The SMARCA2/4 Atpase Domain Surpasses the Bromodomain As A

Published OnlineFirst July 2, 2015; DOI: 10.1158/0008-5472.CAN-14-3798

Cancer Therapeutics, Targets, and Chemical Biology Research

The SMARCA2/4 ATPase Domain Surpasses the Bromodomain as a Drug Target in SWI/SNF- Mutant Cancers: Insights from cDNA Rescue and PFI-3 Inhibitor Studies Bhavatarini Vangamudi1, Thomas A. Paul2, Parantu K. Shah1, Maria Kost-Alimova1, Lisa Nottebaum2, Xi Shi1, Yanai Zhan1, Elisabetta Leo1, Harshad S. Mahadeshwar1, Alexei Protopopov1, Andrew Futreal3, Trang N. Tieu1, Mike Peoples1, Timothy P. Heffernan1, Joseph R. Marszalek1, Carlo Toniatti1, Alessia Petrocchi1, Dominique Verhelle2, Dafydd R. Owen4, Giulio Draetta1, Philip Jones1, Wylie S. Palmer1, Shikhar Sharma2, and Jannik N. Andersen1

Abstract

The SWI/SNF multisubunit complex modulates chromatin target knockdown, the inhibitor fails to display an antiproli- structure through the activity of two mutually exclusive cata- ferative phenotype. Mechanistically, the lack of pharmacologic lytic subunits, SMARCA2 and SMARCA4, which both contain a efficacy is reconciled by the failure of bromodomain inhibition bromodomain and an ATPase domain. Using RNAi, cancer- to displace endogenous, full-length SMARCA2 from chromatin specific vulnerabilities have been identified in SWI/SNF-mutant as determined by in situ cell extraction, chromatin immuno- tumors, including SMARCA4-deficient lung cancer; however, precipitation, and target gene expression studies. Furthermore, the contribution of conserved, druggable protein domains to using inducible RNAi and cDNA complementation (bromodo- this anticancer phenotype is unknown. Here, we functionally main- and ATPase-dead constructs), we unequivocally identify deconstruct the SMARCA2/4 paralog dependence of cancer cells the ATPase domain, and not the bromodomain of SMARCA2, using bioinformatics, genetic, and pharmacologic tools. We as the relevant therapeutic target with the catalytic activity evaluate a selective SMARCA2/4 bromodomain inhibitor suppressing defined transcriptional programs. Taken together, (PFI-3) and characterize its activity in chromatin-binding and our complementary genetic and pharmacologic studies exem- cell-functional assays focusing on cells with altered SWI/SNF plify a general strategy for multidomain protein drug-target complex (e.g., lung, synovial sarcoma, leukemia, and rhabdoid validation and in case of SMARCA2/4 highlight the potential tumors). We demonstrate that PFI-3 is a potent, cell-permeable for drugging the more challenging helicase/ATPase domain to probe capable of displacing ectopically expressed, GFP-tagged deliver on the promise of synthetic-lethality therapy. Cancer Res; SMARCA2-bromodomain from chromatin, yet contrary to 75(18); 3865–78. 2015 AACR.

Introduction ered recurrent somatic mutations and copy-number (CN) changes in histone-modifying enzymes and chromatin remodeling com- Epigenetic dysregulation plays a fundamental role in the devel- plexes supporting a causal role for altered epigenetic states in opment of cancer (1). Large-scale genome sequencing has uncov- tumorigenesis (2–4). Although the mechanistic consequences of these alterations remain poorly understood, it is appreciated that such events promote acquisition of cell oncogenic capabil- 1Institute for Applied Cancer Science, The University of Texas MD Anderson Cancer Center, Houston, Texas. 2Pfizer Oncology Research ities through deregulation of nucleosome-dynamics, gene tran- Unit, La Jolla, California. 3Department of Genomic Medicine, The scription, DNA replication, and repair (5). Indeed, chromatin University of Texas MD Anderson Cancer Center, Houston, Texas. regulators are emerging as therapeutic targets and inhibitors of 4Pfizer Worldwide Medicinal Chemistry, Cambridge, Massachusetts. histone-modifying enzymes, as well as bromodomains, which Note: Supplementary data for this article are available at Cancer Research "read" the histone marks, have recently shown efficacy in pre- Online (http://cancerres.aacrjournals.org/). clinical and clinical settings through their ability to reverse T.A. Paul and P.K. Shah contributed equally to this article. oncogenic transcriptional programs (6–8). Corresponding Authors: Jannik N. Andersen, XTuit Pharmaceuticals, 700 Main TheSwitch/SucroseNonFermentable(SWI/SNF)isamulti- Street, Cambridge, MA 02139. Phone: 617-990-2235; Fax: 617-863-3677; E-mail: subunit chromatin remodeling complex that consists of one of [email protected]; and Shikhar Sharma, Pfizer Oncology Research Unit, two mutually exclusive helicase/ATPase catalytic subunits, 10724 Science Center Drive, San Diego, CA 92121. Phone: 858-526-4172; E-mail: SMARCA2 and SMARCA4. Together with core and regulatory Shikhar.Sharma@pfizer.com subunits, SMARCA2/4 couple ATP hydrolysis to the perturba- doi: 10.1158/0008-5472.CAN-14-3798 tion of histone-DNA contacts. This sculpting of the nucleoso- 2015 American Association for Cancer Research. mal landscape at promoters provides access to transcription

www.aacrjournals.org 3865

Vangamudi et al.

factors and cognate DNA elements facilitating both gene acti- Materials and Methods vation and repression (9). Because various SWI/SNF subunits Bioinformatics are mutated or lost at high frequency in human tumors Genome sequencing data and CN information were down- (2–4, 10), this complex has garnered considerable attention loaded from cBioPortal (Supplementary Table S1). Cell line (11). A tumor-suppressive role has most strongly been dem- genomic annotation was from the Sanger (www.cancerrxgene. onstrated in childhood malignant rhabdoid tumors, in which org) and the Broad Institutes (www.broadinstitute.org/ccle). Out- the SMARCB1 (Snf5) subunit is biallelicaly inactivated in lier sum statistics (26) and standard software packages for nearly all cases (10). Accordingly, knockout of mouse sequence analysis were used. SMARCB1 results in fully penetrant and lethal cancers with 11 weeks median onset (12). In human synovial carcinoma, recurrent chromosomal translocations, which are diagnostic of Cell lines the malignancy, result in oncogenic fusions (SS18-SSX) that Cells obtained from the ATCC were cultured accordingly: alter the composition/function of the SWI/SNF complex (13). RPMI-1640 (A549, H1299, H157, H520, H460, HeLa, and fi Pointing to the broader relevance of SWI/SNF in cancers are THP-1); Iscove's Modi ed Dulbecco's Medium (MV-4-11); fi frequent inactivating mutations in accessory subunits, includ- McCoy's 5a Modi ed Medium (A-204 and G-401) and supple- ing ARID1A in ovarian and endometrial carcinomas (14, 15), mented with 10% FBS (Gibco). The Aska and Yamato cells (Osaka and PBRM1 in renal cell carcinomas (16). Medical Center) were grown in DMEM (20% FBS). All Cell lines Context-specific molecular vulnerabilities that arise during were mycoplasma negative (LookOut Mycoplasma Kit PCR, Sig- < tumor evolution represent an attractive class of drug targets; ma) and maintained at low passage ( 3 month) after thawing fi however, the frequency and spectrum of somatic lesions often from master vials (IACS and P zer BioBanks) subjected to short fi confound efforts to identify such therapeutic targets solely based tandem repeat (STR) pro ling of polymorphic loci (Promega > on genomic information (17). To address this challenge, func- PowerPlex 16 system) with a 80% match criteria for cell line tional, unbiased chemical, and genetic loss-of-function (LOF) authentication. platforms, which use either drug-like small-molecules or siRNA/ shRNA libraries, hold the promise to systematically identify Immunoblotting nonobvious target-genotype interactions that might impact clin- Analysis was performed on whole-cell lysates (Supplementary ical decisions (17–19). Recently, using genetic LOF approaches, Information) using primary antibodies: SMARCA4 (Abcam, three groups have independently identified SMARCA2 as an #108318), SMARCA2 (Abcam, #15597), HA-Tag (Cell Signaling essential gene in SMARCA4-deficient lung cancer (20–22) pro- Technology, #2367), a-Tubulin (Cell signaling Technology, 3873), posing a synthetic lethality therapeutic approach. However, it and secondary antibodies (Li-Cor, #926-68020, #926-32111). remains unclear whether small-molecule inhibitors of the SMARCA2 bromodomain or ATPase domain can mimic the Assays reported RNAi phenotypes resulting from paralog dependency PFI-3 (PF-06687252) is available from SGC (http://www. in SWI/SNF (11, 23). thesgc.org/chemical-probes/PFI-3). Bromodomain selectivity Several subunits in the SWI/SNF complex contain bromo- was measured using ligand binding, site-directed competitive – domains, which are evolutionary conserved protein protein assays (BROMOscan, DiscoverRx; ref. 27). Cell potency was interaction modules that bind acetyl-lysine on proteins and measured using in situ cell extraction, CellTiter-Glo (Promega) histone tails (6, 24). Bromodomains are druggable and fol- and clonogenic assays (Supplementary Information). lowing the antitumor activity of JQ1 (6), there is interest in broadly developing small-molecules inhibitors against other RNAi, plasmids, gene expression, and chromatin family members to dissect their therapeutic potential (1, 6, 24, immunoprecipitation 25). Here, we speculate that SMARCA2/4 bromodomains could SMARCA2/4 shRNA (SIGMA TRC-collection) and siRNA contribute to either assembly or targeting of the SWI/SNF (ON-TARGET PLUS Dharmacon) sequences are listed in Supple- complex to specific genomic loci providing an intervention mentary Table S2. ATPase-dead (K785A) and bromodomain muta- drug target rationale. However, because bromodomains are tions (Y1497F and N1540W) were made in human SMARCA4 frequently found in large protein complexes (and often flanked cDNA (GeneCopoeia #GC-Y3533) using site-directed mutagenesis by additional domains involved in chromatin-binding and (QuickChange, Agilent Technologies) with equivalent mutations protein–protein interactions), RNAi-mediated depletion alone in SMARCA2 (GeneCopoeia #GC-Z4424). All cDNAs, subcloned does not reveal the contribution of individual domains to the into lentiviral vectors, were sequence verified and virus genera- LOF phenotype, representing a specific challenge for drug-target tion, infection, and generation of stable cell lines were conducted validation. following standard procedures (Supplementary Information). In this report, we conduct complementary cDNA rescue and Gene expression (Affymetrix) data and methods have been depos- pharmacologic studies to explore whether the bromodomain ited with NCBI (GSE69088). Chromatin immunoprecipitation of SMARCA2/4 represents a tractable target in SWI/SNF- (ChIP) and qPCR were conducted as previously described (28). mutant cancers. We characterize the PFI-3 bromodomain inhibitor in biochemical assays and across preclinical models with altered SWI/SNF complex (lung, synovial sarcoma, Results leukemia, and rhabdoid tumor cells) and discover that bro- Genomic alterations in SWI/SNF across human tumors modomain function of SMARCA2/4 is dispensable for tumor To build upon recent meta-analysis (2–4), we first examined cell proliferation, while the catalytic ATPase activity is both SWI/SNF mutation and CN variation drawing on a larger set essential. of patient tumors (n ¼ 10,038) from 45 genome sequencing

