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 Introduction

Although DNA hypomethylation was the first epigenetic alteration to be described in human cancers, its effect on gene expression programs and tumor biology has remained enigmatic. Initial examination of cancer genomes identified most losses of DNA methylation in repeated elements [29]. This is not surprising, since these DNA elements are highly abundant and comprise most of the CpG sites that are normally methylated in healthy somatic tissues. A crucial question was whether DNA hypomethylation also affected protein-encoding genes, leading to their ­aberrant expression in tumor cells. It appeared, however, that genome hypomethylation in tumors is not generally associated with the ectopic activation of a multitude of genes [5]. A plausible explanation for this is that most tissue-specific genes use other regulatory mechanisms, including histone modifications, and that DNA methylation, if present, serves merely as secondary layer of repression. Losses of DNA methylation within such genes would therefore not be sufficient to trigger transcriptional activation.

Later work, aiming at isolating genes that code for tumor-specific antigens, led to the identification of a particular group of genes, which are normally expressed exclusively in germline cells but become aberrantly activated in a wide variety of tumors [86]. Given this expression profile, these genes were termed “cancer-­germline” (CG) genes. Interestingly, CG genes were found to rely primarily on DNA methylation for repression in normal somatic tissues, and their activation in tumors was shown to be a direct consequence of genome hypomethylation [22]. These observations highlighted an unexpected link between epigenetic alterations in tumors and cancer immunity. They also provided clear examples of genes that owe their tissue-specific expression to DNA methylation. Moreover, CG genes are being exploited to try to uncover the molecular mechanisms underlying genome hypomethylation in tumors, as this epigenetic process remains largely unexplained.

2 Characterization of CG Genes

Human tumors express specific antigens, as evidenced by the existence in the blood of cancer patients of cytolytic T lymphocytes (CTL) that recognize antigens present on their tumor cells but not on normal cells [10]. Using a gene library transfection approach and a CTL clone isolated from a melanoma patient, Boon and colleagues identified the first human tumor antigen-encoding gene [85]. The gene was named melanoma antigen 1 or MAGE-1 (later renamed MAGEA1). MAGEA1 expression was not found in normal tissues except for testis, but was instead detected in a significant fraction of melanoma samples, as well as in various other tumor types [20, 23]. The same genetic approach led to the identification of other melanoma antigen genes, namely BAGE, GAGE, and MAGEA3, a gene closely related to MAGEA1 [9, 34, 84]. For these genes too, expression among normal tissues was restricted to testis, and activation in tumors was detected among various cancer types. Additional tumor antigen genes were subsequently identified, using an alternative cloning approach, called SEREX (serological analysis of recombinant tumor cDNA expression libraries), and based on the presence of high titers of antitumor IgGs in the blood of tumor-bearing patients [73]. Again, several of the identified genes, including SSX2 and NY-ESO-1, had their normal expression restricted to testis and were activated in a percentage of different tumor types. Later studies indicated that the normal expression of most isolated genes was confined to the germ cells in both testis and fetal ovary [44, 52, 82].

Together, these findings led to the important notion that specific antigens in tumors arise from the aberrant activation of genes that are normally transcribed exclusively in the germline. From an immunological point of view, this dual expression pattern is understandable. Unlike most somatic cells, germ cells lack MHC class I molecules, which are required to present antigenic peptides at the cell surface [37]. Activation of germline-specific genes in tumor cells therefore leads to the expression of truly tumor-specific antigens, which can be recognized as nonself by the immune system.

Further studies using cDNA subtraction procedures or database mining have permitted the identification of additional genes expressed in germ cells and cancer but not in normal somatic tissues [56, 60, 63, 75]. Some genes identified in this way were subsequently shown to encode tumor-specific antigens recognized by CTLs [86]. Altogether about 50 human genes or gene families were identified, which displayed specific expression in the germline and activation in a significant proportion of cancers [2]. These genes appear to exert a variety of cellular functions, but on the basis of their common expression pattern they were grouped under the term cancer-germline (CG) genes. CG genes are dispersed on several chromosomes, with a marked preference for the X chromosome. In human cancers, CG genes are expressed more frequently in specific tumor types, like for instance lung cancer, head and neck cancer, bladder cancer, and melanoma [76]. Other tumor types like colon cancer, renal cancer, and leukemia only rarely show activation of CG genes. An important feature of CG genes is their frequent co-activation in tumors [74]. It was observed indeed that positive tumors often express several CG genes. Clearly, the widespread and concerted expression of CG genes in tumors indicates that their activation in cancer results from a global gene activation process, rather than stochastic individual events.

