Keywords

These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

1.1 Introduction

Originally coined by Conrad Waddington in the 1940s, the term epigenetics currently defines the mitotically and/or meiotically heritable changes in gene expression that take place without changes in the DNA sequence (Berger et al. 2009). More specifically, epigenetics includes a series of cellular mechanisms that, in response to external signals, are capable of modifying chromatin packaging, thus creating transient or permanent, global or local, condensed, and decondensed chromatin states that modulate access of the different enzymatic machineries involved in the main genetic processes (transcription, replication, recombination, and DNA repair). These patterns of more closed or open chromatin (also called heterochromatin and euchromatin, respectively) display special features during stemness. Subsequently, during differentiation, they undergo a series of changes, remain fixed, and become stably transmitted through multiple cycles of cell division. This gives rise to the different gene expression patterns that underlie cell-type identity and lineage specification, as well as adult cell renewal. Epigenetic layers also play a key role in processes like X-chromosome inactivation and genomic imprinting (Portela and Esteller 2010).

The classic epigenetic mechanisms are DNA methylation, a wide range of covalent histone modifications, and nucleosome positioning (through large ATP-dependent chromatin-remodeling complexes and the substitution of canonical histones with different variants) (Fig. 1.1). One should also highlight the role of noncoding RNAs (ncRNAs) such as microRNAs (miRNAs), which directly affect, or are affected by, all the former mechanisms. These are commonly included as part of the epigenetic setting phenomenon although, as with histone modifications, it is possible that not all of them may be self-perpetuating or inherited (Riddihough and Zahn 2010). Dysregulation of any of these mechanisms, which continuously work in close collaboration, is known to provoke misexpression of critical genes, problems in DNA repair, etc., leading to diseases like cancer (Portela and Esteller 2010; Rodriguez-Paredes and Esteller 2011).

Fig. 1.1
figure 00011

Summary of the classical epigenetic mechanisms acting on a typical active gene in normal cells. This figure, depicting a typical transcriptionally active gene in a healthy cell, serves as an introduction to the various classic epigenetic mechanisms. Thus, the promoter and the region around the TSS present different activating covalent histone modifications which, recognized by ATP-dependent chromatin-remodeling complexes (CRCs), lead to a more open configuration of chromatin. Nucleosomes with histone variants like H2A.Z (in brown), upstream and downstream of the TSS, contribute to this transcriptionally permissive state. In the body of the gene, while other histone marks indicate active transcription, DNA methylation aborts spurious initiations

Full size image

Throughout this chapter, we will summarize the current knowledge about the operation of the main epigenetic mechanisms in normal cells, the way their disruption can promote tumorigenesis, and, finally, their importance in stemness. Along the way, we will also describe how all these mechanisms continuously coordinate in a given cell to yield what is currently called its epigenome, the epigenetic status that, starting from a single mammalian genome, gives rise to the diverse cell types and developmental stages of the organism.

1.2 Epigenetic Mechanisms in Normal and Cancer Cells

1.2.1 DNA Methylation

1.2.1.1 In Normal Cells

DNA methylation is historically the most extensively studied epigenetic mechanism, and its main effect is to silence genes and noncoding genomic regions. It normally consists of the covalent addition of a methyl group (-CH3) at the 5′ position of the cytosine ring within CpG dinucleotides, but in embryonic stem (ES) cells, about 25 % of the total DNA methylation occurs in a non-CpG context (Lister et al. 2009). This unusual DNA methylation might be important for the origin and maintenance of pluripotency, as it disappears upon differentiation and is restored in induced pluripotent stem cells (Lister et al. 2009; Laurent et al. 2010).

The majority of the CpG dinucleotides are concentrated either in the so-called CpG islands or in regions of large repetitive sequences like centromeres and retrotransposon elements (Deaton and Bird 2011; Esteller 2008). CpG islands are CpG-rich DNA regions spanning, on average, 1,000 base pairs and are normally located near transcription start sites (TSSs) at approximately 70 % of annotated gene promoters (almost all those corresponding to housekeeping genes and some of the tissue-specific and developmental regulator gene promoters) (Larsen et al. 1992; Saxonov et al. 2006; Zhu et al. 2008). Although recent work has uncovered a large number of new CpG islands, called orphan CpG islands because they are located far away from the annotated TSSs, they may be canonical CpG islands associated with either alternative promoters of nearby annotated genes or with ncRNA promoters (Illingworth et al. 2010; Maunakea et al. 2010).

While the function of DNA methylation at repetitive sequences is to protect chromosomal integrity by preventing chromosome instability and transposition (parasitic sequence elements may represent more than 35 % of our genome), DNA methylation at CpG islands leads to the silencing of the corresponding genes (Esteller 2008). Intriguingly, while most CpG islands remain unmethylated during development and in differentiated tissues, only 6 % of the genes become methylated, and thus silenced, in this process. Among them are pluripotency and germ line-specific genes, different tissue-specific genes depending on each cell type, and, finally, genes involved in genomic imprinting and X-chromosome inactivation (Straussman et al. 2009; Suzuki and Bird 2008; Mohn et al. 2008). In this regard, it has been reported that extensive DNA methylation changes due to differentiation do occur at CpG island shores, regions of comparatively lower CpG density located near canonical CpG islands (Doi et al. 2009; Meissner et al. 2008). The fact that most tissue-specific DNA methylation takes place at these CpG island shores is supported by the observation that they account for 70 % of the differentially methylated regions when reprogramming somatic cells (Doi et al. 2009; Ji et al. 2010). The strong influence of DNA methylation upon cell identity is also underlined by the discovery that the efficiency of this reprogramming is greatly improved when methylation levels are artificially decreased (Huangfu et al. 2008).

On the other hand, it is important to point out that not only CpG dinucleotides within CpG islands may be methylated. Although they are not very abundant, there are also CpG dinucleotides located in the body of the genes, and in the case of those that are ubiquitously expressed, they commonly undergo methylation. Here, the function seems to be related to elongation efficiency and prevention of spurious initiations of transcription (Zilberman et al. 2007). Another interesting issue is the influence of DNA methylation on non-CpG island promoters, as it is known that methylation of the POU5F1 and SERPINB5 promoters, for instance, also has important negative effects on their expression levels (Futscher et al. 2002). In this regard, a recent study showed a clear nucleosome occupancy of these promoters after their methylation (Han et al. 2011).

The enzymes responsible for adding the methyl groups are called DNA methyltransferases (DNMTs). In mammals, there are five DNMTs—DNMT1, DNMT2, DNMT3a, DNMT3b, and DNMT3L—but only DNMT1, DNMT3a, and DNMT3b are able to transfer a methyl group from S-adenosylmethionine, the usual donor, to DNA. DNMT3a and DNMT3b, the so-called de novo DNMTs, are highly abundant in ES cells but very little expressed after differentiation (Portela and Esteller 2010). As a result, they are thought to be in charge of establishing the DNA methylation patterns during development. Despite being catalytically inactive, DNMT3L is expressed during gametogenesis, when it stimulates the de novo DNMTs and has an important role in genomic imprinting (Bourc’his et al. 2001; Holz-Schietinger and Reich 2010). The gene encoding DNMT1, the maintenance DNMT, is transcribed mostly during S phase and has a 30- to 40-fold preference for hemimethylated DNA. Hence, although it also possesses de novo activity, its role is far more important during semiconservative DNA replication, when it methylates the resulting hemimethylated sites. Its interaction with either the DNA polymerase processing factor, PCNA, mainly, or with the ubiquitin-like plant homeodomain and RING finger domain-containing protein 1 (UHRF1) promotes its localization at the replication fork (Chuang et al. 1997; Bostick et al. 2007). In a recent model proposed by Jones and Liang, DNMT3a and DNMT3b, which strongly anchor to nucleosomes containing methylated DNA, also help DNMT1 in this task (Jeong et al. 2009; Jones and Liang 2009). Finally, DNMT2, which does have all the catalytic signature motifs, is known to methylate tRNAAsp (Goll et al. 2006).