3866 Cancer Res; 75(18) September 15, 2015 Cancer Research

Targeting the SMARCA2/4 Bromodomain using RNAi and PFI-3

studies (Supplementary Table S1). Clearly, genomic alterations in and gene expression considerations. Although 3% (19/549) of the 20 canonical SWI/SNF subunits are highly prevalent (Fig. 1A) LUAD patients can be categorized as having low expression of occurring in 15% of all cancers (3, 4, 10). Cancers with the highest both SMARCA2 and SMARC4 (10th percentile; hypergeometric frequency of lesions in SWI/SNF subunits are rhabdoid tumors, distribution), the expression profile for SMARCA2 lacks the female cancers, including ovarian, uterine, cervical and endome- bimodal distribution characteristic for SMARCA4-null cancers trial, lung and gastric adenocarcinoma, melanoma, esophageal, (Fig. 1F). Altogether, the above biomarker assessment (i.e., and renal clear cell carcinoma (Fig. 1A and Supplementary Table SMARCA4 loss) outlines a large, well-defined patient popula- S1). A tumor-suppressive role of the SWI/SNF complex in these tion in need of novel molecularly targeted therapies. contexts has been recognized based on the high frequency of inactivating mutations, which is further underscored by mouse Genotype-specific vulnerability and paralog dependency genetic studies (10, 29). In contrast, SWI/SNF mutations do not examined in SMARCA4-reconstituted cells emerge as significant recurrent alterations in glioblastoma, thy- The recent discovery that SMARCA2 knockdown inhibits the roid cancer, multiple myeloma, and acute myeloid leukemia growth of SMARCA4-deficient cancer cells (20–22) has opened a (AML). In AML, SMARCA4 may instead be an oncogene driving potential therapeutic avenue and received considerable attention cMYC transcription in concert with BRD4 (7, 11, 30, 31). As such, (11, 23, 33). However, it is unknown whether small-molecule it appears that cellular and tissue context defines the tumor- inhibitors against the ATPase or bromodomain can mimic the suppressive or oncogenic functions of the SWI/SNF complex RNAi phenotype. Hence, defining the contribution of each domain (5, 11, 23). to the LOF phenotype is required for prioritizing drug discovery to achieve tangible clinical therapeutic endpoints. To explore this, we SMARCA4 deficiency is prevalent and mutually exclusive to first selected a representative panel of SMARCA4-deficient and SMARCA2 CN loss in lung cancer proficient lung cancer cells (Fig. 2A) and evaluated multiple In primary human lung adenocarcinoma (LUAD), about half of SMARCA2 shRNAs for knockdown efficiency (Fig. 2B) and phe- the SMARCA4 mutations are deleterious (nonsense and frame- notype. Using either viability (Fig. 2C) or long-term clonogenic shift mutations) and occur at a 7% frequency in The Cancer assays (Fig. 2D and Supplementary Fig. S2), we observed robust, Genome Atlas patient samples (Fig. 1B). Overall, SWI/SNF com- genotype-specific growth inhibition when SMARCA2 was depleted plex components are mutated in 71 of 229 patients with an in SMARCA4-deficient cell lines (A549, H1299, and H157). In average mutation rate of approximately 1.7 per sample. In addi- contrast, knockdown in SMARCA4-proficient cells (H460, H520, tion, two copy loss of SMARCA4 is observed in 14 out of 299 and HeLa) had no effect on viability confirming the reported LUAD cases adding to the fraction of SMARCA4-deficient tumors. synthetic-lethal SMARCA2/4 interaction (20–22). Because heterozygous SMARCA4 knockout mice are haploinsuf- Next, we engineered SMARCA4-deficient A549 cells to reex- ficient and tumor prone (29), we also analyzed copy-number press a doxycycline-inducible wild-type SMARCA4 cDNA (Fig. driven mRNA expression and conclude that loss of one allele is 2E). Control cells treated with SMARCA2 siRNAs did not grow, also sufficient to decrease SMARCA4 expression (Fig. 1C). Focus- whereas the expression of SMARCA4 (þDox) completely restored ing on LUAD and lung squamous cell carcinoma (LUSC), map- growth (Fig. 2F and G). Most strikingly, SMARCA4 expression ping of the genomic annotation onto individual patient samples increased 5-fold following SMARCA2 depletion (Fig. 2F) indic- revealed that loss of SMARCA2 and SMARCA4 is largely mutually ative of compensation, a finding that supports the reciprocal exclusive (Fig. 1D and Supplementary Fig. S1A). Moreover, with assembly and stability of SMARCA2 and SMARCA4 into the respect to gene expression, outlier statistics identifies SMARCA4, SWI/SNF complex (22). along with ARID1A, as the most significantly altered subunits in the SWI/SNF complex in LUAD (Fig. 1E) with similar profiles observed for LUSC (Supplementary Fig. S1B). At the genome PFI-3 is a selective, potent, and cell-permeable SMARCA2/4 level, SMARCA4 ranks in the top 5% of all genes with negative bromodomain inhibitor outlier sum statistics and its bimodal expression profile (Fig. 1F) To explore pharmacologic inhibition of the SMARCA2 bromo- clearly defines a SMARCA4-deficient patient population. domain, we next evaluated the small-molecule inhibitor PFI-3 In good concordance with cell line annotation at the Sanger (Fig. 3A) discovered through a collaboration between the Struc- and the Broad Institutes, SMARCA4 protein expression was tural Genomic Consortium and Pfizer. Biochemically, we deter- nondetectable by Western blot analysis in approximately mined that PFI-3 binds avidly to both SMARCA2 and SMARCA4 20% (12/50) of lung cancer lines (Fig. 1G and Supplementary bromodomains (BROMOScan Kd's between 55 nmol/L and 110 Fig. S1C–S1E). Of eleven SMARCA4-mutant cell lines, only one nmol/L) consistent with the binding constant (Kd ¼ 89 nmol/L) (NCI-H2286) displayed measurable protein expression (Sup- measured by isothermal titration calorimetry (www.thesgc.org/ plementary Fig. S1F). However, as previously noted, nonde- chemical-probes/PFI-3). Moreover, using recombinant purified tectable SMARCA4 protein levels (as assessed by immunoblot- bromodomains, we discovered that PFI-3 binds with similar ting) occur more often than predicted from mutation and CN avidity to both the short and long isoform of SMARCA2 revealing analysis, suggesting that promoter methylation and epigenetic that the alternatively-spliced 18 amino acid insert (34) does not silencing may be additional oncogenic mechanisms for impair PFI-3 binding (Fig. 3A). Moreover, profiling against 32 SMARCA4 loss (32). Paradoxically, a few cell lines appear to bromodomains at DiscoveRx (27) confirmed exquisite selectivity have nondetectable expression of both SMARCA2 and versus other subfamilies (Fig. 3C and Supplementary Table S3) SMARCA4 (Fig. 1G). However, such complete dual-loss of expanding the PFI-3 selectivity information obtained using dif- SMARCA2/4 is not observed in primary LUAD and LUSC ferential scanning fluorimetry (DSF). In summary, we find that tumors as indicated by both our CN analysis (using both there is a good concordance between the ligand competition GISTIC and ABSOLUTE algorithms; Supplementary Fig. S1G) (BROMOScan) and the direct biophysical binding (DSF) assays

