Beyond DNA: the Role of Epigenetics in the Premalignant Progression of Breast Cancer

DNA methylation, a covalent and reversible DNA modification, was recognized to play a functional role in gene expression in the 1980s [18, 21]. The formation of 5-methyl-cytosine (5-me-C or m5C), through the addition of a methyl (CH3) group to the 5th carbon of the pyrimidine ring of a cytosine residue, is found to primarily occur at CpG islands. This reversible mark can be transferred by several enzymes including the polycomb repressive complex and DNA methyltransferases (DNMT1, DNMT3A, DNMT3B). CpG islands, ~1000 base pair regions of DNA with a high G-C content, are generally associated with gene promoters and their methylation is tightly linked to gene transcription [22, 23]. Dense methylation of CpG islands functionally silences the associated promoter and represses transcription by blocking transcription factor binding and recruiting chromatin remodeling complexes such as histone deacetylases (HDACs) or polycomb repressive complexes (PRC). Recruitment of HDACs remodels the local chromatin encouraging closed heterochromatin formation. Similarly, the recruitment of the PRC results in histone methylation and heterochromatin formation. The global loss of m5C in a variety of malignant tumor types relative to either normal tissue or benign lesions, was first observed in 1983 and was the first epigenetic change to be described in human cancer [24, 25]. While hypomethylation is a general characteristic of many cancers, including breast [26], a number of individual genes become hypermethylated. These genes are generally tumor suppressors or play critical roles in regulation of a number of hallmarks of cancer including apoptosis and cell-cycle regulation. A number of studies have examined DNA methylation in breast cancer progression.

Verschuur-Maes et al. examined the methylation status of 50 candidate tumor suppressor genes in normal epithelia, columnar cell lesions (CCL), grade I DCIS, CCL paired with DCIS and CCL paired with IDC. When considering all 50 candidates, a significant increase in cumulative methylation was found when comparing normal tissue to either CCL, DCIS or IDC, however there was not a significant global change between DCIS and IDC. However, two tumor suppressors, CACNA1A and MGMT, demonstrated an increase in methylation between DCIS and IDC. Many of the tumor suppressors studied had significant methylation even in CCL underscoring methylation as an early event [27]. Park et al. identified 15 loci differentially methylated between matched IDC and normal tissue. These loci were then tested in an additional patient cohort comprised of normal, flat epithelial atypia (FEA), ADH, DCIS and IDC. This study revealed a stepwise increase in the total number of methylated candidate genes from normal to atypia to DCIS. Several demonstrated increased methylation between normal and ADH (APC, DLEC1, HOXA1, and RASSF1A), however only two were differentially methylated in the transition from DCIS to IDC, HOXA10 and SFRP1 [28].

van Hoesel et al. employed laser capture microdissection to normal epithelia, ductal hyperplasia, ADH, DCIS and IDC samples and identified increased methylation from normal tissue to ADH for both (RARβ, RASSF1A) and from ADH to DCIS for RASSF1A. Both candidates demonstrated a significant linear trend with progression, however did not show a significant change in methylation between DCIS and IDC [29]. Also using lasercapture microdissection, Lehhman et al. found methylation of RASSF1A, and cyclin D2 to increase between normal and DCIS but were similar between intraductal and invasive samples [30]. Fackler et al. also identified increased methylation in cyclin D2, RASSF1A, RARβ, but also in TWIST and HIN-1 when comparing reduction mammoplasties to DCIS [31]. Hoque et al. evaluated methylation status in nine putative tumor supressor genes in synchronous preinvasive (either ADH or DCIS) and IDC lesions as well as pure IDC and normal tissue samples. This study identified APC, CDH1 and CTNNB1 with elevated methylation in IDC relative to normal with only CDH1 showing increased methylation in IDC compared to preinvasive lesions [32]. From these studies and more [33‐36] a model is emerging wherein epigenetic DNA methylation changes are an early event in breast cancer occuring prior to development of invasive growth.