Two mechanisms could explain how DNA methylation leads to the inhibition of expression. First, for transcription factors such as CTCF, among others, DNA methylation represents a physical barrier to access to their binding sites (Rodriguez et al. 2010). Second, methylated DNA can recruit members of both the Kaiso-like or the methyl-CpG-binding domain (MBD) family of proteins (the latter being composed of MeCP2, MBD1, MBD2, MBD3, and MBD4) that, in turn, either interact with repressive histone-modifying enzymes, like histone deacetylases (HDACs) (as is the case of MeCP2), or directly belong to repressive chromatin-remodeling complexes like NuRD, which already includes these kinds of enzymes among its subunits (as is the case of MBD2) (Bogdanovic and Veenstra 2009; Lai and Wade 2011). All the latter actors eventually bring about the silencing.

Another example of the complex interplay between the different epigenetic mechanisms is provided by the way the DNA methylation machinery is probably targeted to particular genes throughout the genome. In this regard, numerous studies describe the recruitment of the DNMTs by histone-modifying enzymes and/or the marks they set on the chromatin (Esteve et al. 2009; Ooi et al. 2007; Tachibana et al. 2008; Wang et al. 2009; Zhao et al. 2009). X-linked CpG islands do not become methylated during X-chromosome inactivation until they acquire silencing marks like trimethylation of lysine (K) 27 on histone H3 (H3K27me3) (Okamoto and Heard 2009), and most CpG islands gaining methylation during differentiation are already silent in ES cells (Mohn et al. 2008), or the interaction between EZH2 (the enzyme that writes this mark) and DNMT3B (Viré et al. 2006). Considering these three points, it was proposed that, as a general rule, target genes were first marked by H3K27 methylation and the sequence was subsequently locked by the DNA methylation machinery (Ohm et al. 2007). Although it seems clear that there must be some connection between H3K27me3 and DNA methylation, the reality is that it has recently been reported in cancer cells that only up to 5 % of promoters containing CpGs are silenced by H3K27me3, independently of DNA methylation (Kondo et al. 2008). Another possibility for targeting the DNA methylation machinery comes from the plant world, where double-stranded small inhibitory RNAs (siRNAs) are known to initiate a stepwise process that targets DNMTs to specific regions—not only promoters but also repetitive sequences (Mosher and Melnyk 2010; Vrbsky et al. 2010). Although some of the components of this mechanism are conserved in mammals, the existence of an equivalent process in humans is still unclear.

Finally, the fact that 5-hydroxymethylcytosine could be an intermediary of DNA demethylation has recently focused considerable attention on either the mark itself, as has already been detected in ES cells and Purkinje neurons, or the enzymes capable of catalyzing it from the usual 5-methylcytosine, the three 2-oxoglutarate- and Fe(II)-dependent oxygenases TET1, TET2, and TET3 (Tahiliani et al. 2009). Further investigation is required to understand the role and nature of this modification, but it seems that while TET1 and, to a lesser extent, TET2 are essential players in maintaining pluripotency, TET3, which is not expressed in ES cells, may work with TET1 or TET2 to control hydroxylation in differentiated cells (Ito et al. 2010; Koh et al. 2011; Williams et al. 2011).

1.2.1.2 In Cancer Cells

Cancer cells show two major types of DNA methylation-related aberrancies that contribute to their malignant state: global hypomethylation of their genomes and CpG island promoter hypermethylation of different tumor suppressor genes.

Global hypomethylation at repetitive sequences, retrotransposons, CpG-poor promoters, introns, and gene deserts was the first epigenetic abnormality detected in cancer cells and is present in benign and malignant tumors (Esteller 2008). Its direct consequence is genomic instability, one of the most characteristic features of cancer cells, as a loss of methylation at repetitive sequences can favor mitotic recombination leading to deletions and translocations, as well as chromosome arrangements (Eden et al. 2003; Esteller 2008). For example, cancer cells from ovarian and breast carcinomas, or Wilms tumors, which display severe hypomethylation of their pericentromeric satellite sequences, also present chromosomal translocations with breakpoints precisely in their pericentromeric DNA (Yeh et al. 2002). Moreover, hypomethylation of parasitic sequence elements can promote their transposition, once again generating chromosome rearrangements. One example is the hypomethylation of L1 retrotransposons in colorectal cancer (Howard et al. 2008). Finally, loss of methylation of latent viral sequences incorporated in the genome, and normally silent through DNA methylation, can activate them, again contributing to cancer. This is the case of genital human papillomaviruses in cervical tumors (Badal et al. 2003).

Global hypomethylation can contribute to tumorigenesis in two other ways. First of all, it can lead to the reactivation of some of the genes normally silenced in healthy cells by DNA methylation. Some examples are SERPINB5, whose aberrant reactivation promotes cell dedifferentiation in some cancers, S100P in pancreatic cancer, SNCG in breast and ovarian cancers, or MAGE and DPP6 in melanomas (Portela and Esteller 2010). Of course, many genes encoding miRNAs are also affected by this abnormal process. A good example is the let-7a-3 miRNA gene, which is reactivated in endometrial and colon cancer (Brueckner et al. 2007). Finally, DNA hypomethylation can promote loss of imprinting (LOI). For example, LOI of the insulin-like growth factor 2 (IGF2) gene has been reported in many different tumor types (Ito et al. 2008).

On the other hand, CpG island promoter hypermethylation leads to cancer by silencing tumor suppressor genes involved in the main cellular pathways related to tumorigenesis, such as cell cycle control, apoptosis, cell adhesion, and metastasis. Importantly, many silenced genes are also involved in DNA repair, which connects this kind of epigenetic dysregulation with all types of mutations, the classic genetic alterations linked to cancer. Some examples are listed in Table 1.1. Likewise, the expression of miRNA genes with tumor suppressor features is also known to be repressed by this epigenetic mechanism. Two examples of this are the silencing of miR-124a, leading to CDK6 activation and Rb phosphorylation, and the silencing of miR-148, miR-34b/c, and miR-9, leading to metastatic phenotypes (Lujambio et al. 2007, 2008). This kind of epigenetic silencing of tumor suppressors has often been proposed as the second of the two essential events described by Knudson in his two-hit model of cancer initiation (Jones and Laird 1999). In this regard, this mechanism could silence the remaining active allele of a previously mutated tumor suppressor.

Table 1.1 Genes typically silenced by CpG island promoter hypermethylation in several frequent cancer types
Full size table

Recent findings suggest that most of the aberrant DNA methylation in cancer does not really occur at CpG islands, but at CpG island shores, although with equivalent consequences (Doi et al. 2009; Irizarry et al. 2009; Ji et al. 2010). In any case, this process is known to affect more genes than do mutations (Jones and Baylin 2002; Schuebel et al. 2007), and in fact, many genes, such as RASSF1A, owe their participation in tumorigenesis solely to this epigenetic alteration, as they are infrequently, if ever, mutated in cancer (Burbee et al. 2001; Dammann et al. 2000).

Hypermethylation patterns are not random. Each tumor type can be defined by its own hypermethylome (Esteller 2007), but what is not so clear is the way cancer cells decide which genes must be aberrantly silenced. One obvious explanation is that silencing of particular genes can provide a growth advantage and, therefore, clonal selection (Esteller 2007). Another possibility comes from the interaction of DNMTs with oncogenic transcription factors, which, in turn, recruit them to their specific target genes. In acute promyelocytic leukemia (APL), for instance, the PML-RARA fusion protein interacts with DNMT1 and DNMT3a leading to the hypermethylation of the RARB2 promoter (Di Croce et al. 2002). Moreover, the observation that large genomic regions become hypermethylated in cancer may mean that, just because of their location within those regions, many CpG islands could gain aberrant methylation (Frigola et al. 2006). Finally, as we saw for healthy cells in the last section, it is clear that histone marks and/or the enzymes that write them must also play a key role in the aberrant targeting of the DNA methylation machinery. In cancer cells, EZH2 also recruits DNMTs to the promoters of subsequently hypermethylated genes, like TBX3 in breast cancer or HOXD10, SIX3, KCNA1, or MYT1 in osteosarcoma (Vire et al. 2006).

