Throwing the Cancer Switch

Histone modification

Covalent modification of specific residues within amino terminal tails of histones alters chromatin structure and function; the unique combination of modifications has been described as the histone code. Both PcG and TrxG proteins affect histone methylation, for example, but these opposing classes of proteins methylate distinct residues, resulting in different biological outcomes. PcG proteins establish histone modifications that repress transcription, whereas TrxG proteins establish histone modifications that activate transcription (box 1).

There are two core PcG complexes, each of which is made up of multiple subunits (fig 2a; table 1). Polycomb repressive complex 2 (PRC2) establishes the histone code, and polycomb repressive complex 1 (PRC1) interprets this code ¹⁹. The mammalian PRC2 complex contains enhancer of zeste homolog 2 (EZH2), embryonic ectoderm development (EED), suppressor of zeste homolog 12 (SUZ12), and retinoblastoma binding protein 4 (RBBP4) or RBBP7 ²⁰. EZH2 contains a SET domain and is a histone methyltransferase (HMT) that trimethylates histone H3 at lysine 27, thereby establishing H3K27me3 characteristic of inactive chromatin ²¹. Although PRC2 had also been reported to establish the H3K9me3 repressive mark ²², this has remained controversial. EED and SUZ12 do not possess HMT activity themselves, but these subunits are required for the HMT activity of EZH2 ²³. In D. melanogaster, PRC2 binds polycomb repressive elements (PREs) throughout the genome. PREs in mammals have been more difficult to identify, although PREs within the Krysler and Hox loci were recently characterized functionally ^{24, 25}. H3K27me3, but not H3K9me3, is lost in PcG mutants, highlighting the importance of H3K27me3 in maintaining cellular memory.

The repressive histone marks established by PRC2 are recognized by PRC1, which stabilizes the inactive state so that transcriptional memory is maintained. There are multiple forms of mammalian PRC1 complexes, which may contain chromobox homolog 2 (CBX2), CBX4 or CBX8, polyhomeotic homolog 1 (PHC1), PHC2 or PHC3, BMI1 polycomb ring finger oncogene (BMI1), and ring finger protein 1 or 2 (RING1 or RNF2) ²⁶ (fig 2a; table 1). CBX proteins contain a CHROMO domain, which recognizes and binds the H3K27me3 mark established by PRC2. BMI1, RING1, and RNF2 have ring-finger domains, a motif crucial for nuclear localization as well as for the oncogenic activity of BMI1 (discussed later) ²⁷.

A working model for how PcG complexes modulate chromatin is that transcription factors and their associated machinery dictate loci destined for silencing, and PRC2 tags this target by methylating lysine 27 of histone H3. This H3K27me3 mark is recognized and bound by the chromodomain of CBX2, CBX4, or CBX8 within PRC1, which mediates transcriptional silencing by binding methylated histones in the vicinity and ubiquitinating H2A ^{28, 29}. Although in vitro assays have suggested that PRC complexes mediate silencing by physically blocking the binding of transcriptional machinery at promoters ³⁰, PRC complexes are more likely to function by impeding RNA polymerase II (pol II) initiation or elongation. Indeed, several groups have shown that pol II and basal transcription factors are present at PcG-silenced loci ^31,32,33. While the hierarchical recruitment model in which PRC2 and PRC1 bind sequentially is attractive, recent evidence indicates that PRC2 and PRC1 bind some loci simultaneously, and that not all PRC2-bound loci are bound by PRC1 ³⁴. Furthermore, PRC1 can bind chromatin in the absence of PRC2 ^{35, 36}.

