Invited Review: Dysregulation of chromatin remodellers in paediatric brain tumours – SMARCB1 and beyond

Mutations in chromatin remodelling genes occur in approximately 25% of all human tumours (Kadoch et al. Nat Genet 45: 592–601, 2013). The spectrum of alterations is broad and comprises single nucleotide variants, insertion/deletions and more complex structural variations. The single most often affected remodelling complex is the SWI/SNF complex (SWItch/sucrose non‐fermentable). In the field of paediatric neuro‐oncology, the spectrum of affected genes implicated in epigenetic remodelling is narrower with SMARCB1 and SMARCA4 being the most frequent. The low mutation frequencies in many of the SWI/SNF mutant entities underline the fact that perturbed chromatin remodelling is the most salient factor in tumourigenesis and could thus be a potential therapeutic opportunity. Here, I review the genetic basis of aberrant chromatin remodelling in paediatric brain tumours and discuss their impact on the epigenome in the respective entities, mainly medulloblastomas and rhabdoid tumours.


Chromatin and its components
Chromatin is a dynamic structure that organizes the genome in eukaryotic cells and thus influences cell type-specific gene expression. The termoriginally coined by the cell biologist Walther Flemming in the 19th century and back then describing various staining patterns of the nucleus as observed under the light microscoperelates to the DNA and the attached histone proteins (as well as their modifications) [1]. Specifically, it comprises the four core histones, H2A, H2B, H3 and H4, which are wrapped into a DNA segment of 147 bp length [2]. The histone H1 links nucleosomes and is not directly involved in the compaction of DNA but is important to facilitate higher order chromatin structure [3]. The spatial organization of this complex as well as its covalent modifications (for example, by the demethylation or methylation of specific histone residues) influence transcriptional programmes and consequently have a direct effect on the physiological differentiation of cells [4].

Histone modifications as targets in tumourigenesis
The N-terminal histone tail, which protrudes out of the nucleosome, is an important regulatory structure and can be subject to several post-translational modifications. These include, for instance, trimethylation of the amino acid lysine in position 27 of histone 3 which exerts a repressive influence on gene expression or its acetylation which usually enhances gene expression. Other amino acids as part of the histone H3 tail, which can be modified either by methylation or by demethylation, are the lysines in position 4 or 27 (H3K4 and H3K27).
Further post-transcriptional modifications which influence the histone code are small protein modifications such as ubiquitin and SUMO (small ubiquitin modifiers). Ubiquitin commonly acts as a signal for proteomic degradation of the marked proteins. The SUMO protein is able to recruit further proteins which carry SUMO-interacting domains (SIM) and can thus act as a scaffold to assemble large protein complexes. It has recently been found that the addition of these residues in turn may influence the presence of other histone marks such as H3K9me3 [5].
The methylation or demethylation of the aforementioned amino acids is mediated by a set of enzymes which in turn can become the target of oncogenic mutations. The KDM family of demethylases catalyses the demethylation of lysine residues and is frequently mutated in non-small cell lung and urothelial cancers [6]. In particular, KDM6A is prototypically affected in acute myeloid leukaemia, where the loss of this protein leads to a higher degree of chemotherapy resistance [7]. While chromatin-modifying enzymes influence gene transcription by directly targeting amino acids of various histones, chromatin remodellers affect the steric conformation of the histone code and are thus less well defined than the focal effects of chromatin modifiers.

Chromatin remodellers in paediatric brain tumoursfamilies and functions
Broadly, chromatin remodelling complexes can be classified into four different families according to the phylogenetic similarities in their ATPase domains [8]: imitation switch (ISWI), chromodomain helicase DNAbinding (CHD), switch/sucrose non-fermentable (SWI/ SNF) and INO80 [9]. These protein families differ by their protein composition but also by their functional effects on chromatin configuration. The ISWI and CHD complexes are pivotal for the maturation of nucleosomes [10] (that is, the proper assembly and configuration of nucleosomes after DNA replication processes) and for correctly spacing these to enable transcription.
The SWI/SNF complexes open up the chromatin by introducing irregular spaces between nucleosomes and by ejecting nucleosomes. This mechanism exposes DNA segments to transcription factors. INO80 is implicated mostly in the exchange of histones in nucleosomes without, however, changing the spacing of the nucleosomes [11,12].