Obviously, any kind of alteration in the normal levels of DNMTs can contribute to the abnormal methylation patterns present in cancer cells. In fact, increased expression of the genes encoding DNMT1, DNMT3a, and/or DNMT3b may be an early and important event in many tumor types (Daniel et al. 2011). Likewise, mutations in DNMT1 and in DNMT3A in acute myeloid leukemia (AML) have also been described (Daniel et al. 2011; Ley et al. 2010). Upregulations and/or mutations of members of the MBD and Kaiso-like families of proteins have also been reported in many human cancers (Kanwal and Gupta 2010). Finally, as MLL-TET1 fusions have been found in some cases of acute myeloid and lymphocytic leukemia, and homozygous null mutations and chromosomal deletions involving TET2 have been observed in several myeloid malignancies, it seems that the disruption of the system in charge of placing 5-hydroxymethylcytosines throughout the genome is also connected to cancer (Mohr et al. 2011).

1.2.2 Covalent Histone Modifications

1.2.2.1 In Normal Cells

The various histones of the nucleosomes can undergo multiple posttranslational covalent modifications, mainly along their N-terminal tails but also in their core and C-terminal regions (Bannister and Kouzarides 2011). Hence, histones may be subject to acetylation, ubiquitination, sumoylation, or propionylation at their lysines; methylation at their lysines (one, two, or three groups) or arginines (one or two); phosphorylation mostly at their serines, threonines, or tyrosines; ADP ribosylation at their arginines or glutamates; β-N-acetylglucosamination of their serines and threonines; deimination of their arginines; and/or isomerization of their prolines (Bannister and Kouzarides 2011; Liu et al. 2009a; Sakabe et al. 2010). Intriguingly, their N-terminal tails can be even clipped (Duncan et al. 2008; Santos-Rosa et al. 2009).

All these modifications are established and removed by different enzymes. So far, histone acetyltransferases (HATs), deacetylases (HDACs), methyltransferases (HMTs), and demethylases (HDMs) have been the most extensively studied systems, although intensive research is currently being devoted to understanding the mode of action, specificity, and function of the other histone-modifying enzymes.

The acetyl groups are added by two types of HATs (Table 1.2). Type-B HATs, like HAT1, acetylate newly synthesized histones H3 and H4 in the cytoplasm to promote their deposition; the marks are subsequently removed (Parthun 2007). Type-A HATs can be classified into three families depending on their amino acid sequence homology and protein conformation: GNAT (HATs hGCN5, PCAF), MYST (TIP60, MYST1–4), and P300/CBP (P300, CBP) (Bannister and Kouzarides 2011). Type-A HATs are normally part of large multiprotein complexes that determine their recruitment and substrate specificity. Thus, while hGCN5 belongs to the SAGA complex (Rodriguez-Navarro 2009), TIP60 is the catalytic subunit of the NuA4 complex (Lu et al. 2009).

Table 1.2 Histone modifications in mammals: related enzymes and functions
Full size table

On the other hand, 18 different HDACs, classified into four classes according to their sequence homology to certain yeast proteins, are capable of removing this mark in mammals (Table 1.2) (Yang and Seto 2007). Class I contains the nuclear enzymes HDAC1–3 and 8; class II is formed by HDAC4–7, 9, and 10, which shuttle between nucleus and cytoplasm; and class IV only has one member, HDAC11; all of them are Zn-dependent HDACs (Yang and Seto 2007). Finally, class III, with seven members, called sirtuins due to their homology with yeast Sir2 (SIRT1–7), is the only one that requires NAD+ as a cofactor (Vaquero 2009; Bannister and Kouzarides 2011). As many HDACs normally belong to different complexes, often together with other HDACs, their recruitment and substrate specificity is not as clear. For instance, HDAC1 has been found together with HDAC2 within the NuRD, Sin3A, and Co-REST complexes (Yang and Seto 2008).

There are two main classes of HMTs (Table 1.2): lysine and arginine methyltransferases (HKMTs and PRMTs, respectively). HKMTs, which catalyze the addition of methyl groups from S-adenosylmethionine (SAM) to the ε-amino group of lysines, can be further subdivided into the SET domain-containing and DOT1L families (Allis et al. 2007). Intriguingly, all the HKMTs methylate histones within their N-terminal tails except DOT1L, which is the only member of the family with the same name. It lacks the SET domain and methylates histones H3 at lysine 79 (H3K79), within their globular cores (Feng et al. 2002). HKMTs are relatively specific enzymes, and in general, it seems that the presence of a particular amino acid (a tyrosine or a phenylalanine) in their catalytic domain may determine their capability of proceeding past the monomethyl-lysine to the di- or trimethylated residues (Cheng et al. 2005; Collins et al. 2005b). On the other hand, PRMTs add methyl groups from SAM to the ω-guanidino group of arginines and can be subclassified into type-I and type-II enzymes; both produce monomethyl-arginines, but while type-I PRMTs also yield asymmetric dimethyl-arginines, type II generates symmetric ones (Bedford and Clarke 2009). The most relevant histone PRMTs are PRMT1, PRMT4, and PRMT6 (type I), as well as PRMT5 (type II) (Bedford and Clarke 2009). Intensive research is also focused on the PR-domain-containing family of proteins (PRDMs), as its members are structurally related to the SET domain-containing HKMTs and possess N-terminal PR domains which are 20–30 % similar to the SET domain (Davis et al. 2006). PRDM2 is already known to methylate histones H3 at lysine 9 (H3K9) (Kim et al. 2003), whereas PRDM6 has been reported to methylate histones H4 at lysine 20 (H4K20) (Funabiki et al. 1994).

Considered for many years to be a stable modification, we currently know a considerable number of HDMs (Table 1.2). LSD1 was the first to be discovered and uses FAD as a cofactor (Shi et al. 2004). Depending on the complex within which it works, LSD1 can be a transcription activator or a repressor. Thus, when associated with the Co-REST repressor complex, it demethylates mono- and dimethyl residues at H3K4 (H3K4me1/2) and acts as a repressor; when associated with the androgen receptor, it demethylates H3K9me1/2 and acts as an activator (Klose and Zhang 2007). Apart from LSD1 and the closely related LSD2, many other HDMs have been described, although their mode of action is completely different because their activity takes places within the shared JmjC jumonji domain; capable of demethylating some trimethyl-lysines, they use Fe(II) and α-ketoglutarate as cofactors and have a high level of substrate specificity (Pedersen and Helin 2010).

Genome-wide studies have confirmed that, like a sort of histone code, diverse combinations of histone modifications result in the different levels of chromatin packaging that, globally or locally, are necessary for (and thus define) the various nuclear functions: transcription, alternative splicing, replication, chromosome condensation, DNA repair, etc. (Kouzarides 2007; Luco et al. 2010; Li et al. 2007a). For instance, trimethylation of lysines 4, 36, or 79 on histone H3 (H3K4me3, H3K36me3, and H3K79me3, respectively), acetylation of its lysines 9 and 14 (H3K9ac and H3K14ac), and monomethylation of lysines 20 and 5 on histones H4 and H2B (H4K20me and H2BK5me), respectively, generally lead to a relaxed, open chromatin state (euchromatin) and gene activation, whereas di- or trimethylation of lysine 9 on histone H3 (H3K9me2 and H3K9me3) as well as trimethylation of its lysine 27 (H3K27me3) lead to a compact, closed chromatin state (heterochromatin) and gene repression (Barski et al. 2007; Li et al. 2007a; Rosenfeld et al. 2009). However, on a small scale, it is well known that histone modifications define the necessary chromatin states for the proper function of all the different sequences involved in transcription. Thus, boundary elements, regions separating heterochromatin and euchromatin that restrict the activity of enhancers, are enriched not only in histone variants like H2A.Z but also in modifications such as H3K9me1 (Barski et al. 2007). Active enhancers contain relatively high levels of H3K4me1 (Hon et al. 2009), transcription start sites (TSSs) are very enriched in H3K4me3, and the entire transcribed regions present a great deal of H3K36me3, among other examples (Barski et al. 2007; Hon et al. 2009; Li et al. 2007a). Together with histone variants and ATP-dependent chromatin-remodeling complexes, the histone code plays a key role in DNA replication and repair, processes that also need special relaxed chromatin conditions.

