Epigenetic profiling of tumor DNAs may reveal important new theranostic targets to improve prognosis and treatment of advanced cancer patients. In this study, we performed a genome-wide profile of DNA methylation patterns in sporadic breast tumors by using the HumanMethylation27 BeadChips to assess relationships between DNA methylation changes and patient tumor characteristics. The arrays identified 264 hypermethylated loci/genes present in genomic CpG islands. Hierarchical clustering based on methylation levels divided the specimens into three distinct groups, within which certain clinical features also clustered. Statistically significant differences were determined between overall methylation levels of these clusters and estrogen receptor and progesterone receptor (ER/PR) status (P = 0.001), tumor relapse (P = 0.035), and lymph node metastasis (P = 0.042). We identified several individual methylated genes associated with clinical features, including six genes (RECK, SFRP2, UAP1L1, ACADL, ITR, and UGT3A1) that showed statistical significance between methylation and relapse-free survival. Notably, the RECK gene in this group has been associated in other cancers with poorest prognosis. Among the leading relapse-associated genes and the genes associated with ER/PR status, we sequenced an independent set of paired normal/tumor breast DNA samples to confirm tumor specificity of methylation. Further, we carried out quantitative real-time reverse transcriptase PCR to confirm reduced expression in methylated tumors. Our findings suggest the utility for the DNA methylation patterns in these genes as clinically useful surrogate markers in breast cancer, as well as new molecular pathways for further investigation as therapeutic targets. Cancer Res; 71(8); 2988–99. ©2011 AACR.
Epigenetic silencing of tumor-related genes due to hypermethylation has recently emerged as one of the pivotal alterations in cancer development (1, 2). Hence, cancer is not only a polygenetic disease, but also a polyepigenetic disease. There are 3 main clinical uses for the DNA methylation patterns associated with cancer development and progression. First, DNA methylation patterns that are cancer cell specific and occur at an early stage of cancer development makes them ideal biomarkers for early cancer detection. Second, DNA methylation patterns can also be used for prognostic purposes such as survival and/or relapse. Third, unlike gene mutations, hypermethylation of gene promoters can be reversed by treatment with demethylating agents and are regarded as optimal targets for what is now known as epigenetic therapy. The clinical use of demethylating agents such as 5-aza-2′-deoxycytidine (decitabine) and 5-azacytidine (Vidaza) have had the greatest success in the field of hematologic malignancies. In 2004, FDA approved the use of 5-azacytidine for the treatment of myelodysplastic syndromes (1).
It is estimated that more than 1 million women are diagnosed with breast cancer globally every year, and a substantial number of these women will die from the disease (3). The economic cost of treating breast cancer patients is substantial. Because of the heterogeneous nature of the disease, clinical progression and response to therapy is difficult to predict with the current prognostic factors and hence treatment regimens are not as effective as they should be. In this study we used the HumanMethylation27 BeadChips to explore associations between breast tumor methylation profiles and important tumor characteristics such as tumor stage, metastasis, relapse, survival, and hormone receptor status.
Materials and Methods
Cell lines and patient DNA samples
We used a total of 9 breast cancer cell lines (HCC1806, HCC1419, HCC1937, HCC1143, HCC1395, MCF-7, T47-D, MDA-MB-231, and HTB19) during this study. All sporadic breast tumor samples were ductal carcinomas, we used one cohort of 39 tumors (tumor characteristics described in ref. 4 and in Fig. 3) and a further independent cohort of 20 tumors with matched normal breast tissue DNA. None of the primary tumor samples were microdissected. All samples were acquired with ethical consent.
We used the Illumina Infinium HumanMethylation27 BeadChips (Illumina). The BeadChip contains 27,578 highly informative CpG sites covering more than 14,000 human RefSeq genes. This allows researches to interrogate all these sites per sample at a single nucleotide resolution. Bisulfite modification of DNA and chip processing and data analysis were carried out by following the manufacturer's manual and was done at the Wellcome Trust Centre for Human Genetics Genomics Lab, Oxford, UK. Array results were run through BeadStudio software (Illumina).