Among these families of chromatin modifiers, mutations in the SWI/SNF complex play a dominant role in human tumourigenesis. In a meta-analysis of 44 whole genome/exome sequencing studies, Kadoch et al. found mutations of SWI/SNF subunits in 19.6% of all cancer types analysed [13]. Among them, ARID1A was the most frequently mutated gene (mainly in endometrioid and clear cell ovarian cancer). Overall, SMARCA4, ARID1A, ARID1B, ARID2 and PBRM1 were mutated significantly above the background mutation rate.
In paediatric oncology, the distribution is remarkably different and narrower. Recently published pan-cancer analyses [14,15] have found SMARCB1 and SMARCA4 as the only significantly mutated genes of this type, which were affected by single nucleotide variants (SNVs) in paediatric tumours. In these studies, 38.2% of all atypical teratoid rhabdoid tumor (ATRT) samples, 6.67% of all Burkitt lymphomas and 4.76 % of all wingless-related integration site (WNT) medulloblastomas displayed SMARCB1 point mutations. SMARCA4 was found to be mutated in 46.67% of all Burkitt lymphomas, 19.05% of all WNT-and 10% of all Group 3 medulloblastomas. No other SWI/SNF complex members were among the significantly mutated genes.
Although these studies are comprehensive in that they comprise samples of most cancer entities, they are particularly enriched in paediatric brain tumours. An independent analysis of several published whole-exome studies from paediatric tumours, which I performed in this review, not only confirms the importance of SMARCA4 and SMARCB1 but also highlights the importance of ARID1A/B in paediatric entities (Figure 1; Table 1). ARID mutations occur recurrently in neuroblastomas, rhabdomyosarcomas and a small fraction of Wilms tumours [16][17][18][19]. In neuroblastoma, up to 10% of all samples harbour mutations in either ARID1A/B or SMARCA4 and many of these tumours belong clinically to the high-risk group. Based on these numbers, it is hard to determine which subunits of the SWI/SNF complex are most often affected in paediatric oncology, but it is safe to state that SMARCB1 and SMARCA4 aberrations occur more often in the tumours of infancy and childhood than in adult entities.
In the setting of paediatric tumours in general or paediatric brain tumours specifically, mutations of SWI/SNF members thus seem to be the prevailing theme. A notable exception is CHD5. The CHD proteins (comprising CHD1-9) have been implicated in the neuroblastomathe 1p36 locus, which is often lost in high-risk neuroblastomas, encompasses the CHD5 locus. Point mutations of this protein have been described in breast and ovarian [33,34] cancers but to the best of my knowledge have not been reported in paediatric malignancies.
In neuroblastoma, both the deletion of CHD5, as part of the complex CHD, as well as a lower expression of CHD5 have independently been identified as markers of a bad prognosis [35]. Unlike the case of SWI/SNF subunit inactivation (which is mostly homozygous), most CHD5 aberrations in neuroblastoma are heterozygous, and homozygous inactivation might be achieved by epigenetic silencing as mutations or structural variants of the second allele are seldom found. Although CHD5 has been ascribed tumour suppressive functions in glioma cell lines [36], a bona fide role as a tumour suppressor in paediatric brain tumours remains to be established.
Similarly, the links of INO80 aberrations to cancer in general and paediatric neuro-oncology specifically are sparse. While it is known that this complex is required for oncogenic transcription in, for example, non-small cell lung cancer and colon cancer [37,38], so far no recurrent oncogenic mutations have been identified. Similarly, although germline mutations of SWI complex members result in more or less well-characterized (neuro)developmental disorders (such as Coffin-Siris syndrome, which is characterized by mutations in ARID1A, ARID1B, SMARCA4, SMARCB1, SMARCE1 or the developmental transcription factor SOX1 [39] or Rett syndrome-like phenotypes in which SMARCA1 is mutated [40]), links between brain tumourigenesis and the ISWI complex are largely missing.