Like PcGs, TrxG proteins are also components of multi-subunit complexes, however, their composition is not as well characterized. TrxG proteins can, however, be broadly classified into two categories: the histone modifiers ³⁷ and the nucleosome remodelers ³⁸ (fig 2; table 1). The histone modifiers include HMTs that establish histone marks favoring transcriptional activation. TrxG complexes contain subunits that, like EZH2, are SET domain-containing HMTs, such as the founding member of the D. melanogaster TrxG family trithorax (TRX), and absent, small, or homeotic 1 (ASH1). However, in contrast to EZH2, which methylates histone H3 at lysine 27 to establish a repressive histone mark, TrxG HMTs methylate histone H3 at lysine 4 to establish active chromatin marks ³⁹ (fig 2, fig 3a). The human TRX homolog myeloid-lymphoid or mixed-lineage leukemia (MLL) occupies the promoters of HOX genes, facilitating H3K4 methylation and transcriptional activation ⁴⁰. Thus, both PcG and TrxG complexes regulate chromatin by directly methylating histones.

In addition to methylation, PcG and TrxG complexes either directly or indirectly recruit proteins that facilitate other covalent histone modifications, such as acetylation (fig 3b). Histone acetylation status is regulated by histone acetyltransferases (HATs) and histone deacetylases (HDACs). Acetylation of histones near promoters is associated with transcriptional activation; therefore, HATs and HDACs favor transcriptional activation and repression, respectively. PcG and TrxG complexes recruit these enzymes (fig 3b). For example, the HDACs sirtuin 1 (SIRT1) and HDAC2 have been purified within PRC complexes ²⁰. By contrast, the HAT MYST1, which acetylates H4K16, a mark of actively transcribed genes, has been purified within MLL-containing complexes ⁴¹. Although ASH1 complexes are not fully characterized, ASH1 interacts with the HAT CREB binding protein (CBP, also known as CREBBP), which is a transcriptional co-activator ⁴². Importantly, acetylation prevents the establishment of H3K27me3. Thus, TrxG proteins facilitate transcriptional activation by establishing activating histone methylation and acetylation marks, while simultaneously preventing PcG complexes from depositing inactivating histone methylation marks. These complexes link histone methylation and histone acetylation. PcG and TrxG complexes are also able to recruit histone demethylases (HDMs), indicating that the reversal of PcG- and TrxG-mediated histone methylation regulates chromatin dynamics and transcription (fig 3c). In addition, PcG complexes can recruit DNA methyltransferases (DNMTs) (fig 3d), indicating that repressive methylation of histones is intimately linked to methylation of DNA—a mechanism by which tumor suppressor genes can be inactivated in human cancer (discussed below). PcG-TrxG not only regulates transcriptional on and off states, it is an adaptable system that allows for graded responses between these extremes. In this dynamic fashion, the bivalent chromatin state of pluripotent embryonic stem cells in which both activating and repressive histone marks are present ⁴³ can be fine-tuned as differentiation proceeds, as was shown in cells of the ectodermal lineage ⁴⁴. Thus, although PcG and TrxG proteins can directly methylate histones, they are also able to recruit additional players that greatly expand their repertoire of covalent histone and DNA modifying activities.

Nucleosome remodeling

The nucleosome repositioning class of TrxG complexes contain ATP-dependent chromatin remodeling that – along with their associated cofactors – modulate nucleosome structure ^{45, 46} (fig 2b, table 1). One of the first hints that TrxG proteins modulate chromatin dynamics by affecting nucleosomes was the realization that D. melanogaster BRM was homologous to Saccharomyces cerevisiae (S. cerevisiae) SWI2-SNF2 ⁴⁷. The observation that S. cerevisiae SWI2-SNF2 mutant phenotypes are repressed by mutations in nuclear histones firmly established a functional connection between SWI2-SNF2 and nucleosomes ^{48, 49}. Subsequent work revealed that SWI2-SNF2 complexes consist of multiple subunits and modulate nucleosome position and or composition.

