Mammalian SWI/SNF Complexes in Cancer: Emerging Therapeutic Opportunities

SMARCA4/SMARCA2

The results of synthetic lethal screening efforts have revealed that cancers bearing protein loss of one mSWI/SNF subunit exhibit uniquely strong dependency on the presence of a remaining, intact paralogous subunit to maintain their proliferative fitness. Synthetic lethal screening efforts using RNAi libraries in SMARCA4-deficient non-small cell lung carcinoma (NSCLC) cell lines identified SMARCA2 as the top synthetic lethal dependency required for cell line proliferation [13,14]. SMARCA4 and SMARCA2 are mammalian homologs of the SNF2-like ATPases in yeast and Drosophila, which use ATP hydrolysis to power the remodeling activity of CRCs [20]. SMARCA4 protein inactivation or loss-of-function point mutations [21] have been reported across neoplasms of various tissue types, including the lung [22–25], blood (Burkitt lymphoma) [21,26], brain (pediatric medulloblastoma) [21,27], ovary [28,29], pancreas [30], and others. Whether this specific intra-complex synthetic lethal relationship between paralog subunits holds true across a wide range of cancer types remains to be determined. However, the discovery of SMARCA2 as an actionable target has informed a potential approach in SMARCA4-deficient lung cancers, especially those that may lack the canonical molecular targets such as EGFR/RAS/PI3K activating mutations.

ARID1A/ARID1B

Dependency on a remaining paralog subunit upon single paralog loss is further demonstrated with the ARID1A-ARID1B pair of subunits, further demonstrating the partially redundant nature of paralog subunits within mSWI/SNF complexes [31]. The AT-rich interactive domain-containing protein 1A (ARID1A) is only found in BAF complexes and not PBAF complexes. Genome wide sequencing efforts have discovered somatic loss-of-function mutations in ARID1A across a range of gynecologic cancers [32], in approximately 30% of ovarian endometrioid carcinomas (OEC) and roughly 60% of ovarian clear-cell carcinomas (OCCC) [33,34]. Furthermore, saturation analysis of cancer genes reveals ARID1A to be one of the top genes to be frequently mutated among three or more tumor types [35]. Interestingly, ARID1A mutations are frequently monoallelic, indicating that haploinsufficiency may also play an important role in the oncogenic process [34,36].

Future Opportunities

Subunit-specific degradation can be realized through modern strategies in chemical biology such as those using ligand-dependent protein degradation of target biomolecules. Such approaches are already being exploited in settings in which a high-affinity small molecule capable of selectively binding the target of interest is tethered to a module that is able to recruit an E3 ubiquitin ligase to induce specific degradation of the targeted protein [37–39]. Efforts are currently underway to modify mSWI/SNF bromodomain-specific small molecules for protein degradation approaches. Such strategies will require additional biochemical studies before they are considered for further preclinical and clinical development, for example, determining whether targeted subunits in isolation or full complex assembles are degraded. While SMARCA2 and ARID1B are the most efficient vulnerabilities of SMARCA4-mutant and ARID1A-deficient cancers, respectively, to date small molecule based inhibitors have been challenging to identify owing to high degree of similarity between paralogs. Moreover, ARID1B loss has been recurrently linked to intellectual disability syndromes [40], thus ARID1B-specific degradation would have to be targeted only to affected tissues. In addition, caution would be required in patient identification; for example, SCCOHT, is a SMARCA4-deficient cancer (genetically) with concomitant loss of SMARCA2 (at the mRNA and protein level) leaving no ATPase subunit to deplete or inhibit [41]. mSWI/SNF-deficient neoplasms seem to recurrently retain dependence on the residual complex, which, in the case of paralogs, is assembled exclusively with the remaining paralog. Targeted chemical biology approaches such as small molecule-inducible protein subunit degradation represent a possible yet currently unexplored avenue for mSWI/SNF-driven cancer therapy.