www.aacrjournals.org Cancer Res; 75(18) September 15, 2015 3867

Vangamudi et al.

A B C complex

3’UTR Frame_Shi_Del Frame_Shi_Del Missense_Mut Nonsense_Mut Splice_SIte (RSEM) – SMARCA4 2 Log Components of SWI/SNF

% Lesions in SWI/SNF Number of mutaons Copy Number (CN) D SMARCA4 6% 7%

SMARCA2 7% 7%

E F n = 598 n = 548 RNA-seq density SMARCA4 0.0 0.2 0.4 0.6 0.8 8 910111213 Outlier score (LUAD) Log2 (RSEM) – SMARCA4 3 G - H1563 H647 H1793* H2023** A549** H2228 H2030** H1792 H1437 H1838 H2444 H520 H1915 Calu H2286 UMC-11 H1703 H358 H2172 MORCPR H1623 A427 H441 H1299***

SMARCA4

SMARCA2

Tubulin

Figure 1. Genomic analysis of the SWI/SNF complex in human cancer. A, percentage distribution of lesions (mutations and CN changes) in SWI/SNF components across tumors proﬁled by The Cancer Genome Atlas and other laboratories (Supplementary Table S1). B, SWI/SNF mutation spectrum in LUAD (n ¼ 229 tumors; 121 mutations). C, correlation of SMARCA4 CN with gene expression (RSEM, RNA-Seq Expression by Expectation Maximization). D, CN loss of SMARCA2/4 is mutually exclusive in LUAD (left; n ¼ 493) and LUSC (right; n ¼ 490). Oncoprint (www.cbioportal.org): blue, high CN loss (GISTIC 2.0 threshold value of 2); red, high CN gain (GISTIC 2.0 threshold value of 2); green, mutations. E, SMARCA4 has the highest negative outlier sum statistics among SWI/SNF components (LUAD; n ¼ 598). F, histogram showing bimodal distribution of SMARCA4 gene expression (LUAD; n ¼ 548) highlighting the predicted patient "responder" population (red). G, protein expression and SMARCA4 genomic annotation across lung cancer cell lines: mutation (), copy-number loss (), and gene silencing ().

3868 Cancer Res; 75(18) September 15, 2015 Cancer Research

Targeting the SMARCA2/4 Bromodomain using RNAi and PFI-3

A B C 9 - -1 150 A549 H460 shS2-2 shS2-4 shS2-5 shS2-7 shS2 shS2 shLuc shS2-10

SMARCA2 100 A549 H1299 H157 H460 H520 HeLa control)

SMARCA4 of Tubulin Figure 2. 50 ﬁ SMARCA4-de cient lung cancer cells SMARCA2 SMARCA2

selectively depend on SMARCA2. A, Growth (%

SMARCA2/4 protein levels in cell lines H460 A549 0 Tubulin Tubulin selected for RNAi studies. B and C, A549 (SMARCA4-deﬁcient) and NCI- H460 (SMARCA4-proﬁcient) cells transduced with control (shLuc) or D SMARCA2-targeting shRNAs (shS2) shLuc shS2-1 shS2-2 shS2-4 shS2-5 shS2-7 shS2-9 shS2-10 and assessed for knockdown (B) and cell viability (C) 1 week after puromycin A549 selection (SD; n ¼ 6). D, clonogenic assay and crystal violet staining of colonies after 10 to 14 days. E, A549 cells with inducible expression/ reconstitution of full-length SMARCA4 H460 cDNA grown in the presence or absence of doxycycline (Dox) and analyzed 4 and 7 days after doxycycline induction. F and G, following doxycycline treatment (5 days), A549 siRNA siRNA (SMARCA2) cells were transfected with either E F G Ctrl S7 S8 nonsilencing control (Ctrl) or SMARCA2 Day 4 Day 7 cDNA: Vector Ctrl SMARCA4 control

siRNAs (S7 and S8) and evaluated for Vector protein knockdown (5 days after siRNA: Ctrl S7 S8 Ctrl S7 S8 Dox

+ + Dox transfection; F) and clonogenicity (G). − Dox − Dox SMARCA4 SMARCA4 SMARCA4 SMARCA2 Tubulin

A549 Tubulin

and note that in addition to targeting SMARCA2/4, PFI-3 also has (Fig. 4A) or long-term clonogenic assays (Fig. 4B and Supple- activity (70% inhibition at 2 mmol/L) against the structurally mentary Fig. S4). Because SWI/SNF is a multisubunit complex related fifth bromodomain from PBRM1, another SWI/SNF containing numerous chromatin-interacting domains, we specu- subunit. lated that selective SMARCA2 bromodomain inhibition by itself In cell-based chromatin-binding assays, using in situ cell extrac- is not sufficient to dislodge the endogenous SWI/SNF complex tion techniques to remove non-chromatin bound proteins, we from chromatin. To elaborate on this, the binding of endogenous observed dose-dependent displacement of GFP-tagged SMARCA2 (full-length) SMARCA2 to chromatin was monitored by immu- bromodomain (i.e., 132 amino acid residues) by PFI-3 (Fig. 3D nofluorescence in A549 cells using SMARCA2 knockdown as a and E). Notably, the inhibitor showed prolonged cell-target specificity control (Fig. 4C–E). Even high concentrations of PFI-3 engagement with similar potency (IC50 ¼ 5.78 mmol/L) fol- (30 mmol/L; 1 and 24 hours) were unable to displace the lowing 2 and 24 hours treatment (Supplementary Fig. S3). As a SMARCA2 protein from chromatin (Fig. 4C and D). Again, we negative control, JQ1 did not inhibit the binding of ectopically cross-validated the in situ cell extraction assay using the reference expressed SMARCA2 bromodomain, but selectively displaced JQ1 inhibitor, which potently inhibited chromatin-binding of GFP-tagged BRD4 (Fig. 3D, and data not shown). Taken endogenous BRD4 (Fig. 4C, bottom) but not SMARCA2 (data not together, our cell-biochemical data cooperate the accelerated shown). Taken together, these data suggest that the bromodo- fluorescence recovery after photobleaching (FRAP) reported main of SMARCA2 is dispensable for chromatin binding and for PFI-3 (35), and we conclude that PFI-3 is a selective, SWI/SNF oncogenic activity in lung cancer. cell-permeable probe suitable to study the inhibition of SMARCA2/4 bromodomains in cells. PFI-3 treatment of synovial sarcoma cells and target gene promoter occupancy studies PFI-3 does not phenocopy the growth-inhibitory effects of As alterations in SWI/SNF have been implicated in disease SMARCA2 knockdown in lung cancer progression of synovial sarcomas (13), we also evaluated the Armed with PFI-3 and motivated by the context-specific phe- pharmacologic activity of PFI-3 in Yamato and Aska cells. These notype of SMARCA2 depletion (Fig. 2), we evaluated PFI-3 in the cells harbor the hallmark recurrent chromosomal translocation SMARCA4-deficient responder lines (A549, H1299, H157), but t(X;18)(p11.2;q11.2), which fuses the SS18 gene, an integral observed no antiproliferative effects in either 3-day cell viability subunit of SWI/SNF complex, to one of the three SSX genes (SSX1,

www.aacrjournals.org Cancer Res; 75(18) September 15, 2015 3869

Vangamudi et al.