ATP-dependent chromatin remodeling enzymes contain ATPase subunits belonging to the SNF2 superfamily; the sequence of the ATPase domain further subdivides this superfamily ⁵⁰. Mammalian SNF2 families are also characterized by domains homologous to D. melanogaster proteins: SWI-SNF subunits have a BROMO domain related to BRM, ISWI subunits have a SANT domain related to ISWI, and the CHROMO domain helicase DNA binding (CHD) family is characterized by CHROMO domains related to CHD1, Mi-2 and Kismet (table 1). Mammalian ATP-dependent nucleosome remodeling proteins enhance gene expression by facilitating transcriptional elongation by (pol II)⁵¹. Although nucleosome remodeling proteins were first associated with transcriptional activation, these complexes can also function as transcriptional repressors (fig 3e) ⁵². Models proposed for the mechanism by which nucleosome remodeling proteins affect gene expression include chromatin looping, nucleosome sliding, nucleosome eviction, and exchange of variant histones; these events facilitate binding of transcription factors ⁵³ and basal transcription machinery ⁵⁴. The histone modifying and the nucleosome remodeling classes of TrxG proteins are not mutually exclusive; TrxG proteins that are HMTs can recruit nucleosome remodeling TrxG proteins. For example, MLL has been purified from a complex containing WD repeat domain 5 (WDR5), a PHD-containing protein required for H3K4me3 that recruits the nucleosome remodeling factor (NURF) complex, indicating that that H3K4me3-mediated transcriptional activation is linked with nucleosome remodeling ⁵⁵. Thus, the interplay between PcG and TrxG complexes integrate histone modifications, DNA methylation and nucleosome remodeling, affecting chromatin structure and gene expression in a complex and dynamic fashion. The varied composition of the complexes may reflect dynamic changes during development or within tissue-specific contexts, adding further diversity to the PcG-TrxG-regulated cellular memory system.

Cellular senescence

BMI1 was the first PcG protein implicated in cancer. It was initially identified as an oncogene that cooperates with MYC in lymphomagenesis ^{56, 57}, and subsequent studies found it robustly expressed in human cancers (table 1). BMI1 facilitates tumorigenesis by mediating escape from cellular senescence, a state of cell cycle withdrawal that provides a potent tumor suppressive mechanism in vivo ⁵⁹. The mechanism whereby BMI1 promotes evasion of senescence involves transcriptional silencing of the cyclin-dependent kinase inhibitor 2A (CDKN2A, which encodes both p16^INK4A and p14^ARF) and CDKN2B (which encodes p15^INK4B) loci, mapping to human chromosome 9p21 ^{20, 60}. Silencing of CDKN2A and CDKN2B simultaneously shuts down expression of three tumor suppressors ^{61, 62} (figure 4a). INK4A (also known as p16) and INK4B (also known as p15) are CKIs that prevent CDK4-dependent phosphorylation of RB, thereby preventing E2F-mediated cell cycle progression. BMI1 coordinately represses CDKN2A and CDKN2B loci via CDC6 loading onto a cis-acting replication origin several thousand nucleotides upstream of the transcriptional start site ⁶³. This is the first example of a replication-coupled transcriptional repression mechanism in mammals. Repression of CDKN2A and CDKN2B loci is a recurring theme in tumorigenesis: mouse models with deletions of this locus are tumor prone, and this locus is frequently inactivated in human cancers by deletion, mutation, or DNA hypermethylation ⁶⁴.

