Epigenomics of cancer – emerging new concepts

Abstract

The complexity of the mammalian genome is regulated by heritable epigenetic mechanisms, which provide the basis for differentiation, development and cellular homeostasis. These mechanisms act on the level of chromatin, by modifying DNA, histone proteins and nucleosome density/composition. During the last decade it became clear that cancer is defined by a variety of epigenetic changes, which occur in early stages of disease and parallel genetic mutations. With the advent of new technologies we are just starting to unravel the cancer epigenome and latest mechanistic findings provide the first clue as to how altered epigenetic patterns might occur in different cancers. Here we review latest findings on chromatin related mechanisms and hypothesize how their impairment might contribute to the altered epigenome of cancer cells.

Highlights

► Genome-wide analyses reveal epigenomic differences in functional regions. ► Epigenetic patterns occur in large domains in the genome. ► Epigenetic mechanisms are interrelated. ► Mutations of epigenetic enzymes are frequently associated with cancer. ► Changes in nuclear architecture are related to epigenomic alterations in cancer.

1. Epigenetic mechanisms and their basal functions

In order to gain access to the fundamental information of the DNA sequence to produce cell type specific gene expression signatures, a highly regulated organization of DNA into chromatin is essential. Long-range silencing of repetitive sequences and formation of silent heterochromatin but also DNA access for transcription in euchromatin or DNA replication are dependent on epigenetic mechanisms including DNA methylation, histone modification and remodeling, and non-coding RNA. These mechanisms are interrelated and need to be stably maintained during cell divisions to conserve cellular identity but also react to cell intrinsic signals during development or to external factors to adopt to altered environmental cues.

1.1. DNA methylation

The modification of the C5 position of the cytosine base (5 mC) is found in approximately 70–80% of CpG dinucleotides in somatic mammalian cells and to some extend in non-CpG sequences in embryonic stem cells (ESC) [1,2]. DNA methylation is the basis for different epigenetic phenomena such as imprinting, X chromosome inactivation or the formation of heterochromatin [3]. Generally, DNA methylation of promoter regions inversely correlates with gene expression (Fig. 1). Exceptions are CpG islands, which are found in about 60% of promoters [4]. They have a high CpG density and are usually kept free of methylation independent of their activity state [5,6]. However, in cancer cells promoter CpG islands tend to become hypermethylated, which then causes silencing of the underlying gene [7]. Intragenic methylation is found at repetitive sequences such as satellite repeats and remnants of retroviral insertions such as LINEs and SINEs in human DNA [1,8]. Gene body methylation is present in highly expressed genes and it has been speculated that this may repress transcriptional noise from alternative start sites, inhibit antisense transcription or direct RNA splicing and relates to replication timing [9–12]. A direct role for alternative splicing has recently been attested to methylation-sensitive CTCF binding and polymerase II pausing at CTCF bound exon boundaries [13]. Comparison of the methylome of different hematopoietic lineages has identified a large number of hypomethylated regions (HMRs), which can be constitutive or lineage-specific and colocalize with transcription factor binding sites [14]. Furthermore, complex patterns of methylation were detected in progenitor cells in HMRs, which resolved in a lineage-specific fashion to methylated or demethylated regions, reminiscent of a bivalent histone methylation state found in progenitor cells [14]. Similarly, the deciphering of the mouse methylome revealed regions of low methylation at distal regulatory elements, which was linked to binding of tissue specific transcription factors [15] (Fig. 1). Thus, recent reports have identified genomic regions, which are associated with distinct DNA methylation patterns regulating gene expression profiles and chromatin compartmentalization.

Fig. 1.

Complex epigenetic patterns demarcate distinct genomic regions. Genome-wide approaches have provided information on the different epigenetic characteristics of defined genomic loci. (Top) DNA methylation levels and chromatin states are indicated for an intragenic region, subdivided in expressed and silent status (red color indicates expressed, blue color silent genes). DNA methylation within promoter regions of genes negatively correlates with expression levels of the gene except in CpG islands, which are usually kept free of methylation (CpG island, CGI within green rectangle; red line, methylation level of expressed gene; blue line, methylation level of repressed gene; TSS, transcription start site). In non-CGI promoters (crossed out green rectangle) transcriptional activity inversely correlates with DNA methylation of the promoter. Upstream enhancers (DRE, distant regulatory element) are demethylated in active genes, whereas they are methylated in repressed genes. Gene body methylation is elevated in expressed genes and shows characteristic spikes at exon-intron boundaries (indicated by black dashed lines). Furthermore, expressed and silent genes are marked by characteristic chromatin marks, nucleosome density and histone variants, which together reflect the on/off state of a gene (see middle panel of figure). Note that only selected histone modification marks are listed. NDR indicates nucleosome depleted regions in enhancers, promoters and 3′ regions of active genes. BLOCs and LOCKs designate large silent gene-rich regions marked by repressive histone marks H3K27me3 and H3K9me2, respectively. (Bottom) DNA methylation levels and chromatin composition including histone modifications, grade of compaction and histone variants are indicated for intergenic regions and heterochromatin. Inverted arrows on top indicate inverted repeats. These regions show generally high levels of DNA methylation and are associated with repressive histone marks.