Thus, the subunits of the SWI/SNF complexor BAF complex ('BRG1-or HBRM-associated factors') are most often affected by genetic alterations in paediatric brain tumours. Three of a kind? Structure and functions of three BAF complexes The SWI/SNF complexinitially identified 20 years ago in yeast [41,42] is a highly conserved protein complex with pleiotropic, mainly activating, functions in the epigenome of mammalian cells. It helps to keep the balance between stem cell properties and the differentiation of cells. An important structural basis for this function is the fact that components of the BAF complex can be exchanged during cellular differentiation. For example, as neuronal progenitors mature into neurons, they lose the ACTL6A/BAF53A and PHF10/ BAF45A components of the npBAF complex which are replaced by ACTL6B/BAF53B and DPF1/BAF45B or DPF3/BAF45C. Thus, turning the neuronal progenitor BAF complex to a neurone-specific BAF complex (see Figure 2a) [43]. This conversion, or subunit switching, seems to be an important characteristic in acquiring cell type-specific functions. Beyond these tissue-specific BAF complexes (cBAF), recent studies have highlighted a further structurally different variant of the SWI/SNF complex, which differs profoundly in its protein composition. Specifically, aside from the cBAF complex and the pBAF complex (polybromo BAF complex), the noncanonical SWI/SNF complex (see Figure 1) has been identified [44,45]. The modular assembly of these complexes has recently been uncovered within the frame of a largescale mass spectrometric study [44]. In the three BAF types, the SMARCC and SMARCD subunits represent the common backbone. By addition of SMARCB and SMARCE, the so called BAF core is formed. The addition of GLTSCR1 then paves the way for further differentiation into the ncBAF complexes, while the addition of SMARCB1 and SMARCE1 builds up the BAF core. This core BAF complex is then further branched into the cBAF complex (by the addition of ARID1, 'ARID-dependent branching') or the pBAF complex (by the addition of ARID2). The addition of the ATPase module (in Table 1. An overview on SWI/SNF aberrations in paediatric entities

Leukaemia and lymphomas
Early T-cell precursor ALL all three types of complexes) finalizes the assembly process.
Although different cellular functions of these complexes have not been fully elucidated, the different composition of the complexes may suggest that a loss of classic SWI/ SNF function (such as for instance in rhabdoid tumours) may be compensated by an increase in, for example, ncBAF functioningwhich would then constitute a putative therapeutic vulnerability. Due to the fact that no mutational or structural aberrations in the ncBAF-specific proteins (such as, for example, GLTSCR1 and Brd9) have been uncovered so far, the suitability of these as potential drug targets is thus an important, potentially clinically exploitable, consequence [45,46].
Along this line, it is noteworthy that the ncBAF complex lacks ARID1, ARID2 and SMARCB1 subunits and may hint towards the use of Brd9 inhibition in both ARID1-altered entities and SMARCB1-altered entities. The identification of dependencies between SWI/ SNF complex members is an active field of investigation. Recently, Schick et al. have comprehensively mapped which proteins become vital for cell survival after, for example, SMARCA4 knock-down. This knowledge may in future be used to target SMARCA4-deficient tumours [47] (see also Figure 2b). Functional insights into how the different BAF complex members exert distinct functional roles are sparse. For the pBAF complex, an interesting contribution came from Wei et al. who found that targeting Brd9 in pancreatic beta cells was able to enhance the interaction between the vitamin D receptor and the pBAF complex, thus increasing beta cell survival and ameliorating hypoglycaemia in animal models [48].
Among other specific roles that have been ascribed to the pBAF complex (or its core unit Brd7) are the regulation of neural crest development in association with subunits of the CHD complex [49,50] as well as the maintenance of embryonal stem cell transcription [51].
Interestingly, the pBAF complex also seems to play a role in increasing the vulnerability of cancer cells to T cell-mediated killing, which may offer the prospect to sensitize tumour cells to immunotherapeutic approaches by using Brd7 inhibitors [52].
It can be assumed that many more of these molecular switches, which result in an increased recruitment of specific BAF subtypes, exist. This may potentially be relevant in the oncological setting.
In the context of cancer, the BAF complex has mainly been recognized as an important tumour suppressor. Jagani et al. showed that the inactivation of SMARCB1, being a key component of the BAF complex, leads to an overactivation of potentially oncogenic sonic hedgehog (SHH) signalling [53]. This experimental finding is backed by studies in primary tumours, which found an overexpression of SHH components in a subgroup of ATRT [25]. Not only MYC-dependent signalling but also the expression of MYC itself are also repressed by an intact SWI/SNF complex [54,55].
Earlier studies have pointed out that the transition from the G1 to S phase is impaired by direct binding of the SWI/SNF complex [56] and can thus induce G1 arrest [57], this goes along with a downregulation of E2F targets (such as E2F1, CDC6, cyclin A and cyclin D1) [58,59]. This function of the SWI/SNF complex is critically dependent on an intact RB protein [59].
