Functional epigenetics approach identifies BRM/SMARCA2 as a critical synthetic lethal target in BRG1-deficient cancers

Deep Coverage Pooled shRNA Screening Reveals BRM/SMARCA2 as a Synthetic Lethal Target in BRG1-Mutant Cancers.

Although RNAi has proven to be a powerful forward genetics approach, the robustness and reproducibility of RNAi screens has been challenged by the prevalence of off-target effects and inability to predict high-potency shRNAs with good confidence (22). In an effort to overcome these limitations, we constructed a deep coverage design shRNA (DECODER) library (Fig. 1A) to yield higher confidence hits through extensive shRNA coverage for each gene. The DECODER epigenome library contained 17 shRNAs per gene against a diverse collection of epigenetic regulators, with a particular focus on druggable classes, including those involved in the covalent modification of histones, and proteins with reader domains recognizing histone marks (SI Appendix, Fig. S1A and Dataset S1). This library was screened across a panel of 58 cancer cell lines from CCLE representing various primary sites and diverse genetic backgrounds (ref. 21; Dataset S2). The growth impact of shRNAs for each cell line was scored by calculating a z score based on the fold change in representation of each individual shRNA relative to its representation in the starting plasmid pool as measured by next generation sequencing (see SI Appendix, Fig. S1B and Methods and Dataset S2 for a detailed description of these calculations). In addition to scoring the individual shRNAs, we derived gene level calls from the 17 shRNAs for each gene by applying the Redundant siRNA Activity (RSA) algorithm, which calculates gene-centric P values (23). To identify genes whose product is selectively required for growth in a subset of cancer lines, we performed k-means clustering (24) of the RSA value for each gene to define groups of “sensitive” and “insensitive” cell lines and subsequently ranked hits based on the difference in cluster centers (SI Appendix, Methods). KRAS, which was included in the library as a positive control, emerged as one of the top differential genes from this analysis, having strong growth-inhibitory effects only in a subset of the cancer cell lines profiled. As expected, the differential activity reflected by the KRAS RSA score strongly correlated with KRAS mutation status (P = 9.65 × 10⁻¹²; Fig. 1B and SI Appendix, Fig. S1C). In contrast to the genotype selective activity of KRAS, some other genes included in this library such as PSMA3, which encodes a proteasome subunit, appeared to be broadly cytotoxic to all cell lines (SI Appendix, Fig. S1D). Notably, application of the ATARIS algorithm (25), which provides a statistical method for identifying shRNAs that share a common activity profile, revealed that 10 of 17 independent KRAS shRNAs displayed similar antiproliferative profiles (SI Appendix, Fig. S1E). Collectively, the assessment of these positive controls demonstrates the robustness of the DECODER lethality screen approach.

Intriguingly, the gene with the strongest robust differential lethal score from this epigenome library screen was BRM, a catalytic subunit of the mSWI/SNF chromatin remodeling complexes, ranking even higher than the KRAS positive control (Fig. 1C and SI Appendix, Fig. S2A). To identify whether any specific genetic or molecular feature correlates with sensitivity to BRM depletion, we performed a systematic interrogation of all features in the CCLE, including gene expression, copy number, and mutation status to identify features enriched in the sensitive cell lines as defined by the k-means clustering for BRM (21). Strikingly, loss-of-function mutations in the mSWI/SNF catalytic subunit BRG1 strongly correlated with sensitivity to BRM shRNAs (Fig. 2A, P = 2.03 × 10⁻⁷). Notably, the ATARIS solution for BRM identified 12 of the 17 BRM shRNAs in the library as showing a similar antiproliferative profile in BRG1-mutant cancer cells, thereby strongly supporting the notion that this differential lethality effect is due to on-target rather than off-target activity (SI Appendix, Fig. S2B).

