Microarray data were identified in the GEO (https://www.ncbi.nlm.nih.gov/geo/) and PubMed (https://www.ncbi.nlm.nih.gov/pubmed/) databases on April 6, 2017 by using keywords “Cholangiocarcinoma” and “Cholangiocarcinoma AND Expression profiling”, respectively (Figure 1(Fig. 1)). Human (Homo sapiens) tissue sample was used as the initial inclusion criterion for search results in GEO. In PubMed, the GEO dataset accession number must be provided and microarray experiment must be performed in humans. The following exclusion criteria were used for removing the remaining datasets: miRNA expression profile, methylation microarray, cell line or tissue culture, biliary stricture samples, and the number of normal tissues was less than 10 % of the sample size.

Data were inspected and evaluated the requirement for the additional pre-processing step. If needed, pre-processing in each dataset was performed separately by using “limma” package from R/Bioconductor (Ritchie et al., 2015[21]; Zhong et al., 2018[31]). In general, the pre-processing step includes target file reading, background correction, data normalization, removal of control and low quality probes, and log2 transformation were performed. The data sets GSE26566, GSE32879, GSE45001, and GSE57555 retained 22182, 18308, 17812 and 19624 probes respectively after pre-processing. Unsupervised analysis and outlier detection was performed by using principal component analysis (PCA). DEGs (false discovery rate; FDR <0.05) between tumor and normal tissue for each dataset were identified by using “limma” method. Fold-change (FC) was calculated between the average intensity of tumor and normal samples. Gene symbol was used to map and combine FC values from each dataset. FC of 1 was assigned for a missing value. In case of a duplicated gene symbol, the probe with higher intensity was used as the representative of that gene symbol. DEGs with FC of at least ± 1.5 in at least 3 datasets were selected and visualized by unsupervised hierarchical clustering heatmap to discover the common expression pattern across all datasets. Dendrogram was cut according to tree height. Gene network analysis was performed by STRING (https://string-db.org/). Enriched biological processes with FDR<0.05 were reviewed.

Human CCA cell lines HuCCT1 and TFK-1 were purchased from the RIKEN BRC (Ibaraki, Japan). Cells were cultured in RPMI-1640 (Gibco, NY) with 10 % fetal bovine serum (FBS) (Gibco, NY) and 100 U/mL penicillin and 100 µg/mL streptomycin (Invitrogen, MA) and incubated at 37 °C in a 5 % CO₂ humidified atmosphere.

Total RNA was isolated using Vivantis GF-1 Total RNA Extraction Kit (Vivantis Technologies, Malaysia) according to the manufacturer's protocol. One µg of RNA was converted to cDNA with the Mastercycler Nexus GX2 (Eppendorf, Germany) using iScript cDNA Synthesis Kit (Bio-Rad, CA) according to the manufacturer's protocol. Quantitative PCR reaction was prepared using SsoFast Evagreen (Bio-Rad, CA, USA) in conjunction with specific primers as follows: CCL20 forward 5'-CAA CTT TGA CTG CTG TCT TGG AT-3' and reverse 3'-ACT TTT TTA CTG AGG AGA CGC AC-5', CCR6 forward 5'-TGC TCT ACG CTT TTA TTG GG-3' and reverse 3'-TTG TCG TTA TCT GCG GTC TC-5' and GAPDH forward 5'-GTG GAC CTG ACC TGC CGT CT-3' and reverse 3'-TGT CGC TGT GGG TGA GGA GG-5'. The reaction was performed through CFX96 Real-Time Thermocycler detection system (Bio-Rad, CA) including initial denaturation at 95 °C for 30 seconds, followed by 40 cycles of denaturation at 95 °C for 30 seconds and annealing/extension at 60 °C for 5 seconds. Melt curve analysis was performed at 65 °C to 95 °C. Relative quantitative expression is calculated by means of fold change (∆Ct) after normalizing with the reference gene GAPDH. The experiment was carried out in three independent sets.

Reverse transfection was carried out using Lipofectamine® 3000 (Thermofisher Scientific, CA) according to the manufacturer's protocol. Briefly, 15 pmol negative control siRNA (Silencer Select Negative Control No.1 siRNA, Thermo Scientific, CA) or CCR6 siRNA (siCCR6, s3214, Thermo Scientific, CA) was diluted in serum-free Opti-MEM medium (Thermo Scientific, CA). Diluted siRNA was incubated at the ambient temperature for 5 minutes before the addition of Lipofectamine® 3000. The complex was incubated at the ambient temperature for 20 minutes and transferred to cell suspended in an antibiotic-free medium. Transfected cells were incubated at 37 °C for 12-16 h until the cells were completely attached to the plate. The RPMI-1640 medium containing 5 % FBS was replenished. The transfection was incubated for 24 h until used in the subsequent experiment.