Replication requires not only access of its enzymatic machinery to DNA but also the removal of parental nucleosomes ahead of the replication fork and the assembly of new ones on the newly synthesized material. In this regard, the HAT MYST2 has been reported to associate with the replication factor MCM2 and the origin recognition complex 1 subunit of the human initiator protein (Burke et al. 2001). Furthermore, a complex containing MYST2 and ING5 (see later in this section) also interacts with the MCM2-7 helicase and seems to be essential for replication in humans (Doyon et al. 2006). On the other hand, one of the first events in DNA double-strand breaks (DBSs) is the phosphorylation of H2A.X in its serine 139 (H2AXS139ph) (Luijsterburg and van Attikum 2011). H2A.X is a histone variant that replaces H2A several megabases upstream and downstream of the break and that, upon phosphorylation, not only leads to massive targeting of repair proteins to the damaged sites (see later in this section) but also recruits HAT complexes such as NuA4, which, through histone acetylation, further enhance access of the DNA repair machinery to damaged chromatin (Luijsterburg and van Attikum 2011).

Although there are many possible combinations of histone modifications, it is important to highlight some peculiarities of their cross talk. Interplay among marks can take place at a single site, within the same histone tail (in cis) or between different tails (in trans). Some modifications cannot coexist at the same residue for steric reasons (Wang et al. 2008b). Some marks promote or block other modifications in cis or trans, such as the ubiquitylation of lysine 120 on histone H2B (H2BK120ub), which promotes trimethylation of H3K4 for eventual gene activation (Kim et al. 2009); the phosphorylation of the serine 10 on H3 (H3S10ph), which results in the acetylation of H3K14 for gene activation (Lo et al. 2000) and, also, in the inhibition of mono- and dimethylation of H3K9 during mitosis (Duan et al. 2008); or the phosphorylation of H3T6 (H3T6ph), which prevents LSD1 from demethylating H3K4 during androgen receptor-dependent gene activation (Metzger et al. 2010).

Histone modifications bring about chromatin remodeling by two main mechanisms.

First of all, marks such as acetylation or phosphorylation neutralize the positive charge of lysines, weakening interactions among nucleosomes and/or between DNA and histones and, thereby, relaxing chromatin (Luger and Richmond 1998; Steger and Workman 1996). For acetylation, this seems to be true, as the large number of possible acetylation sites within histone tails could enable the positive charge of lysines to be neutralized in hyperacetylated regions. Moreover, while hyperacetylation of histones is usually considered a hallmark of open chromatin, deacetylation is a hallmark of closed chromatin, and many acetylation marks are enriched at enhancers and promoters (Wang et al. 2008b). However, for phosphorylation, with fewer available sites within histone tails, this effect is not likely to be of great importance.

Histone modifications can serve as docking sites for recruiting many specific chromatin factors and/or complexes which, eventually, execute the remodeling (Kouzarides 2007). Likewise, regulators of any process related to chromatin, such as transcription factors, can also bind to their target regions through the same mechanism (Sims and Reinberg 2006). Several domains are capable of interacting with the different marks. For instance, PHD fingers and the domains of the Tudor royal family (chromodomains, Tudor, PWWP, and MBT domains) can bind to different methyl-lysines (Champagne and Kutateladze 2009; Kim et al. 2006), bromodomains and, again, PHD fingers interact with acetylated lysines (Mujtaba et al. 2007; Zeng et al. 2010), and, finally, 14-3-3 and BRCT domains can mediate targeting to phosphorylated serines (Macdonald et al. 2005; Stucki et al. 2005). A good example of this type of recruitment is the HP1 recognition of H3K9me2/3 through its single chromodomain; this binding targets the H3K9 HMTs SUV39H1/2 to chromatin (HP1 interacts with it), leading to the methylation of adjacent histones and, eventually, to heterochromatin silencing of sequences like telomeres and centromeres (Bannister et al. 2001; Lachner et al. 2001).

It is important to point out, however, that the same mark can be read by different domains, depending on the cellular context (de la Cruz et al. 2005). For example, H3K4me3 can be recognized not only by the PHD finger of the ING1-5 proteins that, in turn, recruit HATs or HDACs to chromatin (Champagne and Kutateladze 2009) but also by the tandem chromodomains of the ATP-dependent chromatin-remodeling enzyme CHD1 (see next section) (Sims et al. 2005) or by the Tudor domains of the HDM JMJD2A (Huang et al. 2006). To return to the subject of the cross talk between modifications, the binding of an effector to a particular mark can be blocked by other modifications or, even, DNA methylation, as occurs with HP1 binding to H3K9me2/3 during mitosis, due to H3S10ph (Fischle et al. 2005), or with JHDM1A binding to H3K9me3 due to methylcytosines (Bartke et al. 2010). Finally, different marks and, again, DNA methylation can also cooperate to recruit a factor more efficiently. This is the case with H3K4me3 and PHF8, whose interaction is enhanced when H3K9 and H3K14 are also acetylated (Vermeulen et al. 2010), or H3K9me3 and UHRF1, which, in a nucleosomal context, interact much better when CpG dinucleotides are also methylated (Bartke et al. 2010).

1.2.2.2 In Cancer Cells

Cancer cells present aberrant covalent histone modification profiles. It is well established that the disruption of particular histone marks can dysregulate normal expression patterns of oncogenes and tumor suppressor genes, as well as affect genome integrity and/or chromosome segregation, eventually leading to cancer. Typically, cancer cells show a global reduction of H4K16ac and H4K20me3 (Fraga et al. 2005) and also display different alterations in the enzymes in charge of writing, erasing, and reading the histone code (Ellis et al. 2009; Hatziapostolou and Iliopoulos 2011).

Disruption of histone acetylation has been reported in hematological and solid tumors. Chromosomal translocations, leading to aberrant fusion proteins involving HATs like CBP, P300, MYST3, or MYST4 and that eventually dysregulate the expression of multiple genes, have been described in different types of leukemia (Ayton and Cleary 2001; Champagne et al. 1999; Chan et al. 2007; Ida et al. 1997; Liang et al. 1998; Panagopoulos et al. 2001). Moreover, it is known that binding of adenoviral oncoproteins E1A and SV40 T to CBP and P300 leads to transformation due to a decrease in the global levels of H3K18ac, and the subsequent targeting of these HATs to genes promoting cell growth and division (Ferrari et al. 2008; Horwitz et al. 2008). Finally, missense mutations and monoallelic loss of EP300 and KAT5, the genes encoding P300 and TIP60, respectively, have also been found in a range of solid and hematological tumors (Gayther et al. 2000; Gorrini et al. 2007).

In considering the role of histone deacetylation in cancer, it is worth mentioning that some malignant chromosomal translocations give rise to fusion proteins that, although they do not harbor their own deacetylation activity, can associate with HDACs (and other chromatin remodelers) with high affinity and target them to particular subsets of genes that end up repressed. This is the case of the PML-RARA and PLZF-RARA fusion proteins in APL (Minucci and Pelicci 2006), and AML1-ETO and CBFβ-MYH11 in AML (Bhalla 2005; Wang et al. 2007a). In B-cell-derived non-Hodgkin lymphomas, the transcriptional repressor BCL6 is upregulated and, similarly, provokes the silencing of its target genes by recruiting an exaggerated amount of HDACs (Bhalla 2005; Wang et al. 2007a). On the other hand, HDAC1 and HDAC2 are overexpressed in multiple tumor types; HDAC3, HDAC7, and HDAC8 in colon cancer; HDAC4 in prostate and breast cancer; and HDAC6 in AML and breast cancer (Ellis et al. 2009). However, the role of HDAC2 in colon cancer is not as clear because its gene was also found to be mutated in a subset of microsatellite unstable colorectal cell lines and primary tumors (Ropero et al. 2006). Reduced levels of HDACs have also been observed in the disease: HDAC1 and HDAC4 in colon cancer; HDAC5 in AML, lung, and, again, colon cancer; and HDAC10 in lung cancer (Ellis et al. 2009; Osada et al. 2004). With respect to sirtuins, overexpression of SIRT1, SIRT3, and SIRT7 occurs in a wide range of tumors (Saunders and Verdin 2007). Interestingly, when inhibition of SIRT1 was seen to partially reactivate tumor suppressors like CDH1, SFRP1, SFRP2, and MLH1, their promoters remained heavily methylated, thus indicating a prevailing role for deacetylation, against DNA methylation, in gene silencing (Pruitt et al. 2006). Finally, SIRT1 and SIRT4 are downregulated in colon cancer and AML, respectively (Saunders and Verdin 2007).