Beyond these functions at the level of oncogenic signalling, the BAF complex is pivotal in maintaining genome integrity [60]. This may happen via different routes. For example, Kapoor et al. pointed out that Mec1 kinase (a yeast orthologue of the human ATM gene) is positively regulated by the SWI/SNF complex [61]. Moreover, the SWI/SNF complex induces and maintains high levels of cH2A.X, which in turn provides binding sites for the checkpoint factor MDC1 [62].
The importance of the SWI/SNF complex for genome integrity is somewhat counteracted by the observation that many cancers, which display SWI/SNF subunit mutations, do not display a high degree of chromosomal instability.
While the interplay of the BAF subcomplexes is subject to further functional studies, the 'bottom-up' approach of genomic analysis has helped to identify paediatric brain tumours which harbour mutations in SWI/SNF complex members.

SMARCA4 mutations in paediatric brain tumoursthe genomic basis
In the field of paediatric brain tumours, alterations in SMARCA4 have been described in medulloblastoma and rhabdoid tumours, with the cardinal difference being that the alterations in rhabdoid tumours are commonly homozygous, whereas in medulloblastomas only one allele is affected.
A number of studies [63][64][65] found SMARCA4 aberrations in medulloblastomas at varying frequencies between 4% and 11.5% respectively. Notably, most of these tumours belonged to the molecular Group 3 and a minority to the WNT subgroup. It is noteworthy that these mutations in medulloblastoma co-occur with other SNVs, which are typical for the respective MB subgroups such as CTNNB1 in WNT and gains/amplifications of MYC in MB-Group 3 [65]. Mutations in SMARCA4 are relatively depleted in the SHH subgroup of medulloblastomas [66]. This can potentially be explained by a recently described dependency of SHH signalling on wild-type SMARCA4 [67] it can thus be assumed that this oncogenic pathway can only be maintained in the presence of an intact SMARCA4.
The heterozygous mutations in medulloblastoma are thus in contrast to those in rhabdoid tumours where they are presumed to be the sole oncogenic events. For medulloblastomas, this raises the hitherto unanswered question as to whether these mutations are really the initiating event in tumourigenesis or rather bystanders which promote tumour formation at a later stage.
The protein structure of SMARCA4 comprises five functional domains: the QLQ domain (amino acids 171-206) and the HSA domain (amino acids 460-532) at the N-terminal end of the protein, as well as two helicase-binding domains (the SNF2 domain or helicase ATP-binding domain, amino acids 830-995 and the C-terminal helicase domain, amino acids 11480-1310), the Brd domain. The bromodomain of SMARCA4 is localized C-terminally. It has been found to be dispensable for tumour cell proliferation and its pharmacological inhibition did not repress the growth of rhabdoid tumour cells in vitro [68].
The mutations detected in medulloblastomas were (almost) exclusively found in the DEAD-like helicase domain and the helicase C-terminal domain. Unlike in rhabdoid tumours where loss of SMARCA4 (either by a somatic SNV or by cytogenetic alterations) is usually homozygous, the mutations found in medulloblastomas were mostly heterozygous. In rhabdoid tumours, SMARCA4 alterations are seldom foundthey account for approximately 5% of all rhabdoid tumours and are thus exceedingly rare [69]. The complete loss of the protein likely has different functional consequences than the point mutations in medulloblastoma, in which SMARCA4 is still expressed although functionally compromised [69,70].

Mechanistic consequences of SMARCA4 mutations and their consequence on the epigenome
The functional role of SMARCA4 is complemented by SMARCA2both proteins are 75% identical and are expressed in different cell types and stages of differentiation [71]. As a consequence of this functional similarity, SMARCA2 becomes vital for tumour cell survival in SMARCA4 mutant cell lines [72]. The functional consequences of the various SMARCA4 mutations affecting the different protein domains have only recently been investigated. Mutations in the DNA-binding domain of SMARCA4, as would be expected, inhibit its binding to nucleosomes and thus increase the amount of mobile SMARCA4 in a cell, in contrast to mutations in the ATPase-binding domain that prevent the release of SMARCA4 from bound chromatin by decreasing ATP hydrolysis [73].
Despite these opposing mechanistic effects, the broad epigenetic consequences of introducing heterozygous mutants of the ATPase-binding domain and the DNAbinding domain [73] on chromatin accessibility seem to be similar. Consistent with the role of the SWI/SNF complex as an 'opener' of the chromatin, there were more genomic sites losing accessibility than gaining in the ATAC-seq analysis of cells with SMARCA4 mutations.