3 DNA Demethylation in the Activation of CG Genes in Tumors

The marked tendency of CG genes to become co-expressed in tumors suggested that these genes share, at least in part, a common mechanism of transcriptional activation. Initial studies were performed with the MAGEA1 gene in order to identify essential promoter elements and corresponding transcription factors that may contribute to the cell-type-specific expression of the gene. Surprisingly, however, transfection experiments revealed that all cells, including those that do not express MAGEA1, contain transcription factors capable of inducing significant MAGEA1 promoter activity [24]. Transfection experiments with other CG gene promoter constructs led to similar results [17, 89]. This implied that nonexpressing cells have a repression mechanism, probably operating at the chromatin level that protects CG gene promoters against spurious activation.

The initial observation by Weber and colleagues that MAGEA1 could be induced in nonexpressing melanoma cell lines following treatment with the DNA methylation inhibitor 5-aza-2′-deoxycytidine provided a first hint that DNA ­methylation may contribute to the transcriptional regulation of this gene [91]. This was confirmed by studies showing that the promoter of MAGEA1 is invariably methylated in all normal somatic tissues and instead unmethylated in germ cells [26]. Likewise, activation of the MAGEA1 gene in tumors was strictly correlated with demethylation of its promoter [26]. Further studies showed that DNA methylation was similarly involved in the regulation of other CG genes [17, 26, 52, 56, 89]. Altogether, these observations indicated that CG genes rely on DNA methylation for repression in somatic tissues, and that aberrant activation of these genes in tumors results from demethylation of their promoter.

Interestingly, demethylation and activation of CG genes in tumors was found to correlate with global genome hypomethylation [14, 25, 45]. This association was further confirmed by a study on microdissected tumor samples, revealing that intratumor heterogeneity of CG gene expression also correlates with global genome hypomethylation levels [96]. These observations provided therefore the first clear evidence that the process of genome-wide demethylation, common to many cancers, not only affects repeated sequences but also single copy genes, and can lead to aberrant gene activation. The frequent co-activation of CG genes in tumors likely reflects the global process of DNA demethylation, which can simultaneously affect many loci across the cancer genome.

4 DNA Methylation in the Regulation of Germline Genes

Considering the potent effect of DNA methylation on transcriptional repression, it was originally proposed that this DNA modification might serve as a general mechanism to control the programmed expression of tissue-specific genes [39, 72]. Evidence, however, indicates that most tissue-specific genes rely on mechanisms other than DNA methylation for repression in nonexpressing cells [8, 88]. This may be ascribed to the distribution of CpG sequences, where cytosine methylation can occur. Vertebrate genomes show a general depletion of CpG dinucleotides, which was attributed to the high mutability of methylated cytosines, and hence the progressive disappearance of this sequence during evolution [7]. Discrete genomic regions however, which appear generally free of CpG methylation, maintained a high density of CpG sites. These so-called CpG islands often overlap gene promoters [19]. Many tissue-specific genes contain a methylation-free CpG island within their promoter and can therefore not rely on DNA methylation for repression in nonexpressing tissues. On the other hand, genes with few CpG sites within their promoter are only little affected by DNA methylation, and often show an inconstant relationship between promoter methylation and transcriptional silencing [12]. It was therefore proposed that DNA methylation in vertebrates is solely involved in the control of retrotransposable elements, monoallelically expressed imprinted genes, and X chromosome inactivation, the only cases where consistent methylation of CpG-rich regions appeared to exist [101].