SSX2, and SSX4), observed in >95% of patients (36). Incorpo- binding of the SWI/SNF complex to the SOX2 locus, but observed ration of the SS18-SSX fusion protein into the SWI/SNF complex only a minor change in SOX2 promoter occupancy, suggesting results in eviction and degradation of the tumor-suppressor inefficient inhibition of SWI/SNF binding (Fig. 5D). These data SMARCB1. The altered SWI/SNF complex binds to the SOX2 are consistent with the in situ cell extraction results for PFI-3 locus, resulting in aberrant SOX2 expression, which is essential (Fig. 4C and D) showing that SMARCA2/4 bromodomain inhi- for proliferation of synovial sarcomas (13). Accordingly, Yamato bition cannot displace the multisubunit SWI/SNF complex from and Aska cells show high levels of Sox2 expression (37). Hence, we chromatin. hypothesized that PFI-3 may inhibit the altered SWI/SNF complex and impair cell growth, but we did not observe inhibition of cell PFI-3 treatment of SMARCA4-dependent rhabdoid cancer or proliferation in either 4-day viability (Fig. 5A) or long-term leukemia cells proliferation assays (Fig. 5B). We then assessed SOX2 expression Rhabdoid tumors are distinctly characterized by biallelic inac- and found that PFI-3 treatment (day 3 and day 6) failed to reduce tivation of SMARCB1, a core subunit of the SWI/SNF complex SOX2 transcript levels at pharmacologically relevant concentra- (10), and genetic studies have demonstrated that oncogenesis tions (Fig. 5C and Supplementary Fig. S5). mediated by SMARCB1 loss is dependent on the residual activity To elaborate on the lack of inhibitor-induced phenotype, we of SMARCA4-containing SWI/SNF complex (38). To establish a examined SWI/SNF binding at the transcriptionally active SOX2 benchmark for PFI-3 treatment of A-204 and G-401 rhabdoid promoter. SMARCA4 ChIP demonstrated high SWI/SNF complex tumor cells, we identified two shRNAs that produced effective occupancy at the SOX2 promoter in Yamato cells, as previously (>80%) SMARCA4 protein knockdown (Fig. 5E) and confirmed reported (13), while no enrichment was observed at the MYOD1 inhibition of cell viability in both short-term proliferation and locus, a transcriptionally silent locus as confirmed by RNA Poly- long-term clonogenic assays (Fig. 5F and G). In contrast with the merase II ChIP (Fig. 5D and Supplementary Fig. S5D). Impor- RNAi phenotype, pharmacologic bromodomain inhibition did tantly, we also examined whether PFI-3 treatment could impair not impact the growth of rhabdoid cancer cells (Fig. 5H).

A B SMARCA2 SMARCA4 control) of

PFI-3 (%

n Bound bromodomain SMARCA2 isoform A: Kd = 110 nmol/L ( = 2) n SMARCA2 isoform B: Kd = 72 nmol/L ( = 4) SMARCA4: K = 55 nmol/L (n = 2) Figure 3. d PFI-3 Log (µmol/L) PFI-3 is a potent, selective, and cell permeable bromodomain inhibitor of C D SMARCA2/4. A, chemical structure of DMSO PFI-3 (10 µmol/L) PFI-3 (30 µmol/L) PFI-3 and biochemical potency (BROMOScan Kd's). B, BROMOScan dose–response curves using recombinant puriﬁed bromodomains. C, PFI-3 selectivity (2 mmol/L) across 32 bromodomains (DiscoverRx). D, in situ cell extraction of HeLa cells expressing GFP-tagged SMARCA2 bromodomain (green) cotreated with SAHA (5 mmol/L) and PFI-3 (or DMSO control) for 2 hours with Hoescht nuclear counterstain (red). HeLa control cells expressing GFP- tagged BRD4 treated (2 hours) with JQ1. E E, displacement of the SMARCA2 bromodomain from chromatin (IC50) quantiﬁed based on mean GFP signal per nucleus (SD; n ¼ 6). nucleus

per DMSO JQ1 (1 µmol/L) JQ1 (10 µmol/L)

µ IC50 =5.78 mol/L±0.22 normalized to DMSO control) (mean GFP intensity SMARCA2 chroman binding (%) PFI-3 Log (µmol/L)

3870 Cancer Res; 75(18) September 15, 2015 Cancer Research

Targeting the SMARCA2/4 Bromodomain using RNAi and PFI-3

A B

DMSO DMSO 78 nmol/L 156 nmol/L 312 nmol/L 625 nmol/L

A549 H157 H1299 H460 Growth (% of control) Figure 4. PFI-3 Log (µmol/L) 1.25 µmol/L 2.5 µmol/L 5 µmol/L 10 µmol/L 20 µmol/L 40 µmol/L Pharmacologic inhibition of SMARCA2/ 4 bromodomain in lung cancer. A, C D viability of SMARCA4-deficient (A549, DMSO PFI-3 (30 µmol/L) shSMARCA2 H1299 and H157) or SMARCA4- proficient (H460) cells following PFI-3 treatment (72 hours). Error bars, SD; n ¼ 3. B, A549 clonogenic assay (PFI-3 1,000 and media replenished every three days for 1.5 weeks). C, in situ cell extraction 800 (A549 cells) treated with PFI-3 or JQ1 600 (IF) intensity control for 2 hours followed by immunofluorescence staining for 400 endogenous, chromatin-bound nuclear 200 bromodomain (green), and Hoescht nuclear counterstain (red). D and E, 0 Mean immunofluorescence quantification (D) DMSO PFI-3 shSMARCA2 µ using SMARCA2 knockdown as (30 mol/L) specificity control with corresponding immunoblot confirmation (E). IF, E immunofluorescence. shLuc shSMARCA2 µ DMSO JQ1(25 n mol/L) JQ1(10 mol/L) SMARCA2

Tubulin

O 93% Knockdown

Finally, we extended our phenotypic evaluation of PFI-3 to to in situ cell extraction and discovered that the SMARCA2 mutants leukemia as previous RNAi studies have shown that AML cells bound chromatin similarly to that of WT (Fig. 7A and B). Hence, depend on SMARCA4 to support oncogenic transcriptional pro- failure of the ATP-Dead construct to rescue is due to lack of grams (30, 31). Similar to our ﬁndings across lung, synovial catalytic activity and not due to gross impairment in chromatin sarcomas, and rhabdoid tumor cell lines, PFI-3 treatment did not binding. Altogether, our genetic assessment clearly demonstrates afford an anticancer phenotype in THP-1 and MV4-11 leukemic that SMARCA4-deﬁcient cancer cells do not require a functional cells (Supplementary Fig. S6) highlighting the critical importance SMARCA2/4 bromodomain for growth. Instead, we unequivo- of pharmacologic drug target validation as a follow-up to RNAi- cally identify the catalytic activity of the ATPase domain as the mediated knockdown studies. appropriate, albeit more challenging, small-molecule drug target.

Synthetic lethality of SMARCA2 knockdown is linked to the Genome-wide microarray analysis of SMARCA2/4 rescue catalytic ATPase activity experiments To genetically validate the PFI-3 results, we next used a 30UTR To examine the dependency on ATPase activity, we generated targeting shRNA (shS2) to knockdown endogenous SMARCA2 microarray expression data (GSE69088) for the above cDNA in H1299 cells engineered to express either wild-type (WT), rescue experiments. Unsupervised clustering of the top variable ATP-binding pocket deficient (K755A; ref. 20) or bromodomain genes revealed three distinct expression profiles that were robust mutant (N1482W; ref. 39) forms of SMARCA2 (Fig. 6A). Ectopic to the gene set size while clustering (Fig. 7C and D). In the absence expression of either SMARCA2 WT or the bromodomain binding- of SMARCA2 knockdown (shLuc), all H1299 derivative lines deficient mutant (BRD-Mut), but not the ATPase-dead form clustered together (Group 1) irrespective of the nature of the (ATP-Dead), completely rescued the RNAi-mediated LOF pheno- ectopically expressed SMARCA2 constructs. On the other hand, type (Fig. 6B and C). Likewise, A549 cells reconstituted with the SMARCA2 knockdown cells (shSMARCA2) showed strong SMARCA4 WT or BRD-Mut (N1540W, Y1497F), but not ATP- differential gene expression defining two distinct clusters: cells Dead (K785A), were able to grow upon SMARCA2 knockdown rescued with either WT or BRD-Mut (Group 2) versus cells (Fig. 6D–F). We also subjected the isogenic matched-pair cell lines expressing either vector control (Ctrl) or ATP-Dead (Group 3).

www.aacrjournals.org Cancer Res; 75(18) September 15, 2015 3871

Vangamudi et al.

