOBKL2BMOB Mps One Binder kinase activator-like 2B (yeast)6.90 × 10−31.04
19NM_020990CKMT1Acreatine kinase, mitochondrial 1 (ubiquitous)2.46 × 10−42.13
20NM_020897HCN3hyperpolarization activated cyclic nucleotide-gated potassium channel 32.55 × 10−41.16
Figure 2.

Supervised hierarchical clusterings using the expression of genes selected by permutation tests. The color of each cell in the matrix represents the expression level of each gene. Red and green indicate expression levels above and below the median value for that gene across all samples, respectively. Black represents unchanged expression, gray indicates no or insignificant expression. (A) A supervised hierarchical clustering of 40 intrahepatic cholangiocarcinomas (ICCs) using the 49 genes with significantly different expression between Thai and Japanese tumors. (b) A supervised hierarchical clustering analysis of 18 mass-forming (MF)-type ICCs using the 38 genes that were differently expressed between MF-type Thai ICCs and MF-type Japanese ICCs. (C) A supervised hierarchical clustering analysis of 20 Thai tumors using the 17 genes that were differently expressed between MF-type and MF+IDG–type ICCs. (D) A supervised hierarchical clustering analysis of 20 Japanese tumors using the 20 genes that were differently expressed between periductal infiltrating-type (MF+PDI, PDI) and MF-type ICCs.

Identification of Genes Related to Macroscopic Type of ICCs.

Macroscopically ICCs are divided into three types according to growth characteristics related to the mode of spreading: (1) MF type; (2) PDI type; (3) IDG type. To elucidate the different molecular features associated with macroscopic appearance of ICC, we carried out random permutation tests using the data of the 20 Thai tumors containing 10 MF-type and 10 MF+IDG–type (invasive IDG-type) tumors. As a result, we identified 17 genes, among which four were relatively augmented and 13 were relatively downregulated in the MF+IDG–type tumors compared with the MF-type tumors. The four upregulated genes included an intestinal trefoil factor (TFF3), and a mucin gene (MUC5B), and the 13 downregulated genes contained those involved in mRNA splicing (PNN, HNRPH3), transcription (ZBTB20), and metabolism (ATP13A3, DHRS8). Subsequent unsupervised clustering algorithm using the 17 genes correctly classified the 20 Thai tumors into the two groups, with an exception of two MF+IDG–type tumors (Fig. 2C). We also identified 20 genes differentially expressed between the 12 periductal infiltrating-type (10 MF+PDI and 2 PDI) and the eight MF-type Japanese tumors. Subsequent cluster analysis using the expression profiles of these genes almost correctly separated the 20 Japanese ICCs into two groups according to their macroscopic types. Among the 20 genes, eight that were relatively augmented in the MF-type tumors included those associated with cell cycle regulation (BUB1B and BUB1) and growth factor (VEGF), whereas 12 that were relatively enhanced in the 12 periductal infiltrating type contained genes implicated in cytoskeleton and motility (IQGAP1) (Fig. 2D). Altered expression of these genes may reflect biological feature or origin of the three macroscopic types of ICC.

Validation of Microarray Data by Quantitative Reverse Transcription PCR and Immunohistochemical Staining.