A prominent epigenetic pattern, which was present in all heterozygous mutants, was the loss of H3K27Ac while the H3K4me3 mark was still present at the same sites, thus suggesting a transformation of these sites from active to poised enhancers [73]. It is worthwhile noting that the so called super-enhancersclustered enhancers with abundant H3K27Ac signalwere equally affected by this loss of H3K27Ac signal. This is in contrast to the effect of SMARCB1 loss on the epigenome, which was found to alter the distribution of classical enhancers but to leave super-enhancers relatively unaltered [74] (Figure 3). Whether this differential effect on super-enhancers when comparing SMARCB1 and SMARCA4 mutant cells also results in a differential sensitivity to bromodomain inhibition (which mainly affects super-enhancer-driven genes) has not been comprehensively studied so far.
Similar to SMARCB1, the interplay of SMARCA4 and the PRC1 and PRC2 complexes has been actively investigated. Mutations in the ATP-binding domain of SMARCA4 have been shown to increase genome-wide binding of PRC1 by Stanton et al. [75]. The increased deposition of PRC1 components (such as Ring1b, which was specifically investigated in this work) was mostly due to a loss of direct binding between the BAF complex and the PRC1 complex. Furthermore, it was demonstrated that H3K4me3 and other epigenetic marks, which characterize active promoters, were strongly associated with sites that gained PRC1 signals thus suggesting that, in particular, promoter regions were affected by this increase of PRC1. As PRC1 binding prompts the recruitment of PRC2 complex factors, the study also found consecutive increases in H3K27me3, the repressive mark which is deposited by the PRC2 complex. Although a genome-wide increase in H3K27me3 has been suggested in SMARCB1-deleted or mutant entities as well, the deletion of ARID1A, another component of the SWI/SNF complex, did not result in an increase of neither H3K27me3 nor Ring1b.

Putative drug targets in SMARCA4 mutant entities
Only few preclinical studies have been conducted specifically assessing the vulnerabilities of SMARCA4 mutant paediatric brain tumours. However, recent discoveries from various non-CNS entities, with bi-or monoallelic alterations of BRG1, may fuel drug treatment studies in medulloblastoma and rhabdoid tumours as well. For example, the interplay of the PRC2 complex and the SWI/SNF complex has propelled research on the use of EZH2 inhibitors in various SWI/ SNF compromised entities, among them SMARCA4-deficient lung cancers, thoracic sarcomas and small cell carcinoma of the ovary hypercalcaemic type (SCCOHT). A seminal study by Januario et al. [76] found that the expression level of SMARCA2the ATPase subunit which is a paralogue of SMARCA4is predictive of the response to EZH2 inhibition. Notably, the cell line panel, which was investigated in this work, did not contain paediatric brain tumours but nonetheless these results may be generalizable to, for example, medulloblastomas or ATRT. The proof of concept for a specific vulnerability of SMARCA2-and SMARCA4-deficient cancer cells towards EZH2 inhibition was presented by Chan-Penebre et al. showing a differential sensitivity towards the EZH2 inhibitor tazemetostat in SCCOHT [77].
Beyond targeting EZH2, other approaches have aimed to exploit the dependency of SMARCA4-deleted cancers on SMARCA2. The synthesis of orally available SMARCA2/4 inhibitors has paved the way for preclinical studies, which have underlined the dependency of SMARCA4-deficient cancers (such as oesophageal carcinoma [78]) on SMARCA2. It remains to be seen, if these promising preclinical experiments can be translated into clinical success.
Outside the epigenetic field, other putative targets which have been highlighted in SMARCA4-deficient entities are RTKi. In SCCOHT, a CRISPR-Cas9 screening approach detected PDGFR signalling as a putative vulnerability and the use of Ponatinib to inhibit this axis proved to be efficacious preclinically [79]. How far these vulnerabilities represent Achilles heels in all SMARCA4-deficient entities and could thus be expanded to paediatric brain tumours remain to be determined.