Although DNA methylation is a very stable epigenetic mark, which is passed on to subsequent cell generations, reprogramming of DNA methylation occurs during gametogenesis and after fertilization or after artificial reprogramming of somatic cells into induced pluripotent stem cells (iPSC) [16]. Furthermore, differentiation specific demethylation was observed in hematopoietic progenitors [17] and in murine erythroid progenitors, which was associated with high DNA replication rates, arguing for a mechanism of passive demethylation [18]. Modification of 5 mC via oxidation by the TET family of proteins to 5-hydroxymethyl-cytosine (5hmC) or via deamination to thymine by AID/APOBEC1 followed by base excision repair (BER) was proposed as a possible mechanism for DNA demethylation [17]. High levels of 5hmC have been detected in brain tissue and ESCs, which might be a prerequisite for active demethylation by the action of DNA glycosylases or passive demethylation during replication [19].

1.2. Histone modification and remodeling

The coiling of DNA around nucleosome particles is the basis for the organization of eukaryotic genomes. A multitude of different posttranslational modifications of the core histone proteins (H2A, H2B, H3 and H4) allows for demarcation of specific chromatin regions and states, as illustrated by recent genome-wide chromatin modification mapping studies (reviewed in [20]) (Fig. 1). Histone modifications can be dynamically added or removed and associate with both active and repressed regions of chromatin. To date more than a dozen different histone modifications have been detected, which can modify more than 150 conserved residues within histone proteins [21]. This number of different modifications has a high combinatorial potential, which would yield a hugely complex histone code [22] and it is under debate, whether such a code exists or whether histone modifications are a consequence and mere reflection of dynamic processes altering DNA accessibility such as transcription factor or RNA polymerase II (RNAPII) binding or chromatin remodeling [23]. Generally, certain histone modifications such as acetylation or phosphorylation are thought to change chromatin structure by altering the net positive charge of the histone proteins, thereby rendering the underlying DNA sequence information more accessible [24]. Alternatively, histone modifications can be recognized by specific protein domains (e.g. bromodomains, Tudor domains, chromodomains), which in turn might enforce or stabilize the chromatin signature and provide a platform for the recruitment of additional factors [25,26]. Intriguingly, chromatin regulators encompassing histone modifiers and histone modification binding proteins are present in a combinatorial fashion at distinct genomic loci and frequently bring together regulators associated with opposing activities [27]. This might occur counterintuitive, but could provide a dynamic system for fine-tuning gene expression programs or rapid response to altered signals and highlights the importance of a balanced level of chromatin regulators for normal cell function.

Distinct histone modifications correlate with distinct genomic regions (Fig. 1); for example H3K4 trimethylation (H3K4me3) with promoters; H3K4 monomethylation (H3K3me1) with enhancers; H3K9 acetylation and H3K27 acetylation (H3K9ac, H3K27ac) with active regulatory regions; H3K36 trimethylation (H3K36me3), H3K79 dimethylation (H3K79me2) and H4K20 monomethylation (H4K20me1) with transcribed regions and intron/exon usage; H3K27 trimethylation (H3K27me3) with Polycomb repressed regions; or H3K9 trimethylation H3K9me3 with pericentromeric heterochromatin [20,28]. In Drosophila developmental enhancers are marked by heterogeneous histone modifications including H3K4me1, H3K27ac, H3K79me3 and H3K27me3 and the co-occurrence of H3K27ac, H3K79me3 and RNAPII was correlated with the spatio–temporal timing of enhancer activity [29]. The H3K4me1 appeared to represent a general mark present on enhancers irrespective of their activity state and cell type specificity, which was in contrast to earlier reports [30,31], but is in line with a recent report demonstrating that Polycomb repressed genes tend to keep permissive enhancers in differentiated cells, which are marked by the histone variant H2AZ and H3K4me1 [32]. Presumably, developmental plasticity is established through so-called bivalent histone modifications, which combine the active H3K4me3 with the inactive H3K27me3 mark in ESC on silent developmental and differentiation specific genes. During differentiation, these regions can be resolved into either active H3K4me3 or inactive H3K27me3 marks and enable rapid activation or silencing of the underlying genes [33,34].

Aside from gene regulatory functions, which occur in a relatively local and confined chromatin region, histone modifications can also span large regions, as exemplified by X chromosome inactivation in female mammals [35]. Large chromatin blocks of H3K27me3 associated with gene silencing have been identified on mammalian autosomes [36], and H3K9me2 modified regions in the megabase size have been found in differentiated mammalian cells correlating with silenced genes [37] (Fig. 1).