Although a few candidate genes have been identified by multiple studies (RASSF1A, RARβ, CCND2), identification of reproducible, clinically tractable DNA methylation biomarkers from previous studies is limited by differences in methodology (sample collection, statistical cutoffs), small sample size lacking statistical power, and interrogation of limited target gene panels. Improved sequencing technologies have now allowed the interrogation of the methylome. In one of the first studies to look at the global methylation profile, Tommasi et al. interrogated six frozen DCIS and adjacent normal samples and found 108 loci differentially methylated with a significant enrichment of homeobox genes. Further, 53% of identified loci were also Polycomb repressive complex targets [37]. Combining the power of array technology with a longitudinal study, Johnson et al. [38] used the Illumina HumanMethylation450 microarray to evaluate 40 estrogen receptor positive DCIS cases and adjacent normal where available (n = 15). 33% of initial DCIS cases went on to develop invasive disease which were then assessed versus age matched non invasive disease. This longitudinal study revealed 641 progression-associated differentially methylated CpGs. Of these, 276 demonstrated increases in methylation from normal to DCIS to progressed IDC. From the 641 loci, relating to 397 genes, 72 genes demonstrated differential methylation in an independent DCIS-IDC cohort. From this group there was a strong enrichment of homeobox genes, as well as polycomb group gene targets. Interestingly, they also identified HOTAIR, a long noncoding RNA that stimulates invasion and metastasis in breast cancer, as being differentially methylated and demonstrating a positive expression correlation with methylation. HOTAIR (HOX transcript antisense RNA) epigenetically silences genes through redirecting the polycomb repressive complex 2 (PRC2) to many loci including the HOXD cluster thus suggesting one potential mechanism by which both homeobox genes and PRC2 targets may be effected. Two independent studies suggest homeobox genes and polycomb target genes as critically altered during early breast cancer progression. This observation is further strengthened by a recent study where Cai et al. used the MMTV-PyMT mouse model to study DNA methylation during progression. For this study, samples were collected at specific time points corresponding to tumor progression (hyperplasia at 6 weeks, adenoma / mammary intraepithelial neoplasia at 8 weeks, early carcinoma at 10 weeks and late carcinoma with metastasis by 12 weeks). This study revealed that of 374 genes demonstrating increased methylated promoters unique to late stage samples, there was a strong enrichment for PRC2 targets. Furthermore, the authors found significantly reduced expression of PRC2 target genes at all stages of progression, suggesting PRC2 alterations as critical to early progression [39].

Although three independent studies and models have identified polycomb target genes as a group to be altered in progression, this observation requires further validation as it was not observed in all methylome studies. Fleischer et al. interrogated 285 archived tissue samples, including normal, DCIS, DCIS-IDC mixed and IDC using the Illumina Infinium HumanMethylation450 microarray. The authors also correlated methylation with gene expression. While nearly 17,000 CpGs (1011 genes) were differentially methylated between normal and DCIS, only 2000 (154 genes) were altered between DCIS and IDC. These results were validated through over 500 TCGA breast cancer samples as well as an additional set of DCIS / adjacent normal samples. Interestingly, 4 genes demonstrated increased methylation from normal to DCIS and DCIS to IDC (CPA1, CUL7, LRRTM2, and POU2AF1). The authors were also able to study methylation in relation to patient survival This study resulted in development of a prognostic signature of 18 CpG loci,correlating to 26 genes, that can predict survival of patients with IDC, DCIS and mixed DCIS-invasive lesions [34]. Interestingly, these genes were not significantly enriched for canonical signaling pathways and this signature does not include the 4 genes that increased with progression [34].

Several studies suggest that investigating the surrounding normal adjacent tissue may be critical to understanding how DCIS ultimately gain the capacity to invade. Teschendorff et al. [40] recently compared the DNA methylome of 569 breast tissue samples including patient matched DCIS – adjacent normal, as well as 50 samples from cancer free women. This study revealed dramatic changes between normal, cancer free tissue, and normal adjacent tissue demonstrating wide spread DNA methylation field effects in agreement with previous studies [41]. Interestingly, the methylation patterns in the adjacent normal were found strongly enriched at genetic regulator elements, EZH2 and SUZ12, members of the PRC2 complex, as well as CTCF and RAD21, proteins critical to chromatin looping [40]. Together these data suggest that chromatin remodeling in adjacent normal tissue is critical to progression.