This view was challenged by the discovery of CG genes, which were found to be characterized by the presence of a high density of CpG sites within their promoter [26]. Yet, unlike classical CpG islands, CpG-rich promoters of CG genes are methylated in all normal somatic tissues. CG gene promoters appear therefore favorably disposed to DNA methylation-mediated regulation. Consistently, transfection experiments with in vitro methylated CG gene constructs indicated that DNA methylation was sufficient to repress transcription, even in cells that express the corresponding endogenous CG gene, and therefore obviously contain appropriate transcriptional activators [17, 26, 27, 78, 89]. This and the above-mentioned observation that unmethylated CG gene promoters are transcriptionally active in nonexpressing cells provided strong evidence that DNA methylation is an essential component of the repression of this group of germline-specific genes in somatic cells.

More recently, genome-wide studies were conducted in order to identify the distribution of differentially methylated CpG sites across the genome of distinct types of human cells [77, 93]. These studies revealed the existence of novel sets of genes with a CpG-rich promoter that was densely methylated in somatic tissues (in addition to the previously characterized CG genes). Remarkably, most of these genes were specifically demethylated and expressed in testis. It appears therefore that DNA methylation has a particular role in the regulation of germline-specific genes.

Why would DNA methylation be particularly suitable for the regulation of genes with specific expression in germline cells rather than in other cell types? A plausible explanation may be that methylation-dependent germline genes have the advantage of being little exposed to the evolutionary loss of methylated CpGs, because they are unmethylated precisely in the cells that transmit their genome to the offspring. As a result, such genes maintain a high density of CpG sites within their promoter and remain therefore fully responsive to DNA methylation.

5 Mechanisms Leading to Hypomethylation of CG Genes in Cancer

CG genes have served as model sequences to investigate the distribution and dynamics of methylation losses in tumor genomes. Detailed analysis of the MAGEA1 locus revealed preferential hypomethylation of a restricted region surrounding the transcription start site of the gene in expressing tumor cells, suggesting that hypomethylated CpG sites are unevenly distributed across cancer genomes [27]. Consistently, recent genome-wide DNA methylation studies confirmed that DNA hypomethylation in tumors adopts mosaic patterns, with defined hypomethylated domains (between one kilobase and several megabases in size) surrounded by normally methylated regions [66, 71, 92]. These observations indicate that certain genomic regions, including CG promoters, are particularly susceptible to DNA hypomethylation in tumors.

The possibility that MAGEA1-expressing tumor cells possess a DNA demethylation activity targeted towards the 5′-region of the gene was investigated [27, 58]. Thus, a large genomic fragment comprising the MAGEA1 gene was methylated in vitro and then stably transfected into several human tumor cell lines, where the endogenous MAGEA1 gene is hypomethylated and active. The newly integrated MAGEA1 transgenes did not undergo demethylation, indicating that the process that once led to demethylation of the endogenous MAGEA1 gene was not preserved in these cells. Remarkably, when unmethylated MAGEA1 constructs were introduced into such cells, de novo methylation of the transgenes occurred except in a region overlapping the MAGEA1 promoter [27]. This mechanism of protection against de novo DNA methylation was lost when mutations that impair the MAGEA1 promoter activity were introduced into the transgene, or when the transgene was transfected into tumor cells that induce only little MAGEA1 promoter activity. Altogether, these data suggest that site-specific hypomethylation of MAGEA1 in tumors results from a past event of transient DNA demethylation and is maintained locally by the presence of potent transcriptional activators that prevent remethylation.

In vivo studies, evaluating global genome methylation levels in colon and breast cancers, demonstrated that DNA hypomethylation is present in the early stages of the disease, and does not progress towards later stages, adding support the transient nature of the DNA demethylation process [30, 41]. Other studies, however, reported a higher prevalence of genome hypomethylation and an increased frequency of CG gene activation in more advanced tumor stages [53, 100]. This was interpreted as an indication that DNA demethylation might instead be a continuous process leading to progressive methylation losses with tumor development. Other interpretations for the increased hypomethylation in advanced tumor genomes, which implicate a transient DNA demethylation process, are however possible: (1) transient demethylation would initially produce a mixed population of precancerous cells with varying levels of DNA hypomethylation, and cells with the most hypomethylated genome would later be selected to contribute to the more advanced stages of the disease; or (2) the transient demethylation process could occur at varying time points during tumor progression and would therefore be more likely to have already occurred in late stage tumor samples [22]. Additional support for a transient DNA demethylation process comes from the observation that tumor cell lines with a hypomethylated genome do not show further CpG methylation losses during culturing [32, 55, 94]. Of note, many tumor cells display instead de novo methylation activities [3, 43].