Further, exchange of canonical histones by variants is connected to transcriptional activity as well as chromatin structure. Examples include macroH2A on the inactive X chromosome [38], CENP-A at centromeres [39] or γH2AX at DNA double strand breaks [40]. The histone variants H2AZ and H3.3 are enriched at active promoters and enhancers [41,42] (Fig. 1). Additionally, H3.3 can be incorporated into telomeres and pericentromeric chromatin [43–45]. Deposition of histone variants can be replication coupled or independent and an interesting function has been attested to H3.3 recently. H3.3 deposition via the chaperone HIRA was linked locally to RNAPII occupancy at sites of active transcription and a more global gap-filling mechanism to protect genome integrity during transcription or replication was proposed [46].

DNA accessibility can be affected by the structure of nucleosomes and their interaction with DNA by exchanging canonical histones with histone variants or by histone modifications, respectively. Additionally, the position and the density of nucleosomes on the DNA string can determine the level of accessibility. Active genes have characteristic nucleosome depleted regions (NDRs) flanked by positioned nucleosomes upstream of their transcription start sites, which contain binding sequences for transcription factors [47]. Repressed genes usually lack a NDR, but DNA sequence, binding of transcription factors and the action of chromatin remodeling complexes has been suggested to act in a multistep process to determine local nucleosome composition and density [48] (Fig. 2).

Fig. 2.

Setting the epigenetic marks. Establishment of a specific chromatin environment can be dependent or independent on DNA sequence. The intrinsic DNA sequence shows different affinities for nucleosomes. A/T sequences are generally repelling nucleosomes, whereas G/C sequences are more favorable for nucleosome incorporation in vitro (left, top). Regular phasing of A/T dinucleotides might help bend the DNA and facilitate positioning of nucleosomes (left, top). Additionally, sequence-specific transcription factors are involved in establishing a defined chromatin state by recruiting chromatin remodelers (CM) and influence nucleosome positioning/depletion and histone modifications of adjacent nucleosomes (left, middle). Alternatively, non-coding RNA can serve as a scaffold to recruit chromatin remodelers in a sequence dependent fashion (left, bottom). As a second route, trans acting factors can bind to methylated DNA or specific chromatin modifications and recruit chromatin modifying complexes, which is essential for maintenance or spreading of epigenetic information (right, top). Also, non-coding RNA can span large regions of chromatin (e.g. XIST on the inactive X chromosome) in a sequence independent fashion and enforce a specific chromatin environment (right, bottom). Gray cylinders, nucleosomes; black lines, DNA; red line, specific DNA motif; TF, transcription factor; CM, chromatin modification/remodeling complex; black circles, methylated DNA; pink line, non-coding RNA; MBP, methyl DNA binding protein; HMBP, histone modification binding protein; small colored circles and hexagons indicate different histone modifications as indicated in Fig. 1.

1.3. Non-coding RNA

In recent years, it has become increasingly clear that non-coding RNAs are important modulators of chromatin regulation and gene expression. Whole genome and transcriptome sequencing demonstrated that at least 90% of the genome is actively transcribed, although less than 2% represent protein-coding genes. Thus, the non-coding part of the transcriptome became a new focus in gene expression and regulation [49–52].

Currently, two major groups of non-coding RNA players can be distinguished: small ncRNAs and long ncRNAs (lncRNAs) [52,53]. Small ncRNAs are processed from longer precursors and comprise, in addition to transfer RNAs (tRNAs) and ribosomal RNAs (rRNAs), the well-studied microRNAs (miRNAs), piwi interacting RNAs (piRNAs), small nuclear RNAs (snoRNAs) and other less well-characterized RNAs (for a detailed review on these types of ncRNA and their function see [54,55]). Long ncRNAs (lncRNAs), on the other hand, are a heterogenous class of mRNA-like transcripts from 200 nt up to 100 kb which do not code for proteins [53]. They can be transcribed from sense or antisense strand, in a bidirectional way so that a coding transcript of the opposite strand is initiated in close proximity, from intronic sequences processed from a transcript and from intergenic regions between two genes [56]. Recent studies suggest that they make up the bulk of the human transcriptome, excluding ribosomal and mitochondrial RNA [57,58]. Functional lncRNAs fulfill important regulatory roles in gene expression and regulation by assembling protein complexes and localizing them to their genomic target DNA sequence [59]. Long intergenic non-coding RNAs (lincRNAs) have been identified at intergenic sites outside protein-coding genes containing H3K4me3 and H3K36me3 domains [60]. Many of these biological active lincRNAs can physically associate with chromatin remodeling complexes such as PCR2 [61] and are needed for pluripotency and differentiation [62] (Table 1).

Table 1.