BRM and BRG1 are closely related paralogs that function as mutually exclusive ATP-dependent catalytic subunits of the mSWI/SNF complexes (26). Although BRM and BRG1 are significantly conserved at the protein level, they display overlapping and distinct functions (27–29). The identification of BRM as a synthetic lethal hit in the context of BRG1 mutations raises the possibility that BRM is substituting for essential functions of the mSWI/SNF complex in BRG1-deficient cancer cells and, thus, creating a cancer-selective vulnerability. A prediction of this model would be that complete (i.e., homozygous) loss of BRG1 should lead to more pronounced BRM dependency compared with heterozygous loss of BRG1. Indeed, cancer cell lines with complete loss of BRG1 were highly sensitive to BRM shRNAs, whereas BRM shRNAs had little or no impact in cells that were either BRG1 wild type or with heterozygous BRG1 mutations (Fig. 2A and Dataset S3). The status of BRG1 protein expression in BRG1-mutant and wild-type (WT) lines was further confirmed by immunoblotting (Fig. 2B and SI Appendix, Fig. S3). Our data also suggest that sensitivity to BRM shRNAs in the BRG1-mutant setting is not only confined to lung cancer (as noted in the case of A549, H1299, and H838 lung cancer cell lines), which is the predominant indication in which BRG1 mutations have been reported, but also include ovarian (TYKNU) and liver (SKHEP1) cancer cell lines with BRG1 loss (Fig. 2A and Dataset S3). Collectively, these findings demonstrate that cells lacking a functional copy of BRG1 become exquisitely dependent on residual BRM containing mSWI/SNF complexes for their survival.

BRM Depletion Selectively Inhibits the Growth of BRG1-Mutant Cancer Cells.

To further examine the impact of BRM depletion on BRG1-deficient cells, we engineered several BRG1-deficient and WT cell lines with doxycycline (dox)-inducible shRNA constructs targeting BRM. In all three BRG1-mutant/deficient lung cancer cell lines tested (NCI-H838, NCI-H1299, and A549), induction of BRM shRNAs produced highly efficient depletion of BRM protein and led to profound growth inhibition in short-term proliferation and colony formation assays (Fig. 3 A, C, and E and SI Appendix, Fig. S4). Consistent with the results from the screening data, BRM knockdown with the same shRNAs that impacted growth in BRG1-mutant cancer cells, did not affect proliferation of cells with intact BRG1, such as the BRG1-WT lung cancer cell line NCI-H460 (Fig. 3 B, D, and F) and BEAS2B, a nontumorigenic immortalized lung epithelial cell line (Fig. 4 A and B and SI Appendix, Fig. S5). When we examined the effects of BRM depletion in CORL23 lung cancer cells, which harbor a heterozygous BRG1 lesion, we detected a modest impact on cell growth (SI Appendix, Fig. S6). Although this growth inhibitory effect was significantly less pronounced than in cells with homozygous loss-of-function BRG1 mutations, these findings raise the interesting possibility that heterozygous loss of BRG1 may already partially sensitize cells to BRM inhibition. Alternatively, this heterozygous BRG1 mutation, which results in an in-frame deletion, may lead to mild dominant negative effects. To further investigate the observed synthetic lethality, we tested whether expression of either BRM or BRG1 is sufficient to sustain cancer cell proliferation and whether cells can tolerate combined inactivation of BRM and BRG1. Indeed, although depletion of BRM or BRG1 did not impact the proliferation of two BRG1-WT cell lines, simultaneous knockdown of BRG1 and BRM led to marked growth inhibition in both of these BRG1-WT cell lines, strongly supporting the synthetic lethal relationship between BRM and BRG1 (Fig. 4 A–D and SI Appendix, Fig. S5).

BRM Knockdown Does Not Disrupt Association of Other Core and Accessory mSWI/SNF Subunits.

