Global alterations of DNA methylation in cholangiocarcinoma target the Wnt signaling pathway

Authors


  • Potential conflict of interest: Nothing to report.

  • Supported by grants from the Deutsche Forschungsgemeinschaft (DFG) to P.S. and K.B. (SFB/TRR77; Liver Cancer – From Molecular Pathogenesis to Targeted Therapies) and from the Helmholtz Foundation to C.P. The work was supported by the Biobank of the National Center of Tumor Diseases (NCT) Heidelberg.

Abstract

The molecular mechanisms underlying the genesis of cholangiocarcinomas (CCs) are poorly understood. Epigenetic changes such as aberrant hypermethylation and subsequent atypical gene expression are characteristic features of most human cancers. In CC, data regarding global methylation changes are lacking so far. We performed a genome-wide analysis for aberrant promoter methylation in human CCs. We profiled 10 intrahepatic and 8 extrahepatic CCs in comparison to non-neoplastic biliary tissue specimens, using methyl-CpG immunoprecipitation (MCIp) combined with whole-genome CpG island arrays. DNA methylation was confirmed by quantitative mass spectrometric analysis and functional relevance of promoter hypermethylation was shown in demethylation experiments of two CC cell lines using 5-aza-2′deoxycytidine (DAC) treatment. Immunohistochemical staining of tissue microarrays (TMAs) from 223 biliary tract cancers (BTCs) was used to analyze candidate gene expression at the protein level. Differentially methylated, promoter-associated regions were nonrandomly distributed and enriched for genes involved in cancer-related pathways including Wnt, transforming growth factor beta (TGF-β), and PI3K signaling pathways. In CC cell lines, silencing of genes involved in Wnt signaling, such as SOX17, WNT3A, DKK2, SFRP1, SFRP2, and SFRP4 was reversed after DAC administration. Candidate protein SFRP2 was substantially down-regulated in neoplastic tissues of all BTC subtypes as compared to normal tissues. A significant inverse correlation of SFRP2 protein expression and pT status was found in BTC patients. Conclusion: We provide a comprehensive analysis to define the genome-wide methylation landscape of human CC. Several candidate genes of cancer-relevant signaling pathways were identified, and closer analysis of selected Wnt pathway genes confirmed the relevance of this pathway in CC. The presented global methylation data are the basis for future studies on epigenetic changes in cholangiocarcinogenesis. (Hepatology 2014;59:544–554)

Abbreviations
BilIN 3

biliary intraepithelial neoplasia grade 3

BTC

biliary tract cancer

CC

cholangiocarcinoma

CGI

CpG island

DAC

5-aza-2′-deoxycytidine

DMR

differentially methylated region

ECC

extrahepatic cholangiocarcinoma

FFPE

formalin-fixed paraffin-embedded

GBAC

adenocarcinoma of the gallbladder

ICC

intrahepatic cholangiocarcinoma

IRS

immunoreactive score

MA

MassARRAY

MCIp

methyl-CpG immunoprecipitation

qPCR

quantitative polymerase chain reaction

TMA

tissue microarray

Cholangiocarcinomas (CCs) are malignant epithelial tumors that arise from bile ducts. Based on their anatomic location, CCs are divided into two major clinical phenotypes: intrahepatic (ICC) and extrahepatic (ECC). In the U.S. and Europe, risk factors for CCs include chronic biliary tract diseases, especially primary sclerosing cholangitis (PSC), hepatolithiasis, and several anatomical malformations of the bile ducts. Nonbiliary diseases such as heavy alcohol use, obesity, nonalcoholic fatty liver disease, chronic hepatitis C virus infection, and cirrhosis are more prevalent in ICC patients compared to the general population. For CC patients the prognosis is poor, with 5-year survival of ∼30% in ICC and ∼50% in ECC. Curative therapeutic options are restricted to surgery; however, CC patients frequently present in an unresectable state and respond poorly to conventional chemotherapy. Therefore, there is an urgent need for new therapeutic options that are based on an improved understanding of the molecular pathogenesis of CC.[1-3]

The molecular mechanisms underlying the development and progression of CC remain to be described. Pathologically altered gene functions including activation of KRAS, BRAF, or silencing of tumor suppressor genes, such as TP53, SMAD4, and p16ink4a have been described.[4]