SMARCB1-loss and EZH2 inhibition

The discovery of recurrent, biallelic inactivation of SMARCB1 in malignant rhabdoid tumor (MRT) by Versteege and colleagues in the late 1990’s provided the original evidence for the uniquely potent tumor suppressive function of mSWI/SNF complexes [42,43]. MRT is a rare, highly aggressive pediatric cancer with a remarkably low burden of secondary mutations, that to date lacks effective therapeutic approaches[44]. Roberts and colleagues reported elevated expression of EZH2 (enhancer of zeste homolog 2) in SMARCB1-deficient cells [9] and showed that depletion of EZH2 in MRT cell lines resulted in apoptosis [9]. Over the past several years, studies have shown that xenograft-based mouse models of MRT respond to small molecule inhibition of EZH2 [45]. Potent and specific EZH2 inhibition has also been demonstrated in various preclinical studies of hematopoietic malignancies such as lymphoma [46–49]. Preclinical proof of concept of small molecule-mediated EZH2 inhibition has been validated in SMARCB1-deficient solid tumors [45,50]. EZH2 functions as the core enzymatic subunit of the polycomb repressor complex 2 (PRC2), which is responsible for the trimethylation of histone H3 at lysine 27 (H3K27me3), a mark associated with genomic transcriptional repression [51–54]. Both the mSWI/SNF and the PRC2 complex are crucial effectors of genomic transcription; their activating and repressive functions, respectively, are tightly regulated during development [55]. The mSWI/SNF complex has been shown to antagonize the polycomb complex, hinting at a co-evolved transcriptional role [9,56–59]. The synthetic lethal relationship between SMARCB1 loss and PRC2 inhibition represents one of the most actionable dependencies that can be rationally targeted with current inhibitory agents against both EZH2 and EED [49,60]. As the ongoing human clinical trials come to term, the prospect of using EZH2 inhibitors as a mainstay in SMARCB1-deficient cancers will become clear (Table 1).

In addition to MRT, genetic studies have uncovered SMARCB1-loss in a number of other solid tumors including atypical teratoid/rhabdoid tumors (AT/RT), epithelioid sarcomas (ES), epithelioid malignant peripheral nerve sheath tumors (EMPNST), various SMARCB1-deficient carcinomas (e.g. myoepithelial carcinomas, sinonasal basaloid carcinoma), schwannomas, and very rare cases of multiple meningiomas [61–65]. Whether the loss of SMARCB1 is universally concomitant with elevated expression of EZH2 in every SMARCB1-deficient neoplasm remains to be determined as new studies emerge. Interestingly, an exhaustive survey of the genomic and epigenetic landscape of AT/RT uncovered three epigenetically distinct AT/RT subgroups in addition to inactivation of SMARCB1 [62]. The primary AT/RT samples showed wide genetic uniformity with specific epigenetic differences in three major pathways termed ATRT-TYR, ATRT-SHH and ATRT-MYC. As expected, EZH2 expression was elevated in all three subgroups [62]. Most importantly, this degree of granularity in deciphering the heterogeneous nature of AT/RT can further diversify targeted therapies in the clinic. This finding suggests that EZH2 inhibition could potentially synergize with pathway-dependent inhibitors in each respective AT/RT subgroup where drugs already exist. Of note, two small molecule inhibitors of smoothened (SMO), a key effector regulating SHH signaling, were recently approved by the federal drug administration (FDA) [66]. AT/RT-MYC on the other hand, showed MYC target gene upregulation with bromodomain-containing protein 4 (BRD4), a transcriptional co-activator [62], targeted to MYC enhancers. Bromodomains are protein modules that bind acetyl lysines on chromatin serving as readers of epigenetic states and recruiters of transcriptional apparatus [67–70]. Bromodomain inhibition has been established and validated in various hematopoietic cancers [71,72] and has been shown to reduce MYC target gene expression [73–75]. Hence, BRD4 inhibition in the ATRT-MYC subgroup may represent a viable path forward. SMARCB1-deficient AT/RTs are thus particularly primed for drug combination studies pairing EZH2 inhibitors with molecular probes of the SHH pathway to downregulate hedgehog signaling or BRD4 inhibitors to attenuate expression of MYC target genes.

SMARCA4-deficient thoracic sarcomas bear concomitant loss of SMARCA2

Over the past two years, sequencing studies of solid tumors revealed that two tumor types, small cell carcinoma of the ovary, hypercalcemic type (SCCOHT) and SMARCA4-deficient thoracic sarcomas (SMARCA4-DTS) feature genetic loss of SMARCA4 [28,76–79] with concomitant silencing of SMARCA2. Immunohistochemical staining of SCCOHT tumor samples showed reduced or absent nuclear SMARCA2 protein [79], indicating that not all cancers that feature loss of SMARCA4 will benefit from a SMARCA2 ATPase inhibitor. Intriguingly, unsupervised hierarchical clustering of transcriptional profiles of these tumor types denoted that SMARCA4-DTS profiles were most closely aligned with MRT and SCCOHT, highlighting the mechanistic similarity of mSWI/SNF ATPase loss and BAF47 (SMARCB1/hSNF5) loss [79]. However, cell of origin may also contribute to the striking similarities in gene expression profiles. These findings represent the first demonstrations of SMARCA4-deficient cancers (SMARCA4-DTS and SCCOHT), which correlate with SMARCB1-deficient MRT. SMARCB1 loss is hallmark to ~98% of MRT; the remaining 2% have been reported to have SMARCA4 genetic loss with SMARCA2 silencing. These results suggest that EZH2 inhibition should be explored in SMARCA4-DTS and SCCOHT cell lines and murine pre-clinical models to establish whether this approach may show efficacy in these tumor types (Table 1).