SMARCB1 alterations in paediatric brain tumours
Rhabdoid tumours represent the prototypic example of SMARCB1 compromised entitiesthe first description of SMARCB1 as genetic target in rhabdoid tumours dates back to the 1990s. After the initial description of rhabdoid tumours as a series of case reports in the 1980s [80][81][82], Versteege et al. found deletions of SMARCB1 in established rhabdoid tumour cell lines [83]. Since then, several studies have confirmed this gene as the most important genetic hit in rhabdoid tumours, which are otherwise genomically very stable [84,85]. While all ATRT share a loss of INI-1 (the protein encoded by the SMARCB1 gene), the genetic basis of this loss differs between tumours. Rhabdoid tumours of the periphery (eMRT) more often than CNS rhabdoid tumours display homozygous losses affecting SMARCB1, which also cover several kBP beyond this gene. [86]. In contrast, rhabdoid tumours of the central nervous system have a more complex deletion pattern. For example, the recently identified major ATRT subgroups, ATRT-TYR, ATRT-SHH and ATRT-MYC, [25,87] differ in their inaxctivation pattern of SMARCB1 (see Table 2).
The functional consequences of these findings are still largely unknown but codeletions in genes other than SMARCB1 may well contribute to the divergent transcriptomic profiles found in the three ATRT subgroups.
The spectrum of SMARCB1-deleted entities in paediatric neuro-oncology goes beyond rhabdoid tumours and also comprises a set of rare entities, specifically CRINETs (cribriform neuroepithelial tumours) and SMARCB1-deleted chordomas. Antonelli et al. found an absence of INI-1 staining in four out of eight paediatric chordomas [88]. As was later shown, these INI-1-negative chordomas indeed form a separate molecular subgroup when compared to ATRT and INI-1 intact chordomas. The INI-1-negative chordomas differ from classic chordomas also by the absence of further cytogenetic aberrations such as chromosome 9 or chromosome 10 deletions, which are commonly observed in classic chordomas [89]. In line with the evidence from other INI1 deleted tumour entities, INI-1-deficient chordomas seem to have a particularly bad prognosis [90]. The immunohistochemical absence of INI-1 thus prompts the question if these tumours are susceptible to the same therapeutic approaches as ATRT such as, for example, EZH2 inhibition, which have been studied widely in MRT. A clinical trial using the EZH2 inhibitor tazemetostat [91] has shown two partial responses in children with INI-1-deficient chordomas.
In contrast to INI-1-deficient chordomas, CRINETs are indeed highly similar to ATRT, in particular to subgroup ATRT-TYR, [92] and it is conceivable that this entity represents a histological variant of this ATRT subgroup with an excellent survival.
Beyond these entities, SMARCB1 mutations are rarely found in paediatric neuro-oncology and evidence is circumstantial. Recently, El-Ayadi et al. have described a case of a medulloblastoma with concurrent IDH1 and a (heterozygous) SMARCB1 mutation [93]. However, unlike SMARCA4 and Arid1a/Arid1b, the protein does not seem to be a protagonist in the pathogenesis of these tumours.

The epigenetic consequences of SMARCB1 alterations
Unlike SMARCA4, SMARCB1 is not a part of the ATPase subunit of the SWI/SNF complex and the functional consequences of its deletion may thus differ from the aforementioned SMARCA4 aberrations. As mentioned previously in the context of SMARCA4, a wellstudied mechanistic consequence of SWI/SNF complex perturbations is the activation of the PRC2 complex. Initial evidence towards an antagonism between the PRC2 complex and the SWI/SNF complex came from studies in Drosophila melanogaster. PcG proteinsthe analogues of the PRC2 complex in flywere found to block SWI/SNF-induced nucleosome mobilization [94,95]. The human analogue, the PRC2 complex, is composed of four subunits [96,97]: SUZ12, RBBP4/7, AEBP2 and EZH2with the latter two being the methyltransferases that are responsible for placing methyl groups at the lysine residue 27 of Histone 3, thus inducing H3K27me3.
With the advent of inhibitors of the EZH2 methyltransferase activity, the interplay between the SWI/SNF complex and the PRC2 proteins once more came into focus. McCabe et al. found that the EZH2 inhibitor GSK126 could be used to abrogate mutation-induced hyper-trimethylation of H3K27 in EZH2 mutant lymphoma [98]. Subsequently, Knutson et al. showed a regression of rhabdoid tumour flank xenografts when being treated with the EZH2 inhibitor EPZ-6438 [99]. Along a similar line, it was described that EZH2 inhibition was able to abrogate self-renewal and sensitize to radiation [10].