It is now becoming evident that in addition to genetic mutations, epigenetic aberrations (e.g., DNA methylation and histone tail modifications) also play an important role in cancer development and progression.[5] As reversible epigenetic alterations can significantly affect gene activity without changing the primary DNA sequence, interference with the DNA methylation machinery discloses an outstanding opportunity to change expression and activity of cancer-relevant genes. Aberrant DNA methylation is an early and stable event in carcinogenesis and identification of specific DNA methylation signatures has great potential to become diagnostic markers for early detection and development of therapeutic regimens. DNA methylation is characterized by the modification of cytosine residues at the carbon 5 position.[6] In mammalian genomes, this modification almost exclusively occurs on cytosine residues in CpG dinucleotides, which are not equally distributed across the human genome but are concentrated in distinct GC-rich regions called "CpG islands" (CGIs). Approximately 60% of protein-coding mammalian genes harbor CGIs in their promoter region. Epigenetic patterns of the genome are dramatically different between normal and cancerous cells.[6] Abnormal changes in methylation patterns found in human cancers include global demethylation of DNA with simultaneous hypermethylation of CGIs. Hypomethylation at repetitive DNA elements is associated with genomic instability and can induce overexpression of oncogenes, whereas hypermethylation of CGIs located in promoter regions or in the vicinity of transcription start sites of tumor suppressor genes results in gene silencing.[7]

Anecdotal promoter hypermethylation of several tumor suppressor genes including RASSF1A, APC, CDKN2A, CDH1, DAPK1, FHIT, and MLH1 has been associated with CC; however, the degree of methylation and associated detection sensitivity and specificity varied between different studies.[8-11] These studies were restricted to candidate gene approaches using methylation-specific polymerase chain reaction (PCR) or array-based screening of limited target panels. To date, no attempt has been made to identify aberrant gene methylation and alterations of specific regulatory pathways in human CC on a global scale. Therefore, genes that may function as indicators for progression to a more aggressive CC phenotype remained unidentified.

In this study we performed a comprehensive screening for global patterns of aberrant DNA methylation in human CCs and identified genes in cancer relevant pathways including the Wnt axis as one of the major targets in this tumor type.

Materials and Methods

Patient Samples

Fresh-frozen tissues were obtained from CC patients that received surgery treatment at the University Hospital of Heidelberg between 2006 and 2010. All tissue specimens were instantly snap-frozen and stored in the Biobank of the National Center of Tumor Diseases Heidelberg. Ten ICCs (mass-forming type) and eight perihilar ECCs (ductal type) were included in the MCIp analysis. Due to the lack of patient-matched normal tissue specimens, one male ECC, five male ICC, and three female ICC samples were cohybridized with the respective common reference pools (CR1, male only; CR2, female only) (Table 1). Respective control samples were collected from normal bile duct tissue located at least 3 cm from the tumor. For verification, histopathologic evaluation by two experienced pathologists (B.G., W.W.) was performed on all samples used in this study. Only ductal adenocarcinomas were included in this study. For MassARRAY (MA) validation, formalin-fixed paraffin-embedded (FFPE) samples from an independent cohort of 93 CC (50 ECC, 43 ICC) patients (Supporting Table 1) were used. Tumors were staged and graded according to the 7th TNM Classification of Malignant Tumors and classified according to the current World Health Organization (WHO) tumor classification system (4th ed., 2010). The study was approved by the Institutional Ethics Committee (206/05).

Table 1. Clinicopathological Data of Cholangiocarcinoma Patients Analyzed With CGI Microarrays
PatientNormal SampleGenderAgeTumor TypeClinicopathological DataUICC Stage
  1. Abbreviations: CR1 and CR2, common reference 1 and 2, respectively.

1Yesmale68ECCpT2b, pN0, M0, G2, R0II
2Yesmale66ECCpT2b, pN0, M0, G2, R0II
3Yesfemale43ECCpT2b, pN0, M0, G2, R0II
4Yesmale71ECCpT2b, pN0, M0, G2, R0II
5Yesfemale63ECCpT2b, pN0, M0, G2, R0II
6Yesmale64ECCpT2b, pN0, M0, G2, R0II
7Yesfemale57ECCpT2b, pN0, M0, G2, R0II
8CR1male60ECCpT2b, pN0, M0, G3, R0II
9CR1male42ICCpT2b, pNx, M0, G2, R0II
10CR2female55ICCpT2b, pN1, M0, G2, R0IVa
11CR1male77ICCpT2a, pN0, M0, G2, R0II
12CR2female64ICCpT2b, pN0, M0, G2, R0II
13CR2female45ICCpT2b, pN1, M0, G3, R0IVa
14CR1male54ICCpT2a, pN0, M0, G2, R0II
15CR1male76ICCpT1, pNx, M0, G1, R0I
16CR1male54ICCpT2b, pNx, M0, G2, R0II
17Yesmale73ICCpT2a, pN0, M0, G2, R0II
18Yesfemale43ICCpT2b, pN0, M0, G2, R0II