Abnormal activity of the histone methylation enzymatic systems is also common in tumorigenesis. In the case of the methylation of H3K4, partial tandem duplication and more than 50 different fusions of the MLL gene are the cause of most infant leukemias, as well as of 5–10 % of AML and lymphoid leukemias (Krivtsov and Armstrong 2007). In some cases, the malignancy occurs due to an increase in activating marks like H3K4me2 and bulk histone acetylation at leukemia-promoting genes, but in other cases, these critical genes become activated by the anomalous recruitment of DOT1L, the H3K79 HMT (Chi et al. 2010; Dorrance et al. 2008). Another H3K4 HMT, SMYD3, is typically upregulated in colorectal and hepatocellular carcinoma cell lines (Hamamoto et al. 2004). Meanwhile, upregulation of the H3K27 HMT EZH2 provokes the silencing of key tumor suppressor genes in prostate, breast, colon, skin, and lung cancer (Bracken and Helin 2009). The increase of endothelial EZH2 promotes angiogenesis, at least in ovarian cancer, by silencing the VASH1 gene (Lu et al. 2010). However, the mutation of EZH2 can also yield malignant phenotypes, as was observed in some B-cell lymphomas (Lu et al. 2010; Morin et al. 2010). On the other hand, while high levels of the H3K9 HMTs EHMT2 and SETDB1 have been described in liver cancer and melanoma, respectively (Kondo et al. 2007; Ceol et al. 2011), NSD1, an HMT for H3K36 and H4K20, is involved in leukemogenic translocation and lost in gliomas and neuroblastomas by CpG island promoter hypermethylation (Wang et al. 2007c). The heterozygous mutation/loss of heterozygosity of the corresponding gene causes Sotos syndrome, a combination of symptoms that includes a higher risk of tumorigenesis (Berdasco et al. 2009). Highly similar to NSD1, the HMT NSD2 is involved in aggressive t(4;14)-associated multiple myelomas (MMs) (Marango et al. 2008). Finally, inactivation of PRDM2, an H3K9 HMT originally identified as a pRB-binding protein, either by mutations or promoter hypermethylation, is also found in many tumors (Gibbons 2005). The alteration of PRDM5, PRDM16, and the arginine methyltransferase PRMT4 is linked to cancer, too (Lahortiga et al. 2004; Majumder et al. 2006).

The role of HDMs in tumorigenesis is no less important. LSD1 is upregulated in many tumors (Kahl et al. 2006; Lim et al. 2010), and the disruption of many different JmjC domain-containing HDM genes has also been observed: while KDM5A (encoding JARID1A) is translocated in myeloid leukemia, overexpression of KDM5B (encoding JARID1B) was found in advanced stages of breast and prostate cancers, where it facilitates the G1/S transition in the cell cycle, attenuates its mitotic checkpoint, or inactivates tumor suppressors like BRCA1 and CAV1 (Scibetta et al. 2007). In a recent study, inactivation of KDM5C (encoding JARID1C) by truncating mutations appeared in ≈3 % of renal cell carcinoma (RCC) samples (Dalgliesh et al. 2010). Moreover, JHDM1B, another HDM capable of repressing the tumor suppressor locus CDKN2A-CDKN2B by erasing both H3K36me1/2 and H3K4me3 marks, is known to be upregulated in T-cell lymphomas (Tzatsos et al. 2009). Mutations of the gene encoding the H3K27me2/3 HDM UTX become inactivated in a subset of MMs, esophageal squamous cell carcinomas, and RCC, and the expression of the related JMJD3, H3K27me2/3 HDM, is downregulated in many cancers (Agger et al. 2009; van Haaften et al. 2009). Finally, KDM4C, which encodes the H3K9me2/3 HDM JHDM3C/GASC1, is frequently overexpressed by gene amplification in esophageal squamous cell carcinoma, lung sarcomatoid carcinoma, desmoplastic medulloblastoma, and breast cancer (Italiano et al. 2006; Liu et al. 2009b).

Further research is needed to determine the importance of histone phosphorylation in cancer. For instance, it is known that JAK2, the tyrosine kinase for H3Y41, is frequently activated by chromosomal translocations or point mutations in different hematological malignancies (Dawson et al. 2009). Also, the upregulation of aurora kinase B, responsible for the H3S10ph mark, has been reported in several solid tumors (Dar et al. 2010; Hirota et al. 2005), and its collaboration with aurora kinase A is essential for the progression of Myc-driven B-cell lymphomas (den Hollander et al. 2010).

Malfunction of histone modification readers is also a significant event in tumorigenesis. Thus, the members of the ING family of H3K4me3 readers are components of different transcriptional regulation complexes controlling many cellular cancer-related processes: while ING1 and ING2 recruit HDAC-Sin3A repressive complexes, ING3, ING4, and ING5 recruit HATs to induce gene activation (Chi et al. 2010). Dysregulation of their expression or mutations affecting their binding capacity to the H3K4me3 mark has been described in many tumors (Chi et al. 2010).

1.2.3 Nucleosome Positioning

1.2.3.1 In Normal Cells

As previously stated, DNA methylation and covalent histone modifications determine, globally and locally, the levels of chromatin packaging, thereby controlling access to DNA of the enzymatic machineries in charge of the different genetic functions. However, in close collaboration with these, histone variants and ATP-dependent chromatin-remodeling factors/complexes are the cellular mechanisms that, ultimately, are capable of generating the different chromatin compaction patterns by physically executing the appropriate nucleosome displacement and/or removal. For example, it is well established that, in transcriptionally active genes, nucleosomes are positioned at fixed locations around promoters while they are more randomly distributed in the interior of the genes (Mavrich et al. 2008). Moreover, the 5′ and 3′ ends of these genes must also adopt nucleosome-free regions (NFRs) in order to expose binding sites for the different necessary transcription factors (Jiang and Pugh 2009; Schones et al. 2008). The loss of the nucleosome located directly upstream of the TSSs is strongly correlated with gene activation, whereas its presence is normally associated with gene repression (Portela and Esteller 2010).

The best studied histone variants are expressed outside S phase and incorporated into chromatin in a replication-independent manner (Santenard and Torres-Padilla 2009). They differ from canonical histones in their amino acid sequence and structure (Li et al. 2007a). Histone H2A may have three main variants: H2A.Z, H2A.X, and macroH2A. Preferentially enriched at promoters of active genes or genes poised for activation, H2A.Z contributes to their transcription by destabilizing the nucleosomes (Jin and Felsenfeld 2007). It is also known to protect those genes against possible repression by DNA methylation (Zilberman et al. 2008). As explained previously, the phosphorylation of H2A.X at its serine 139 (H2AXS139ph) occurs soon after a DNA double-strand break and promotes the recruitment of DNA repair complexes (Luijsterburg and van Attikum 2011). Finally, macroH2A is specifically placed throughout the inactive X chromosome in female mammals (Chadwick and Willard 2002). Histone H3 has two possible variants: CENP-A, which is essential for the highly condensed chromatin of centromeres and so for chromosome segregation, and H3.3, which is enriched at promoters of active genes like H2A.Z (Sarma and Reinberg 2005).

It is important to point out that not only H2A.X can undergo covalent modifications, as the acetylation of H2A.Z is commonly observed in active genes, whereas its ubiquitylation is associated with facultative heterochromatin (Svotelis et al. 2009); furthermore, H3.3 normally presents di- and trimethylation of lysine 4, as well as acetylation of lysines 9, 14, 18, and 23, marks which reflect transcriptional competence (Sarma and Reinberg 2005).