DNA methylation clearly plays a role in early breast cancer progression, with the majority of changes occuring between normal and even the earliest atypical lesions. Now advances in microarray technology, and an increased emphasis on understanding progression, changes between DCIS and IDC are beginning to emerge. Globally, homeobox genes and polycomb complex genes may be critical targets to understanding progression, however this needs to be further validated in additional patient cohorts. With a focus on understanding DNA methylation in progression, and the possibilty of longitudinal patient studies, the prospect of building a diagnostic signature to predict invasive potential that incorporates DNA methylation seems within reach.

Histone modification is critical to gene regulation both through working hand in hand with DNA methylation as described above, and independently. The chromatin compaction model illustrates how DNA must be organized in nucleoprotein complexes to fit within the nucleus. Negatively charged DNA associates with an eight histone protein complex called a nucleosome. Each nucleosome is composed of two sets of histones H2A, H2B, H3, H4 and a linked histone protein H1. DNA connects nucleosomes, like beads on a string, which must then become further compacted to form tight chromatin fibers. Tightly wound DNA of heterochromatin is unavailable to transcriptional machinery thus compaction regulates gene expression. Post-translational modification of histones that alter the association of DNA with nucleosomes can alter gene expression without changing the underlying DNA [42].

Histone proteins have tails extending out of the nucleosomes that are available for post-translational modification. Histone tails are the subjects of several modifications including mono- di- or tri-methylation, acetylation, phosphorylation and ubiquitination. Each modification has a unique effect on its surroundings. Modifications can create docking sites for recruiting proteins to the nucleosome, or remodel the chromatin to either relax or tighten the DNA changing the chromatins accessibility. Multiple methylation marks at different locations are associated with different outcomes. For example, mono-methylation of H3K9, H3K27 or H3K79 is generally associated with gene activation whereas di-methylation is generally repressive on H3K9 and H3K27, but activating on H3K79. Further, several modifications can occur on the same tail and on several histones within the same nucleosome, each having an independent effect on how the local DNA associates with the histones. This complex system is governed by a series of “writers” to deposit the modifications, “readers” to interpret and “erasers” as histone modifications are reversible allowing for genome plasticity. Similar to the genetic code, we are now learning how the histone code functions in the cell to regulate gene expression and signaling networks [43, 44].

A number of studies of histone modification are focused on specific histone readers, writers or erasers to determine their impact on breast cancer. The Polycomb Repressive Complex 2 (PRC2) is a histone modification writer that plays an important role in development, stem-cell plasticity, differentiation and maintaining cell identity. PRC2 is recruited to chromatin where the catalytic subunit, enhancer of zeste homolog 2 (EZH2), a histone lysine N-methyltransferse, tri-methylates histone H3 at lysine 27, a repressive mark. EZH2 can then also serve as a scaffold for the additional recruitment of DNA methyltransferases functionally reinforcing a closed chromatin state and transcriptional repression [45, 46]. Inappropriate chromatin remodeling can result in dysregulated signaling that can cause a cell, in effect, to “forget” appropriate cell identity.

Interrogating EZH2 as the functional PRC2 subunit, tissue microarray [47] and immunohistochemistry analysis [48] of normal, atypia, DCIS, invasive and metastatic samples found EZH2 expression to increase with each stage of progression. Consistent with observations with DNA methylation, EZH2 transcripts and protein were found elevated in DCIS lesions while largely absent from normal tissue, again suggesting this is an early event. Kleer et al. interrogated the relationship between EZH2 expression and clinical outcomes and found elevated EZH2 to be associated with a more aggressive breast cancer and shorter metastasis, disease free, and overall survival [47]. Furthermore, exogenous expression of EZH2 in normal immortalized H16N2 cells increased anchorage independent growth and invasion [47]. Together this suggests that EZH2 expression and chromatin remodeling drives breast cancer progression.

Lysine-specific demethylase 1 (LSD1) is a histone methylation eraser that specifically works on H3K4 and H3K9 [49‐51]. LSD1 is found highly expressed in several tumor types including prostate [52], lung [53], bladder [54] and breast [50]. Serce et al. extended this analysis to examine early breast cancer progression. When comparing low, intermediate and high grade DCIS to invasive breast cancer, LSD1 expression increases with grade, again an indication of an early epigenetic event in breast cancer progression [51]. LSD1 expression was found highest in those tumors that were estrogen receptor negative (ER-) and progesterone receptor negative (PR-) [50].