These in vitro and in vivo data support the use of EZH2 inhibitors in preclinical studies. However, more recent findings from both functional and descriptive studies, which characterize epigenetic features of SMARCB1-deficient tumours, suggest that the role of EZH2 may be more complex. Kim et al. confirmed that SWI/SNF mutant cell lines were in general more susceptible to EZH2 inhibition than SWI/SNF intact cell lines and virtually all cell lines were sensitive to EZH2 knock-down [11]. However, the introduction of a noncatalytically active, mutant EZH2 in these cell lines, which were unable to restore H3K27me3, rescued the effect of EZH2 knock-down in cell lines insensitive to pharmacological EZH2i. This speaks for a role of the non-catalytic subunit of EZH2 in tumourigenesis and indicates that pharmacological decrease of H3K27me3 may not be the only mechanism by which EZH2 exerts pro-tumourigenic influence. Along a similar line, it has been found previously that the reduction of H3K27me3 does not correlate with the tumour response when treating rhabdoid tumour xenografts with EZH2 inhibitors [12].
Adding to these mechanistic and preclinical insights, immunohistochemical and ChIP sequencing studies have shed a new light on EZH2 as a drug target. Hasselblatt et al. [13] and Kakkar et al. [14] found that about 50% of all ATRTs do not display increased but rather diminished H3K27me3 signals. In a large epigenetic study examining the presence of the ENCODE marks in ATRT primary samples, Erkek et al. [15] did not observe a global increase of the H3K27me3 mark when comparing ATRT to other embryonal brain tumours. However, specifically, neuronal differentiation genes were found to be affected by increased H3K27me3 deposition. Notably, it was found that in many sites which bound EZH2, no or only a very faint signal of H3K27me3 was detected. At these genes, abundant SMARCA4 binding was present, thus suggesting that residual SWI/SNF complex binding may play a role in maintaining the expression of, for example, cell cycle genes despite the overexpression of EZH2. This may suggest that EZH2 inhibition as the sole treatment in the clinical setting could not be enough to abrogate all oncogenic effects present in the epigenome of rhabdoid tumours and that the inhibition of SMARCA4 may represent a rationale to tackle this disease. Further support for this approach came from a histopathological study, showing that SMARCA2 staining is often absent in malignant rhabdoid tumours [16]. Thus, SMARCB1-deficient tumours may be critically depending on the SMARCA4 (the paralogue of SMARCA2) and SMARCA4 inhibition as a concept to exploit synthetic lethality could be a concept of preclinical studies.
In addition, the putative dependency on SMARCA4 in the absence of SMARCA2 may hold the promise of testing substances or substance combinations, which have been used in the context of SMARCA2/4 mutant entities such as CDK inhibitors (which have been tested successfully in SCCOHT). An additional interesting concept which so far has not been fully exploited when treating rhabdoid tumours in the clinic is the use of EZH2 as a sensitizer for other therapeutics. In SMARCA4 mutant lung cancer, the combination of EZH2 inhibition and topoisomerase inhibition has shown efficacy both in vitro and in vivo. This dependency is likely mediated by a SWI/SNF-induced repression of EGFR. It can be speculated that SMARCB1deleted entities may show the same vulnerability [17].
With regard to the direct influence of SMARCB1 on the epigenome, Nakayama et al. [18] demonstrated that SMARCB1 is important for mediating the stability of SWI/SNF chromatin binding without, however, influencing the stability of the complex itself. Beyond that SMARCB1, re-expression in MRT cell lines contributed to a widespread gain of the activating marks H3K27Ac and H3k4me1, whereas the levels of H3K4me1 did not change significantly. These findings have been paralleled by a study from Wang et al. [74] confirming the importance of SMARCB1 for sharpening the enhancer landscape. Adding to that, they showed that super-enhancer activity is maintained by residual SWI/SNF complex members in the SMARCB1-deficient setting, thus retaining the expression of genes which form cell identity. However, the landscape of typical enhancers is drastically changed by the deletion of SMARCB1 thus affecting mainly tumour suppressor genes which are downregulated.
On the contrary to these reports which dwell on the activating function of SMARCB1, there is also growing evidence on its repressive role in specific cell types or developmental stages. Langer et al. [19] recently studied the role of INI-1 in human embryonic stem cells (hESC). Performing ATAC-seq, they found that SMARCB1 represses the accessibility of key pluripotency genes, thus contributing to differentiation. In line with its roles as an important protagonist in neurogenesis, differentiation of hESC into neuronal cells was not possible after deletion of SMARCB1. These findings substantiate further publications that have shown the indispensability of the SWI/SNF complex as a whole for self-renewability and neuronal differentiation [73,110,111].