Background normalized average β values were considered hypermethylated if no normal had a β value greater than 0.25 and if either (a) had a β value greater than 0.5 in more than 20% of samples, or (b) had a difference of at least 0.4 between sample β value and mean normal β value in more than 20% of samples. For subsequent list analysis by Fisher's exact tests, samples were given a value of 0 or 1 representing unmethylated or methylated, respectively, as determined by the above criteria.
Hierarchical clustering was carried out in Cluster3. When clustering all samples for all probes, Euclidean distance and single linkage were used, when clustering the 291 hypermethylated loci for the 39 tumors, Euclidean distance and average linkage were used. Clustering results were viewed by using the JavaTree program.
Assessment of overall methylation levels with clinical features were analyzed by ANOVA in MedCalc, significance was taken to be P < 0.05. Association of single loci with clinical features were assessed by Fisher's exact test, all P values were corrected by false discovery rate (FDR) and taking significance to be P < 0.05. Kaplan–Meier analysis was carried out in MedCalc taking into account Bonferroni correction when assessing significance, samples were considered significant if P < 0.05. Student's t test was used to determine whether there was significance between quantitative expression of normals and tumors, again significance was taken as P < 0.05.
The Database for Annotation, Visualization and Integrated Discovery (DAVID) was used to identify groups of genes within the selected list of 263 that were within groups based on related functions and KEGG (Kyoto Encyclopedia of Genes and Genomes) pathways relevant to carcinogenesis by using the functional annotation chart and pathways analysis functions within the functional annotation tool.
Methylation analysis of individual genes
Combined bisulfite restriction analysis (CoBRA) was used to assess the methylation status of genes in cell lines and a separate tumor/normal cohort. All DNA were modified by using an EpiTect kit (Qiagen) according to the manufacturers' instructions. All CoBRA PCR were semi-nested, except ACADL, which was fully nested because of unsatisfactory amplification by using seminested primers. CoBRA primer sequences are shown in Supplementary Table S1. Primary PCRs were touchdown PCRs with a gene-specific final annealing temperature. Secondary PCRs used 5 μL primary product as starting material and were also touchdown PCRs with a longer cycle number (45 instead of 40). PCR products were digested with BstUI (Fermentas UK) overnight at 37°C prior to visualization on a 2% agarose gel.
Samples were cloned and sequenced to validate the array and confirm CoBRA results. PCR products were cloned into the pGEM-T easy vector (Promega) according to the manufacturers' instructions. Up to 12 colonies were picked per sample and sequenced following single-colony PCR by using the following primers: forward, 5′-TAATACGACTCACTATAGGG-3′; reverse, 5′-ACACTATAGAATACTCAAGC-3′. Methylation indexes were calculated as a percentage of the number of methylated CpGs out of the total CpGs sequenced.
Cell line expression analysis
Breast cell line genomic DNA was demethylated by treatment with 5 μmol/L 5′-aza-2′-deoxycitidine (5-azadc) over 5 days. Daily media changes were made with the addition of fresh 5-azadc each time. All cell lines were maintained in Dulbecco's modified Eagle's medium (Sigma-Aldrich) supplemented with 2 mmol/L glutamine and 10% fetal calf serum at 37°C, 5% CO2. RNA bee (AMS Biotechnology) was used to extract total RNA from cell lines according to the manufacturers' instructions. One microgram total RNA was converted to cDNA by using Superscript III (Invitrogen) and random hexamer primers (Fermentas UK). Reverse transcriptase (RT)-PCR was then carried out by a touchdown PCR program with a final annealing temperature of 56°C. All expression primer sets are shown in Supplementary Table S2.
Quantitative real-time RT-PCR was carried out as described previously (4). Briefly, cDNA was made from total RNA extracted from frozen normal and tumor breast tissues. We quantified transcripts of the TBP (TATA box-binding protein) gene as the endogenous RNA control. Results, expressed as N-fold differences in target gene expression relative to the TBP gene (termed Ntarget), were determined by the following formula: Ntarget = 2ΔCtsample, where the ΔCt value of the sample was determined by subtracting the average Ct value of the target gene from the average Ct value of the TBP gene. The Ntarget values of the samples were subsequently normalized such that the mean ratio of the normal breast samples would equal a value of 1. Primer sequences provided on request. PCR was carried out by using the SYBR Green PCR Core Reagents kit (Perkin-Elmer Applied Biosystems). The thermal cycling conditions comprised an initial denaturation step at 95°C for 10 minutes and 50 cycles at 95°C for 15 seconds and 65°C for 1 minute. Experiments were conducted with duplicates for each data point.