Another family of histone modifying proteins are the acetylation readers, writers and erasers. Acetylation by histone acetyltransferases (HATs) occurs on the ε-amino groups of lysine residues, effectively neutralizing the positive charge on lysine. Strong acetylation induces an open chromatin confirmation allowing transcription factor binding. Acetylated histones can also serve as a mark for recruiting bromodomain proteins (proteins that bind acetylated lysine residues, “readers”) which in turn recruit the mediator complex required for transcription activation [55‐57]. Histone deacetylases (HDACs) oppose this process, remove the acetylation, and generally support a closed chromatin confirmation. To date, there are 18 HDAC enzymes divided into 4 classes based on their mechanism of action (zinc or NAD+ dependent) and location [58]. Following the discovery of HDACs in 1996 [59], the HDAC inhibitors that followed were found to attenuate cell proliferation, induce differentiation of both normal and breast cancer cell lines, and regulate expression of cell cycle genes [60‐62].

Given the dramatic cell biological phenotypes of the HDAC inhibitors, many groups began investigating the functional role and expression of HDACs. Suzuki et al. investigated HDAC1, HDAC2, HDAC6, acetylated H4, and ac-H4K12 in pure normal (from reduction mammoplasties), adjacent normal, DCIS and IDC patient samples. Acetylation was largely consistent between pure normal and adjacent normal, however there was global hypoacetylation of histone H4 as well as a decreased expression of HDACs 1, 2, and 6 with progression. The most dramatic loss of ac-H4 occurred between normal and DCIS, again suggesting an early event in disease progression [63]. Recently, Lapierre et al. examined HDAC9 expression in the MCF10a breast cancer progression series which models the transitions between normal, ADH, DCIS and IDC [64‐68]. HDAC9 expression was found to increase 10–12 fold between normal and DCIS model cell lines with no change between DCIS and IDC, again indicating an early event in progression [69]. While these studies suggest that HDAC expression and activity may be altered in early lesions, how HDACs may support progression remains unclear. Future studies that focus on HDACs in early lesions should also identify changes in where modifications are occurring as opposed to global changes, to inform what genes are impacted by chromatin remodeling.

To date there are very few studies focused on histone modifications in early breast cancer lesions. The majority of histone modification studies have been performed on invasive disease where many groups have found specific modifications, and associated proteins, differ based on breast cancer subtype and can correlate with tumor phenotypes and clinical outcomes. Elsheikh et al. used 880 well-characterized breast carcinomas to perform tissue microarrays for a number of epigenetic marks. Acetylation of H4K16ac was found absent from the majority of samples suggesting that loss may be an early event in breast cancer [70]. The authors were able to couple modification status with clinical outcomes and found that low detection of several modifications (H3K18ac, H3K9ac, H4K12ac, H4K16ac, H4R3me2, H3K4me2, H4K20me3) was associated with adverse overall and disease free survival [70]. Expression HDAC1 is positively correlated with positive ER/PR status, and luminal A subtype [71‐74]. Seo et al. found no correlation between HDAC1 and overall survival using univariate analysis, however high HDAC1 expression was positively correlated with overall survival in ER+ patients [73]. Concordantly, while Zhang et al. also found patients with high HDAC1 expression had better patient outcomes, expression was not an independent prognostic indicator for overall or disease free survival [74]. HDAC2 expression is significantly correlated with HER2 overexpression and a negative hormone receptor status [72], and HDAC6 expression is associated with luminal B subtype [73]. Although few studies are focused on early progression, these data demonstrate that histone modification and chromatin remodeling play a clear role in breast cancer, likely as an early event in disease progression.

Advances in high throughput array and sequencing technology (ChIP-seq, RNA-seq, DNAse1-seq, ChIA-Pet, Hi-C chromatin capture…) have made interrogating the transcriptome and genomic architecture more accessible, resulting in an explosion of genome wide studies. The advanced genomic approaches have both increased our understanding of the complex systems regulating transcription, and added new players to the transcriptional models. Although enhancers were originally defined in 1987 [75], our understanding of how they function and regulate gene expression is continuing to evolve.