Pathway vulnerabilities in ARID1A-deficient cancers

In addition to the striking intra-complex dependency (ARID1B) in ARID1A-deficient tumors, novel synthetic lethal targets that are not members of the mSWI/SNF complex have also been identified, with studies exploring the tractability of EZH2 [10], YES1 [11], PARP [80], ATR [81] and PIK3CA [82] pathway inhibition. It is interesting to note that SMARCB1-deficient tumors such as MRT and AT/RTs showed increased sensitivity to EZH2 inhibition. However, the ovarian clear-cell carcinoma (OCCC) cell lines lacking ARID1A described thus far do not show upregulation of EZH2 [10], suggesting that the molecular dependencies caused by SMARCB1 loss in MRT are different in ARID1A-loss cancers. In addition, EZH2 was not ranked among the top vulnerabilities identified in shRNA-based dropout screening set published by Helming et al (Figure 2B) [31]. Strikingly, Kim and co-authors showed that the viability of mSWI/SNF-deficient (ARID1A, SMARCA4 and PBRM1 mutated) cancer cell lines primarily depended on a structurally intact PRC2 complex as opposed to the enzymatic activity of EZH2 [83]. On that account, further studies are warranted to determine whether EZH2 enzymatic inhibition or PRC2-specific destabilization will amount to a specific vulnerability across a broader set of ARID1A-deficient cancers.

mSWI/SNF complexes have long been implicated in DNA damage repair [84,85]. Recently, various studies have demonstrated that ARID1A deficient cancer cells exhibit enhanced vulnerability to inhibitors of DNA repair pathways [80,81]. While DNA checkpoint inhibition warrants greater pre-clinical exploration, the PIK3/AKT pathway has garnered significant interest in the field. A recent report has converged on the PIK3/AKT pathway as a vulnerability for ARID1A-mutant neoplasms [82]. Notably, both catalytic subunits of PIK3, PIK3CA (p110) and PIK3R2 (p85), are featured among the highest vulnerabilities across a set of ARID1A-deficient cells within shRNA-based dropout screening data sets (Figure 2B). PIK3CA activating mutations have been recurrently seen in ARID1A-deficient OCCC patients [86,87]. This raises the possibility that ARID1A-loss cooperates with activating PIK3CA mutations for OCCC oncogenesis. Furthermore, knock down of ARID1A in breast cancer and fibroblast cell lines increased sensitivity to PIK3 and AKT inhibition. Interestingly, OCCC lines that have retained proper expression of ARID1A have been shown to be less sensitive to PIK3 and AKT inhibition [82]. PIK3CA is known to play an essential role in tumor development across various tissue types [88]; PIK3CA inhibition is a clinically tractable option due to the wide availability of PIK3 pathway drugs [89].

In fact, there are more than 30 inhibitors targeting various nodes of the PIK3/AKT pathway with some rapidly moving toward FDA-approval [90]. Whether PIK3CA is the most tangible target to treat ARID1A-deficient cancers is still unknown; consequently, further research will be required to mechanistically exploit the role of ARID1A-loss in tumorigenesis.

ATPase and bromodomain inhibition

The increasingly apparent role of mSWI/SNF complexes in human disease, coupled with recent successes in the development of bromodomain inhibitors, have inspired the discovery of PFI-3, a potent and specific small-molecule inhibitor of the bromodomain modules on SMARCA2, SMARCA4 and PBRM1 (Figure 3) [91]. However, despite engaging various key subunits of the mSWI/SNF complex (in both BAF and PBAF complexes), PFI-3 has failed to show anti-proliferative responses in cancer cell lines [92]. PFI-3 demonstrated proper engagement of its cellular targets but was unable to disengage endogenous SMARCA2-containing mSWI/SNF complexes from chromatin, likely owing to the number of other DNA-binding and chromatin-binding domains on mSWI/SNF complexes [93]. Vangamudi et al., established the ATPase activity of SMARCA2/4 as the pertinent biologically active target by demonstrating that an ATPase-dead mutant of SMARCA2/4 fails to rescue the proliferation of SMARCA4-deficient lung cancer cell lines, while both a SMARCA2/4 wild-type and a bromodomain-dead mutant construct were able to maintain cellular viability [93]. With these results, targeted inhibition of the ATPase domain of SMARCA2 has become widely recognized as the most viable approach for the treatment of a subset of SMARCA4-deficient cancers (e.g. 10-20% of NSCLC cases) [93].

