Here we report the DNA methylation profiles of a breast cancer progression series, including normal breast tissue, DCIS, IBC and mixed lesions. Interestingly, most of the aberrations in the epigenetic profile were observed already in the pre-invasive DCIS stage. The affected pathways suggested that many of the changes may not occur in the tumor, but in infiltrating cells or at least in genes that enable cross-talk to such cells. Also of interest was that DNA methylation profiles of the basal type of breast cancer were more similar to normal tissue than were the luminal-like tumors. These data suggest that the methylation profiles may be a function of the cell of origin as much as a marker of progression. We also report a signature comprising DNA methylation levels of 18 CpGs that was prognostic for breast cancer patients with invasive tumors as well as for patients with DCIS and mixed lesions of DCIS and IBC. The signature was discovered in a training data set of 176 patients and validated in 583 patients from the TCGA. In the validation patient group the prognostic signature and lymph node status were complementary, potentially providing valuable information for clinical decision-making. The patients that were classified with good prognosis by DNA methylation and additionally were lymph node-negative might benefit from reduced or no adjuvant treatment, while patients that were classified with adverse prognosis by DNA methylation and were lymph node-positive could potentially benefit from more aggressive treatment.
A great advantage of DNA methylation is that it is relatively easy to design an assay that may be used in a clinical setting. DNA methylation can be measured on an absolute scale (from 0 to 100%), is stable in the cell over time, and is relatively insensitive to handling in the laboratory. This work clearly shows the potential of DNA methylation-based signatures for clinical utilization.
With data from two independent cohorts of normal tissue and DCIS, we report that the DNA methylation profiles of DCIS were radically changed compared with normal breast tissue, involving more than 5,000 genes. One cohort consisted of fresh frozen tissue and normal controls from healthy women, while the other cohort consisted of formalin-fixed paraffin-embedded (FFPE) DCIS samples and matched adjacent normal breast tissue. Thus, the reported changes in methylation levels across these genes appear to be independent of tissue preparation and the normal tissue’s proximity to tumor tissue. Comparably, the changes between DCIS and IBC were more modest, involving around 1,000 genes. These findings suggest that the epigenome is severely altered in the early neoplastic setting in the breast. Previous studies of breast cancer progression have also reported early aberrant DNA methylation in DCIS, but they characterized fewer genes (summarized in ). The current study has the advantage of a high coverage methylation assay (Illumina HumanMethylation450) and leverages true normal controls from healthy women. Our observation that extensive epigenetic alterations occur early in cancer progression has been reported for other cancer types, including colorectal cancer. For example, studies using the HumanMethylation450 assay reported that precancerous adenomas demonstrate heterogeneity similar to invasive tumors, and that aberrant DNA methylation occurs early in colorectal cancer formation .
Classification of breast cancer by hierarchical clustering showed that basal-like tumors clustered with the normal samples in one cluster, and luminal A and luminal B tumor clustered together in the second cluster (Figure 1). This observation largely recapitulates and extends the results from a previous study . Since DNA methylation aberrations occur early in carcinogenesis, it is possible that DNA methylation changes may play a role in the development of molecular subtypes of breast cancer, although it is also possible that the correlation with methylation is a consequence of subtype. Future studies are needed to define the mechanistic effects that DNA methylation and other epigenetic marks may have on early development of cancer.
DCIS lesions tend to grow slower and show less inter-tumor heterogeneity than IBC lesions. Consequently, it would be pertinent to perform subtype-specific analyses of differences between DCIS and IBC. In the present study, however, the number of DCIS samples was too few to perform subtype-specific analyses. Future studies should aim to collect enough DCIS samples to divide both DCIS and IBC samples into intrinsic subtypes of breast cancer while including enough samples for statistical analyses. The inter-sample heterogeneity in the normal samples (mammoplastic reductions and needle biopsies from healthy women) was low compared with the neoplastic lesions (Figure 1).