HumanMethylation27 BeadChip data were obtained from 39 sporadic breast tumors and 4 matched breast tumor/normal pairs. Comparison of overall DNA methylation between normal breast DNA samples showed little variance, with high correlation coefficient values (R2) ranging from 0.974 to 0.9302 (Supplementary Fig. S1). Additionally, hierarchical clustering (Euclidean, single linkage) of all the probes grouped all 4 matched normals together. Of 4 of the matched tumor/normal pairs, 3 showed low correlations (R2 = 0.724, 0.834, and 0.909), indicative of varying levels of aberrant methylation, whereas the fourth sample showed a correlation similar to normals (R2 = 0.9381). This suggested that altered methylation levels could be important in a large number of the sporadic breast tumor cohorts analyzed whereas for some tumors other genetic factors may be involved.
Identification of hypermethylated CpG loci
To identify the CpG loci showing the most significant tumor-specific changes in methylation all probes with one or more normal sample showing a β value greater than 0.25 were excluded (n = 11,211). Remaining probes were only considered hypermethylated if more than 20% of the tumor samples had either (a) β values of 0.5 or more (n = 261 probes, 242 loci) or (b) a difference between the tumor sample β value and the mean normal β value of 0.4 or more (n = 202 probes, 182 loci). All imprinted and X chromosome genes were removed from the analysis and a final selection of 291 probes representing 264 genes was chosen (Supplementary Table S3; Supplementary Fig. S2). The selected genes were analyzed by PubMeth, identifying 34 genes out of the 264 (12.9%) that have been previously shown to be methylated in cancer, including 8 genes in breast cancer representing 10.3% of the genes in PubMeth methylated at a frequency of more than 20% in breast cancer (Supplementary Table S4).
Functional analysis of this selection by the DAVID bioinformatics resource (http://david.niaid.nih.gov) showed the presence and enrichment of genes involved in functions related to tumorigenesis: cell adhesion (24 genes), regulation of cell proliferation (21 genes), negative regulation of cell death (15 genes), cell migration (12 genes), cell–cell adhesion (11 genes), regulation of cell cycle (10 genes), and tumor suppressor (6 genes; Supplementary Fig. S3). Similarly, DAVID KEGG pathway analysis showed genes present from relevant pathways: pathways in cancer (12 genes), cell adhesion molecules (6 genes), and the Wnt signaling pathway (7 genes; Supplementary Fig. S3).
Further confirmation and analysis of the 291 selected probes was performed by bisulfite sequencing of (a) selected genes previously known to be associated with breast cancer, (b) analysis of most highly methylated genes, and (c) analysis to identify methylation specific to clinical features.
Bisulfite sequencing confirmation of probe methylation
For 3 selected hypermethylated genes previously known to be associated with breast cancer (SFRP5, CLDN6, and SIPA1; refs. 5–8), bisulfite sequencing was done on a region surrounding the CpG island probe for examples of high and low β values. Sequencing showed that β values greater than 0.5 tended to be both highly methylated at the probe position and be indicative of wider CpG island methylation, whereas low β values showed little methylation at the probe position, although in some cases, a greater level of methylation was observed in the surrounding region compared with the probe (Fig. 1). These results confirm the correlation of the β value with methylation levels at the specific CpG being measured by the array.
Analysis of breast tumor hypermethylated genes
The selected genes were further filtered into 2 groups; the top 10 most highly methylated genes with true CpG islands (amenable to CoBRA; Supplementary Fig. S4a) and genes with more than a single selected probe (Supplementary Fig. S5). Selected genes were then analyzed for methylation in breast cancer cell lines, expression in breast cancer cell lines after treatment with 5-azadc and demonstration of tumor-specific methylation in an additional independent set of 20 tumor/normal matched pairs, where appropriate.