Senescence limits proliferation, therefore proliferating cells must inhibit senescence to maintain proliferative potential; PcG-mediated transcriptional repression of CDKN2A plays a pivotal role in senescence inhibition (fig 4b) ⁶⁵. Renewal capacity of both neural and hematopoietic stem cells is dramatically compromised in Bmi1^−/− mice ^{66, 67}, concomitant with homeotic phenotypes such as the 2^nd cervical vertebrae having the identity of the 3^rd, reminiscent of loss of cellular memory observed in D. melanogaster PcG mutants (fig. 1b) ⁶⁸. These phenotypes are partially rescued by depletion of p16^INK4A, underscoring the critical role of BMI1-mediated repression of this senescence inducer. Renewal capacity is also maintained in MEFs via PcG-mediated repression. Activated oncogenes such as H-RAS or BRAF—signals that cause senescence of proliferating MEFs—result in reversal of PcG-mediated repression of p16^{INK4A 69, 70}. Oncogene-induced senescence coincides with loss of PcG protein (EZH2, SUZ12, BMI1 and CBX8) binding at the CDKN2A and CDKN2B loci, as well as transcriptional repression of EZH2, and coincident transcriptional activation and recruitment of the H3K27me3 HDM, lysine (K)-specific demethylase 6 (KDM6B, also known as jumonji domain containing 3, JMJD3). The repressive H3K27me3 marks are replaced with activating ones (H3K4me3), pol II is recruited and p16^INK4A expression is induced, culminating in senescence. The presence of AP-1 consensus binding sites in the JMJD3 promoter suggests that this transcription factor plays a role in BRAF-mediated transcriptional activation of JMJD3 ⁷⁰. This indicates that the ability of normal primary cells to resist oncogene-mediated transformation depends on a switch between PcG-mediated repression and TrxG-mediated activation.

PcG-mediated repression is also reversible in cancer cells, as TrxG proteins can override PcG-mediated repression, leading to reactivation of tumor suppressors. SWI-SNF chromatin remodeling complexes contain an ATPase subunit (either BRM or BRG1) as well as the non-catalytic subunit (SNF5) that is common to various SWI-SNF-like complexes (fig 2) ^{71, 72}. Loss of SNF5, BRM, and BRG1 has each been associated with human cancer (table 1). Malignant rhabdoid tumors (MRTs)—a rare but extremely aggressive form of childhood cancer—are caused by bi-allelic deletion or truncating mutations of SMARCB1 (which encodes SNF5).

Reintroduction of SNF5 into human MRT cells (which are deficient for this non-catalytic subunit of SWI-SNF) can alleviate PcG-mediated repression and reverse the tumorigenic phenotype (fig 4c) ⁷³. Expression of SNF5 in MRT cells induces cell cycle arrest and senescence ^{74, 75}, increasing expression of p16^INK4A, increasing hypophosphorylated Rb, and compromising expression of E2F target genes ^{76, 77}. SNF5 recruits the catalytic BRG1 subunit to the CDKN2A and CDKN2B promoters, facilitating binding of pol II. SNF5-mediated recruitment of BRG1 coincides with removal of PRC1 and PRC2 repressive complexes, as well as dissociation of DNMT3B and hence decreased DNA methylation ⁷³. MLL is simultaneously recruited, repressive histone marks are replaced with active ones, and senescence is induced. SNF5–mediated senescence requires the ATPase activity of BRG1. This demonstrates that contrary to previous reports obtained in vitro ⁷⁸, SWI-SNF activity can override PcG-mediated silencing, exemplifying the dynamic interplay between PcG and TrxG complexes.

The above studies in MRT cells showed that SNF5 induces senescence by activating p16^INK4A-RB-mediated pathways. Although Smarcb1 (which encodes SNF5)^+/− mice develop spontaneous MRTs of the brain ^79–81, tumorigenesis was not increased in either Rb1 or Cdkn2a^INK4A deficient backgrounds ^{82, 83}. Similarly, although Smarca4 (which encodes BRG1)^+/− mice develop mammary gland carcinomas ⁸⁴, tumorigenesis was not increased in an Rb1^+/− background ⁸⁵. Although these findings implicate SWI-SNF proteins in tumor suppression in vivo, they do not support the findings that p16^INK4A-Rb-mediated pathways are involved.