Several large multiprotein complexes have been found to move, destabilize, eject, or restructure nucleosomes using the energy of ATP hydrolysis (Clapier and Cairns 2009; Hargreaves and Crabtree 2011). They all have in common the presence of a DNA-dependent ATPase of the Snf2 family as their catalytic subunit. More precisely, these ATP-dependent chromatin-remodeling complexes always include an ATPase coming from up to seven different subfamilies of the Snf2 family: Snf2, ISWI, INO80, SWR1, CHD1, Mi-2, and CHD7 (Flaus et al. 2006). The ATPases of all these subfamilies are characterized by their particular combination of domains. Thus, apart from the SNF2_N and HELICc domains, which together form the catalytic center of all them, the ATPases of the Snf2 subfamily also have a bromodomain (Wilson and Roberts 2011), the ATPases of the ISWI subfamily present both a SANT and a SLIDE domain (Corona and Tamkun 2004), and the ones belonging to the CHD1 subfamily have two N-terminal chromodomains and a C-terminal DNA-binding domain (Hall and Georgel 2007; Marfella and Imbalzano 2007); the ATPases of the Mi-2 and CHD7 subfamilies also possess a couple of N-terminal chromodomains, but while the first ones also have two PHD domains, the others display a SANT domain and one or two BRK domains (Hall and Georgel 2007; Marfella and Imbalzano 2007). Finally, the very similar ATPases of the INO80 and SWR1 subfamilies do not have extra domains, but they are characterized by a typical spacer fragment within the amino acid sequence of their catalytic centers (Morrison and Shen 2009). The role of all these domains consists not only of stabilizing the binding of the entire complexes to the covalently modified histones and/or nucleosomal DNA typical of their target regions throughout the genome but also of mediating the necessary interactions between the different subunits of the complexes themselves, which include histone modification enzymes (such as TIP60 in the TRRAP/TIP60 complex or HDAC1 and HDAC2 in the NuRD complex) and, logically, histone modification readers (like ING3 or MBD3, in the same complexes, respectively), among other proteins with a wide variety of functions (Clapier and Cairns 2009; Bao and Shen 2007; Hargreaves and Crabtree 2011).

Thus, in humans, BAF and PBAF are the ATP-dependent chromatin-remodeling complexes with ATPases of the Snf2 subfamily. The ATPase of the BAF complex can be either hBRM or BRG1, whereas the ATPase of the PBAF complex can only be BRG1 (Bao and Shen 2007). These complexes are key regulators of gene expression and alternative splicing, controlling important processes such as cell cycle progression, organ development, and immune responses, among others (Wilson and Roberts 2011). NURF, hCHRAC, hACF/WCRF, or WICH, some of the complexes with ATPases of the ISWI subfamily, have been reported to repress transcription (rather than activate it), promote chromatin assembly during mitosis, and participate in DNA replication (Bao and Shen 2007; Corona and Tamkun 2004; Hargreaves and Crabtree 2011). Complexes with ATPases of the Mi-2 subfamily are mainly involved in transcription repression and DNA repair (Hargreaves and Crabtree 2011), and much less well-known, putative complexes harboring ATPases of the CHD7 subfamily may contribute to transcription, both its activation and repression, depending on the exact context (Lutz et al. 2006; Nishiyama et al. 2009; Rodriguez-Paredes et al. 2009; Schnetz et al. 2009; Shur et al. 2006; Takada et al. 2007). However, CHD1, ATPase of the subfamily with the same name, seems to work alone, without any complex, and its functions may be the promotion of transcription and splicing, as well as the deposition of the histone variant H3.3, as it does in Drosophila (Konev et al. 2007; Sims et al. 2007). Finally, complexes with ATPases of the INO80 subfamily have important roles in transcription, DNA replication and repair, chromosome segregation, and telomere regulation (Morrison and Shen 2009). Most importantly, the SRCAP complex, with the ATPase of the SWR1 subfamily of the same name, is essential for the deposition of the histone variant H2A.Z (Wong et al. 2007).

1.2.3.2 In Cancer Cells

In comparison with DNA methylation and covalent histone modifications, very little is known about the influence of aberrations in histone variants and/or ATP-dependent chromatin-remodeling factors on cancer.

First of all, although the histone variant H2A.Z is overexpressed in several cancer types where it seems to favor cell cycle progression, its loss has also been associated with the spread of repressive chromatin domains and de novo promoter hypermethylation of tumor suppressors (Svotelis et al. 2009); moreover, the H3 variant CENP-A becomes upregulated in colorectal cancer and may be responsible for aneuploidy (Tomonaga et al. 2003). On the other hand, bi-allelic loss of SNF5, encoding one of the core subunits of the BAF and PBAF complexes, is common in most human malignant rhabdoid tumors (MRTs) and leads to cancer through inactivation of the p21 and p16INK4A pathways (Chai et al. 2005). Nonetheless, it remains unclear whether its tumor suppressor function requires BAF-/PBAF-mediated chromatin remodeling. Loss of BRG1 and hBRM, the two mutually exclusive ATPases of these complexes, has been reported in up to 15–20 % of primary non-small-cell lung cancers, and bi-allelic loss of SMARCA4 (encoding BRG1) takes place in prostate, lung, breast, and pancreatic cancer cell lines (Reisman et al. 2003; Medina and Sanchez-Cespedes 2008). Intriguingly, a recent study reveals that BRG1 also has an oncogenic role based on the destabilization of the p53 protein (Naidu et al. 2009). The NuRD complex was first connected to cancer through its metastasis-associated subunits MTA1, MTA2, and MTA3, but it has recently been found that, in leukemia, it also favors the targeting of the Polycomb repressive complex 2 (PRC2) and DNMT3A to the oncogenic transcription factor PML-RARA target promoters, leading to their permanent silencing (Morey et al. 2008; Wang et al. 2007b). It is well known that, in general terms, aberrant gene silencing through CpG island promoter hypermethylation normally involves the presence of a nucleosome that blocks the TSS (Lin et al. 2007). Finally, it is important to highlight the role in cancer of CHD5, which is required for the normal expression of CDKN2A (encoding p16INK4A and p19ARF) and is usually deleted in epithelial, neural, and hematopoietic malignancies (Bagchi et al. 2007).

1.3 The Epigenetic Role of miRNAs and Its Disruption in Cancer

miRNAs are the best studied type of ncRNAs. Generally speaking, these molecules, with an approximate length of 22 nucleotides, finely downregulate the expression of probably more than 60 % of protein-coding genes (Friedman et al. 2009b). However, in particular cases, miRNAs have also been found to increase the expression of their target genes (Vasudevan et al. 2007). Thus, miRNAs are involved in the majority of cellular functions, including proliferation, differentiation, and apoptosis (Davalos and Esteller 2010).

With respect to their biogenesis and mode of action, miRNAs are initially transcribed by RNA polymerase II as a pri-miRNA, a long, capped, polyadenylated transcript, which is further processed into a hairpin RNA precursor, the pre-miRNA, by the double-stranded RNA-specific ribonuclease Drosha, in collaboration with the microprocessor protein DGCR8 (Cai et al. 2004; Lee et al. 2004; Han et al. 2006). This pre-miRNA, a 70–100-nucleotide-long molecule, is then transported to the cytoplasm by Exportin-5 (XPO5) and, once there, cleaved by a large complex composed of Dicer, a ribonuclease III, and TRBP, a double-stranded RNA-binding protein (Hutvagner et al. 2001; Yi et al. 2003). The result is a duplex, 18–24 nucleotides long, which is then loaded into the RNA-induced silencing complex (RISC). In this complex, which includes proteins of the Argonaute family, one strand of the duplex remains stably associated, becoming the mature miRNA, and promotes RISC targeting normally (but not always) to the 3′ untranslated region (UTR) of the corresponding complementary mRNA (Davalos and Esteller 2010). If the miRNA and its target mRNA sequence match perfectly, the consequence is the cleavage and, hence, destruction of the latter. If they do not match perfectly, then mRNA translation is blocked (Davalos and Esteller 2010).

It is known that while each mRNA can be targeted by more than one miRNA (Vatolin et al. 2006), a single miRNA can regulate many different transcripts. In this regard, miR-15a and miR-16 have been shown to regulate expression of ≈14 % of the human genome in a leukemic cell line (Calin et al. 2008).