BRM and BRG1 are catalytic components and part of the core mSWI/SNF complex. We therefore sought to ascertain the impact of BRG1 mutations and BRM depletion on mSWI/SNF complex composition and stability. Purification of the mSWI/SNF complex via coimmunoprecipitation of core subunits such as SNF5 or BAF155, and size-exclusion chromatographic separation, showed that a subcomplex containing core and accessory subunits remained intact in BRG1-deficient cells (Fig. 5A and SI Appendix, Figs. S7 and S8). Moreover, knockdown of BRM in BRG1-mutant cancer cells did not appear to destabilize the complex (Fig. 5A and SI Appendix, Figs. S7 and S8). Although the majority of the complex stays intact, we noted that BAF53A no longer associates with the complex after BRM depletion (Fig. 5A). Because previous studies have shown that BAF53A directly interacts with the ATPase subunit of the complex (i.e., BRG1) (30), we speculate that BAF53A may interact with the “residual” BRM ATPase in BRG1-deficient cells, but dissociate from the complex once both ATPase subunits are absent. Overall, these findings indicate that the observed synthetic lethality cannot simply be explained by destabilization of the entire mSWI/SNF complex, but rather suggests that the specific inhibition of the redundant activity of BRM and BRG1 suffices to produce a marked growth defect. To further investigate complex composition in the absence of both ATPases, we examined SW13 cells, which lack BRG1 and BRM expression. Consistent with prior results, we detected robust association of the core and accessory SWI/SNF subunits, with the exception of BAF53A (SI Appendix, Fig. S9).

BRM Depletion Results in a Growth Arrest and Induction of H3K9me3 in BRG1-Mutant Cancer Cells.

We next sought to investigate the mechanism for growth inhibition in response to BRM depletion. Examination of cell cycle profiles in the BRG1-mutant cell lines NCI-H838 and NCI-H1299 indicated that BRM knockdown led to a prominent G₁ arrest (Fig. 5B and SI Appendix, Fig. S4 C and I) without appearance of a sub-G₁ population that would be indicative of cell death. Consistent with these results, we did not observe any signs of apoptosis upon BRM knockdown, as judged by Caspase 3 cleavage (SI Appendix, Fig. S10). The G₁ arrest was accompanied by the appearance of senescent cells as evidenced by flattened cell morphology and positive staining for acidic β-galactosidase (Fig. 5C and SI Appendix, Fig. S4J), suggesting that the growth inhibitory effect of BRM is mediated, at least in this subset of BRM-dependent lines, through induction of G₁ arrest and senescence. The growth inhibitory effect upon BRM knockdown appears to be irreversible as cells continued to remain growth arrested even upon withdrawal of dox (SI Appendix, Fig. S11).

Given the critical role of the mSWI/SNF complex in chromatin structure and function, we reasoned that global chromatin profiling may provide potential insights toward the molecular mechanisms associated with the synthetic lethal relationship between BRM and BRG1. Using quantitative mass spectrometry-based methods, we surveyed a variety of histone modifications in response to BRM depletion. Although most global histone marks remained unaffected, BRM knockdown induced a significant increase in H3K9me3 levels in NCI-H1299 cells (SI Appendix, Fig. S12). Immunofluorescence-based detection further confirmed the substantial increase in H3K9me3 staining upon BRM knockdown (Fig. 5D). Of note, H3K9me3 is a repressive histone mark that is characteristic of heterochromatic gene regions and can be associated with cells undergoing senescence (31). Thus, the marked increase in repressive H3K9me3 in response to BRM depletion in BRG1-mutant cells may be reflective of cells entering a growth arrest/senescent state.

BRM Knockdown Leads to Selective Growth Inhibition of BRG1-Mutant Tumors in Vivo.