Cell Lines and Drug Treatment

Human CC cell lines EGI-1 and TFK-1 were cultured in high glucose Dulbecco's modified Eagle's medium (DMEM) (Life Technologies) or RPMI 1640 (PAA), respectively, supplemented with 10% fetal calf serum (FCS). DNA demethylation treatment was performed with 0, 0.5, 1, or 2.5 μmol/L 5-aza-2′deoxycytidine (DAC) (Sigma-Aldrich, St. Louis, MO) for 72 hours. The medium was changed every 24 hours to maintain DAC concentrations.

DNA and RNA Isolation

DNA and RNA was isolated from fresh-frozen tissue with the AllPrep DNA/RNA Mini kit (Qiagen); DNA from FFPE samples was isolated with the QIAamp DNA FFPE Tissue Kit (Qiagen).

Methyl-CpG Immunoprecipitation (MCIp)

Three μg genomic DNA were sonicated in 125 μL H2O to a fragment size range of 200-800 bp with Biorupter NextGen (Diagenode, Belgium) using the following settings: 2 × 15 cycles with 30 sec ON/30 sec OFF at low energy. MCIp was performed as described[12] with minor modifications using the SX-8G IP-Star robot (Diagenode). In brief, fragmented DNA was incubated with 60 μg Methyl-CpG-binding domain 2 (MBD2)-Fc protein coupled to SiMAG protein A magnetic beads (Chemicell, Berlin, Germany). DNA was eluted from beads by incubation with increasing NaCl concentrations (300-1,000 mM). Fractions were desalted using the MinElute Purification Kit (Qiagen). Enrichment of methylated DNA fragments was controlled by quantitative PCR (qPCR) analyzing the differentially methylated sequence in the imprinted gene SNRPN.

Labeling, Hybridization, and Microarray Scanning

See the Materials and Methods section of the Supporting Material.

MCIp Microarray Data Preprocessing

Microarray images were analyzed using the Feature Extraction Software 10.5 (Agilent) and the standard ChIP protocol. The resulting data were processed using the R statistical environment, v. 2.13 (http://www.R-project.org) and bioconductor.[13] Background correction and log2-ratio transformation were performed according to the NormExp method with offset = 50 for two-color microarrays.[14] To reduce variations between cohybridized samples, intensity-based LOESS normalization on rank-invariant probes and negative controls was applied.[15] The data values resulting from these preprocessing steps are called M-values.

Identification of Differentially Methylated Regions and Global Methylation Analysis

Differentially methylated regions (DMRs) were defined and identified using the following stepwise criteria: 1) a "region" consists of a coherent sequence with more than one CGI microarray probe, and in which two vicinal probes are separated by ≤500 bp; 2) per tumor sample (array), only the top 5% probes with respect to positive or negative M-values are considered hypermethylated or hypomethylated probes, respectively; 3) a hypermethylated or hypomethylated region, i.e., a DMR, of a tumor sample is represented by at least two vicinal top 5% probes of same sign allowing a gap of a single probe with different sign or not belonging to the top 5%. According to our criterion (1), the CGI array covers 25,888 regions represented by 199,399 probes in total. To visualize the genomic location of DMRs and CGI array probes (refer to Fig. 1C,D; reference genome hg18/build 36), custom tracks were generated and loaded into the Integrated Genomics Viewer (IGV 2.1).[16] Patterning of global methylation was analyzed by counting for each hypermethylated array probe the number of samples in which the probe was among the top 5% probes. We compared the resulting distribution with the expected distribution of randomly selected probes (null distribution). The expected distribution is a binomial distribution with parameters sample number (n = 18) and P = 0.05 (fraction of top 5% probes).

Figure 1.