Total protein was extracted by RIPA lysis buffer (Cell Signaling Technology, MA) containing protease inhibitor cocktail (Cell Signaling Technology, MA). Protein concentration was measured by Bradford protein assay (Bio-Rad, CA). Cell lysate in the amount of 15 µg (for all proteins) or 50 µg (for N-cad detection in TFK-1) was loaded onto 10 % sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transferred onto nitrocellulose membrane (Bio-Rad, Germany). The membrane was blocked with 5 % skim milk and later probed overnight at 4 °C with primary antibodies including 1:1000 mouse anti-human CCR6 (14-1969-82, Thermo Fisher Scientific, CA), 1:10000 rabbit anti-human α-tubulin (2144, Cell Signaling Technology, MA), 1:5000 rabbit anti-human E-cadherin (3195, Cell Signaling Technology, MA), 1:5,000 rabbit anti-human N-cadherin (13116, Cell Signaling Technology, MA), 1:5,000 rabbit anti-human β-actin (4967, Cell Signaling Technology, MA). The membrane was washed 3 times and probed with appropriate secondary antibody including goat 1:20,000 anti-mouse-horseradish peroxidase (626520, Invitrogen, MA) and 1:5,000 goat anti-rabbit-horseradish peroxidase (7074, Cell Signaling Technology, MA). The chemiluminescence intensity was measured by ECL Western blotting detection (GE Healthcare, UK) under Gel Doc^TM XP⁺ Gel Documentation System equipped with Image Lab^TM Software version 6.0.0 (Bio-Rad, CA). The experiment was carried out in three independent sets.

Cells seeded in 12-well plate were incubated until it reaches 90 % confluency. The cell monolayer was scratched with a yellow pipette tip. The wound area was photographed at 0, 6, 9 and 12 h under the light microscope and measured by ImageJ software version 1.8.0 (National Institutes of Health, USA). Relative wound area compared to 0 h was calculated. The experiment was carried out in three independent sets.

HuCCT1 cells were transfected with siRNA in 96-well plate and incubated for 12 h. The medium was replaced with RPMI-1640 supplemented with 10 % FBS and further incubated for 24, 30 and 36 h followed by the addition of CellTiter 96 Aqueous One solution (Promega, WI). The reaction was incubated for 2 h and the absorbance at 490 nm was measured by Synergy HTX multi-mode reader (BioTek, VT). The assay was performed in three independent sets of triplicate wells.

HuCCT1 cells were transfected with siRNA for 24 h before collected by trypsinization. TFK-1 were collected directly without siRNA transfection. 1.5 x 10⁵ cells were resuspended in a FBS free-RPMI-1640 and plated inside the Transwell insert (Corning Incorporated, NY). The lower chamber containing a RPMI-1640 supplemented with 5 % FBS or 5 % FBS with recombinant CCL20 (rCCL20, Thermo Fisher Scientific, CA). After 12 h (HuCCT1) or 16 h (TFK-1) incubation at 37 °C, non-migrated cells inside the upper membrane were removed by cotton swab and washed twice with phosphate buffered saline. The membrane was fixed with absolute ethanol for 10 minutes at room temperature and stained with 0.5 % crystal violet. The number of migrated cells was counted from 10 randomly selected areas. The number of migrated cells in relation to the respective control was calculated. The experiment was carried out in three independent sets.

Statistics analysis was performed using student's t test from GraphPad Prism 7 (GraphPad Software, CA). The p-value of < 0.05 was considered statistically significant.

Microarray datasets were searched and selected from both GEO repository and PubMed literature to include as much as possible all available datasets (Figure 1(Fig. 1)). After a systematic selection, 4 datasets were identified including 2 datasets from GEO searching, GSE45001 (Sulpice et al., 2016[26]) and GSE57555 (Murakami et al., 2015[18]), and 2 datasets from PubMed searching, GSE26566 (Andersen et al., 2012[1]) and GSE32879 (Oishi et al., 2012[19]). All of these microarray datasets were obtained from CCA tissues and a minimum sample size of 10 % non-tumor tissues.

Differential expression analysis between tumor and normal tissues identified a total of 13649 unique DEGs including 1988, 2895, 12110, and 6464 from GSE45001, GSE57555, GSE26566 and GSE32879, respectively. 2899 DEGs were selected by filtering with an average fold-change of ± 1.5 in at least 3 datasets, of which 1610 were over-expressed and 1289 were under-expressed (Figure 2(Fig. 2)). The heatmap shows consistent trends in the expression of these DEGs in all datasets. As expected, DEGs were separated into clusters of over-expression (O) and under-expression (U). To aid the gene enrichment analysis, we further divided these major clusters into 5 (O1 - O5) and 5 (U1 - U5) sub-clusters, respectively (Supplementary Figure 1 and Supplementary Table 1).