A B Yamato (PFI-3) DMSO 3 µmol/L 10 µmol/L Yamato 30 µmol/L Aska number Aska (PFI-3)

HeLa cell DMSO 3 µmol/L 10 µmol/L Viable Growth (% of control) 30 µmol/L

PFI-3 Log (µmol/L) Time (days)

CDSOX2 locus MYOD1 locus 2.0 0.010

1.5 0.008 IgG

0.006 SMARCA4 expression 1.0 0.004 Enrichment

(Sox2/GAPDH) 0.5 0.002 (relave to 5% input)

Relave mRNA 0.0 0.000 DMSO 3 10 30 DMSO 3 10 DMSO 3 10 PFI-3 (µmol/L) PFI-3 (µmol/L) PFI-3 (µmol/L)

E F shRNA (SMARCA4) G shS4-4 shS4-5 shLuc shLuc shS4-4 shS4-5 SMARCA4 A-204

shLuc

A-204 shS4-4 Tubulin shS4-5

SMARCA4 401 - G G-401 A-204 Tubulin

H DMSO DMSO 78 nmol/L 156 nmol/L 312 nmol/L 625 nmol/L G-401 shLuc Growth (% me = day 0) shS4-4 shS4-5 G-401

1.25 µmol/L 2.5 µmol/L 5 µmol/L 10 µmol/L 20 µmol/L 40 µmol/L Time (days)

Figure 5. Evaluation of PFI-3 in synovial sarcoma and rhabdoid tumor cells. A, viability of synovial sarcoma (Aska and Yamato) and HeLa cells treated with PFI-3 (96 hours) relative to DMSO-treated controls (SEM; n ¼ 3). B, long-term (2-week) proliferation assay. Cells were split and replenished with fresh media/PFI-3 every 3 or 4 days counting viable cells (SEM; n ¼ 3). C, PFI-3 treatment (3 days) does not repress SOX2 expression in Yamato cells. SOX2 transcript levels (RT-qPCR) normalized to GAPDH (SEM; n ¼ 12). D, control (DMSO) and PFI-3-treated Yamato cells (day 3) subjected to anti- SMARCA4 ChIP followed by qPCR for SOX2 promoter regions (target gene) or MYOD1 exon1 locus (negative control). The decrease in occupancy at theSOX2locus(10mmol/L) is small but signiﬁcant. , P 0.05 (SEM, n ¼ 9). E–G, A-204 and G-401 rhabdoid cells transduced with SMARCA4-targeting (shS4-4, shS4-5) or control (shLuc) shRNAs and analyzed for protein knockdown (1 week after puromycin selection; E), colony formation (2–3 weeks after puromycin; F), and viability (CellTiter-Glo; 6 days after puromycin; G). Error bars, SD; n ¼ 6. H, PFI-3 does not impair growth of G-401 cells (clonogenecity 1.5 weeks; similar data for A-204 not shown). Media/PFI-3 was replenished every 3 days.

3872 Cancer Res; 75(18) September 15, 2015 Cancer Research

Targeting the SMARCA2/4 Bromodomain using RNAi and PFI-3

A SMARCA2 cDNA B shLuc shSMARCA2 Control WT BRD-Mut ATP-dead RNAi: Luc shS2 Luc shS2 Luc shS2 Luc shS2 Vector Control SMARCA2 (An-HA)

SMARCA2 SMARCA2 Tubulin WT C

Figure 6. SMARCA2

Rescue experiments highlight the BRD-Mut importance of ATPase activity for cancer-speciﬁc vulnerability. A, SMARCA2 cDNA rescue experiments in

H1299 cells transduced with SMARCA2- (% of control) targeting (50UTR) shRNA (shS2) or SMARCA2

control shRNA (shLuc) and analyzed by Colony-forming units (CFU) ATP-dead immunoblotting 10 days after Control WT BRD-Mut ATP-dead puromycin selection. B and C, in parallel, shLuc shSMARCA2 the isogenic cell lines were seeded in 6- well plates (24 hours after puromycin selection) and after 2 weeks stained by D SMARCA4 cDNA E shLuc shSMARCA2 crystal violet (B) and colony forming units (CFU) were quantiﬁed (C). D, Control WT BRD-Mut ATP-dead SMARCA4 cDNA rescue/reconstitution RNAi: Luc shS2 Luc shS2 Luc shS2 Luc shS2 experiments in A549 cells transduced Vector with indicated shRNAs and analyzed by SMARCA4 Control immunoblotting 10 days after puromycin selection. E and F, clonogenic assay (crystal violet SMARCA2 staining; 1.5 weeks; E) and SMARCA4 ﬁ quanti cation of colony forming Tubulin WT units (SD; n ¼ 3; F). F

SMARCA4 BRD-Mut

(% of control)

SMARCA4 Colony-forming units (CFU) Control WT BRD-Mut ATP-dead ATP-dead shLuc shSMARCA2

Notably, expression/rescue using either SMARCA2 or SMARCA4 whether similar gene expression programs may account for the showed identical clustering behavior, and differences in mRNA observed functional complementation. To establish a frame- levels within the three groups were nonsignificant. Hence, the work for this analysis, we first derived gene expression signa- transcriptional profiles reinforce the view that BRD-Mut is able to tures for SMARCA4 expression in A549 cells (in the context of perform similar functions to the WT gene while ATP-Dead, despite SMARCA2 knockdown) comprising the top-100 upregulated retaining its ability to bind chromatin (Fig. 7A and B), cannot. and downregulated genes, respectively (Supplementary Table Consistent with the phenotypic responses (Fig. 6), gene set S4). Using GSEA, we then looked for enrichment of these enrichment analysis (GSEA) revealed upregulation of apoptosis signatures with gene lists from SMARCA2 rescue experiments and death pathways in Group 3 versus the rescued cell lines as queries (Fig. 7E and F and Supplementary Table S5). When (Group 2) highlighting the observed context-specific synthetic comparingATP-DeadtoWT,theenrichmentprofiles suggest lethality (Supplementary Fig. S8). that the ATPase enzymatic activity preferentially reverses expression of genes that are upregulated upon RNAi-mediated The ATPase activity common between SMARCA2 and SMARC4 (synthetic lethal) knockdown of SMARCA2 (Fig. 7E). Genes shares a suppressive function on gene expression programs downregulated upon SMARCA2 knockdown were not fully Next, focusing on the requirement for ATPase activity, we reversed by the rescue (Fig. 7F and Supplementary Table S5). compared the SMARCA2 and SMARCA4 rescue profiles to see Therefore, our microarray data clearly show that the ATPase

www.aacrjournals.org Cancer Res; 75(18) September 15, 2015 3873

Vangamudi et al.

A Control WT BRD-Mut ATP-dead B -SMARCA2 HA % chroman bound Mean nuclear intensity (HA) Hoechst 3334 Ctrl WT BRD-Mut ATP-dead

Group 1 Group 3 Group 2 C Group 1 Group 2 Group 3 D

SMARCA2: WT Ctrl BRD* ATP* WT BRD* Ctrl ATP* SMARCA4: Ctrl ATP* WT BRD* Ctrl ATP* WT BRD* shLuc shSMARCA2 shLuc shSMARCA2 E SMARCA2-KD signature (up)/SMARCA4 rescue (down) G SMARCA4 SMARCA2 (P FDR) (0,0) G2M_Checkpoint (0.04, 0.11) (0, 0.006) P53_Pathway (0.001, 0.009) (0.001, 0.012) Mitoc_Spindle (0.001, 0.007) (0.067, 0.137) EM_Transion (0, 0.002) (0.051, 0.045) ATP-Dead WT KRAS.50_UP.V1_DN (0.154, 0.212) (0.073, 0.139) Cyclin_D1_KE_V1_UP (0.06, 0.166) (0, 0) KRAS.300_UP.V1_DN (0.071, 0.168) (0.035, 0.091) KRAS.DF.V1_UP (0.002, 0.041 (0, 0.001) F SMARCA2-KD signature (down)/SMARCA4 rescue (up) P53_DN.V1_DN (0, 0.016) (0.001, 0.002) K_DNA Replicaon (0.007, 0.241 (0.004, 0.03) Integrin3_Pathway (0.006, 0.284) (0, 0.008) DNA_Replicaon (0.004, 0.26 (0, 0.011) NABA_ECM_Glycoproteins (0, 0.225) (0.01, 0.07) NABA_Basement_Membranes (0.003, 0.193) ATP- daeD WT (0.001, 0.013) Lagging_Strand_Synthesis (0.004, 0.199) (0.001, 0.017) Integrin1_Pathway (0, 0.035)