Global aberrant DNA methylation in CC. (A) Venn diagrams showing numbers of all hypermethylated and hypomethylated regions (DMRs) in MCIp-enriched methylated DNA of 8 ECC and 10 ICC samples analyzed with CGI microarrays. (B) Venn diagrams showing numbers of DMRs which are commonly present in more than two of eight (≥37.5%) ECC and more than three of ten (≥40%) ICC samples. Numbers in brackets refer to promoter DMRs. (C) IGV browser, genome-wide view (hg18/build 36; extents of the individual chromosomes shown on top) of DMRs common in both ECC and ICC (track CC) and present in more than two ECC (track ECC) or more than three ICC (track ICC) samples. In these tracks, red bars indicate hypermethylated, green bars hypomethylated regions. Tracks below indicate positions of RefSeq genes (black bars) and CGIs (green bars). (D) Zoomed-in IGV browser views of common, CC-related promoter DMRs associated with Wnt pathway genes SFRP2 (hypermethylated, upper part) and CCND1 (hypomethylated, lower part). Bars in tracks ECC probe and ICC probe indicate positions of array probes aberrantly methylated in more than two ECC or more than three ICC samples, respectively. Note that the upstream region of SFRP2, transcribed from right to left, harbors three CGIs of which the middle represents a DMR only in ECC.

MassARRAY (MA) Methylation Analysis

Quantitative DNA methylation analysis of candidate regions was performed with the MA technique. Briefly, genomic DNA was chemically modified with sodium bisulfite using the EZ methylation kit (Zymo Research, Orange, CA), PCR-amplified (primers in Supporting Table 2), in vitro transcribed, cleaved by RNase A, and subjected to matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF) mass spectrometry analysis.[17] DNA methylation standards (0%-100% methylated genomic DNA) were used to control for potential PCR bias. Methylation results are displayed as a heat map using the Multiple Experimental Viewer software (v. 4.3).[18]

cDNA Synthesis and qPCR

See the Materials and Methods section of the Supporting Material and Supporting Table 2.

Statistical Analyses

The statistical comparison of MassARRAY or qPCR data between two groups was done by the nonparametric Mann-Whitney U test, and P-values (two-tailed) were calculated in GraphPad Prism v. 5.0c (GraphPad Software). Differences in methylation levels of matched-pairs of tumor and normal tissue were analyzed with the nonparametric Wilcoxon matched-pair test. Stars indicate significance level (*P < 0.05, **P < 0.01, ***P < 0.001). A Mann-Whitney U test was used to compare the TMA expression in normal tissue with BilIN 3 and BTC. To test the correlation between SFRP2 expression and pT stage, an asymptotic Linear-by-Linear Association Test was used. Survival curve estimation for different subgroups of patients based on TMA expression was done using the Kaplan-Meier method. The Log-rank test was used to compare survival curves for subgroups. All P-values were two sided. P < 0.05 was considered significant.

Pathway Enrichment Analysis

A total of 1,850 genes associated with common, CC-related DMRs (see Results, below) were subjected to DAVID[19, 20] (release 6.7) and IPA (Ingenuity Systems, www.ingenuity.com) pathway analysis to identify CC-specific pathways. Statistical enrichment analyses were done using the default parameters of the pathway identification tools.

Clinicopathological Characteristics of the Biliary Tract Cancer (BTC) Cohort (TMA)

Tissue samples from 223 patients (median age 65.5 years) who underwent bile duct and/or liver surgery in the University Hospital Heidelberg between 1995 and 2010 were included in this study. Only patients with primary adenocarcinomas of the biliary tract and without other known malignancies at the time of diagnosis were included. BTCs consisted of 98 ECCs, 56 ICCs, and 69 adenocarcinoma of the gallbladder (GBAC). For 88 cases, corresponding histologically normal tissue, and for 66 cases samples of corresponding biliary intraepithelial neoplasia grade 3 (BilIN 3) were included. For most patients, complete clinicopathological data including sex, age, tumor grade, lymph- and hemangiosis carcinomatosa, perineural tumor invasion, TNM/UICC status, as well as overall survival times were included in this study. Patients who received radiochemotherapy prior to surgery were excluded. Detailed clinicopathological data of the cohort is given in Supporting Table 3.

TMA Preparation, Immunohistochemistry, and TMA Analysis

See the Materials and Methods section of the Supporting Material.

Data Access and Supporting Material

The global methylation data from this study have been submitted to the NCBI Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo/) under accession number GSE44965.