To prove the robustness of our strategy for the identification of genes showing altered expression, we selected three genes from the 77 commonly upregulated genes (BIRC5, CDH3, and FOXM1), 3 from the 325 commonly downregulated genes (APCDD1, DOC1, and CLDN10), and performed quantitative reverse transcription (RT)-PCR analysis using the same aRNAs subjected to the microarray analysis. As a result, the median values of cancer to non-cancerous biliary epithelial cell ratio of the six genes were similar between the microarray data and the quantitative RT-PCR data (Pearson's correlation coefficients > 0.5) (Fig. 3A). We additionally analyzed four genes expressed differentially between the Thai and Japanese tumors (IGFBP3 and SULT1C1 from unsupervised clustering; TGFBI and ALDH1A from permutation analysis), and two genes discriminating MF+IDG–type tumors from the MF-type tumors (TFF3 and MUC5B). Quantitative PCR analysis of the six genes showed consistent change of expression with the data from cluster or permutation analyses (P < .05, Student t test) (Fig. 3B–C). Additionally, we performed immunohistochemical analysis of four commonly upregulated genes (BIRC5, CDH3, ADAM8, and IGFBP3), and four genes differentially expressed between the Thai and Japanese ICCs (TFF1, MUC1, EPHA4, and PGF) using the 20 Japanese ICCs and tissue microarrays containing 273 Thai ICCs. In our previous study, we showed that expression of BIRC5 and CDH3 were elevated in 18 and 13 of 23 Japanese ICCs, respectively.25 In line with these data, BIRC5 and CDH3 were also accumulated in 272 (99.87%) and 224 (82.0%) of the 273 Thai tumors, respectively. ADAM8 and IGFBP3 were expressed in 258 (94.55%) and 151 (55.34%) of the Thai ICCs, respectively, and in 17 (85%) and 11 (55%) of 20 Japanese ICCs, respectively. BIRC5 and IGFBP3 were stained in cytoplasm, and CDH3 and ADAM8, in plasma membrane of cancer cells. Notably, the four proteins were negative for immunohistochemical staining in noncancerous biliary epithelium (Fig. 4A). These data confirmed the elevated expression of BIRC5, CDH3, ADAM8, and IGFBP3 in both Thai and Japanese ICCs. Additionally, immunohistochemical staining of MUC1 and TFF1 showed that MUC1 was accumulated in 95.24% of Thai tumors and 50.0% of Japanese tumors (P < .05, Fisher's exact test), and that TFF1 was abundantly expressed in 28.57% of Thai tumors, whereas none was expressed (0%) in Japanese tumors (P < .05, Fisher's exact test). These data corroborated relatively enhanced expression of MUC1 and TFF1 in Thai tumors compared with Japanese ICCs. Although PGF and EPHA4 did not show statistical difference in expression between Thai and Japanese tumors, they showed tendency of elevated expression (70.83%, and 94.44%, respectively) in Japanese tumors compared with Thai tumors (46.03% and 68.57%, respectively) (Fig. 4B–C). MUC1, TFF1, and PGF were stained in cytoplasm, whereas EPHA4 was stained in plasma membrane. These results are in good agreement with our microarray data, and demonstrate that these proteins are differently involved in carcinogenesis of Thai and Japanese ICC.

Figure 3.

(A) Expression of three commonly upregulated genes (FOXM1, BIRC5, and CDH3), and three commonly downregulated genes (APCDD1, DOC1, and CLDN10) analyzed by cDNA microarray (open boxes) and quantitative reverse transcription polymerase chain reaction (RT-PCR) (hatched boxes). The box chart represents distribution of log2-transformed relative expression ratios of tumor to normal by microarray and real-time PCR using the same RNA samples. Expression level of FDFT1 serves as an internal quantitative control. (B) Quantitative RT-PCR analysis of four genes (ALDH1A1, SULT1C1, TGFBI, and IGFBP3) expressed differentially between Thai (open boxes) and Japanese (hatched boxes) ICCs, and (C) two genes (MUC5B and TFF3) expressed differently between mass-forming (MF) type (shaded boxes) and MF+intraductal growth (IDG) type (striped boxes) tumors.

Figure 4.

(A) Immunohistochemical staining of BIRC5, CDH3, IGFBP3, and ADAM8, four commonly upregulated genes. Representative tissue cores from a total of 293 (273 Thai and 20 Japanese) tumors and 10 non-cancerous liver tissues are presented. All proteins were not or very slightly expressed in normal bile duct epithelium. (B) Immunohistochemical staining of MUC1, TFF1, PGF, and EPHA4, four discriminating genes between liver fluke–associated (Thai) tumors and non–liver fluke–associated (Japanese) tumors. (C) Frequency of different (strong, weak, and negative) expression levels of MUC1, TFF1, PGF, and EPHA4 by immunohistochemical staining in the 273 Thai and 20 Japanese intrahepatic cholangiocarcinoma (ICC) tissues. Strong staining (hatched box), positive staining in more than 50% of tumor cells; weak staining (shaded box), positive staining between 10% and 50% of tumor cells; negative staining (open box), no detectable staining or positive staining in less than 10% of tumor cells.