Selected lncRNAs and interaction partners.

lncRNA	Interaction/Target	Function	References
Polycomb repressive complexes (PCR)
XIST	PCR2	Silencing of X chromosome	[84]
HOTAIR	PCR2	Targets PCR2 complex to HOXD locus	[118]
ANRIL	PCR2 and PCR1	Targets PCR complexes to INK4b-ARF-INKa locus	[119–121]
Histone methyltransferases (HMTs)
AIR	HMT G9a	Silencing of maternally expressed Igf2r/Slc22a2/Slc22a3	[87]
Kcnq1ot1	PCR2 and HMT G9a	Lineage-specific silencing of the Kcnq1 locus	[88]
HOTTIP	WDR5/MLL	Activation of the HOXA cluster	[170]
Histone demethylases
HOTAIR	LSD1/CoREST/REST	HOXD locus regulation, silencing of neuronal specific genes	[171]
Transcription factors/mRNA processing
lincRNA-p21	Repressive hnRNP-K complex	Targets hnRNP-K to repress p53 target genes	[90]
NRON	NFAT	Affects transcription factor acitvity	[172]
PANDA	NF-YA	Delimits apoptosis by inhibiting NF-YA	[91]
miRNAs
PTENP1	miR-17, miR-19	Decoy for miRNAs targeting tumor suppressor PTEN	[124]
HULC	miR-372	Upregulated in liver cancer	[125]
linc MD1	miR-133	Regulation of muscle differentiation	[173]

Another class of non-coding RNAs constitute the recently discovered transcribed ultraconserved regions (T-UCRs) [63,64], which are transcribed from evolutionary ultraconserved regions found in the human, rat and mouse genome [65]. They show aberrant expression in several cancers, including adult chronic lymphocytic leukemia, colorectal and hepatocellular carcinomas and neuroblastomas [64–66]. Although distinct T-UCR expression signatures are associated with specific cancer types and their high sequence conservation across species argues for functional properties, their precise mode of action in the cell is still unknown [64].

Noteworthy, a recently introduced method called ChIRP (Chromatin Isolation by RNA Purification) has the potency to shed light on the complexity of RNA-chromatin or RNA-DNA-protein interactions on a genome-wide scale. This technology was used to map genome-wide interaction sites of three different lncRNAs including HOTAIR, and has revealed focal, sequence-specific binding at numerous sites in the genome [67].

1.4. Interrelations of different epigenetic mechanisms and targeting of epigenetic modifications

The different epigenetic layers described above are interrelated and can both reinforce each other and inhibit opposing functions. In order to establish a repressive chromatin environment DNA methylation and repressive histone modifications are combined within the same chromatin regions. Methylation of H3K9 is found in regions of DNA methylation, whereas H3K4me3 and DNA methylation occur mutually exclusive. Different HMTs including G9a, SUV39H1, EZH2 can direct DNA methylation via direct or indirect recruitment of DNMTs [68–70], which might be a mechanism for de novo methylation of DNA in ESC but appears to be nonessential for maintenance of methylation in differentiated cells [71].

The ubiquitin multi-motif protein UHRF1 is a central player in targeting repressive chromatin marks. It contains a SRA domain, which binds to hemimethylated DNA, a Tudor domain binding to methylated H3 (H3K9me3) and a PHD finger interacting with an unmodified arginine residue within H3 (H3R2) [72–74]. Furthermore, UHRF1 interacts with DNMTs, G9a and HDAC1 and thereby unites various enzymes that can provide a repressive chromatin environment [75–77]. Interestingly, UHRF1 also recruits the H2AK5 actetyltransferase TiP60 thus integrating a multitude of different epigenetic signals [78]. A further example for the link between DNA methylation and histone modifications represent methyl C binding proteins such as MeCP2, which interact with co-repressor complexes including HDACs and HMTs [79,80]. Interestingly, a recent report shows that components of the piRNA pathway are required to target de novo DNA methylation to an imprinted region of the mouse genome implicating that selective methylation of imprinted regions can be regulated by non-coding piRNAs [81].

Although we can dissect different functional genomic areas by their chromatin modification pattern, we still don't understand the mechanism underlying this patterning and whether epigenetic marks such as posttranslational histone modifications are cause or consequence of chromatin states [23] (Fig. 2). Evidence is accumulating, that the DNA sequence per se is at least in part able to direct chromatin structure and modification. This involves positioning of nucleosomes in a sequence dependent manner [82]. Additional binding of transcription factors and recruitment of chromatin remodeling complexes are essential to demarcate active regions and to adjust nucleosome composition, occupancy and positioning [48]. Interestingly, de novo methylation of DNA is also mediated by genetic elements, which are found in promoters and contain binding motifs for transcription factors [83].