Supporting material is available for this article and consists of: (1) Supporting Figs. S1-7, (2) Supporting Tables S1-S22, (3) Track files for UCSC browser for the CGI array.

Results

Global Aberrant CpG Island Methylation in CC

Using CGI microarrays, we analyzed genome-wide CGI methylation in 18 CC samples, including seven ECC and two ICC matched pairs. By applying stringent criteria for definition of DMRs (refer to the Materials and Methods), we identified a total of 7,179 hypermethylated and 8,892 hypomethylated regions (DMRs), of which 3,454 and 4,860, respectively, were shared by the ECC and ICC samples (Fig. 1A; Supporting Tables 4-7). In order to narrow down the most characteristic CC-related DMRs, we confined the selection of DMRs to those that were common to at least 3/8 ECC (37.5%) and 4/10 (40%) ICC cases. We identified 1,371 common hypermethylated CC-related regions, of which 400 were promoter DMRs. Furthermore, we found 911 common hypomethylated CC-related regions, of which 159 were promoter DMRs (Fig. 1B-D; Supporting Tables 8, 9). Specific for ECC (present in ≥37.5% ECC and absent in ≥40% ICC cases) were 558 and 1,332 hyper- and hypomethylated regions, respectively. We discovered 571 and 897 hyper- and hypomethylated regions specific for ICC (Fig. 1B; Supporting Tables 10-15).

At the chromosomal level, the majority of autosomes, 16/22, showed more hypermethylated than hypomethylated CC-related regions (Supporting Table 16); chromosome 2 displayed the highest density of hypermethylated regions with 89.3% (142 from 159 DMRs), while chromosome 9 showed the highest preponderance of hypomethylated regions with 75% (75 from 100 DMRs). The most striking imbalance between hyper- and hypomethylated regions, however, was observed on the X-chromosome, with 94.8% (165 from 174 DMRs) hypomethylated regions (Fig. 1C; Supporting Fig. 1). This imbalance was weakest in the five matched male samples but consistently strong in both matched and unmatched female and in unmatched male samples (data not shown), ruling out a technical or sampling bias.

We also observed autosomal regions with skewed aberrant methylation, the most striking example being the distal 30 Mb of chromosome 1p with 133 hypomethylated but only two hypermethylated regions (Supporting Table 17, see Supporting Fig. 1). Distal chromosome 1p contains various suspected tumor suppressor genes and is often deleted in cancer.[21] The vast majority, 107 of 133 hypomethylated regions, could be ascribed to intragenic hypomethylation, which is a suspected cause or consequence of dysregulation of the affected genes.[22] In CC, intragenic hypomethylation affected the frequently deleted 1p genes CHD5, CAMTA1, KIF1B, and CASZ1, and 13 hypomethylated regions alone resided in the gene body of the transcriptional regulator PRDM16; only six of 133 were promoter-associated DMRs. Other chromosomal stretches of several Mb in length with strong bias to either hyper- or hypomethylation were identified on chromosomes 6, 9, 16, and 19, three of these areas residing in distal chromosomal q-arms (Supporting Table 17, Supporting Fig. 1). The affected chromosomal stretches contained a variety of genes encoding regulatory factors such as CREBBP, EGFL7, and zinc-finger proteins.

Previous work has demonstrated that human malignancies display nonrandom patterns of aberrant CGI methylation.[23] We performed a binomial test for global methylation patterns with the CC-related hypermethylated array probes for all n = 18 samples (matched and nonmatched) and observed a nonrandom distribution (Supporting Fig. 2) among the tested cases, indicating CC-related occurrence of aberrant DNA methylation.

Wnt Signaling Pathway Is Affected in CC

For the identification of CC-related signaling pathways which might be affected by aberrant DNA methylation, we combined the lists of common hypermethylated (1,371 DMRs) and hypomethylated (911 DMRs) regions (refer to Fig. 1B; Supporting Tables 8, 9) and obtained a compilation of 1,850 nonredundant, DMR-associated genes (Supporting Table 18). Analysis of this gene list with the functional annotation tool DAVID and with the Ingenuity pathway tool (see also Materials and Methods) revealed various significantly enriched cancer-relevant pathways like the Wnt, TGF-β, PI3K, MAPK, and NOTCH signaling pathways (Supporting Tables 19, 20). The Wnt pathway scored as common best at the third, fourth, and fifth position in the Panther (www.pantherdb.org), KEGG (www.kegg.jp) (both screened by DAVID), and Ingenuity pathway databases, respectively. The three databases together identified 65 aberrantly methylated Wnt pathway genes, of which 13 genes were shared by all three databases and 22 genes by at least two (Supporting Fig. 3, Supporting Table 21). With the exception of only two genes, the corresponding DMRs were either located in a core promoter CGI or in a CGI associated with an alternative promoter. Among the 20 promoter DMRs, 16 were hyper- and four were hypomethylated (Supporting Table 22).