In this study, we performed gene expression profile analysis of ICCs associated with OV in Thai patients, and compared them with profiles of non–OV-associated ICCs that were previously analyzed in Japanese patients.25 By means of unsupervised hierarchical clustering algorithm and random permutation tests, we revealed the different expression profiles between Thai and Japanese ICCs, and identified a set of genes differently expressed between the two groups. These data have facilitated our comprehensive understanding of ICCs with different causative backgrounds. Genes whose expression was expressed at relatively higher levels in Thai tumors, namely, the OV-associated group, compared with Japanese tumors included those involved in xenobiotic and endobiotic metabolisms: UDP-glucuronosyltransferases (UGT1A10, UGT2B11) and sulfotransferases (CHST4, SULT1C1). Glucuronidation and sulfate conjugation plays an important role in the detoxification of carcinogens such as aromatic hydrocarbons and nitrosamines.29, 30 Therefore, epithelial cells in bile duct of Thai patients might be more exposed than those of Japanese patients to microenvironment that activates these enzymes. Possible explanations for the elevated expression of detoxification enzymes in Thai ICCs include the exposure to carcinogens. Daily intake of salt-fermented fishes exposes Thai people in the northeast region to carcinogens such as nitrosamines. The second possibility is undetermined toxic or mitogenic substance(s) secreted by OV. A recent study demonstrated that excretory/secretory (ES) products from OV increased NIH3T3 cell proliferation in vitro.31 Further identification of carcinogenic substance(s) produced from OV may clarify the mechanisms underlying OV-associated tumors. The third possibility is increased endogenous 8-oxodG or nitric oxides synthesis induced by OV-associated chronic inflammation.18 Experimental models using hamsters demonstrated that OV-associated antigens induced strong local inflammatory response,32 and re-infection with OV resulted in early inflammatory changes of the liver and alterations of liver enzymes.33 Taken together, induction of genotoxic stress by carcinogens, unknown substance(s) produced by OV, or chronic inflammation attributable to the OV infection may play a crucial role in the carcinogenesis of Thai ICCs. Polymorphisms in UDP-glucuronosyltransferases and sulfotransferases were reportedly linked with the risk of ER-negative breast cancer tumors.34 Hence, investigating whether polymorphisms in these genes are associated with ICCs will be interesting.

Most Japanese ICCs exhibited relatively abundant expression of genes associated with growth factor signaling and inflammatory response compared with Thai tumors. This result may reflect that growth factors and immune response to tumor cells play more roles in carcinogenesis of Japanese tumors than Thai tumors. We also showed that genes involved in regulation of actin cytoskeleton (RND3, TMSB4X, GSN, and TAGLN) are more abundantly expressed in Japanese tumors than Thai tumors. In addition, we demonstrated here that two proteins associated with angiogenesis signaling, EPHA4 and PGF, showed relatively higher expression in Japanese ICCs compared with Thai ICCs by microarray and immunohistochemical staining. EPHA4, a receptor tyrosine kinase, is involved in neuronal circuit development and angiogenesis through regulation of cell adhesion and migration.35 PGF, a member of the vascular endothelial growth factor (VEGF) family, is a key player in pathological, but not physiological, angiogenesis. PGF activates VEGFR1, leading to intermolecular phosphorylation of VEGFR2, which results in increased VEGF-driven angiogenesis through VEGFR2.36 Loss of PGF impairs angiogenesis and plasma extravasation in many pathological conditions, including cancers, without causing any normal vascular defect.37 Therefore, EPHA4 and PGF may serve as novel molecular targets for anti-angiogenic therapy, and treatment with anti-VEGFR antibody may be more effective in Japanese ICCs than in Thai ICCs. These data also suggest that the samples we analyzed might have different biological features such as growth or angiogenesis between Thai and Japanese tumors. Future clinical investigations on phenotype and prognosis of Thai and Japanese ICCs may support the molecular findings demonstrated in this study.