Considering the suggested dynamics of DNA demethylation in tumors, it is reasonable to propose that hypomethylation of CG genes in tumors is mediated by two groups of factors: those that contribute to the transient DNA demethylation process and those that are required to protect the CG gene promoter region against subsequent remethylation.

5.1 Process of DNA Demethylation

Factors contributing to the DNA demethylation process during cancer development remain unknown. The apparent transient nature of this process suggests that activation of such demethylation-inducing factors might occur in association with one (or several) of the multiple steps through which precancerous cells are ­progressing before acquiring full malignancy. Interestingly, a recent study evaluating genome methylation levels in an isogenic series of human mammary epithelial cell cultures transitioning from normal to malignantly transformed revealed that most losses of DNA methylation occurred at the stage of acquisition of indefinite lifespan [67]. Another study reported that genome hypomethylation and CG gene activation is more prevalent in tumors displaying the alternative telomere (ALT) maintenance phenotype rather than telomerase activation, the two possible mechanisms by which cancer cells stabilize their telomeres and acquire immortality [83]. These observations establish therefore a possible link between DNA demethylation and cellular immortalization. Underlying molecular mechanisms remain, however, to be identified.

Theoretically, DNA demethylation in tumor cells could possibly occur through two distinct processes commonly referred to as active demethylation and passive demethylation [16]. Active demethylation would involve the activation of demethylating enzymes, which can remove methylation marks from the DNA in a replication-independent manner. Enzymes contributing to active DNA demethylation in animal cells are beginning to be characterized [16], but their potential involvement in cancer genome demethylation has not yet been reported. Passive demethylation on the other hand, would rely on the inhibition of DNA methyltransferases, which normally preserve the DNA methylation marks through the successive replication cycles. Three DNA methyltransferases exist in mammals: DNMT1, DNMT3A, and DNMT3B [6]. DNMT1 is primarily involved in DNA methylation maintenance, as it appears to be specialized in copying preexisting methylation sites onto the newly synthesized strand during replication. DNMT3A and DNMT3B instead have de novo DNA methylation activity and are responsible for the establishment of new DNA methylation marks in the developing embryo. For CG genes in particular, studies based on targeted depletion of the distinct DNMTs indicate that DNMT1 is the principal enzyme for methylation maintenance [42, 57]. It is therefore likely that passive DNA demethylation of CG genes in tumors would necessarily involve factors that decrease the amount or impair proper functioning of DNMT1. In certain tumor cells, however, combined depletion of DNMT1 and DNMT3 enzymes was required to obtain efficient demethylation and activation of CG genes [42, 95]. This indicates that de novo methyltransferases can be targeted to these genes, where they might restore lost methylation sites, and underscores the importance of acquiring mechanisms of protection against remethylation for long-term activation.

5.2 Factors that Protect Against Remethylation

Studies with the MAGEA1 promoter suggest that protection of the promoter against DNA remethylation is dependent on the level of transcriptional activation [27]. It is therefore likely that maintenance of CG gene promoter hypomethylation in tumor cells relies on the presence of appropriate transcription factors, as well as on the activation of such factors by upstream signaling pathways.