Of the 3 most highly methylated genes with true CpG islands (ABCA3, COX7A1, and SST), COX7A1 and SST showed increased expression after 5-azadc treatment in methylated breast tumor cell lines (Supplementary Fig. S4b). In an independent cohort of breast tumor/normal pairs, COX7A1 showed increased methylation in 70.6% (12 of 17) of tumors compared with corresponding normal tissue and a further 5 showed equal amounts of low-level methylation in tumor and corresponding normal tissue. SST showed tumor-specific methylation in 31.6% (6 of 19) of pairs and increased methylation in a further 31.6% (6 of 19) of tumors compared with corresponding normal tissue DNA. One tumor showed equal amounts of methylation in the tumor and normal and no methylation in the remaining tumors (Supplementary Fig S4b).
A total of 24 genes had more than a single selected probe, 4 of which (17%) had been previously shown to be methylated in breast cancer (APC, CCND2, GSTP1, and PTPRO; Supplementary Fig. S5; refs. 9–12). Further 2 genes (CDKL2 and ZNF154) from the above 24 genes were analyzed in this study. Both CDKL2 and ZNF154 showed increased expression after 5-azadc treatment in methylated breast tumor cell lines and also showed altered methylation patterns in matched tumor/normal-paired DNA (Fig 2). CDKL2 and ZNF154 showed tumor-specific methylation in 25% (4 of 16) and 56.3% (9 of 16), respectively, and increased methylation in tumors compared with low-level methylation of matched normal samples in 12.5% (2 of 16) and 25% (4 of 16), respectively. CDKL2 and ZNF154 also showed equal methylation in tumor and normal DNA in 18.8% (3 of 16) and 6.3% (1 of 16) of samples. No methylation was detected in the remainder of the tumors.
Identification of gene hypermethylation specific to clinical features
Initially, to investigate whether hypermethylation seemed to be associated with any clinical feature, hierarchical clustering was conducted on the 291 selected probes (Fig. 3). This split the sporadic tumor samples into 3 distinct methylation-based clusters: high methylation (18 tumors; mean β value of 0.433), intermediate methylation (7 tumors; mean β value of 0.317), and low methylation (12 tumors; mean β value of 0.176), with 2 outliers (Fig. 3). The high methylation cluster associated with samples that relapsed (high, 12 of 18; intermediate, 1 of 7; and low, 3 of 12) and retained both the estrogen and progesterone receptor (ER/PR; high, 14 of 18; intermediate, 1 of 7; and low, 2 of 12), whereas the low methylation cluster associated with the triple-negative tumors (low, 7 of 12; intermediate, 2 of 7; and high, 0 of 18; Fig. 3).
ANOVAs were carried out by using either the mean β value (methylation) or the disease-free survival for each sample to compare between clinical features. The clusters showed no statistically significant variation between disease-free survival (P = 0.162) and no statistically significant variation was observed between methylation and tumor stage (P = 0.214) or menopause (P = 0.414; Supplementary Fig. S6). Statistically significant variation was observed between methylation and tumor relapse (P = 0.035), ER/PR status (P = 0.001), and lymph node metastasis (P = 0.042; Fig. 4).
To identify individual gene loci whose methylation status correlated with clinical features, Fisher's exact tests with FDR correction were conducted for all 291 selected probes between groups defined by a single clinical feature. This identified 9 methylated probes (9 genes) significantly associated with tumor relapse, 19 methylated probes (18 genes) significantly associated with presence of both estrogen and progesterone receptors, and 2 probes specifically methylated in tumors with lymph node metastasis (Supplementary Table S5). Probes specifically methylated in samples that relapsed should show strong correlations with disease-free survival and thus Kaplan–Meier analysis was done. The 9 identified probes/genes (RECK, ACADL, SFRP2, ITR, UGT3A1, UAP1L1, SYDE1, UGT3A2, and PRKCB1) all showed significant association with relapse-free survival and 6 remained significant after Bonferroni correction (underlined genes; Fig. 5; Supplementary Table S6).