Another targeting mechanism of epigenetic modifications relies on lncRNAs, which function as “adaptor platforms” for interactions between chromatin and chromatin remodeling complexes and are able to recruit and direct chromatin remodeling complexes to specific loci in the genome [59] (Fig. 2). The longest-known examples for this are lncRNAs involved in epigenetic silencing and imprinting, such as the X-inactivation promoting lncRNA XIST, which recruits the Polycomb repressive complex (PRC) to silence the X chromosome from which it is transcribed [84] and TSIX, which is transcribed from the opposite strand and regulates XIST levels during X-inactivation [85]. Other lncRNAs involved in imprinting processes are the paternally expressed lncRNA AIR, which is required for silencing the maternally expressed protein-coding genes Igf2r/Slc22a2/Slc22a3 [86] and inhibits expression by targeting the H3K9 histone methyltransferase G9a to the Slc22a3 promoter [87], and paternally expressed lncRNA Kcnq1ot1, which mediates lineage-specific silencing of the Kcnq1 locus by interaction with the PRC2 complex and G9a HMT in placenta, but not in fetal liver [88].

Furthermore, loss of function studies and analyses of custom-designed microarrays demonstrate involvement of functional lncRNAs in development and differentiation [62,89]. Specific sets of lncRNAs are needed to keep embryonic stem cells in a pluripotent state and promote differentiation by association with not only chromatin modifiers but also specific transcription factors such as SOX2 and REST [89].

lncRNAs can also contribute to modulation of cell cycle networks in the cell, such as the recently discovered lincRNA-p21, which is induced by p53 and mediates gene repression of p53 target genes by associating with repressor complex heterogeneous nuclear ribonucleoprotein K (hnRNP-K) and targeting the complex to previously active genes [90]. Other examples for lncRNAs in the p53 network are the p53-induced lncRNA PANDA and the maternally expressed gene 3 (MEG3). PANDA is transcribed from the CDKN1A locus and delimits apoptosis after DNA damage [91] by specifically binding and inhibiting NF-YA, a nuclear transcription factor that is responsible for activating genes related to apoptosis. MEG3 can activate both p53-dependent and p53-independent pathways and has tumor suppressive functions [92]. Interestingly, it was shown that partial replacement of MEG3 RNA with unrelated sequences did not alter p53 activation, indicating that lncRNA function largely depends on secondary structures and that lack of sequence conservation, which is observed for many lncRNAs, does not affect functionality.

2. Alterations in cancer

2.1. Altered epigenetic patterns

The epigenome of cancer cells displays numerous alterations in comparison to the epigenome of their normal counterpart. Changes in DNA methylation include a genome-wide loss and a regional gain of DNA methylation. This causes on one hand genomic instability and deregulation of tissue specific and imprinted genes and on the other hand silencing of tumor suppressor genes – controlling cell cycle, apoptosis or DNA repair – by hypermethylation of their promoter CpG islands [3,7]. Interestingly, no global hypomethylation but rather a directed hypomethylation at satellite repeats could be detected in malignant peripheral nerve sheath tumors using a genome-wide approach [93].

Unusually high frequency of DNA methylation at CpG rich sites was termed CIMP (for CpG island methylator phenotype) and was first identified in colorectal cancer [94]. CIMP is associated with diverse clinicopathological characteristics such as patient age, gender, tumor location, microsatellite instability and genetic mutation in the BRAF gene [95]. Interestingly, CIMP can also be found in other tumor entities such as glioma or breast cancer, where it also allows for a sub-classification of tumors and determines metastatic potential, respectively [96,97]. In analogy to genetic mutation, tumors seem to accumulate higher levels of DNA methylation during tumor progression and genome-wide profiling of DNA methylation has proven useful for tumor classification of different tumor types [98–100]. These defined alterations are currently evaluated for their use as biomarkers for diagnosis, prognosis and prediction of therapy response for different cancers [101].

Another target of aberrant DNA methylation in cancer are CpG island shores, defined as approximately 2 kb regions surrounding CpG islands. These regions were initially identified hypermethylated in colorectal cancer and represent regions of tissue specific methylation in normal tissues [102]. Methylation of shore regions is related to gene expression and has also been detected in different tumor cell lines and nerve sheath tumors [93,103]. In contrast, no differential methylation of shore regions was detected in different lineages of the murine immune system [104].

Tumor heterogeneity presents an obstacle for therapeutic intervention and cure. Stochastic methylation variability was detected in cancer specific differentially methylated regions (cDMRs), which might contribute to tumor heterogeneity [105]. Further, large hypervariable blocks covering half the genome show differences in gene expression patterns, involving genes that regulate tumor-associated processes such as cell division and matrix remodeling [105]. This heterogeneity of methylation patterns might be the foundation for the selective advantage of tumor cells and provide a cellular mechanism of evolution [106].

Additional insight into the cancer methylome has been gained from a recent study in primary colon cancer [107]. Single CpG resolution genome-wide bisulfite sequencing enabled the identification of large hypomethylated regions, covering more than half of the genome. These regions coincided with late replication foci and nuclear lamina associated domains. This study thus proposes the 3D chromatin architecture to be involved in epigenetic reprogramming in cancer cells. Another link between chromatin architecture and DNA hypermethylation of tumor suppressor genes stems from the finding that loss of CTCF binding in multiple tumor cell lines coincides with silencing of p16^INK4a and the loss of an upstream chromatin boundary [108].