Several DNA-binding factors have been identified, which appear to induce ­activation of CG gene promoters. Transcriptional activation of several genes of the MAGEA family has been shown to depend on the binding of ETS transcription factors within their promoter [21, 24]. Interestingly, ETS-binding sequences in MAGEA promoters contain a CpG site, and it was shown that methylation of this site inhibits binding of the corresponding factor [25]. In the promoter of MAGEA1, two ETS-binding sites were shown to be essential to maintain hypomethylation of the promoter in expressing tumor cells, as evidenced by remethylation of transfected MAGEA1 constructs containing mutations within these two essential promoter elements [27]. The ETS family of transcription factors comprises about 30 members in humans, which all bind a similar DNA motif with a central GGAA/T sequence [68]. The precise member(s) involved in the regulation of MAGEA genes remain(s) to be characterized.

SP1 is another transcription factor, which was shown to contribute to the activation of several MAGEA genes, as well as the CTAG1 gene (also termed NY-ESO-1) [24, 46]. The ubiquitously expressed SP1 factor acts as a transcriptional activator and recognizes a consensus DNA sequence (GC box element), which includes a CpG site [80]. SP1-binding elements are therefore often present in CG-rich promoter sequences. Binding of SP1 to the CTAG1 gene was shown to occur only in cells where the promoter is unmethylated [46]. Interestingly, SP1-binding elements were previously shown to be involved in preserving the methylation-free status of classical CpG-island promoters [13, 62]. It is therefore likely that, once bound to the demethylated promoter of CG genes, SP1 proteins contribute to protect the region against remethylation.

BORIS (also known as CTCFL) is a testis-specific paralog of the ubiquitously expressed DNA-binding protein CTCF, which is involved in various aspects of epigenetic regulation, including gene imprinting and X chromosome inactivation [59]. Both proteins share a highly similar central DNA-binding domain, and recognize therefore overlapping DNA sequences, but contain divergent amino- and carboxy-terminal domains. The gene-encoding BORIS belongs to the CG group of genes, as its expression is regulated by DNA methylation and becomes activated in a wide variety of tumors [38, 49, 87, 95]. Remarkably, it has been demonstrated that in expressing tumors cells, BORIS is targeted to the promoters of other CG genes, namely MAGEA1 and CTAG1, where its recruitment coincides with loss of CTCF binding [40, 87]. BORIS exerts transcriptional activation of CG genes, possibly in cooperation with SP1 transcription factors [46, 87]. In one study, forced overexpression of BORIS led to demethylation (albeit only partially) and activation of various CG genes in normal human fibroblasts, suggesting that BORIS activation in tumors might represent a primary triggering event for the epigenetic de-repression of other CG genes [87]. However, similar experiments from other groups did not confirm CG gene demethylation and activation resulting from BORIS overexpression [49, 97]. Moreover, it was found that many tumors display activation of various CG genes in the absence of BORIS expression. It is therefore unlikely that BORIS is a necessary factor for the derepression of other CG genes in tumors. Its presence in certain tumor cells may, however, facilitate maintenance of the hypomethylated and active state of CG gene promoters.

Many more transcription factors involved in CG gene regulation remain to be identified, and it is likely that each particular CG gene is controlled by a distinct combination of transcription factors. Tissue-specific differences in the content of transcription factors probably account for the fact that, while CG genes tend to be co-activated in hypomethylated tumors, some of them nevertheless show preferential activation in specific tumor types [36, 56].

Cell signaling through tyrosine kinase receptors appears to represent an additional level of control of CG gene regulation. A study in mast cell lines reported that signaling through KIT, an oncogenic receptor hyper-activated in several types of cancers, increases transcription of MAGE genes [99]. Other studies revealed that signaling through FGFR2, an FGF receptor often down-regulated in thyroid and pituitary cancers, exerts a negative effect on MAGEA3 and MAGEA6 transcription [51, 102]. It is therefore possible that particular dysregulations in cancers, such as those affecting cell signaling pathways, increase the activity of transcription factors that target CG genes, and thereby facilitate long-term activation of these genes in hypomethylated tumor cells. This may partially explain the observation that experimental DNA demethylation, by the use of DNMT inhibitors, often induces CG gene activation more efficiently in tumor cells than in normal cells [47].