Selected genes associated with tumor relapse and the presence of both hormone receptors were further analyzed for breast cancer cell line methylation, expression in breast tumor cell lines after treatment with 5-azadc, demonstration of tumor-specific methylation in an additional 20 tumor/matched normal pairs and decreased expression in the original sporadic primary tumors (shown by HumanMethylation27 array to be methylated) by quantitative real-time RT-PCR, where appropriate.
For tumor relapse association, the 3 genes with the highest association with disease-free survival were selected; ACADL (P < 0.0009), RECK (P = 0.0009), and SFRP2 (P = 0.0009; Fig. 6). All the 3 genes first showed increased expression or were reexpressed (after 5-azadc treatment) in methylated breast tumor cell lines and then showed methylation in tumor/normal-paired samples at a frequency of 30% (6 of 20), 22.2% (4 of 18), and 57.1% (8 of 14) of samples for ACADL, RECK, and SFRP2, respectively [Fig. 6A and B (ii and iii)], via CoBRA. CoBRA showed no methylation in any of the corresponding normal breast tissues for RECK and some methylation in adjacent normal tissues of methylated tumors in ACADL (2 of 6) and SFRP2 (4 of 8), in all cases methylation observed in the associated normal tissue was less than that observed in the tumor tissue. Bisulfite sequencing of ACADL and RECK showed tumor-specific methylation throughout the selected area of the CpG island [Fig. 6A and B (iv); Supplementary Fig S7]. Quantitative real-time RT-PCR showed significant downregulation of ACADL (P = 0.043) and RECK (P > 0.0001) in methylated primary sporadic tumors compared with a panel of normal breast tissues [Fig. 6A and B (v)]. In general, expression in unmethylated tumors was found to be higher than that in methylated tumors for both ACADL and RECK (Supplementary Fig. S8).
For association between methylation and the presence of both hormone receptors (ER+/PR+), 3 genes were selected (TNFRSF10D, C1orf114, and COL1A2) for analysis that were also present within the list of genes with multiple hypermethylated probes (Supplementary Fig. S5). Again, all 3 genes first showed increased expression in methylated breast tumor cell lines after 5-azadc treatment and then showed specific methylation in tumors of tumor/normal-paired samples. COL1A2 showed tumor-specific methylation in 76.5% (13 of 17) of the paired samples (Fig. 7A). TNFRSF10D showed tumor-specific methylation in 50% (8 of 16), increased methylation in the tumor compared with normal tissue in 18.8% (3 of 16) and 12.5% (2 of 16) with similar levels in the tumor and normal tissue. C1orf114 showed tumor-specific methylation in 35.3% (6 of 17), and increased methylation in the tumor compared with normal tissue in 42.2% (7 of 17) of tumors. To confirm the above CoBRA methylation results we selected C1orf114 for bisulfite sequencing. Cloning and bisulfite sequencing of C1orf114 showed tumor-specific methylation throughout the selected area of the CpG island, hence confirming the CoBRA results (Fig. 7B). To discern further biological relevance of the methylation found in the above 3 genes, we selected TNFRSF10D gene for expression analysis in breast tumor samples. Quantitative real-time RT-PCR analysis showed significant downregulation of TNFRSF10D (P = 0.016) in methylated primary sporadic tumors compared with a panel of normal breast controls (Fig. 7C). TNFRSF10D showed higher expression levels in unmethylated tumors compared with methylated tumors in general, although not in every case, suggesting alternative mechanisms of downregulation (Supplementary Fig. S8).