We selected six promoter DMRs associated with the Wnt pathway genes SOX17, WNT3A, DKK2, SFRP1, SFRP2, and SFRP4 for validation by MA analysis. All DMRs except that of SFRP4 were confirmed as being significantly hypermethylated (Fig. 2A). In an extended MA validation with 93 matched sample pairs of an independent ICC patient cohort, we confirmed the same DMRs as in the previous validation (Fig. 2B). Correlation of methylation levels and patient survival did not reach statistical significance (data not shown).

Figure 2.

CC-related hypermethylation of Wnt pathway gene promoters. (A) Technical validation of the Wnt pathway genes SOX17, WNT3A, DKK2, SFRP1, SFRP2, and SFRP4 by MassARRAY. The methylation heatmap is shown on the left. Lines represent normal and tumor samples as indicated, columns depict single CpG units. The right panel shows the average methylation levels of normal (N, gray circles) and tumor (T, red circles) samples. Bar indicates mean methylation value of the respective group (n.s. = not significant). (B) Methylation levels of the six Wnt pathway genes in an independent cohort of 93 CC patients. Plots show average methylation levels of tumor (ICC as blue and ECC as green circles) and matched normal (N, gray circles) samples. Bar represents mean methylation value of the respective group (n.s. = not significant).

Aberrant Expression of Wnt Signaling Genes in CC Cell Lines

We evaluated the impact of promoter methylation on the expression of selected Wnt pathway genes in CC cell lines EGI-1 and TFK-1, with and without treatment, using the DNA demethylating agent DAC. Upon DAC treatment, we observed a significant decrease of DNA methylation levels in LINE repetitives, DNA elements usually monitored for the success of global DNA demethylation, by 21% and 11%, respectively (Fig. 3A). Treatment also resulted in considerably reduced (22%-45%) promoter methylation of Wnt pathway genes SOX17, WNT3A, DKK2, SFRP1, SFRP2, and SFRP4 in at least one of the cell lines compared to untreated cells (Fig. 3A). Moreover, we observed a strong reactivation of the genes SFRP1 and SFRP2 in both cell lines and of SFRP4 and SOX17 in at least one of the two cell lines upon DAC administration (Fig. 3A). The suppressing influence of promoter DNA methylation on gene expression could be particularly demonstrated by genes SFRP1 and SFRP2, whose decreasing promoter methylation levels, as a consequence of rising DAC concentrations, were paralleled by their reactivation (Fig. 3B; Supporting Fig. 4). Comparing by qPCR the messenger RNA (mRNA) levels of the above chosen candidate genes in four normal and 13 CC specimens revealed no significant differences between the two groups. Moreover, there was no correlation between methylation and expression for this sample set (data not shown).

Figure 3.

Reactivation of Wnt pathway genes in CC cell lines EGI-1 and TFK-1 by demethylating drug treatment. (A) Gene promoter demethylation upon treatment with 2.5 μM DAC was paralleled by an increase of expression for genes SFRP1 and SFRP2 (in both cell lines) and for genes SOX17 and SFRP4 in at least one of the cell lines. Average methylation in amplicons and relative gene expression levels are indicated by color codes. Numbers in the respective lines indicate differences in methylation (%) or fold expression in DAC-treated versus untreated cells. (B) Decreasing promoter methylation levels (right y-axis) of SFRP2 in CC cell line EGI-1 by increasing DAC concentrations were associated with increasing expression levels (left y-axis) of the gene. Left y-axis: 3xHK = three housekeeping genes.