It is widely accepted that an aberrant miRNome, with mainly downregulated but also upregulated miRNAs, is a hallmark of cancer (Croce 2009). In principle, these ncRNAs can regulate any gene and so in each tissue some of them act like real proto-oncogenes and some like real tumor suppressors. Therefore, it is not surprising that different chromosomal abnormalities, mutations, and/or polymorphisms (SNPs) affecting miRNA sequences have been described in many tumors, especially as miRNA genes are usually located in cancer-associated regions or fragile sites (Calin et al. 2004). Some examples include the deletion of the cluster miR-15a / miR-16-1 (miRNAs are often found in clusters) in chronic B-cell lymphocytic leukemia (Calin et al. 2002), the amplification of miR-26a in glioblastoma (Huse et al. 2009), and the translocation of the miR-17-92 cluster in T-cell acute lymphoblastic leukemia (Mavrakis et al. 2010). Moreover, the dysregulation of transcription factors like p53 or Myc, which are well established in cancer, promotes the misexpression of different miRNAs (in this case, the miR-34 family and the miR-17-92 cluster, respectively), thereby contributing to tumorigenesis (Davalos and Esteller 2010). Intriguingly, apart from particular examples, the pathological activation of Myc seems to globally repress miRNA expression and should be further investigated (Chang et al. 2008). Finally, defects in the miRNA processing machinery (see above) can also account for the altered miRNA levels related to cancer. For instance, frameshift mutations in the TARBP2 gene, which encodes TARBP, lead to a decrease in normal miRNA levels, favoring tumorigenesis in sporadic and hereditary carcinomas with microsatellite instability (Melo et al. 2009). Likewise, mutations in DICER1, or the downregulation of the corresponding protein, have been shown to produce the same effect in some tumors (Bahubeshi et al. 2010; Merritt et al. 2008; Rio Frio et al. 2011).

However, most importantly, tumorigenic miRNA levels can also arise from the disruption of epigenetic mechanisms. Considering that approximately half of all miRNA genes are associated with CpG islands, it is not surprising that many of them have already been found to be silenced in cancer through CpG island promoter hypermethylation (Weber et al. 2007a). A good example is miR-127, which regulates the proto-oncogene BCL6 and is silenced in several tumors (Saito et al. 2006). Others include miR-9-1 in breast cancer, and miR-34b and miR-34c in colon cancer (Lehmann et al. 2008; Toyota et al. 2008). Additionally, this kind of aberrant silencing of miR-9, miR-34b, miR-34c, and miR-148a, which modulate MYC, E2F3, CDK6, and TGIF2 expression, is known to give rise to metastatic capacities (Lujambio et al. 2008). Conversely, miRNAs controlling real tumor suppressors are activated in cancer by DNA hypomethylation, like miR-21 in epithelial ovarian cancer (Iorio et al. 2007). As an example of the tissue specificity of epigenetic regulation, while let-7a-3 is transcribed in lung adenocarcinoma due to DNA hypomethylation, it has also been found to be silenced by promoter hypermethylation in ovarian cancer (Brueckner et al. 2007; Lu et al. 2007). However, DNA methylation cannot be the only epigenetic mechanism capable of leading to miRNA misexpression in cancer, given its strong relationship with covalent histone modifications and ATP-dependent chromatin-remodeling complexes. Thus, it is known that, upon inhibition of histone deacetylation, miRNA levels are profoundly altered in SKBR3 breast carcinoma cells (Scott et al. 2006). Moreover, it has been demonstrated in AML that the AML1/ETO fusion oncoprotein silences transcription of miR-223 by recruiting HDAC activities (Fazi et al. 2007). For further examples, see Table 1.3.

Table 1.3 Epigenetically regulated miRNAs altered in human cancer
Full size table

Finally, it is worth noting that epi-miRNAs, miRNAs which regulate the expression of members of the epigenetic machineries, can also be dysregulated in cancer. Good examples are miR-29a, miR-29b, and miR-29c, regulating DNMT3a and DNMT3b, in AML and lung cancer (Fabbri et al. 2007; Garzon et al. 2009) and miR-101 , which regulates EZH2, in prostate and bladder cancer (Friedman et al. 2009; Varambally et al. 2008). Other examples are listed in Table 1.4.

Table 1.4 Epi-miRNAs altered in human cancer
Full size table

1.4 Epigenetic Control of Stem Cell Behavior

1.4.1 Epigenetic Status in ES Cells

The definition of stemness includes two main capacities: indefinite self-renewal and pluripotency. The best experimental model for studying stemness is probably that of ES cells; derived from the inner cell mass of blastocyst-stage embryos, these cells are able to differentiate into any possible cell type, either fetal or adult, both in vivo and in vitro (Jaenisch and Young 2008). In general, and in contrast to what happens in differentiated cells, the chromatin of ES cells is in an open state; in fact, its constitutive heterochromatin (centromeres, telomeres, etc.) appears rather dispersed and there is a highly dynamic exchange of both histones and nonhistone proteins (including HP1) (Orkin and Hochedlinger 2011). These special chromatin features reflect the unusual nature of these cells; they are devoted not only to preserving their identity but also to maintaining their genomes in a flexible state capable of giving rise to any cell type. Although it is well established that stemness mainly requires a key set of transcription factors (Oct4, Nanog, Sox2, c-Myc, among others), it is also obvious that epigenetic mechanisms contribute decisively to providing the special chromatin characteristics of ES cells.

In the human genome, promoters can be divided into two main categories according to their CpG content: those with high CpG density (the vast majority), normally corresponding to housekeeping and tightly regulated developmental genes, and those with low CpG density, which are mainly associated with tissue-specific genes. The first group displays low DNA methylation levels and, in principle, should be transcriptionally active (Meissner et al. 2008; Mikkelsen et al. 2007), while the other tends to be hypermethylated and, therefore, silenced (Meissner et al. 2008; Weber et al. 2007b). Nevertheless, to maintain stemness, the subgroup of developmental genes (the ones able to trigger the different differentiation programs) cannot be active. This problem is solved by additional covalent histone modifications. Thus, while housekeeping and tissue-specific genes present typical active and inactive histone marks, respectively, developmental genes present both (Bernstein et al. 2006; Mikkelsen et al. 2007). The so-called bivalent domains are characterized by the simultaneous presence of H3K4me3 and H3K27me3 marks, are frequently located on the binding sites of the transcription factors in charge of the decision stemness/differentiation (including Oct4, Nanog, and Sox2), and keep these genes repressed but poised to become fully active or silenced, depending on the triggering of particular differentiation programs (Bernstein et al. 2006; Mikkelsen et al. 2007). While the H3K4me3 mark is mediated through the hSET1A and/or diverse MLL complexes (Christophersen and Helin 2010), the H3K27me3 mark is placed by EZH1 or, mainly, EZH2, core members of the Polycomb repressive complexes 1 and 2 (PRC1 and PRC2), respectively (Fig. 1.2) (Margueron et al. 2008; Margueron and Reinberg 2011). These latter complexes are recruited to their target genes either by proteins like MTF2 and JARID2, a catalytically inactive HDM, or through short ncRNAs previously transcribed from them (Kanhere et al. 2010; Landeira et al. 2010; Walker et al. 2010). Their action sets not only the H3K27me3 mark but also the H2AK119ub (via PRC1), which blocks RNA polymerase II activity and is essential for maintaining ES cell identity (Endoh et al. 2008; Stock et al. 2007). The H2A variant H2A.Z has also been found enriched at some of these target genes (Creyghton et al. 2008).