Numerous changes in modification patterns have also been observed at the level of posttranslational histone modifications [109]. The loss of repressive heterochromatin is reflected by a depletion of repressive histone marks such as H4K20me3 and H4K16ac in these regions [110]. Further, H3K27me3 seems to occur mutually exclusive to DNA methylation and promote de novo silencing of genes in different cancers [6,111]. Many developmental genes that are silenced by H3K27me3 in embryonic stem cells are silenced by DNA methylation in cancer cells, establishing an epigenetic switch from a dynamic to a more stable silencing system that promotes a stem cell-like signature of cancer cells [112–114]. Generally, changes in histone modification patterns were reported in a variety of tumors and correlated to tumor stage and prognosis, however with contradicting results [109]. Since histone modification patterns are very dynamic and can be easily erased by opposing enzymes, it seems that a regulated balance of these modifiers needs to be present within a cell. Shifting the balance to either side might therefore result in different outcomes and either promote or restrict tumor proliferation. This is highlighted by the recent finding that the histone deacetylase HDAC1, which was previously associated with cell cycle arrest, can induce tumor proliferation in a teratoma mouse model [115,116].

Altered epigenetic patterns in cancer cells are also associated with the deregulation of lncRNAs and subsequent re-positioning of chromatin modifying complexes. For example, increased expression of the lincRNA HOTAIR was found in primary and metastatic breast tumors [117]. Overexpression of HOTAIR, which is normally expressed antisense to the HOXC locus during development and targets the PCR2 complex to the HOXD locus [118], leads to different PCR2 occupancy at chromatin sites, altered H3K27 methylation patterns resembling those of embryonic fibroblasts and increased cancer invasiveness in breast cancer cells [117]. Another lncRNA involved in targeting of PCR complexes to tumor suppressor genes is antisense non-coding RNA in the INK4 locus (ANRIL) [119,120]. ANRIL is transcribed from the antisense strand at the INK4b-ARF-INK4a locus, which is an important regulator of cell cycle progression, apoptosis and cellular senescence [121,122]. By recruiting PCR1 and PCR2 complexes to form heterochromatin surrounding the INK4b-ARF-INK4a locus, ANRIL mediates silencing of these tumor suppressor genes.

According to an interesting new hypothesis non-coding RNAs can, besides assembling chromatin complexes and modulating cell cycle networks, act as decoys for miRNAs when they contain specific microRNA binding sites. This in turn has implications on the expression of other mRNAs and regulatory networks normally targeted by the respective miRNA [123]. An example for this competing endogenous RNA (ceRNA) is the transcribed PTENP1 pseudogene, which contains many microRNA response elements (MREs) also present in the tumor suppressor PTEN. PTENP1 has been shown to be able to regulate cellular levels of PTEN by detracting miRNAs from PTEN mRNA and is selectively lost in cancer, indicating a tumor suppressor function [124]. Another example for an oncogenic “endogenous sponge” is the highly upregulated in liver cancer (HULC) lncRNA, which sequesters miR-372 and in turn induces its own transcriptional up-regulation in liver cancer [125].

2.2. Mutations in epigenetic enzymes

Chromatin modifying enzymes such as HMTs, histone acetyltransferases (HATs) and HDACs have been implicated in the pathology of leukemia, either as direct or indirect partners of oncofusion proteins [126]. The HATs MOF, MOZ or p300/CBP and the HMT MLL are frequently found in translocations in acute myeloid leukemia (AML) [127], whereas indirect recruitment of co-repressor complexes including HDACs has been found in PML/RARalpha and PLZF/RARalpha fusions in acute promyelocytic leukemia (APL) [128,129].

Advances in next generation sequencing technologies have resulted in the discovery of novel somatic mutations driving different cancers [130]. Intriguingly, numerous chromatin related enzymes and proteins were among the newly identified genes [109,131]. For example, the HMT EZH2, which has previously been attested oncogenic potential in different solid cancers such as prostate or breast cancer, is frequently mutated in hematological malignancies together with other members of the PRC2 such as EED and SUZ12 [132]. Interestingly, inactivating mutations were identified in myeloid disorders [133,134] and in T-ALL (acute lymphoblastic leukemia) [135], whereas activating mutations leading to hyper-trimethylation of H3K27me3 were associated with two mutations (Y641, A6779) in follicular lymphoma and other B-cell lymphomas [136–138].