In contrast, evidence for a genetic connection between SNF5 and RB-regulated pathways was gained when it was found that inactivation of cyclin D1 inhibits tumorigenesis of Smarcb1^+/− mice, placing cyclin D1 downstream of SNF5 ⁸⁶. To determine whether SWI-SNF complexes inhibit tumorigenesis through RB-related proteins, tumor incidence was monitored in Smarcb1^+/− mice that also carried the T121 transgene—a truncated version of SV40 T-antigen that inhibits each of the three Rb family pocket proteins (Rb, p107, and p130) ⁸⁵. Mice expressing the T121 transgene under control of the lymphotrophic papovavirus (LPV) promoter develop choroid plexus carcinomas (CPCs), a tumor type that also develops in patients with inactivating mutations in SNF5. Although CPC development is not enhanced in Smarcb1^+/−; LPVT121 mice and the wild type Smarcb1 locus was retained, these mice develop MRTs with increased penetrance and decreased latency. Thus, SNF5 and RB family members cooperate to suppress cancer in a cell type that leads to MRTs, but not in a cell type that leads to CPCs ⁸⁵.

TrxG proteins of the nucleosome remodeling class also regulate senescence by transcriptional modulation of CDKN2A. CHD5 was shown to be a tumor suppressor that regulates expression of p16^INK4A and p19^{ARF 58}. Mammalian CHD proteins are homologues of the D. melanogaster TrxG proteins Mi-2 and kismet (KIS), and based on structural homology to these TrxG proteins, CHD5 is proposed to function as a nucleosome remodeler ⁸⁷. CHD3 and CHD4 (the CHD proteins most similar to CHD5) are homologs of D. melanogaster Mi-2α and Mi-2β, respectively; these proteins define the nucleosome remodeling and histone deacetylation (NURD) complex. The function of CHD5 in cancer is striking. Compromised CHD5—either by heterozygous deletion of the region of the chromosome encompassing Chd5 using chromosome engineering or by specific depletion of Chd5 in MEFs using RNAi—compromises p16^INK4A and p19^ARF expression and enhances proliferation ⁵⁸. Whereas expression of oncogenic Ras (H-Ras^G12V) in wild type MEFs induces senescence and prevents tumor formation, Ras-mediated senescence is bypassed in MEFs in which Chd5 has been knocked down, leading to robust tumour formation in vivo ⁵⁸. Conversely, gain of the chromosomal region encompassing Chd5 augments expression of p16^INK4A and p19^ARF, causing reduced proliferation and senescence in MEFs, as well as homeotic defects and stem cell depletion in vivo ⁵⁸; (A.A.M. personal observations).

As was the case for Bmi1^−/− mice, phenotypes resulting from gain of the chromosome region encompassing Chd5 were alleviated when levels of p16^INK4A and p19^ARF were reduced ⁵⁸. The reciprocal nature of the cancer, homeotic and stem cell phenotypes of mouse models with altered dosage of Chd5 with those of Bmi1 ^{58, 88}, support the idea that CHD5 and BMI1 have opposing roles in vivo (fig 1). Recent work in D. melanogaster indicates that KIS counteracts PcG-mediated repression by recruiting TRX and ASH1 to chromatin, reversing the H3K27me3 repressive mark and facilitating transcriptional elongation by pol II ⁸⁹. Whether CHD5 functions to counteract PcG-mediated repression awaits further study. These observations provide yet another example of the dynamic interplay between PcG-TrxG proteins in cancer and underscore the critical role that gene dosage plays in this process.

Cell cycle

Whereas the findings discussed above indicate that TrxG proteins are tumor suppressors, the role of MLL in cancer is more complex. MLL interacts with a number of proteins, including PcG (BMI1, HPC2), TrxG (SNF5, WDR5), and their accessory proteins (RBBP5, histone deacetylase, histone acetylase p300/CBP. Thus, translocations involving human chromosome 11q23—the most frequent genetic lesion in leukaemia ^{90, 91} that result in MLL fusions—likely disrupt PcG-TrxG-mediated chromatin dynamics. This idea is supported by the observation that MLL fusions lose their HMT domain yet retain the PcG-interacting regions ⁹². MLL is normally expressed in a cell cycle-dependent manner; it is dynamically regulated by SCF^SKP2 and anaphase promoting complex (APC)^CDC20-mediated degradation, resulting in two peaks of expression that facilitate G1-S and G2-M transitions ⁹³. In contrast to wild type MLL, MLL fusions are insensitive to cell cycle-coupled degradation, resulting in constitutive levels of MLL throughout the cell cycle and inappropriate expression of target genes ⁹⁴.