5.3 Histone Modifications

Active CG gene promoters in tumors usually display a hypomethylated region that comprises one to several kilobases [27]. It is therefore likely that the protective influence of transcription factors against DNA remethylation extends beyond their narrow-binding site. Consistently, impaired binding of ETS transcription factors to MAGEA1 transgenes, as caused by mutations in their recognition sites, resulted in de novo methylation of CpG sites within the entire promoter region, not just those located nearby the mutated ETS-binding sites [27]. This regional, rather than site-specific effect, might be related to the presence of modifications on the chromatin, such as histone modifications, which after being initiated by specific transcription factors often propagate themselves over larger domains [31]. Histone modifications can indeed influence DNA methylation states [15]. Repressive histone marks, such as methylation of lysine 9 and 27 of histone H3 (H3K9 and H3K27), favor local DNA methylation, whereas active marks, such as histone acetylation or methylation of lysine 4 of histone H3 (H3K4), appear to exclude the DNA methylation machinery. Studies from several groups have shown that demethylation and activation of CG genes in tumor cells is always associated with gains in histone acetylation and H3K4 methylation [42, 70]. The repressed state of human CG genes instead has been associated to a certain extent with the presence of H3K27 and H3K9 methylation marks [42, 70]. The exact relationship between histone modifications changes and DNA demethylation in CG gene promoters remains unclear. A crucial question is whether the varying histone modifications in CG gene promoters are a cause or a consequence of DNA methylation alterations. Studies using inhibitors of histone-modifying enzymes showed that these were on their own unable to induce significant demethylation and activation of CG genes. Only in combination with inhibitors of DNA methylation, did they significantly modulate the level of activation of CG genes [35, 54, 70]. These observations support the notion that DNA methylation exerts a dominant role in the epigenetic repression of CG genes. But it remains possible that histone modifications assume the responsibility of maintaining the active status of the promoter following its demethylation.

5.4 Multiple Factors Determining CG Gene Activation in Tumors

Considering the above, it appears that activation of a particular CG gene in a tumor cell will depend on several factors: (1) the extent of CpG methylation losses resulting from the transient DNA demethylation process; (2) the level of de novo DNA methylation activities in the cell, which might induce remethylation of the promoter; (3) the presence of transcriptional activators and histone-modifying enzymes capable of counteracting remethylation activities. The likelihood that a CG gene becomes activated in a tumor cell probably depends on a complex balance between these different factors (Fig. 7.1).

Fig. 7.1
figure 1

Proposed model of demethylation and activation of CG genes during tumor development. The activation of CG genes in tumors depends on several factors: the extent of the transient DNA demethylation process, occurring at some step of tumor development; the level of counteracting de novo methylation activities in the cell; and the presence of transcriptional activators that protect the CG gene promoter against remethylation, for instance by increasing (+) or decreasing (−) distinct histone marks locally. Filled circles represent methylated CpG, empty circles unmethylated cytosines

Full size image

6 Oncogenic Function of CG Genes

Activation of CG genes in tumor cells raises the possibility that their proteins might have oncogenic activities. The biological function of most of these genes, which encode very diverse proteins, remains however poorly understood. One extreme possibility is that the main contribution of DNA hypomethylation to tumor progression resides in its repercussions on genomic instability [33], and that the accompanying activation of CG genes is merely a side effect with no impact on malignancy (other than inducing the expression of tumor antigens). Another possibility has been proposed, in which the concerted expression of CG genes in cancer would correspond to the activation of a gametogenic program, thereby bestowing tumor cells with germ cell properties, including the capacity to self-renew (a feature of spermatogonial stem cells) and increased motility (a feature of sperm cells) [79]. Activation of CG genes in tumors is however only partial, making it very unlikely that all genes necessary for inducing a gametogenic program become expressed at the same time. Nevertheless, it remains possible that some CG genes contribute to tumor progression. Several MAGE proteins were found to inhibit p53 transactivation function, thereby exerting antiapoptotic properties [28, 64, 98]. GAGE proteins were also shown to render cells resistant to apoptosis [18]. Other studies reported that MAGEA11 serves as a co-stimulator for the androgen receptor and might therefore contribute to the development of prostate tumors that have become independent of the presence of androgen for their growth [4, 48]. Moreover, it was noted that several CG genes, including BORIS, BRDT, and ATAD2, encode nuclear proteins that have a potential impact on chromatin structures and might therefore be involved in the epigenetic alterations commonly affecting cancer genomes [90]. Altogether, these observations support the notion that the activation of several CG genes in tumors, resulting from DNA demethylation, might be associated with the acquisition of oncogenic properties.