Mutations associated with DNA methylation have also been described in hematological disorders. The de novo methyltransferase DNMT3a was found mutated in acute myeloid leukemia (AML) and myelodysplastic syndrome [139–141]. The family of TET proteins has been implicated in DNA demethylation in ES cells and is frequently found mutated in myeloid disorders [142]. This is related to lower levels of 5-hydroxymethyl-Cytosine (5hmC) but surprisingly also to DNA hypomethylation in affected patients [143]. TET function can also be affected by mutations in IDH1 and IDH2 genes, which cause accumulation of 2-hydroxyglutarate inhibiting TET2 [144]. Somatic mutations of IDH1 have been associated with a CIMP phenotype in glioma [96]. An oncogenic cooperation was suggested recently for DNMT3a and TET2 in T-cell lymphoma indicated by a frequent co-occurrence of mutations in both genes, where 73% of patients with DNMT3a mutations also harbored TET2 mutations [145].

Two recent publications independently demonstrated that mutations at two critical sites of posttranslational modifications in the histone H3 variant H3.3 are frequently detected in different forms of brain tumors [146,147]. Schwatzentruber et al. further found somatic mutations in the H3.3-ATRX-DAXX chromatin remodeling pathway, which is needed for the incorporation of histone H3.3 into pericentromeric heterochromatin, in 44% of pediatric glioblastoma multiforme (GBM). This was accompanied by alternative telomere lengthening highlighting the important role of H3.3 for genome integrity and chromatin architecture [46].

An unexpected role in chromatin regulation was recently ascribed to the tumor suppressor BRCA1 [148]. Using nestin-Cre specific deletion of Brca1 in murine neural stem cells, Zhu et al. discovered an impaired heterochromatin structure in knockout cells, which was associated with increased transcription of satellite, repeats. This was due to the loss of histone H2A ubiquitylation at pericentromeric heterochromatin, which was dependent on the ubiquitin ligase function of BRCA1. Increased transcription of satellite repeats was associated with DNA damage and genomic instability and was also detected in BRCA1 deficient murine and human breast cancers. Thus, the authors propose that the tumor suppressive function of BRCA1 might be largely dependent on its regulation of pericentromeric heterochromatin, which allows for controlled cell division and genome integrity [148]. Overexpression of satellite repeats was also identified in pancreatic and other epithelial cancers lately, indicating that disruption of heterochromatin may result in genomic instability in a variety of human cancers [149].

2.3. Possible causes of aberrant epigenetic patterns

In order to recognize the source of aberrant epigenetic patterns, we need to understand how epigenetic pathways work in their normal environment. Recent literature has provided us several options (Fig. 3). First, alterations of epigenetic patterns such as aberrant DNA methylation could arise stochastically due to loss of fidelity or mutation of epigenetic enzymes [109]. This is supported by the finding that aging results in a loss of the global DNA methylation content [150] and by a large heterogeneity in genome-wide methylation patterns of different cancers [105].

Fig. 3.

Altered nuclear and epigenomic structure of cancer cells. In normal cells, the nucleus is subdivided into active (yellow) and inactive (gray) territories. Inactive territories are located close to the nuclear lamina (blue circle) and contain inactive, hypermethylated regions (black circles) that are tightly packed into nucleosomes with repressive histone marks (red cylinders). Active regions are found in the nuclear center, and contain “transcription factories” (green shapes), which are composed of specific transcription factors, RNAPII, chromatin remodelers and CTCF/cohesin proteins for looping chromatin into these areas. Promoters of active genes (red lines) are unmethylated (white circles) and marked by positive histone marks (green cylinders). Permanent stress (such as chronic inflammation (CI), UV exposure, reactive oxygen species (ROS)) might induce a stable epigenetic switch via mutation in epigenetic enzymes (I), mistargeting or altered composition of epigenetic complexes (II) or altered nuclear architecture linked to stress specific gene expression patterns (III). The altered nuclear morphology of cancer cells is associated with re-arranged active and inactive territories and chromatin, global DNA demethylation (white circles) and local DNA hypermethylation at CpG islands, which results in silencing of underlying genes (indicated by blue lines, black circles within blue lines indicate methylated promoter CpG islands). Transcription factories contain a tumor cell specific set of proteins (red shapes) such as transcription factors and maintain a tumor cell specific gene expression profile.

The nucleus has to accommodate DNA in an ordered fashion, to allow for regulated gene expression. Using Hi-C to resolve the 3-dimensional genome architecture, Lieberman–Aiden et al. constructed spatial proximity maps identifying regions of segregated open and closed chromatin states [151]. This suggests a territorial arrangement of nuclear architecture, which is supported by the finding that heterochromatic regions and repressed genes tend to be associated with the nuclear lamina, whereas active chromatin is moved away from the lamina [152,153]. Clustering of Polycomb regulated HOX genes and the organization of repressive chromatin into Polycomb bodies lends further support to these findings [154]. Actively transcribed regions of the genome have been thought to cluster in so-called “transcription factories” [155]. Recent data substantiate this model, which suggests that genes regulated by specific transcription factors are clustered and looped around factories of concentrated RNAPII [156]. The first 3D interaction map of RNAPII occupied sites was recently established using chromatin interaction analysis by paired-end-tag sequencing (ChIA-PET) [157]. RNAPII dependent interaction sites consist of extensive promoter–promoter interactions between proximal and distant genes yielding multi-gene complexes that cooperatively regulate their activity and are enriched for active chromatin marks. One might envision a model, in which specific transcription factors assemble genes via promoter/enhancer binding assisted by proteins involved in chromatin looping and organization such as CTCF, cohesin or chromatin remodelers [156–160]. Nuclear structure including chromatin texture is altered in tumor cells and used by pathologists for diagnosis of malignancy. Molecularly this would hold a considerable potential for large-scale rearrangements of genomic organization causing deregulated gene expression patterns and aberrant chromatin structure, which is in line with recent findings related to aberrant DNA methylation in cancer cells and provides a second route to altered epigenetic patterning in cancer [37,105,107,161].