Correlation between DNA methylation and gene expression was found throughout the genome and involved almost 3,000 genes. CpGs whose methylation level correlated with expression were enriched close to TSSs, but also found at distances up to 100 kb from them. Interestingly, about a quarter of the genes whose expression level correlated with methylation level showed a positive correlation, meaning that a higher methylation level was associated with higher expression. Viewed in relation to functional regions in genes, 70% of the positive correlations between methylation level and expression were found in the 3’ UTR or the gene body. Similar findings have been reported in chronic lymphocytic leukemia  and support that promoter hypermethylation is an important mechanism for gene silencing, while DNA methylation elsewhere may have more complex functions that are yet to be fully understood. Possible mechanisms for regulation of gene expression by non-promoter methylation include interplay between nucleosome positioning and chromatin structure, regulation of enhancer region availability, and/or gene body regulation of alternative promoters [25,26]. Statistical significance of correlation between DNA methylation and gene expression was corrected for multiple testing by Bonferroni correction. This method is very strict, and may underestimate the association between DNA methylation and gene expression.
The survival signature segregated patients with DCIS and mixed DCIS-IBC into two groups with significantly different prognosis. The signature classified most of the patients with mixed DCIS-IBC that experienced breast cancer-specific death into the bad prognosis group. Additionally, the single patient with pure DCIS that experienced breast cancer-specific death was also classified into the bad prognosis group. Since only one of the patients with pure DCIS died of breast cancer, it was not possible to perform the analysis on only patients with pure DCIS. Taken together, the signature may have great potential to classify patients with DCIS or mixed lesions according to prognosis, but more patients must be studied to further validate the clinical value.
Several of the genes in the survival signature have roles in tumor suppressive functions. The protein product of IRF6 has been shown to function synergistically with the tumor suppressor maspin to regulate mammary epithelial differentiation , and has also been shown to have tumor suppressor activity in squamous cell carcinoma . TBX5 is a transcription factor that has been implicated as a tumor suppressor in colon cancer and has been found silenced by DNA methylation . A SNP (rs1265507) located between TBX5 and TBX3 was also associated with mammographic density in a genome-wide association study . In the present study, high methylation levels of CpGs in TBX5 were associated with lower expression levels of TBX5 and adverse prognosis. DIEXF is thought to be involved in the turnover of p53 , and CEND1 has been shown to affect cyclin D1 levels . ZNF259 has been shown to be involved in regulation of the cell cycle through interactions with the epidermal growth factor receptor , and KCTD21 is thought to act as a tumor suppressor in medulloblastoma by modulating Hedgehog signaling through degradation of histone deacetylase 1 .
Some genes in the survival signature have also been associated with functions related to motility and invasion: EPN3 over-expression has been shown to promote cancer cell invasion , MACF1 has been shown to be involved in cell mobility and steering by interactions with HER2 , and CSNK1G2 is thought to modulate the activity of metastasis-associated MTA1 while itself a target of ER . Taken together, many of the genes associated with the survival signature have tumor suppressive functions or are involved in regulation of motility and ability to invade.
A strong immune component in breast tumors observed by measuring DNA methylation has previously been reported . The genes that were differentially methylated between DCIS and IBC were borderline significantly enriched in the agranulocyte and granulocyte adhesion and diapedesis pathways, suggesting that many of the observed changes may occur in infiltrating cells or in genes that enable cross-talk to such cells.
CUL7 (cullin 7) methylation levels increased from both normal to DCIS and DCIS to IBC. CUL7 encodes a ubiquitin ligase that forms complexes with p53 and Parc. It was shown to regulate apoptosis independently of p53 . In another report , CUL7 was shown to function as an antiapoptotic oncogene through cooperation with Myc in a p53-dependent manner. Also, CUL7 has been shown to be involved in liver carcinogenesis . Importantly, CUL7 has not previously been reported in breast cancer. ICAM2 (Intercellular adhesion molecule 2) methylation levels decreased between normal and DCIS, and increased between DCIS and IBC. ICAM2 is involved in cell adhesion and thought to play a role in immune response. In pancreatic cancer, ICAM2 has been reported to have tumor suppressor function through immune surveillance .