Protein Expression of SFRP2 Is Stepwise Down-Regulated in All BTC Subtypes

In order to confirm that our results from the screening approach and the in vitro data are of relevance in human cholangiocarcinogenesis, we examined the expression of the Wnt pathway protein SFRP2 in a well-characterized BTC cohort, including information on anatomical subtypes (ICC, ECC, and GBAC). Expression data were correlated with comprehensive clinicopathological and survival data of the patients. SFRP2 showed a strong cytoplasmic staining in normal cholangiocytes of gallbladder and bile ducts (Fig. 4A,B) but only weak staining in neoplastic tissue (Fig. 4C-F). Statistical analyses of the TMAs revealed a significantly higher immunoreactive score (IRS) of SFRP2 in human normal biliary epithelium compared to BTC (Fig. 5). According to our scoring index, 62.5% (n = 55) of normal samples showed a high or moderate SFRP2 expression, while in the BTC group only 11.6% (n = 18) of samples showed similar expression levels. Overall, SFRP2 expression dropped below the 25th percentile (IRS = 0-4) compared to normal tissue in 70.1% (n = 108) of the BTC samples. Considering diverse disease stages, we observed a significant down-regulation of SFRP2 in BilIN 3 followed by an even more pronounced SFRP2 reduction in invasive tumors (Fig. 5A). These findings were consistent in all BTC subgroups (Fig. 5B-D).

Figure 4.

Immunohistochemistry of SFRP2 in CC. (A,B) Normal cholangiocytes exhibited a strong cytoplasmic staining of SFRP2, while staining was reduced in (C,D) ICCs, and (E,F) ECCs. Original magnification: 50×, 200× (inset).

Figure 5.

Stepwise reduction of SFRP2 expression in cholangiocarcinogenesis. (A) SFRP2 staining was significantly reduced in preinvasive BillN 3 and decreased even more in invasive BTC. (B) Subgroup analysis showed the same picture for ICC, (C) ECC, and (D) GBAC. Due to technical reasons, some TMA dots could not be evaluated, thereby decreasing the total number of cases (see Materials and Methods).

A hallmark of activated canonical Wnt signaling is the stabilization of cytoplasmic β-catenin and its translocation to the nucleus.[24] We analyzed the BTC TMAs for β-catenin expression (nuclear and cytoplasmic) and correlated the data with SFRP2 expression. Approximately 50% of the BTC cases (n = 76) showed absent or low SFRP2 (IRS 0-4) and high cytoplasmic β-catenin expression levels (IRS 9-12) (Supporting Fig. 5A). Nuclear β-catenin expression (score 1-3) was observed in 36.4% (n = 56) of all BTC cases. Within this group of 56 BTC cases, the vast majority, 78.6% (n = 44), showed low or absent (IRS 0-4) SFRP2 expression levels (Supporting Fig. 5B). However, correlation analysis showed no significant correlation between SFRP2 expression and either nuclear or cytoplasmic β-catenin expression.

By correlating SFPR2 expression with clinicopathological data, we found a significant inverse correlation between SFRP2 expression and pT-status (P = 0.01), while no significant correlation was observed for pN- or M-status and SFRP2 expression (Supporting Fig. 6). We correlated overall survival with SFRP2 expression in 126 BTC patients. The vast majority showed absent to low (n = 112, IRS 0-4) or intermediate (n = 10, IRS 5-8) SFRP2 expression. These two groups displayed very similar survival curves. Only four cases, two ECC and two GBAC, showed strong SFRP2 expression (IRS 9-12) which correlated with improved overall survival, yet without reaching statistical significance (Supporting Fig. 7).

To further investigate the SFRP2 expression profile in cholangiocarcinogenesis, we included normal biliary tissue samples from patients operated on for a nonmalignant disease. Additionally, SFRP2 expression of biopsy material from inoperable CC patients was examined. Immunohistochemistry showed a strong cytoplasmic staining of SFRP2 in normal biliary tissue samples and a markedly reduced SFRP2 expression in biopsies from inoperable CC patients, thereby mirroring the TMA results of the BTC cohort (Supporting Figs. 8, 9).

Discussion

In human CC, previous molecular studies have identified only a few cancer-relevant genes, and our understanding of the underlying molecular mechanisms involved in initiation and progression of CC are still fragmentary. Known CC-related genomic changes are, for example, activating somatic mutations in oncogenes KRAS, BRAF, ERBB2, and PIK3CA as well as inactivating alterations of tumor suppressor genes such as SMAD4, CDKN2A, and TP53.[4] Furthermore, transcriptomic analyses of human CC samples demonstrated drastic changes of gene expression.[25, 26] As genome-wide epigenetic alterations in CC have not been examined so far, we analyzed by means of a CGI array approach global DNA methylation changes in 18 CC, 8 ECC and 10 ICC, patient samples. Although our experimental setting even enabled us to distinguish between the methylomes of ECC and ICC, as demonstrated by confirmative validation of ICC-specific promoter hypermethylation, we focused here on aberrant DNA methylation common in CC. Our systematic approach identified thousands of CC-related, aberrantly methylated genes and, hence, provides a valuable repository for forthcoming studies on the methylome of CC in general, and specifically on the methylome of ECC and ICC. Exemplary validation of DMRs by an independent quantitative technique, mass spectrometry, contributed to enhanced reliability of the array-based results.