Various SMARCA4 deficiencies manifest as tumor suppressor mechanisms [28,79], but a recent finding in neuroblastoma (NB) highlighted the ATPase as an oncogene [94]. Jubierre et al., showed overexpression of SMARCA4 in stage 4 NB patients was associated with worst outcomes [94]. Experiments both in cell culture and xenograft mouse models using a range of NB cell lines demonstrated apoptotic cell death upon shRNA-mediated knockdown of SMARCA4 [94]. This discovery did not evaluate whether the oncogenic effects were mediated by the catalytic activity of SMARCA4 or within the context of an intact and fully functional mSWI/SNF complex. As future work establishes this distinction, the toxicity and potential therapeutic window of SMARCA4 or SMARCA2 ATPase inhibition needs to be explored in pre-clinical murine models. Moreover, the ATPase domains of SMARCA4 and SMARCA2 are essentially identical (Figure 3), making identification of target-specific small molecule chemical probes exceedingly challenging. Small molecule probes that are potent but unable to discriminate between the ATPases of SMARCA4 and SMARCA2, may still be clinically effective in specific mSWI/SNF-driven cancer types.

The bromodomain-containing accessory subunits, BRD9, PBRM1 and BRD7 (PBAF-specific) have been linked to acute myeloid leukemia (AML), clear-cell renal cell carcinoma (ccRCC) and breast cancer respectively [95–97]. Moreover, BRD9/7 and PBRM1 feature bromodomains that are liable to small molecule inhibition (Figure 3). Unlike the acetyl readers found on SMARCA4 and SMARCA2, the bromodomain of BRD9 has been demonstrated to be a tangible target in AML [95]. Although BRD9 is rarely mutated in human cancers and is expressed rather ubiquitously, it is vital to leukemia maintenance, particularly in AML [95]. While studies have begun to shed light on the role of the mSWI/SNF complex in leukemia maintenance, due to the combinatorial complexity of mSWI/SNF complexes, the proper subunit module responsible for AML maintenance remains unresolved [98]. A number of potent small molecule inhibitors of BRD9 have been recently reported (Figure 3), and future work in this area is warranted to establish whether BRD9 is critical to cells other than AML cells. This will further resolve whether BRD9 inhibition impedes remodeling activity of the mSWI/SNF complex across the genome or whether this is a specific response, occurring at MYC-driven enhancers loaded with BRD9 protein not tethered to the mSWI/SNF.

Targeting the unique oncogenic property of SS18-SSX in synovial sarcoma

The mSWI/SNF complex is broadly depicted as a tumor suppressor [99,100], with genetic loss of one subunit often perturbing the complex integrity and composition, which results in transcriptional deregulation of genes that regulate cellular proliferation [101]. However, mSWI/SNF complexes can also be oncogenes, noted most strongly in synovial sarcoma, in which a unique gain-of-function mutation confers oncogenic functionalities. Synovial sarcoma is a highly aggressive and rare cancer which occurs in the extremities of pediatric and young adult patients [102] and is uniformly characterized by an in-frame fusion of the mSWI/SNF subunit SS18 to the C-terminal end of SSX1 or SSX2 and rarely SSX4 [103– 105]. SS18-SSX is most often the only genetic aberration (in an otherwise stable genomic background) and specific knockdown of the fusion protein rapidly induces cell death in SS cell lines. Moreover, recent studies have shown that SS18-SSX competes the endogenous SS18 protein and consequently evicts SMARCB1 away from the mSWI/SNF complex [102]. In the presence of SS18-SSX, SMARCB1 is unable to bind the complex and is degraded by the proteasome [102], which can be detected by immunohistochemical analyses. SS18-SSX-specific degradation represents a challenging yet highly impactful therapeutic avenue. Ongoing clinical trials are investigating the tractability of EZH2 inhibition in SS due to the low levels of SMARCB1. In addition, further synthetic lethal screening efforts, both with shRNA/CRISPR-Cas9 and small molecule libraries represent additional opportunities toward the identification of unique pathway sensitivities that result from the activity of the oncogenic SS18-SSX-containing SWI/SNF complex.