Going beyond the canonical BAF complex, recent literature has underlined the dependency of SMARCB1deficient cells on the non-canonical BAF complex. In the absence of SMARCB1, Brd9 becomes critical in stabilizing the remaining SWI/SNF complex [46] and displays a distinct binding pattern which differs from the canonical BAF complex. Remarkably Michel et al. found a tight colocalization of Brd9 and SMARCA4 in the absence of SMARCB1 underlining that Brd9 plays a pivotal role in stabilizing the residual SWI/SNF complex.
Targeting Brd9 is possible and in vivo active inhibitors are available [112]. It has been attempted in vitro in rhabdoid tumours [113] and in vivo in other entities [114].
In summary, the impact of SMARCB1 (or its absence) on the epigenome remains complex. While recent years have seen a surge in publications studying the epigenetic landscape of these tumours, a clearly amenable epigenetic target remains yet to be identified. The paralogue dependency on SMARCA2, which has been described in SMARCA4-deficient entities, may be a concept that also applies to SMARCB1-deficient entities, given the emerging role of the residual SWI/SNF complex. EZH2arguably the best studied epigenetic target in rhabdoid tumourscontinues to be in the focus of further investigations. However, both the limited effect of monotherapies with EZH2 inhibitor in the clinical setting and the mechanistic data which suggest non-canonical functions of this enzyme may demand sophisticated combinations with other epigenetic therapies or immunotherapeutic approaches.

ARID1A/ARID1B and ARID2 mutations
While ARID1A and ARID1B mutations are very common among solid neoplasms in adulthood (for example, in oesophageal carcinomas, lung cancers reviewed in [115]), their occurrence in paediatric neuro-oncology seems to be limited to medulloblastomas. Northcott et al. [66] and Jones et al. [64] found mutation frequencies of 2% and 8%, respectively, in both cases tightly associated with the WNT subgroup. The vast majority (10/11 mutations) are stop-gain mutations and possibly contribute to expressing a truncated protein in the cell.
As described previously, ARID1A and ARID1B are close paralogues and in analogy to the paralogues SMARCA2 and SMARCA4, inactivating ARID1A mutations render ARID1B essential. While this dependency so far cannot be exploited clinically, AURKA inhibition seems to be synthetically lethal with ARID1A mutations at least in colorectal cancer cells [116] and thus merits further investigation.
ARID2 is present only in the pBAF complex and not the cBAF complex. Mutations have been described in non-small cell lung cancer and hepatocellular carcinoma but not in paediatric brain tumours [117,118].
Although there are no substances available that interfere directly with ARID expression or chromatin binding, a number of surrogate targets have been described in the context of ARID mutant tumourssuch as the kinase ATR, which was found to be essential for cell survival in ARID mutant entities [119]. Thus, although ARID mutations are rare in paediatric neuro-oncology, the findings from other entities may hold the promise of being transferred to ARID mutant medulloblastomas as well.

Summary and perspective
Mutations of the SWI/SNF complex subunits and the consecutive changes in epigenetic and transcriptional programmes are important genetic contributors to tumourigenesis. In a remarkably high proportion of tumours, it is clear that these aberrations represent the initiating event for the formation of cancer cells. However, translating this knowledge into therapeutic possibilities so far has been difficult, owing to the fact that the majority of these mutations constitute loss of function events which impact widely on gene transcription. The strategies which have been pursued to tackle these diseases have so far aimed at limiting or reversing the epigenetic consequences which result from SWI/SNF perturbation. This path is further strengthened by recent investigations which have revealed functional dependencies between members of the SWI/SNF complex [47]. With the advent of the first available SMARCA4 inhibitors, the concept of paralogue dependency and the idea of targeting the residual SWI/SNF complex may become more prominent in the clinic as well. Beyond epigenetic therapies, further drug targets have recently emerged. Despite the fact that rhabdoid tumours display low mutation frequencies, two recent studies found a substantial degree of immune cell infiltration and a response of ATRT to checkpoint blockade [86,120]. Thus, novel and extratumoural therapeutic opportunities exist and it is worthwhile studying them together with classical chemotherapy or epigenetic treatments as a future prospect.