In addition to being tightly coupled with cell cycle transitions, MLL is cleaved by the endopeptidase taspase 1 (TASP1) to generate amino- and carboxy-terminal peptides that heterodimerize to form functional MLL ^95–97. MLL is not cleaved in Tasp1−/− cells, compromising HMT activity ⁹⁸. The HMT activity of MLL appears critical for its ability to transactivate its targets, as MLL increases H3K4me3 and activates transcription of E2F-regulated genes required for cell cycle progression ⁹⁹. The co-regulator host cell factor C1 (HCF1) interacts with E2F1 and recruits MLL, thereby serving as a switch during the transition between G1- and S-phase. Mature MLL transcriptionally modulates both cyclins and CKIs; in the absence of TASP1-mediated cleavage of MLL, expression of cyclins A, E, and B are reduced and p16^INK4A and p19^ARF are induced, inhibiting cell cycle progression ⁹⁸. Thus, MLL is intimately associated with the cell cycle on two different levels: it directly regulates expression of key cell cycle regulators, and its level oscillates with cell cycle transitions.

The current view is that the oncogenic activity of MLL fusions are through a gain-of-function mechanism. Mll-deficient mouse models are inviable ^{100, 101}. Mll+/− mice are not tumor prone, but have aberrant expression of Hox genes and homeotic defects such as cervical vertebrae with an identity that is inappropriate for that portion of the skeleton, reminiscent of the phenotypes observed in D. melanogaster. On the other hand, leukemia cells with MLL fusions have expression patterns typical of embryonic stem cells, consistent with the idea that cellular identity has been lost.

How can MLL fusions activate gene expression when the HMT-encoding C-terminus is lacking? ^{102,90,103,91}. Intriguingly, the MLL fusion partner AF10 interacts with DOT1-like histone H3 methyltransferase (DOT1L) ¹⁰⁴, a non-SET domain-containing HMT that trimethylates H3K79, a mark of actively transcribed genes ^105–107. An artificial MLL-DOT1L fusion not known to occur in patients transforms myeloid cells in a manner similar to the tumor-derived MLL-AF10 fusion, and transformation activity is dependent upon DOT1L-mediated H3K79me3 and enhanced HOXA9 expression ¹⁰⁴. The finding that MLL-DOT1L and MLL-AF10 fusions have similar transactivation activities suggested a model in which the MLL-AF10 fusion causes abnormal recruitment of DOT1L, changing H3K4me3 to H3K79me3 at specific target loci, including HOXA9 ¹⁰⁴. DOT1L activity is required for both establishment and maintenance of the transformed state in leukaemia cells with the MLL-AF10 fusion, suggesting that targeting this HMT may pose a new therapeutic strategy for leukemias caused by this MLL fusion. It is possible that other MLL fusions are also capable of recruiting “moonlighting” HMTs, an idea that requires further study.

Other MLL fusions can also recruit additional proteins that facilitate malignancy. The MLL fusion partner AF9 interacts with YEATs domain containing 4 (YEATS4), a protein aberrantly upregulated in neuroblastoma ¹⁰⁸ and required for repression of p53 ¹⁰⁹. The ability of AF9 to recruit YEATS4 is retained in the MLL-AF9 fusion, providing a mechanism for aberrant inactivation of p53. The finding that YEATS4 can recruit SNF5, suggests that this MLL-AF9-YEATS4 triad promotes tumorigenesis by recruiting SWI-SNF complexes ¹¹⁰. Aberrant recruitment of SWI-SNF complexes may also be a strategy used by the MLL-ENL fusion protein, as ENL has been purified from complexes that include SWI-SNF subunits ¹¹¹. In support of this model, MLL-ENL recruits SWI-SNF to the HOXA7 promoter resulting in increased expression of HOXA7—a protein essential for MLL-ENL-mediated oncogenesis ¹⁰³.