The tumor microenvironment can, in some settings, profoundly impact the therapeutic response to chemotherapy and targeted agents. Hence, we wanted to investigate whether the selective BRM dependency of BRG1-mutant cancers translates in vivo. We compared the effects of BRM knockdown in BRG1-mutant NCI-H1299 and BRG1-WT NCI-H460 xenograft models (SI Appendix, Fig. S13A), containing either dox-inducible control (CTL) nontargeting shRNA or two distinct BRM-targeting shRNAs (sh2025 or sh5537). Upon dox treatment, BRM expression was markedly decreased in the BRM shRNA tumors but not in the CTL shRNA tumors (Fig. 6 A–C and SI Appendix, Fig. S13 B–F). The variability in BRM levels in the CTL shRNA tumors and dox-treated BRM sh2025 tumors is attributed to intratumoral and intertumoral variability in the extent of necrosis (SI Appendix, Fig. S14). Efficient BRM knockdown was maintained through the end point of the studies (SI Appendix, Fig. S13 B–F). Dox treatment of mice bearing BRG1-mutant NCI-H1299 xenografts with either BRM sh2025 or BRM sh5537 led to significant inhibition of tumor growth (T/C = 29% and T/C = 5%, respectively) (Fig. 6D). This effect was due to depletion of BRM rather than dox treatment alone, as NCI-H1299 CTL shRNA tumors progressed rapidly despite dox treatment (Fig. 6D). BRM depletion led to a marked decrease in the proliferation marker Ki67 in dox-treated NCI-H1299 BRM shRNA tumors but not in NCI-H1299 CTL shRNA tumors (Fig. 6 F and G and SI Appendix, Fig. S15 A and B). Moreover, BRM inhibition in NCI-H1299 tumors increased expression of the senescence marker acidic β-gal (SI Appendix, Fig. S16 A and B), but we did not observe increased apoptosis or an inflammatory response (SI Appendix, Figs. S16 C and D and S17 A–D). Together, these findings suggest that similar to the in vitro findings, the in vivo growth inhibition is mediated by G₁ arrest and induction of senescence. Importantly, knockdown of BRM did not impact the growth of BRG1-WT NCI-H460 tumors (Fig. 6E and SI Appendix, Fig. S15 C and D), demonstrating the selective effects of BRM depletion in vivo.

Library Design, Construction, and Screening.

A custom 6,500 element shRNA library focused on enzymes involved in epigenetic regulation was constructed by using chip-based oligonucleotide synthesis and cloned as a pool into the the pRSI9 lentiviral plasmid (Cellecta). Viral packaging was carried out according to the manufacturers recommended protocol. Each cell line was screened in duplicate, maintained an average minimal representation of 1,000 cells per sRNA, and harvested after five-population doublings. The representation of each barcode in the library was measured by next generation sequencing on an Illumina GA2X. Detailed protocols for the viral packaging, transduction, screening, and data analysis are provided in SI Appendix, Methods.

Cell Culture, Immunoprecipitation, and Western blotting.

NCI-H1299, NCI-H460 NCI-H838, and A549 cells were cultured in recommended media. Nuclear lysates were prepared by using the NE-PER Nuclear and Cytoplasmic Extraction kit (Thermo) by following manufacturer’s recommendations. Immunoprecipitaiton was performed with an anti-BRM antibody (Abcam) and associated SWI/SNF subunits detected by Western blot using standard protocols. Chemiluminescent signal was detected by using SuperSignal West Femto Maximum Sensitivity Substrateor Li-Cor Odyssey. Additional details on methods and antibodies used for immunprecipitation and Western blotting are provided in SI Appendix, Methods.

Functional Characterization of BRM Knockdown Using Inducible shRNA Constructs.

shRNA sequences targeting BRM cloned into the pLKO-Tet-On inducible vector system. The sequences of the oligonucleotides used and details on the cloning and lentiviral production are provided in SI Appendix, Methods. Lung cancer cell lines were infected with lentiviruses carrying BRM shRNAs, and the effect of BRM knockdown in growth and focus formation assays was determined in the presence or absence of dox-induced shRNA expression. Details for the cell cycle, senescence, and immunofluorescence assays are provided in SI Appendix, Methods.

In Vivo Efficacy Studies.

All animal studies were carried out according to the Novartis Guide for the Care and Use of Laboratory Animals. Mice were inoculated s.c. with NCI-H1299 or NCI-H460 cancer cells stably expressing dox-inducible CTL nontargeting shRNA or two distinct BRM-targeting shRNAs, and tumor volume was measured twice weekly. At termination of each study, tumor tissue was collected from each group and processed for immunohistochemistry by using standard methods. A detailed description of these xenograft and the immunohistochemistry (IHC) studies is provided in SI Appendix, Methods.