Finally, alterations in the cancer epigenome might result from mistargeting or altered composition of epigenetic complexes. By inducing cellular oxidative stress O'Hagan et al. showed, that a complex consisting of DNMTs and HDACs is formed and recruited to the sites of damaged DNA. Components of the complex are recruited from non-GC-rich to GC-rich areas. The authors detect similar changes in an in vivo inflammatory model and suggest, that delocalization of key epigenetic enzymes upon cellular stress might be the cause of global and local epigenetic alterations of cancer cells [162].

2.4. Signaling to the cancer epigenome

Epigenetic patterns are initially established as a consequence of developmental cues and are maintained even after the initiating signal is removed. For example, early differentiation programs are turned on in response to specific transcription factors and are upheld in the determined lineage. Nonetheless the epigenome retains some level of plasticity and can be shaped by environmental factors. Polycomb group (PcG) proteins are essential to maintain the cellular memory by acting in these aforementioned ways of both stabilizing cell fate decisions but also regulating gene expression patterns in response to extrinsic signals [163]. This has been exemplified by the induction of transdetermination in Drosophila via suppression of PcG proteins by JNK kinase signaling [164], which links upstream kinase signaling with chromatin modifications. A further connection between kinase signaling in response to external factors and PcG proteins has been made by the finding that the PRC2 component EZH2 can be directly phosphorylated at different sites by a number of cell cycle dependent and stress induced kinases [165]. Although the effects of site specific phosphorylation of EZH2 are somewhat controversial, it can be anticipated that they regulate EZH2 interaction with other PRC members, enzymatic activity or targeting to chromatin sites. Aside from modification of chromatin related proteins, stress induced kinases can directly modify histone tail residues as recently demonstrated for JNK during stem cell differentiation into neurons [166]. Recent work has identified numerous upstream kinases that directly act on chromatin in response to cytokines, growth factors or ultraviolet light and are involved in transcriptional regulation, chromatin condensation, apoptosis and DNA damage repair [167]. Combinatorial histone modification patterns have been observed among the many possible modifications of histone tails. Related to histone phosphorylation, a crosstalk between phosphorylation and acetylation of neighboring residues has been detected. For example, H3S10 phosphorylation is facilitating the acetylation of the nearby H3K9/K14 marks and has an important function for the induction of different genes in response to stress (for details see also review by Sawicka and Seiser this issue). Additionally, phosphorylation of the H3 tail can alleviate the repressive function of histone methylation on H3K9 and H3K27 by releasing HP1 or PRC2 from chromatin, respectively [167]. Thus, kinase signaling targets transcription factors, chromatin modifiers and chromatin itself to robustly induce transcriptional responses to external stimuli and if deregulated might lead to local and global changes in chromatin structure and gene expression patterns as observed in various cancers.

3. Conclusions

Research of the last decade has highlighted the essential role of epigenetic alterations in cancer development and progression [168]. Latest technologies have allowed for the analysis of the cancer epigenome and have resulted in important discoveries, which provide the basis for new concepts of how epigenetic alterations might emerge. We are now able to look at the cancer epigenome from a bird's eyes view and we are just beginning to understand how alterations in nuclear architecture and global chromatin, as observed by pathologists since more than 150 years, are related to epigenetic mechanisms [169]. Based on latest literature we can envision different pathways that might cause epigenetic alterations in cancer cells as observed by genome-wide epigenomic profiling (Fig. 3). (I) Mutations in chromatin related enzymes such as DNMTs, histone modifiers or chromatin remodelers might induce stochastic changes in the epigenome causing global changes in chromatin. (II) Faulty targeting (e.g. by lncRNAs or transcription factors) or altered composition of epigenetic complexes can lead to global and gene specific changes in the epigenetic signature. (III) Changes in the spatio–temporal organization of nuclear architecture or loss of boundaries might cause altered nuclear territories, disrupt ordered epigenetic patterns and induce altered gene expression programs. All three pathways could result from altered external signaling due to permanent stress and act in concert to lock in the cancer epigenome.