Fig. 1.2
figure 00012

Basic features of bivalent domains in ES and differentiated cells for a typical developmental gene. In ES cells, developmental genes are repressed but poised for full activation or repression upon differentiation. These genes present both activating (H3K4me3) and repressive (H3K27me3) covalent histone modifications around their TSSs. If differentiation involves a definitive activation of the gene, the HDMs UTX and JMJD3 erase the H3K27me3 mark; if it involves permanent repression, the HDM JARID1A removes the H3K4me3 mark. The figure also shows the enzymes and complexes in charge of placing H3K4me3 (hSET1A and different complexes with MLL proteins) and H3K27me3 (PRC1 and, mainly, PRC2 complexes)

Full size image

As described in the DNA methylation section, upon differentiation, only a few genes become silenced by CpG island promoter hypermethylation, including germ line-specific and pluripotency genes (such as POU5F1 and NANOG, which encode Oct4 and Nanog, respectively), different tissue-specific genes depending on each cell type, and genes involved in genomic imprinting and X-chromosome inactivation (Mohn et al. 2008; Straussman et al. 2009; Suzuki and Bird 2008). In fact, it also explained that most of the changes in DNA methylation were found, on the one hand, at CpG island shores and, on the other hand, in a non-CpG context (Doi et al. 2009; Laurent et al. 2010; Lister et al. 2009; Meissner et al. 2008). Genes with bivalent domains become less frequent as differentiation progresses (Fisher and Fisher 2011), usually ending up fully activated or repressed (Hawkins et al. 2010). In the former situation, the H3K27me3 mark is removed by the action of the HDMs UTX and JMJD3 (Christophersen and Helin 2010); in the latter, the H3K4me3 mark disappears by means of JARID1A (Fig. 1.2) (Pasini et al. 2008).

At this point, it is worth mentioning that other enzymes in charge of different covalent histone modifications have also been linked to stemness, for instance, the HMTs CARM1, EHMT1, EHMT2, and SETDB1 and the HDMs LSD1, JARID1B, JHDM2A, and JHDM3C (Lessard and Crabtree 2010; Orkin and Hochedlinger 2011).

Some ATP-dependent chromatin-remodeling complexes, on the other hand, play an important role in ES cells. Containing the ATPase BRG1 and subunits such as BAF155 and BAF60a, but never hBRM, BAF170, or BAF60c (like in differentiated cells), the ES cell-specific BAF complex, esBAF, is essential for self-renewal and pluripotency (Fisher and Fisher 2011; Ho et al. 2009). It is present in about the 25 % of promoters in ES cells, where it interacts with key transcription factors like Oct4, Sox2, Nanog, STAT3, and SMAD1 (Ho et al. 2009; Kidder et al. 2009); although it promotes the activation of many genes, its main functions related to stemness seem to be the repression of genes involved in differentiation, as well as the prevention of the overexpression of pluripotency-specific genes (Saladi and de la Serna 2010). Also essential for stemness, the TIP60-p400 complex is recruited to more than half of the promoters in ES cells by either the H3K4me3 mark (directly) or Nanog (indirectly); despite its normally activating HAT activity on histone H4, it mainly represses developmental genes in the context of other repressive marks (Fazzio et al. 2008). On the other hand, the ATPase CHD1 is key to maintaining the typical open chromatin of ES cells. Without it, an abnormal accumulation of heterochromatic foci makes correct differentiation impossible (Gaspar-Maia et al. 2009). It is tempting to speculate that CHD1 may accomplish its mission through the deposition of the histone variant H3.3, which is already found enriched at bivalent domains, as it does in Drosophila (Goldberg et al. 2010). Other ATP-dependent chromatin-remodeling factors involved in stemness are BPTF and MBD3, which are subunits of the NURF and NuRD complexes, respectively, and are mainly linked to the maintenance of pluripotency (Landry et al. 2008; Zhu et al. 2009). Finally, it has also been reported that CHD7 modulates ES cell-specific gene expression by targeting enhancers of active genes (Schnetz et al. 2010).

In mouse ES cells, the disruption of any enzyme involved in their biogenesis impairs stemness (Pauli et al. 2011). The ES cell-specific miRNAs, miR-290-295, were found to be able to complement the proliferative defects that appeared in Dgcr8 knockout cells (Wang et al. 2008a). It has also been discovered that Oct4, Nanog, Sox2, or Tcf3 bind to the promoters of these and many other miRNAs, such as let-7 and miR-145 (Marson et al. 2008). However, the latter two remain normally repressed in ES cells by means of the PRC2 complex, and upon differentiation, both are transcribed and downregulate genes encoding key pluripotency factors like lin-28 and SALL4 (let-7) or Oct4, Sox2, and Klf4 (miR-145) (Melton et al. 2010; Xu et al. 2009). Consistently, repression of let-7 in somatic cells contributes to their reprogramming (Melton et al. 2010). Related to the latter, it seems that the overexpression of the miR-302-367 cluster alone could be enough as to induce pluripotency in fibroblasts (Anokye-Danso et al. 2011; Lin et al. 2011).

1.4.2 The Cancer Stem Cell Theory

Tumorigenesis is considered to be a complex multistep process whereby cells, after accumulating epigenetic and/or genetic alterations, display uncontrolled proliferation and aberrant patterns of differentiation due to the activation of oncogenes and/or silencing of tumor suppressors. However, the issue of cancer initiation remains under intense discussion. In the previous century, the clonal genetic model, which was based merely on subsequent genetic mutations in differentiated cells, was widely supported, but currently, the possibility that epigenetic dysregulation may precede genetic alterations and that stem cells may be the origin of some (if not all) tumors is an increasingly popular idea.

The cancer stem cell model postulates that epigenetic changes occurring in normal stem or progenitor cells may be the earliest events in tumorigenesis (Feinberg et al. 2006). Such an idea is consistent with several observations. First, it is known that certain epigenetic alterations indeed occur very early in various types of cancer and that, additionally, normal tissues have altered progenitor cells in cancer patients (Cui et al. 2003; Matsubayashi et al. 2003; Peters et al. 2007). Second, tumors normally include different cell populations with diverse tumorigenic and metastatic capacities (Al-Hajj et al. 2003). Third, it is well established, as we have already seen, that epigenetic mechanisms are essential for maintaining stemness (Surani et al. 2007; Wang et al. 2008b). Finally, human ES cells with cancer cell characteristics have already been found (Werbowetski-Ogilvie et al. 2009). Therefore, epigenetic disruption in a stem/progenitor cell of a particular tissue may create a different population with enhanced stemness, a high-risk cell population capable of undergoing transformation more easily after gaining subsequent genetic and epigenetic changes (Jones and Baylin 2007). Once generated, this resulting small cancer stem cell population would be able not only to self-renew but also to produce the phenotypically diverse cancer cells responsible for the usual massive tumor proliferation. Although this model remains a hypothesis, in the last decade, numerous studies have already described cancer-initiating cells in diverse solid tumors (Al-Hajj et al. 2003; Collins et al. 2005a; Dalerba et al. 2007; Li et al. 2007b; O’Brien et al. 2007; Prince et al. 2007; Ricci-Vitiani et al. 2007; Singh et al. 2004; Suetsugu et al. 2006). Most interestingly from a clinical point of view, others have reported that, unlike their derivatives, cancer stem cells are resistant to conventional chemo- and radiotherapy and could be responsible for eventual tumor repopulation and treatment failure (Bao et al. 2006; Matsui et al. 2008).

1.5 Concluding Remarks

Epigenetics is going through its golden age. Modern high-throughput technologies based on the combination of either bisulfite treatment of methylated DNA or chromatin immunoprecipitation with next-generation sequencing (methylC-seq and ChIP-seq, respectively) allow us to determine complete individual DNA methylomes and the exact position along the genome of any covalent histone modification or member of the epigenetic machineries. The information obtained by these kinds of epigenomic approach is now making the difference not only in basic epigenetic research but also to our understanding of stemness and tumorigenesis. The objective is to fully establish the role of epigenetic mechanisms in the origin and progression of the disease in order to be able to offer a more personalized treatment. However, there is still much work to do in many very exciting areas. For instance, we also need to expand our knowledge about DNA demethylation and the degree of participation of ncRNAs in the positioning of covalent histone modifications. In fact, there are still many marks of the histone code to identify and many combinations to understand in the different cellular contexts. The composition and function of some ATP-dependent chromatin-remodeling complexes also remain unknown, and moreover, connections between the epigenetic machineries and miRNAs still need to be much better understood. Finally, further research is needed to shed light on the epigenetic processes underlying stemness.