MLL fusions have also been found in complexes containing JMJD3, suggesting that MLL fusions recruit machinery that removes repressive histone marks ^{112, 113}. These findings exemplify the different means by which leukemogenic TrxG fusions exploit various components of the epigenetic machinery. Thus, MLL fusions drive lymphomagenesis through a gain-of-function mechanism involving misregulation of MLL during the cell cycle or aberrant reactivation of specific developmental programs through opportunistic recruitment of PcG-TrxG-associated machinery.

Apoptosis

PcG-TrG-mediated chromatin dynamics affects expression of genes that induce apoptosis (fig 5a). The general theme is that PcG proteins inhibit- whereas TrxG proteins induce-, apoptosis. For example, the ability of PcG and TrxG to work in opposing ways to regulate the CDKN2A locus discussed above, has a dramatic affect on ARF expression. As an inhibitor of MDM2-mediated ubiquitylation and degradation of p53, ARF facilitates apoptosis; therefore, PcG and TrxG repress and evoke apoptosis, respectively.

Loss of PcG repression causes excessive p19^ARF-p53-mediated apoptosis. Indeed, Bmi1−/− mice have increased apoptosis in lymphoid tissues, a phenotype that is rescued by depleting p16^INK4A and p19^ARF or by overexpressing the apoptosis inhibitor Bcl2 ⁶¹. Apoptosis is also enhanced in Bmi1+/−; EµMyc mice relative to Bmi1+/+; EµMyc mice, compromising lymphomagenesis. Lymphoma development in Bmi1+/−; EµMyc mice requires repression of p19^ARF, as apoptosis is compromised in Bmi1+/−; EµMyc; Cdkn2a^+/− mice, enhancing lymphomagenesis ⁶¹. Gain of PcG-mediated activity has the opposite effect: BMI1 cooperates with oncogenic H-Ras in breast cancer cells by suppressing apoptosis and attenuating p53, p21 and caspase 3, leading to highly malignant tumors that metastasize to the brain ¹¹⁴. BMI1 is essential for the survival and proliferation of both normal stem cells and leukemia-propagating cells; depletion of BMI1 causes apoptosis, as well as senescence ¹¹⁵. These findings underscore the role of PcG-mediated repression in modulating apoptosis via p19^ARF-mediated pathways.

PcG proteins can also affect apoptosis by modulating expression of genes besides p19^ARF. PRC2-mediated recruitment of DNMTs can silence specific loci, such as those encoding the apoptotic machinery ¹¹⁶. An ‘apoptotic methylation signature’ has been suggested as a strategy for diagnosing cancer ¹¹⁷. EZH2 suppresses expression of the apoptosis mediator DAB2 interacting protein (DAB2IP) in prostate cancer, thereby inhibiting tumour necrosis factor (TNF)-mediated apoptosis ¹¹⁸. Yet, this repression is reversible, as HDAC inhibitors can induce apoptosis ^119,120, in some instances sensitizing tumor cells to TNF-related apoptosis inducing ligand (TRAIL)-induced apoptosis ¹²¹. The reversibility of this repression is underscored by the ability of the global histone methylation inhibitor DZNep to remove PRC2 components (EZH2, EED, SUZ12) from chromatin, reduce H3K27me3, and activate expression of the apoptosis effector F-box-only protein 32 (FBXO32) in breast cancer but not in normal cells ¹²².