Normalized enrichment score

3874 Cancer Res; 75(18) September 15, 2015 Cancer Research

Targeting the SMARCA2/4 Bromodomain using RNAi and PFI-3

domain of SMARCA2/4 complements each other at the tran- SMARCA4-deficient patient population predicted to depend scriptional level exerting similar suppressive function on spe- exclusively on SMARCA2 activity (Fig. 1). Because SMARCA2- cific gene expression programs. deficient mice are viable showing no overt phenotype (42), while Next,usingGSEA,weidentified biologic pathways/states SMARCA4 inactivation is embryonic lethal (43), one might shared by both rescue experiments leveraging annotated path- anticipate a significant therapeutic window, if selective small- ways from the Molecular Signature Database (mSigDB; The molecule SMARCA2 inhibitors can be developed that mimic the Broad Institute). GSEA identified enrichment of cell prolifera- RNAi knockdown phenotype. tion (cell cycle, cyclin D1, G2–M checkpoint), chromatin remo- SMARCA2 contains an ATPase and bromodomain, suggesting deling (mitotic spindle, DNA replication and synthesis), and at least two tractable avenues for inhibitor development. Target- tumorigeneis (EMT, integrin, KRAS and P53 pathway signa- ing the acetyl-lysine recognition function of SWI/SNF bromodo- tures). These gene expression programs, which are enriched mains (e.g., the SMARCA2/4, PBRM1, BRD7, and BRD9 subunits) in rescued cells compared with cells lacking ATPase activity represents an unexplored opportunity for perturbation of the common to SMARCA2/4, have bonafide tumorigenic functions SWI/SNF complex. Recently, the anticancer activity of BET bro- (Fig. 7G). modomains inhibitors has fueled the development of novel In conclusion, the SMARCA4 cDNA complementation (i.e., re- chemical scaffolds that selectively target other bromodomains expression) and the SMARCA2 RNAi rescue experiments are (6, 24, 25), and the PFI-3 inhibitor exemplifies one such novel consistent with the observed lack of pharmacologic activity of chemical probe. However, despite being broadly available from PFI-3 in SWI/SNF-mutant cancers, and we demonstrate for the SGC, no phenotypic data have yet been reported. Hence, we first time that selective SMARCA2/4 bromodomain inhibition is subjected PFI-3 to rigorous biochemical and cellular characteri- not a feasible therapeutic strategy for targeting aberrant SWI/SNF zation confirming its exquisite selectivity, potency and cell per- activity in SWI/SNF-mutant cancers. Instead, drug discovery meability (Fig. 3). Such pharmacodynamics studies are a critical efforts should be focused on inhibiting the ATPase catalytic component of drug target validation studies as they provided activity to deliver on the promise of robust, cancer-specific syn- confidence that a compound-induced phenotype (or lack thereof) thetic lethal therapy. correlate with biochemical target engagement in cells. Surprisingly, in contrast with SMARCA2 knockdown, PFI-3 did not display any antiproliferative phenotype in SMARCA4-defi- Discussion cient lung cell lines across a variety of biologic assays. Likewise, in A common theme has emerged from genetic studies where models harboring defined SWI/SNF alterations, including syno- imbalances between various paralogous subunits within SWI/ vial sarcoma (SSX-fusion), rhabdoid tumors (SMARCB1-null), SNF (e.g., SMARCA2/4 and ARID1A/B) can render cells more and leukemia (SMARCA4-dependent), PFI-3 did not mimic the tumorigenic and simultaneously hypersensitive to targeting of the anticancer phenotype observed upon RNAi-mediated knock- residual complex (13, 20–23, 40, 41). However, despite the high down of SMARCA2/4 (31, 38, 44). Mechanistically, and consis- prevalence of genomic lesions in SWI/SNF, studies have not tent with the lack of cellular phenotype, we discovered that PFI-3 addressed how this observation can be translated into effective, cannot displace endogenous SMARCA2 (i.e., lung) or SMARCA4 drug discovery endpoints. (i.e., synovial sarcoma) from chromatin potentially due to the In this study, we first demonstrated context-specific antiproli- activity of other chromatin-interacting SWI/SNF subunits ferative phenotype of SMARCA2 depletion in SMARCA4-deficient highlighting challenges in targeting large protein complexes. This lung cancer using multiple, non-overlapping hairpins, as well as result is in sharp contrast with efficient chromatin displacement of independent siRNAs, validating the synthetic–lethal relationship endogenous BRD4 by JQ1 (Fig. 4C), and we note the contrasting between SMARCA2 and SMARCA4 (20–22). We further showed feature of the BET family of tandem bromodomains, which are that expression of either SMARCA2 or SMARCA4 completely not flanked by other known regulatory or conserved domains. rescued the effects of SMARCA2 knockdown in SMARCA4-defi- Recent studies have highlighted the role of residual SWI/SNF cient cells, indicating paralog dependence and reciprocal role of complex along with paralog dependence, indicating a potential these two subunits in tumorigenesis. The functional complemen- combinatorial role of chromatin-interacting domains in SWI/SNF tation of SMARCA2/4 was also evident at the transcriptional level recruitment (5). Our data further highlight the need to conduct where the ATPase activity appears to control gene programs similar target identification/validation studies of other paralog related to proliferation, cell cycle, and chromatin remodeling. subunits like ARID1A and ARID1B that form mutually exclusive Furthermore, genomic analysis revealed mutual exclusivity of SWI/SNF complexes and display a synthetic lethal relationship SMARCA2 and SMARCA4 mutations in LUSC and LUAD carci- (40). Additional vulnerabilities like antagonism between noma and expression-based biomarker analysis outlined a SMARCB1 and EZH2, which renders rhabdoid tumors dependent

Figure 7. Chromatin binding and gene set enrichment highlights the importance of ATPase catalytic activity for cancer-specific vulnerability. A, immunofluorescent images (H1299 cells) expressing either vector control or HA-tagged SMARCA2 wild-type, bromodomain-mutant or ATPase-dead constructs (red), and Hoechst counterstain (blue). B, quantification of chromatin binding (normalized to non-extracted immunofluorescent signal). C and D, clustering of 1,000 most variable genes for SMARCA2 (C) and SMARCA4 rescue experiments (D). E and F, SMARCA2 knockdown signatures (derived from A549 cells reconstituted with SMARCA4) comprising upregulated (up; E) and downregulated (down; F) genes. The ranked gene list (x-axis) was derived fromtheSMARCA2rescueexperimentscomparingATP-Dead with WT as query. Genes responding differently (interaction contrast, Group 3 vs. Group 2) were ranked according to their P values with direction provided by the fold change. G, significantly enriched gene sets (mSigDB) shared between SMARCA2 and SMARCA4 focusing on their ATPase activity (i.e., WT vs. ATP-Dead cDNA expression).

www.aacrjournals.org Cancer Res; 75(18) September 15, 2015 3875

Vangamudi et al.