The use of gene expression data from patient tumor samples to determine better treatment options is becoming increasingly common in clinical practice. The instability of mRNA makes expression profiling a challenging endeavor for utilization in routine clinical practice outside of major hospitals or commercial laboratories. Although DNA methylation patterns are more stable and can be detected by using a number of high-throughput and sensitive techniques which require little patient material. Using a high-throughput quantitative methylation approach, we established aberrant methylation patterns for a clinically well-characterized cohort of breast tumors. The generated methylation patterns showed strong differentiation between tumor and normal DNA, with many tumor DNA samples showing a far greater number of hypermethylated CpG loci compared with normal samples, as well as differing patterns in relation to specific clinical features. The hypermethylated CpG loci identified in our study include both previously reported and novel methylated genes that fall into a wide range of functional categories. Tumors with lymph node metastasis, ER+/PR+ tumors and patients that had relapsed showed overall higher methylation versus patients with none of the above. There were 3 distinct methylation profiles generated within the tumor samples, ranging from the high methylation to intermediate to low methylation clusters. Although patient survival was not significantly different among the clusters (P > 0.05), statistically significant (P = <0.05) variation among the clusters was found for tumor relapse, ER/PR status and lymph node metastasis. Examination of individual loci/genes revealed that (i) methylation of 6 genes (RECK, SFRP2, ITR, UGT3A1, ACADL, UAP1L1) significantly associated with worst survival; (ii) methylation of 18 genes/loci (B3GAT1, TNFRSF10D, GRIA4, C1ORF114, DNAJC6, C6ORF145, CORIN, LAMA2, ITR, MCAM, C12ORF34, NPY, COL1A2, RSNL2, ZNF660, ADAMTSL1, DPP4, MAT1A) associated with ER+/PR+ receptor status; and (iii) methylation of 2 genes (LDHD and ZNF660) associated with lymph node-positive tumor status.
To show biological relevance for these methylated genes, we next examined the relationship between gene expression and gene methylation for 2 genes most associated with poor survival (RECK and ACADL). In both cases, expression analysis showed that methylated breast tumors had lower gene expression versus normal breast tissue samples. We confirmed tumor-specific and/or increased methylation in tumors compared with corresponding normal breast tissue for RECK and ACADL in an independent set of 20 breast normal/tumor-paired DNA. RECK is a membrane-anchored metalloprotease regulator, expression of RECK is frequently reduced in various tumor types and reduced expression correlates with worst prognosis. Forced expression of RECK in malignant cells and in mouse xenograft models leads to reduced angiogenesis, invasion and reduced metastasis (13, 14). RECK−/− mice die around embryonic day 10.5 with reduced tissue integrity and defects in various organs including blood vessels and central nervous system. RECK has been shown to be methylated in NSCLC, gastric cancer, oral squamous cell carcinoma, colon cancer, and melanoma. RECK methylation is associated with poor prognosis and with lymph node metastasis in oral squamous cell carcinoma and NSCLC cancer, respectively (15, 16). But to our knowledge this is first report of RECK methylation in breast cancer and that methylated RECK is significantly associated with relapse and worst survival.
Effect of nutrition and lifestyle choices have been recognized to influence breast cancer risk and survival but there is a lack of studies investigating these breast cancer risk factors and tumor DNA methylation. A very recent study showed that not only breast tumor size but also alcohol and folate intake were associated with DNA methylation profiles (17).
Hence, it is interesting that we found ACADL (acyl-CoA dehydrogenase, long chain) methylation is associated with poor survival in our cohort of breast cancer patients. The protein encoded by the ACADL gene belongs to a family of mitochondrial falvoenzymes that are involved in fatty acid and branched-chain amino acid metabolism. This protein is 1 of the 4 enzymes that catalyze the initial step of mitochondrial β-oxidation of straight-chain fatty acid. Yanagitani and colleagues, showed that transgenic mice expressing retinoic acid receptor-α–dominant-negative form in hepatocytes exhibited severe fatty acid deposition (steatohepatitis) and liver tumors (18). The expression of mitochondrial enzymes involved in fatty acid β-oxidation, including CAPT II, VLCAD, ACADL, and HCD was greatly reduced in the transgenic mice compared with the wild-type mice, thus hepatic steatosis was induced in transgenic mice due to impaired β-oxidation of fatty acids in mitochondria. In addition, Wnt signaling was activated in transgenic mice. Histologic and biochemical abnormalities were reversed and occurrence of liver tumors was inhibited by feeding the transgenic mice with a high retinoic acid diet. The authors suggest that retinoic acid plays an important role in preventing hepatocellular carcinoma in association with fatty acid metabolism and Wnt signaling (18).