TrxG proteins also modulate apoptosis. For example, conditional inactivation of SNF5 in MEFs compromises cell survival, induces expression of p53 and p21, leading to apoptosis ^{82, 123}. Whereas p53 deficiency inhibited apoptosis of SNF5^−/− MEFs, p53 loss did not reverse their proliferation defect or rescue the embryonic lethality of Smarcb1^−/− mice ¹²³. However, tumour onset was extremely rapid when Snf5 and p53 were co-inactivated using a Smarcb1 inverting conditional-p53 conditional-Mx-Cre model, presumably because apoptosis was suppressed ⁸².

CHD5 also regulates p53-mediated apoptosis. Whereas loss of Chd5 compromises p53 activity and leads to tumorigenesis in vivo, gain of Chd5 increases senescence of MEFs in culture and triggers excessive apoptosis and perinatal lethality in vivo ⁵⁸. In contrast to the lethal apoptotic phenotype caused by SNF5 deficiency ¹²³, the apoptotic phenotypes caused by increased dosage of the region encompassing Chd5 were rescued in a p53 compromised background ⁵⁸. It remains to be determined whether p53 deficiency exacerbates tumorigenesis in mice with compromised Chd5.

In contrast to other TrxG proteins that induce apoptosis, some MLL fusions inhibit apoptosis. This further supports the idea that MLL fusions work by a gain of function mechanism.

DNA damage response and genomic integrity

SWI-SNF complexes suppress cancer by maintaining the genome ⁷⁷ (fig 5b). Proliferation of human MRT cells could still be inhibited by expression of a tumor-derived SNF5 mutant, indicating that loss of cell cycle control was not the only thing responsible for tumorigenesis ¹²⁴. A subset of SNF5 deficient tumor cells grown in culture had multi-lobed nuclei and were polyploid or aneuploid, indicating that the mitotic checkpoint was defective ¹²⁴. Wild type SNF5 inhibited proliferation of these abnormal cells, thereby establishing a population of cells that were essentially all diploid. By contrast, expression of tumor-derived SNF5 mutants exacerbated genomic instability. Although wild type SNF5 inhibited proliferation of aneuploid cells, this was not the case when a version of CDK4 that could not be inhibited by p16^INK4A was simultaneously expressed, indicating that p16^INK4A-mediated pathways are essential for SNF5 to re-establish genomic integrity ¹²⁴. MEFs in which SNF5 was conditionally ablated were hypersensitive to DNA damage caused by cross-linking (UV irradiation) and double strand breaks (doxorubicin) ¹²³. MEFs in which SNF5 was ablated had altered nuclear structure and extra centrosomes, indicating that SNF5 deficiency leads to cytokinesis failure, polyploidy, and genomic instability (fig 5b). E2F1 and its downstream target, Mad2—a protein implicated in mitotic defects and aneuploidy ¹²⁵—were downregulated by SNF5 expression ¹²⁴. Further, SWI-SNF complexes facilitate double-strand break repair by inducing the activity of gamma-H2AX at DNA lesions ¹²⁶. Yet, results of a separate study—in which the effects of SNF5 ablation was assessed at an earlier time point—indicated that SNF5 loss does not cause hypersensivity to DNA damage and does not compromise the DNA damage checkpoint ¹²⁷, in agreement with the finding that human MRT cell lines have a normal cell cycle checkpoint in response to DNA damage ¹²⁸.

In vivo evidence has also pointed to a role for SWI-SNF in regulating genomic stability. Mammary carcinomas of Smarca4^+/− mice are genomically unstable, but in contrast to MRT cells maintained in culture, these tumors are not aneuploid ⁸⁴. Human MRTs as well as MRTs that develop in Smarcb1^+/− mice are diploid and genomically stable ¹²⁷. This implies that disruption of SWI-SNF-mediated chromatin remodeling activity can substitute for genomic instability in cancer. Thus, perturbation of epigenetic events can contribute to cancer in the absence of genetic lesions ¹²⁹. These findings support the hypothesis that SWI-SNF proteins maintain genomic stability by coupling cell cycle and DNA damage checkpoints.