on EZH2 for disease maintenance (45), present another promis- clusters in the human genome— related SMARCA proteins and ing approach to target SWI/SNF-mutant cancers. The antiproli- a class of DNA helicases, suggesting possibility for achieving ferative response to SMARCA4 knockdown and EZH2 inhibitor exquisite selectivity over other ATPases. An obvious challenge treatment highlights clear dependencies on SWI/SNF activity (Fig. would be to obtain selectivity over SMARCA4 as these enzymes 5E–Hand Supplementary Fig. S7A). However, it remains to be are highly homogenous in their active site and dual inhibition investigated whether other SWI/SNF-mutant cancers are sensitive in normal cells could limit the therapeutic window (29). to EZH2 inhibition since we did not see activity of the EZH2 Structural insights often guide the design of selective inhibitors; inhibitor in SMARCA4-deficient lung cancer cells (Supplementary however, very few X-ray crystal structures for ATPase domains Fig. S7B). are currently available for SMARCA2/4-related proteins with For SWI/SNF-mutant cancers, our target validation approach most being in open inactive conformation, like the yeast Chd1 has focused on dissecting the functional contribution of an ATPase domain, highlighting the need for furthering structural impaired SMARCA2/4 bromodomain or ATPase domain to biology. Another barrier for pharmaceutical development of cellular phenotype through parallel cDNA complementation selective ATPase inhibitors is the current lack of commercial and rescue experiments. The observation that expression of high-throughput screening assays and selectivity panels against BRD-Mut, but not ATP-Dead, can rescue the SMARCA2 knock- the large family of ATP-binding proteins. Nevertheless, struc- down phenotype is consistent with the pharmacologic PFI-3 tural diversity in the vicinity of the nucleotide-binding sites, inhibitor data. Thus, our genetic and chemical findings con- including possible allosteric sites, should enable SMARCA2 verge and we unequivocally conclude that small-molecule ATPase drug discovery supported by prior identification of inhibition of the bromodomain is dispensable for the ability potent and selective inhibitors of the ATPase activity of KIF11, of the SWI/SNF complex in controlling tumor growth. As such, Hsp90, and VCP (49, 50). The recent development of selective the present study is the first to deprioritize SMARCA2/4 bro- and cell-potent covalent inhibitors that block ATP binding, as modomain inhibition as a tractable target in genetically defined well as allosteric inhibitors that impair nucleotide turnover for lung, synovial sarcoma, leukemia, and rhabdoid tumors. How- the VCP ATPase (50), is also an encouraging avenue for the ever, we cannot exclude that compounds with a selectivity developmentofinhibitorstargeting the SMARCA2 ATPase profile that simultaneously inhibits additional bromodomains catalytic activity. in SWI/SNF (e.g., PBRM1, BRD7, and BRD9 subunits) could be Taken together, our target validation studies identify the an efficacious strategy (although pleiotropic bromodomain SMARCA2 ATPase domain, but not the bromodomain, as a inhibition in normal cells could be a potential concern). tractable, albeit more challenging therapeutic target for a well- Although the importance of ATPase activity has previously defined SMARCA4-deficient patient population representing been shown for chromatin remodeling (20), our studies pin- more than 20,000 patients a year in the United States alone point the ATPase activity as the molecular synthetic-lethal (i.e., 10%–20% of NSCLC cases). Moreover, the SMARCA4- target providing a genetically validated strategy for targeting deficient patient population generally lacks targetable oncogenes SWI/SNF-mutant lung cancer. (such as mutant EGFR or ALK translocations; ref. 20), which ATPases represents a large and diverse family of proteins, many further emphasize the potential medical impact of developing of which perform chaperone-like functions assisting in the assem- inhibitors of the ATPase domain of SMARCA2/4. bly, operation, and disassembly of protein complexes (46). Not surprisingly, because numerous cellular processes are driven by energy-dependent conformational changes in multisubunit com- Disclosure of Potential Conflicts of Interest plexes, ATPases have been implicated in various human diseases T.A. Paul reports receiving commercial research grant from Pfizer, Inc. No with several inhibitors in clinical use. However, most of these do potential conflicts of interest were disclosed by the other authors. not directly engage/bind the nucleotide-binding site (47). Devel- oping potent inhibitors that must compete with intracellular concentrations of ATP (2–10 mmol/L) have been challenging. Authors' Contributions Phosphate groups contribute significant nucleotide-binding affin- Conception and design: B. Vangamudi, T.A. Paul, P.K. Shah, E. Leo, M. Peoples, ity, but to overcome poor cell-permeability of negatively-charged T.P. Heffernan, D. Verhelle, P. Jones, S. Sharma, J.N. Andersen phosphate groups, the majority of synthetic ATP analogues are Development of methodology: B. Vangamudi, T.A. Paul, M. Kost-Alimova, X. Shi, E. Leo, H.S. Mahadeshwar, T.N. Tieu, M. Peoples, S. Sharma, J.N. Andersen devoid of highly charged phospho-mimetic groups (47). More- Acquisition of data (provided animals, acquired and managed patients, over, the high sequence homology between ATP-binding sites provided facilities, etc.): B. Vangamudi, T.A. Paul, M. Kost-Alimova, L. Notte- (among ATPases and other ATP-binding proteins) represents a baum, Y. Zhan, A. Protopopov, M. Peoples, A. Petrocchi, S. Sharma profound selectivity challenge. Overall, there is a need for the Analysis and interpretation of data (e.g., statistical analysis, biostatistics, development of novel, potent, and bioavailable ATPase inhibi- computational analysis): B. Vangamudi, T.A. Paul, P.K. Shah, M. Kost-Alimova, tors, and the successful design of ATP-competitive kinase inhibi- L. Nottebaum, E. Leo, H.S. Mahadeshwar, A. Protopopov, A. Futreal, C. Toniatti, D. Verhelle, P. Jones, J.N. Andersen tors, yet another class of ATP-binding enzymes, supports at least in Writing, review, and/or revision of the manuscript: B. Vangamudi, T.A. Paul, principal, the feasibility of targeting the ATP-binding site of P.K. Shah, M. Kost-Alimova, L. Nottebaum, X. Shi, A. Futreal, T.P. Heffernan, SMARCA2. J.R. Marszalek, C. Toniatti, A. Petrocchi, D. Verhelle, D.R. Owen, G.F. Draetta, SMARCA2 belongs to the SNF2 family of chromatin remo- P. Jones, W.S. Palmer, S. Sharma, J.N. Andersen deling ATPases and contains most of the conserved motifs Administrative, technical, or material support (i.e., reporting or organizing found in SF2 helicases (48). However, SMARCA2/4 share little data, constructing databases): B. Vangamudi, T.A. Paul, Y. Zhan, G.F. Draetta Study supervision: T.A. Paul, D. Verhelle, G.F. Draetta, S. Sharma, J.N. Andersen overall sequence homology with other helicases and even less Other (synthesis of PFI-3 used for the studies as well as obtained the profiling homology with other ATPases (46, 48). Moreover, the sequence data): W.S. Palmer homology of the SMARCA2 ATPase is limited to only two SNF2 Other (bioinformatics): P. Shah.

3876 Cancer Res; 75(18) September 15, 2015 Cancer Research

Targeting the SMARCA2/4 Bromodomain using RNAi and PFI-3

Acknowledgments advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate The authors thank Dr. Daniel K. Treiber at DiscoveRx for custom assay this fact. development, Drs. Norifumi Naka and Kazuyuki Itoh for the Aska and Yamato cells, and Dr. Chang-gong Liu for microarray services. The costs of publication of this article were defrayed in part by the Received January 5, 2015; revised May 31, 2015; accepted June 15, 2015; payment of page charges. This article must therefore be hereby marked published OnlineFirst July 2, 2015.