We also showed that SFRP2 methylation is significantly associated with poor survival. Secreted frizzled proteins (SFRP) act as soluble modulators of Wnt signaling. There are 5 SFRP genes, and the gene family has been shown to be methylated in various cancers including breast cancer. Methylation of SFRP1 and SFRP5 in breast cancer is associated with unfavorable prognosis (5, 19). SFRP2 methylation has previously been reported in breast, bladder, colon, liver, lung, esophagus, stomach cancer and leukemia. Although detection of SFRP2 methylation in fecal DNA has been shown to be a marker for colorectal cancer and for detection of colorectal polyps (20–22), in breast cancer SFRP2 methylation was not associated with clinical patient outcome or other clinicopathologic factors, but loss of SFPR2 protein expression was weakly linked to poor patient overall survival (23). Overexpression of SFRP2 in mammary cells leads to inhibition in cell proliferation. A recent article linked methylation of sFRP2 and sFRP5 with adverse clinical outcome in patients with normal karyotype acute myeloid leukemia (24). In this report we have shown that methylation of SFRP2 in breast cancer is associated with poor patient survival.
The remaining 3 genes [UGT3A1 (UDP glycosyltransferase-3 family, polypeptide A1); UAP1L1 (UDP-N-acteylglucosamine pyrophosphorylase 1–like 1); ITR/GPR180 (G protein-coupled receptor 180)] associated with survival in our cohort of breast cancer patients are novel genes and to our knowledge epigenetic inactivation of these genes have not been described previously in cancer; further work will be required to understand their role in breast cancer development.
In a recent genome-wide DNA methylation study in Korean breast cancer patients, hyper/hypomethylation of 5 genes was linked to ER+/PR+ status (25). We have identified 18 hypermethylated genes associated with ER+/PR+ status in Caucasian breast cancer patients. Li and colleagues identified FAM124B and ST6GALNAC1 hypermethylation associated with ER+/PR+ status in Korean breast cancer patients (25). These 2 genes were not linked to ER/PR status in our cohort of breast tumor cases from Caucasian individuals. This could be because of the ethnical differences between the 2 sets of breast cancer patients; in addition, we used a high-stringency criterion to identify differentially methylated genes in tumor versus normal DNA. Hence, according to our selection criteria the above 2 genes were not differentially methylated in tumor versus normal DNA.
We identified TNFRSF10D (DCR2) gene among the 18 genes/loci significantly hypermethylated in ER+/PR+ breast tumors compared with receptor-negative breast tumors. TNFRSF10D (DCR2), the protein encoded by this gene is a member of the TNF-receptor superfamily. TNF-related apoptosis-inducing ligand TRAIL can induce apoptosis in cancer cells but has little effect on normal cells, hence it is a promising anticancer agent. However, cancer cells are often resistant to TRAIL-induced cell death. It was recently shown that depletion of adenine nucleotide translocase-2 in breast cancer cells sensitized these cells to TRAIL-induced apoptosis by activating c-jun NH kinase and modulating the expression of TRAIL receptors, including downregulation of DCR2 expression both in in vitro and in vivo studies (26). DCR2 methylation has been previously reported in several cancers including NSCLC, breast cancer, and prostate cancer (27). COL1A2 methylation was also associated with ER/PR receptor status in our breast cancer cohort, was cancer specific, and could be reversed in methylated breast tumor cell lines following treatment with 5-azadc. Using computational protocols, Loss and colleagues recently showed COL1A2 methylation in breast tumor cell lines (28). In medulloblastoma COL1A2 methylation was found to be subgroup specific, and in bladder cancer COLIA2 gene inactivation due to methylation was shown to contribute to cell proliferation and migration of bladder cancer cells (29, 30).
In summary, we have identified a set of genes hypermethylated in breast cancer that are linked to patient clinical outcome, confirmation of our results in a larger breast cancer cohort and subsequent biological analysis of the proteins encoded by these genes may lead to a better understanding of breast cancer pathogenesis and development of clinical applications.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
F. Latif's laboratory was funded by Breast Cancer Campaign and Cancer Research UK.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).
- Received November 5, 2010.
- Revision received January 28, 2011.
- Accepted February 18, 2011.
- ©2011 American Association for Cancer Research.