Surprisingly, however, two independent studies indicate that MAGEA4 displays instead tumor-suppressor functions. In one study, MAGEA4 was shown to interact with gankyrin and to inhibit anchorage-independent growth in vitro and tumor formation in mice [65]. In the other study, MAGEA4 was found to promote tumor cell death and to increase their sensitivity to apoptotic stimuli [69]. Clearly, more studies will be required before we can evaluate the full spectrum of consequences of CG gene activation in tumors.

7 DNA Hypomethylation in Cancer: An Immunological Paradox

There is now compelling evidence that the immune system is able to identify and destroy tumor cells [81]. This immune surveillance of cancer is believed to provide a barrier to cancer development, even though progressing tumors eventually escape this obstacle by activating a variety of immune evasion strategies. Evidence for the existence of such surveillance of cancer by the immune system is provided for instance by the observation that solid tumors are often infiltrated by lymphocytes. Not surprisingly, several of these tumor-infiltrating lymphocytes were shown to be directed against antigens encoded by CG genes [50]. This suggests therefore that DNA hypomethylation and the consequent activation of CG genes has, at least at some stage of oncogenesis, a detrimental effect on tumor development. Yet, DNA hypomethylation is observed in most tumors, suggesting that it must otherwise have a strong tumor-promoting effect that outweighs this negative immunogenic effect.

8 Epigenetically Assisted Cancer Immunotherapy

Clinical trials of therapeutic vaccination of cancer patients using antigens encoded by CG genes are underway. Noticeable clinical responses were observed, albeit in only a fraction of the treated patients [11]. An interesting possibility to increase vaccination efficiencies would be the use of epigenetic drugs, such as the DNA methylation inhibitor decitabine, which should increase the number of expressed CG genes in the tumors, thereby rendering them more visible to the immune system. Importantly, decitabine is expected to induce reactivation of epigenetically silenced tumor-suppressor genes as well, and hence to reduce the growth rate of the tumors at the same time. Clinical trials combining decitabine and vaccination against antigens encoded by CG gene have been initiated [1].

There are, however, several points concerning the efficiency and safety of such approaches, which remain to be addressed. The first point concerns the specificity of decitabine-induced expression of CG genes in tumor cells rather than normal cells. Although studies have found that tumor cells are more sensitive to decitabine [47], it is obvious that the drug also induces CG genes in normal cell cultures, including fibroblasts and blood lymphocytes [25, 56, 61]. It will therefore be crucial to monitor decitabine/vaccine-treated patients for potential autoimmune reactions directed against their healthy tissues. Another concern relates to the duration of CG gene expression following decitabine treatment. Several studies have shown that CG gene expression in tumor cells was only transient following exposure to decitabine [26, 91]. This may be related to the absence of appropriate transcription factors, and hence lack of protection of the promoters against remethylation. The duration of CG gene expression in tumor cells may be critical to allow complete rejection by the immune cells. In this particular immune context, tumor cells that lose CG gene expression might be strongly selected. Prolonged decitabine treatment or combination with another epigenetic drug favoring protection of CG promoters against remethylation (e.g., drugs affecting histone marks) might be a solution to the problem. Finally, as genome hypomethylation is obviously associated with tumor development, there is a concern that decitabine treatment may generate strongly hypomethylated tumor cells with increased malignancy [33]. This is particularly problematic if it is confirmed that CG genes themselves exert oncogenic functions.

Clearly, a better understanding of the mechanisms of activation and of the biological functions of CG genes should help to resolve these questions, and may help to design the most efficient and safest ways to epigenetically augment tumor immunogenicity, thereby rendering cancer cells more vulnerable to vaccination.