References 1. Geutjes EJ, Bajpe PK, Bernards R. Targeting the epigenome for treatment of 23. Helming KC, Wang X, Roberts CW. Vulnerabilities of mutant SWI/SNF cancer. Oncogene 2012;31:3827–44. complexes in cancer. Cancer Cell 2014;26:309–17. 2. Garraway LA, Lander ES. Lessons from the cancer genome. Cell 2013;153: 24. Filippakopoulos P, Picaud S, Mangos M, Keates T, Lambert JP, Barsyte- 17–37. Lovejoy D, et al. Histone recognition and large-scale structural analysis 3. Kadoch C, Hargreaves DC, Hodges C, Elias L, Ho L, Ranish J, et al. of the human bromodomain family. Cell 2012;149:214–31. Proteomic and bioinformatic analysis of mammalian SWI/SNF complexes 25. Garnier JM, Sharp PP, Burns CJ. BET bromodomain inhibitors: a patent identifies extensive roles in human malignancy. Nat Genet 2013;45: review. Expert Opin Ther Pat 2014;24:185–99. 592–601. 26. Tibshirani R, Hastie T. Outlier sums for differential gene expression 4. Shain AH, Pollack JR. The spectrum of SWI/SNF mutations, ubiquitous in analysis. Biostatistics 2007;8:2–8. human cancers. PLoS ONE 2013;8:e55119. 27. Fabian MA, Biggs WH III, Treiber DK, Atteridge CE, Azimioara MD, 5. Wilson BG, Roberts CW. SWI/SNF nucleosome remodellers and cancer. Nat Benedetti MG, et al. A small molecule-kinase interaction map for clinical Rev Cancer 2011;11:481–92. kinase inhibitors. Nat Biotechnol 2005;23:329–36. 6. Filippakopoulos P, Knapp S. Targeting bromodomains: epigenetic readers 28. Komashko VM, Acevedo LG, Squazzo SL, Iyengar SS, Rabinovich A, O'Geen of lysine acetylation. Nat Rev Drug Discov 2014;13:337–56. H, et al. Using ChIP-chip technology to reveal common principles of 7. Shi J, Vakoc CR. The mechanisms behind the therapeutic activity of BET transcriptional repression in normal and cancer cells. Genome Res 2008; bromodomain inhibition. Mol Cell 2014;54:728–36. 18:521–32. 8.WeeS,DhanakD,LiH,ArmstrongSA,CopelandRA,SimsR,etal. 29. Bultman SJ, Herschkowitz JI, Godfrey V, Gebuhr TC, Yaniv M, Perou CM, Targeting epigenetic regulators for cancer therapy. Ann N Y Acad Sci et al. Characterization of mammary tumors from Brg1 heterozygous mice. 2014;1309:30–6. Oncogene 2008;27:460–8. 9. Tolstorukov MY, Sansam CG, Lu P, Koellhoffer EC, Helming KC, Alver BH, 30. Shi J, Whyte WA, Zepeda-Mendoza CJ, Milazzo JP, Shen C, Roe JS, et al. et al. Swi/Snf chromatin remodeling/tumor suppressor complex estab- Role of SWI/SNF in acute leukemia maintenance and enhancer-mediated lishes nucleosome occupancy at target promoters. Proc Natl Acad Sci U S A Myc regulation. Genes Dev 2013;27:2648–62. 2013;110:10165–70. 31. Zuber J, Shi J, Wang E, Rappaport AR, Herrmann H, Sison EA, et al. RNAi 10. Kim KH, Roberts CW. Mechanisms by which SMARCB1 loss drives rhab- screen identifies Brd4 as a therapeutic target in acute myeloid leukaemia. doid tumor growth. Cancer Genet 2014;207:365–72. Nature 2011;478:524–8. 11. Hohmann AF, Vakoc CR. A rationale to target the SWI/SNF complex for 32. Medina PP, Romero OA, Kohno T, Montuenga LM, Pio R, Yokota J, et al. cancer therapy. Trends Genet 2014;30:356–63. Frequent BRG1/SMARCA4-inactivating mutations in human lung cancer 12. Roberts CW, Leroux MM, Fleming MD, Orkin SH. Highly penetrant, rapid cell lines. Hum Mutat 2008;29:617–22. tumorigenesis through conditional inversion of the tumor suppressor gene 33. Thompson KW, Marquez SB, Reisman D. A synthetic lethality-based Snf5. Cancer Cell 2002;2:415–25. strategy to treat cancers harboring a genetic deficiency in the 13. Kadoch C, Crabtree GR. Reversible disruption of mSWI/SNF (BAF) com- chromatin remodeling factor BRG1-letter. Cancer Res 2014;74: plexes by the SS18-SSX oncogenic fusion in synovial sarcoma. Cell 4946–7. 2013;153:71–85. 34. Yang M, Sun Y, Ma L, Wang C, Wu JM, Bi A, et al. Complex alternative 14. Jones S, Wang TL, Shih Ie M, Mao TL, Nakayama K, Roden R, et al. Frequent splicing of the smarca2 gene suggests the importance of smarca2-B variants. mutations of chromatin remodeling gene ARID1A in ovarian clear cell J Cancer 2011;2:386–400. carcinoma. Science 2010;330:228–31. 35. Philpott M, Rogers CM, Yapp C, Wells C, Lambert JP, Strain-Damerell C, 15. Wiegand KC, Shah SP, Al-Agha OM, Zhao Y, Tse K, Zeng T, et al. ARID1A et al. Assessing cellular efficacy of bromodomain inhibitors using mutations in endometriosis-associated ovarian carcinomas. N Engl J Med fluorescence recovery after photobleaching. Epigenetics Chromatin 2010;363:1532–43. 2014;7:14. 16. Varela I, Tarpey P, Raine K, Huang D, Ong CK, Stephens P, et al. Exome 36. Clark J, Rocques PJ, Crew AJ, Gill S, Shipley J, Chan AM, et al. Iden- sequencing identifies frequent mutation of the SWI/SNF complex gene tification of novel genes, SYT and SSX, involved in the t(X;18)(p11.2; PBRM1 in renal carcinoma. Nature 2011;469:539–42. q11.2) translocation found in human synovial sarcoma. Nat Genet 17. Kim HS, Mendiratta S, Kim J, Pecot CV, Larsen JE, Zubovych I, et al. 1994;7:502–8. Systematic identification of molecular subtype-selective vulnerabilities in 37. Naka N, Takenaka S, Araki N, Miwa T, Hashimoto N, Yoshioka K, et al. non-small-cell lung cancer. Cell 2013;155:552–66. Synovial sarcoma is a stem cell malignancy. Stem Cells 2010;28: 18. Wassermann AM, Camargo LM, Auld DS. Composition and applications of 1119–31. focus libraries to phenotypic assays. Front Pharmacol 2014;5:164. 38. Wang X, Sansam CG, Thom CS, Metzger D, Evans JA, Nguyen PT, et al. 19. Blagg J, Workman P. Chemical biology approaches to target validation in Oncogenesis caused by loss of the SNF5 tumor suppressor is dependent on cancer. Curr Opin Pharmacol 2014;17:87–100. activity of BRG1, the ATPase of the SWI/SNF chromatin remodeling 20. Oike T, Ogiwara H, Tominaga Y, Ito K, Ando O, Tsuta K, et al. A complex. Cancer Res 2009;69:8094–101. synthetic lethality-based strategy to treat cancers harboring a genetic 39. Shen W, Xu C, Huang W, Zhang J, Carlson JE, Tu X, et al. Solution structure deficiency in the chromatin remodeling factor BRG1. Cancer Res 2013; of human Brg1 bromodomain and its specific binding to acetylated histone 73:5508–18. tails. Biochemistry 2007;46:2100–10. 21. Hoffman GR, Rahal R, Buxton F, Xiang K, McAllister G, Frias E, et al. 40. Helming KC, Wang X, Wilson BG, Vazquez F, Haswell JR, Manchester HE, Functional epigenetics approach identifies BRM/SMARCA2 as a critical et al. ARID1B is a specific vulnerability in ARID1A-mutant cancers. Nat Med synthetic lethal target in BRG1-deficient cancers. Proc Natl Acad Sci U S A 2014;20:251–4. 2014;111:3128–33. 41. OrvisT,HepperlaA,WalterV,SongS,SimonJM,ParkerJS,etal.BRG1/ 22. Wilson BG, Helming KC, Wang X, Kim Y, Vazquez F, Jagani Z, et al. Residual SMARCA4 inactivation promotes non-small cell lung cancer aggres- complexes containing SMARCA2 (BRM) underlie the oncogenic drive of siveness by altering chromatin organization. Cancer Res 2014;74: SMARCA4 (BRG1) mutation. Mol Cell Biol 2014;34:1136–44. 6486–98.

www.aacrjournals.org Cancer Res; 75(18) September 15, 2015 3877

Vangamudi et al.

42. Reyes JC, Barra J, Muchardt C, Camus A, Babinet C, Yaniv M. Altered control 46. Neuwald AF, Aravind L, Spouge JL, Koonin EV. AAAþ: A class of chaperone- of cellular proliferation in the absence of mammalian brahma (SNF2al- like ATPases associated with the assembly, operation, and disassembly of pha). EMBO J 1998;17:6979–91. protein complexes. Genome Res 1999;9:27–43. 43. Guidi CJ, Sands AT, Zambrowicz BP, Turner TK, Demers DA, Webster W, 47. Chene P. ATPases as drug targets: learning from their structure. Nat Rev et al. Disruption of Ini1 leads to peri-implantation lethality and tumor- Drug Discov 2002;1:665–73. igenesis in mice. Mol Cell Biol 2001;21:3598–603. 48. Ryan DP, Owen-Hughes T. Snf2-family proteins: chromatin remodellers 44. Buscarlet M, Krasteva V, Ho L, Simon C, Hebert J, Wilhelm B, et al. for any occasion. Curr Opin Chem Biol 2011;15:649–56. Essential role of BRG, the ATPase subunit of BAF chromatin 49. Rickert KW, Schaber M, Torrent M, Neilson LA, Tasber ES, Garbaccio R, et al. remodeling complexes, in leukemia maintenance. Blood 2014;123: Discovery and biochemical characterization of selective ATP competitive 1720–8. inhibitors of the human mitotic kinesin KSP. Arch Biochem Biophys 45. Wilson BG, Wang X, Shen X, McKenna ES, Lemieux ME, Cho YJ, 2008;469:220–31. et al. Epigenetic antagonism between polycomb and SWI/SNF 50. Magnaghi P, D'Alessio R, Valsasina B, Avanzi N, Rizzi S, Asa D, et al. complexes during oncogenic transformation. Cancer Cell 2010;18: Covalent and allosteric inhibitors of the ATPase VCP/p97 induce cancer 316–28. cell death. Nat Chem Biol 2013;9:548–56.

3878 Cancer Res; 75(18) September 15, 2015 Cancer Research

The SMARCA2/4 ATPase Domain Surpasses the Bromodomain as a Drug Target in SWI/SNF-Mutant Cancers: Insights from cDNA Rescue and PFI-3 Inhibitor Studies

Bhavatarini Vangamudi, Thomas A. Paul, Parantu K. Shah, et al.

Cancer Res 2015;75:3865-3878. Published OnlineFirst July 2, 2015.

Updated version Access the most recent version of this article at: doi:10.1158/0008-5472.CAN-14-3798

Supplementary Access the most recent supplemental material at: Material http://cancerres.aacrjournals.org/content/suppl/2015/07/02/0008-5472.CAN-14-3798.DC1

Cited articles This article cites 50 articles, 14 of which you can access for free at: http://cancerres.aacrjournals.org/content/75/18/3865.full#ref-list-1

Citing articles This article has been cited by 25 HighWire-hosted articles. Access the articles at: http://cancerres.aacrjournals.org/content/75/18/3865.full#related-urls

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department at Subscriptions [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://cancerres.aacrjournals.org/content/75/18/3865. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.