We further identified genes differently expressed between MF-type and MF+IDG–type tumors. Intraductal papillary mucinous neoplasms of the pancreas, whose macroscopic feature closely resemble that of IDG type, may share common mechanisms of carcinogenesis as IDG-type ICCs.38 Genes differentially expressed higher in MF+IDG–type than in MF-type ICCs included TFF3 and MUC5B, whose overexpression in intraductal papillary mucinous neoplasms were previously documented.39 Interestingly, PNN, a gene with relatively decreased expression in MF+IDG–type, is markedly downregulated in a variety of human cancers; ectopic expression of this gene induced apoptosis in human cancer cell lines, suggesting that PNN may act as a tumor suppressor gene.40 Hence, its reduced expression in MF+IDG–type may contribute to the neoplastic transformation of this unique type of ICCs. Although expression of MUC1 was reported to be more commonly upregulated in non–IDG-type Taiwanese ICCs compared with IDG-type tumors,41 MUC1 was relatively enhanced in Thai ICCs containing 10 non–IDG-type tumors, compared with Japanese ICCs containing 20 non–IDG-type tumors. Because our microarray data showed that MUC1 expression was approximately 1.3-fold higher in the 10 non–IDG-type Thai tumors than in the 20 non–IDG-type Japanese tumors (data not shown), a different cause is likely to affect the expression of MUC1. Consistent with this finding, another study showed that MUC1 expression was not associated with histological subtypes in OV-associated Thai ICCs.42 We also identified genes expressed differentially between MF-type and periductal infiltrating-type ICCs. Notably, IQGAP1 was in the list of genes relatively enhanced in periductal infiltrating-type ICCs compared with MF-type ICCs. Because increased IQGAP1 expression has been reported to promote tumor cell invasion and migration,43 elevated IQGAP1 expression may play a role in the invasion of periductal infiltrating-type ICCs, which have the poorest prognosis among the three macroscopic types.

In addition, we identified a total of 77 genes that were commonly upregulated in both Thai and Japanese ICCs. The 77 genes include a number of genes associated with G2/M transition or chromosome segregation, such as BUB1B, BUB1, STK6, and CDC20. BUB1B regulates kinetochore-microtuble attachments with Aurora B kinase,44 and BUB1 protects centromeric cohesion during mitosis.45 STK6, in other words, Aurora kinase A, STK15, or BTAK is essential for normal chromosome segregation. Enhanced expression of STK6 is observed in a wide range of human neoplasms, including colorectal cancer, breast cancer, and gastric cancer.46 Localized at mitotic spindle poles, CDC20 interacts with STK6 and regulates chromosome segregation.47 Because these genes are enhanced in both types of ICC, deregulated mitotic or spindle checkpoints may result in chromosome mis-segregation leading to aneuploidy, a common feature of ICC. Additionally, we have demonstrated that ADAM8 and IGFBP3 are enhanced in most Thai and Japanese ICCs. ADAM8 is a secreted protein that plays a role in cell–cell and cell–extracellular matrix interaction. A recent study showed that ADAM8 expression was enhanced in most lung cancers and that ADAM8 may be a novel serological marker for lung cancer.48 Because our immunohistochemical analysis of ADAM8 showed its elevated expression in more than 90% of ICCs, ADAM8 may be used as a diagnostic marker for ICCs as well. IGFBP3 acts as a mitogen that potentiates EGF (epidermal growth factor) or interacts with HER-2 in breast epithelial cells.49 A high level of cytoplasmic IGFBP3 is associated with poor prognosis of patients with breast cancer.49 Therefore, elevated IGFBP3 may exert a mitogenic role in various human cancers through the EGF signaling pathway. We also identified a total of 325 genes that were commonly downregulated in both Thai and Japanese ICCs. Among them, some genes were previously reported to be genetic or epigenetically inactivated in human cancers. For example, decreased expression of DOC1 resulted in inactivation of cellular senescence, leading to immortalization of human prostate epithelial cells.50 Future studies on these genes may identify novel tumor-suppressor genes for intrahepatic biliary epithelia.

In summary, the comprehensive expression profiles reported here should contribute to the better understanding of carcinogenesis of ICCs with different causes and macroscopic type. Further investigations on the function of genes identified in this study may provide a more profound understanding of ICC carcinogenesis and facilitate the development of novel diagnostic markers and more effective therapeutic modalities for ICCs.


The authors thank Tae Makino, Kiyomi Jindo, and Noriko Ikawa for technical assistance, Emi Okutsu for the analysis of microarray data, and Drs. Artit Jinawath, Ryo Takata, and Yan Liang for helpful discussions.