Classic enhancers are stretches of DNA to which transcription factors bind to induce expression of target genes. Enhancers can promote local gene transcription or undergo chromatin looping to promote distal gene transcription, potentially up to several thousand base pairs away. These elements are identified by regions of DNase1 hypersensitivity, demonstrating open chromatin confirmation, associated protein binding (CTCF, Med1, CBP300), and histone modification (H3K4me1, H3K27ac). Multiple enhancers can be clustered together creating a region of extremely high occupancy of transcriptional machinery such as the Mediator complex, and an abundance of transcription factor binding sites. These clusters of enhancers (or super-enhancers) play essential roles in driving expression programs that define cell identity and are frequently tissue type specific. Importantly, their normal function can become co-opted during tumorigenesis [76, 77].

Several groups have recently reported the acquisition of super-enhancers at oncogenes during tumorigenesis [76‐80]. Hnisz et al. sought to understand if acquired super-enhancers amounted to background noise or if they functionally contributed to oncogenic signaling networks. To ask this question the authors examined normal mammary epithelia compared to the ER+ MCF7 cell line through ChIP-seq for H3K27ac and found an active super-enhancer at the estrogen receptor (ESR1) locus only in the MCF7 cells [76]. Similarly, Rhie et al. identified global similarities and differences between poised and active enhancers in normal immortalized human mammary epithelial cells (HMEC) and triple negative breast cancer MDA-MD-231 cells [81]. The authors found the cell type specific enhancers in the HMECs to be associated with epidermis development, cell adhesion, wound healing, chemotaxis and proteolysis, whereas the enhancers in the TNBC cells were related to driving chromatin modification, DNA replication, cell division, cell cycle, and mitosis. The authors further examined their identified gene signatures in TCGA and found 60 of the MDA-MB-231 specific genes were in the top 10% of genes overexpressed in tumor versus normal, while 53 of the HMEC specific genes were in the top 10% of the underexpressed genes [81]. Together these data suggest a penetrant and dramatic shift in enhancer rewiring as a critical step in breast cancer progression.

The rewiring of super-enhancers, has recently become a new target of epigenetic therapeutic approach. Active enhancers, marked by strong chromatin acetylation, recruit bromodomain and extraterminal (BET) family proteins to facilitate gene transcription. The BET inhibitor JQ1, targeted against the bromodomain protein BRD4, has been demonstrated to preferentially bind super-enhancers [77]. As many oncogenes are found driven by super-enhancers it has been proposed that BET inhibitors could be an epigenetic drug to selectively inhibit inappropriate oncogene expression. Demonstrating how important enhancer activity is to supporting tumor cells, JQ1 and other BET inhibitors have dramatic impacts on tumor cell viability and cell cycle progression in multiple tumor types [82‐84]. Shu et al. identified that BET inhibition also holds promise for treatment of triple negative breast cancers for which current treatment options are limited [85].

Supporting the emerging concept of enhancer hijacking in early stage breast cancer, Shen et al. recently reported the tumor suppressor activity of RACK7, a histone reader and component of a demethylase complex with KDM5C. RACK7 ChIP-seq found a strong overlap of RACK7 binding to regions of high H3K4me1 and H3K27Ac occupancy. Analysis with the ROSE (ranking ordering of super enhancers [77, 86]) algorithm, a published algorithm in which enhancers can be ranked by occupancy of common identifiers (Mediator, H3K27ac, H3K4me1, DNase hypersensitivity) found that RACK7 occupied nearly all active super-enhancers in MCF7 and mouse embryonic stem cells. Functionally, RACK7 was found to recruit KDM5C to super-enhancers. Loss of RACK7, or KDM5C, results in loss of H3K4me1, gain of H3K4me3, and increased transcription of associated enhancer targeted genes, including the S100A oncogenes [87]. RACK7 knockout drives tumorigenic phenotypes including increased soft agar growth, invasion, migration, and tumor volume in mammary fat pad xenograft models, suggesting RACK7 functions as a tumor suppressor. Finally, RACK7 expression was examined in 6 patient matched DCIS and IDC samples and was significantly decreased in IDC relative to DCIS [87]. While these observations need to be further validated on a larger patient cohort, together these data suggest that epigenetic chromatin remodeling and enhancer activation as a critical step in the transition to invasive disease.