We observed an increase in hyper- over hypomethylated regions in the highly methylated DNA fraction of the CC methylome and identified skewed patterns of aberrant methylation affecting complete chromosomes, such as highly predominant hypermethylation on chromosome 2 and highly predominant hypomethylation on the X-chromosome. These patterns were also observed for larger genomic stretches of up to 30 Mb in length on chromosomes 1, 6, 9, 16, and 19. Some of these areas are located in distal p- or q-arms, which might be particularly prone to cancer-related, long-range epigenetic remodeling, a now more frequently encountered phenomenon due to refined technical advances in the characterization of the cancer epigenome.[27, 28] Our findings of long-ranging epigenetic remodeling in CC, based on the CGI array data, may now be scrutinized by interrogating the complete CC methylome through whole-genome bisulfite sequencing.

Extensive biological pathway analysis screening three pathway databases, with a list of 1,850 CC-related aberrantly methylated genes as input, clearly identified the Wnt signaling pathway. This is a well-known cancer-related pathway[24] and has been described recently, for example, to be epigenetically affected in hepatocellular carcinoma[29]; yet an involvement of Wnt signaling in Western world-CC was unreported thus far. Functional studies by administration of the demethylating agent DAC to two CC cell lines underscored that several of the identified Wnt signaling genes might be epigenetically dysregulated in CC. Previous reports described reduced membranous expression of α-catenin, β-catenin, and E-cadherin in the vast majority of CCs, thus implicating activation of the Wnt/β-catenin signaling pathway during progression of CC.[30, 31] In ICC, altered expression of β-catenin without genetic mutation has been reported.[30, 32] Molecular mechanisms underlying this phenomenon remain elusive. One of the candidate genes identified and highlighted in our study, SFRP2, encodes a member of the secreted frizzled-related protein family. Epigenetic silencing of secreted frizzled-related proteins has been found in other human malignancies, including colon cancer[33] and chronic lymphocytic leukemia.[34] SFRP2 shares at its amino-terminus homology with the cytosine rich domain of Fz receptors but lacks the transmembrane or cytoplasmic domain which is required for signal transduction into the cell.[35] As SFRP2 competes with Fz receptors for WNT binding and inhibits signaling, its reduction in CC may be associated with stabilization of β-catenin. Using TMAs of BTC patients, we tested a correlation between the expression of SFRP2 and β-catenin (nuclear and cytoplasmic), showing no significant result. Epigenetic alterations are suspected to facilitate the manifestation of somatic mutations in key signaling pathways, as exemplified by the Wnt pathway in gastrointestinal cancer.[36] Epigenetic silencing of SFRPs can provoke abnormal activation of Wnt signaling and, consequently, cell proliferation. Survival and proliferative advantages of affected cells enable the accumulation of genetic mutations in other components of the Wnt signaling pathway.[37, 38] In cholangiocarcinogenesis, epigenetic silencing of SFRP2, as shown in this study, may be involved in the described cancer genetic/epigenetic interplay, thereby increasing the likelihood for mutation in genetic gate keeper genes.

In summary, we provide a comprehensive repository of CC-related DMRs suitable for the identification of novel biomarkers, signaling pathways, and tumor suppressor genes. We identified the Wnt pathway and SFRP2 in particular as a putative tumor suppressor in cholangiocarcinogenesis. Future studies should integrate the presented epigenetic alteration patterns with results of high-resolution genomic and expression data in CC, which would be pivotal for a molecular classification of human cholangiocarcinogenesis.

Acknowledgment

We thank Oliver Mücke (Division of Epigenomics and Cancer Risk Factor, German Cancer Research Center), Andrea Hain, and John Moyers (Institute of Pathology, University of Heidelberg) for their excellent technical assistance. We also thank the NCT tissue bank Heidelberg for its support.

Ancillary