Gene enrichment analysis was performed for each subcluster. Due to a large number of significantly enriched biological processes (FDR < 0.05) and redundancy of gene ontology terms, manual reviewed was performed on enriched processes. Networks, which cover high percentage of gene list were considered the representative of enriched networks for each subcluster. After overall analysis, we found that cell cycle, cell migration, response to cytokine and blood vessel development were the common enriched biological processes for subclusters O1 to O5. While lipid metabolic process, drug metabolic process, blood coagulation and regulation of complement activation were commonly enriched for each subclusters U1 to U5 (Table 1(Tab. 1)). Based on this result, over-expressed genes might play an important role for cancer characteristics such as cell cycle and cell migration. Therefore, genes in cluster O5 were considered for subsequent analysis because of their highest magnitude of over-expression. To link to the tumor microenvironment, genes in networks of cell migration and response to cytokine were reviewed. Ten genes including ADAM9, CCL20, COL1A1, COL1A2, CXCL5, FOXC1, LEF1, MIF1, MMP1, and WNT5A were enriched in both cell migration and response to cytokine. Among these genes, the role of CCL20 has not been studied in CCA. Moreover, this chemokine has only one specific receptor CCR6, their specific effect in CCA might be investigated by modulating their interaction. Therefore, CCL20 and CCR6 were chosen for functional validation in our study.

To investigate the role of CCL20 and CCR6 in CCA, mRNA expression of these genes were screened in HuCCT1 and TFK-1 cells using real-time RT-PCR. As shown in Figure 3a(Fig. 3), different expression levels of CCR6 and CCL20 was observed in these cell lines. Markedly high CCL20 expression was observed in HuCCT1, while the expression of both genes was comparable in TFK-1. Next, we examined the role of CCL20 in EMT process. The constitutive expression of both E-cadherin (E-cad) and N-cadherin (N-cad) was detected in HuCCT1 with 15 μg of protein used in Western blot assay, whereas only E-cad could be detected in TFK-1 (Figure 3b(Fig. 3)). Nevertheless, it was possible to detect N-cad with 50 μg cell lysate in TFK-1 (see Figure 5c). Altogether, the results highlighted the difference in expressions of CCL20 and EMT markers in HuCCT1 and TFK-1. Constitutive expression of CCL20 in HuCCT1 may be responsible for its higher expression of mesenchymal markers such as N-cad. To further validate the involvement of CCL20/CCR6 in EMT process, CCA cell lines were treated with siCCR6 or rCCL20 and migration assays were performed.

CCR6 knockdown assays were performed in both HuCCT1 and TFK-1. Western blot analysis showed a decrease in CCR6 expression of approximately 40 % and 30 % after 24 and 48 h reverse transfection with siCCR6 in HuCCT1, respectively (Figure 3c(Fig. 3)). The knockdown efficiency in TFK-1 was < 20 % (data not shown), therefore, HuCCT1 was selected for CCR6 knockdown assay and related functional study.

To examine the role of CCL20/CCR6 signaling in cell migration and proliferation, wound healing assay was performed in siCCR6 transfected HuCCT1. Compared to siNeg transfected cells, two-fold slower wound closure was observed in siCCR6 transfected HuCCT1 (Figure 3d(Fig. 3)). MTS assay was performed to determine the effect of siCCR6 on cell proliferation, there was no significant difference in proliferation rate between the siCCR6 and siNeg transfected HuCCT1 (Supplementary Figure 2). Nevertheless, we could confirm the effect of siCCR6 on cell migration ability through transwell assay in which the number of migrated cells decreased significantly (p < 0.05) compared to siNeg transfected HuCCT1 (Figure 4a and b(Fig. 4)). Therefore, the EMT markers, E-cad and N-cad, were further analyzed. Although the difference in E-cad expression was not observed, N-cad expression in siCCR6 transfected cells was slightly decreased suggesting the partial contribution of CCL20/CCR6 signaling in cell migration through EMT process (Figure 4c and d(Fig. 4)). We also analyzed the migration ability in rCCL20 treated HuCCT1, however, only slightly increased in the number of migrated cells could be observed (Supplementary Figure 3).

To avoid the effect of high constitutive CCL20 expression, we performed rCCL20 treatment and employed transwell assay to determine its effect on cell migration in TFK-1. Here we verified that the presence of rCCL20 in the lower chamber of the transwell plate substantially increases the number of migrated cells (Figure 5a and b(Fig. 5)). In addition, the increase in N-cad expression was observed in rCCL20 treated cells suggesting its effect on EMT process (Figure 5c and d(Fig. 5)).