Reactivation of the G1 enhancer landscape underlies core circuitry addiction to SWI/SNF

Abstract

Several cancer core regulatory circuitries (CRCs) depend on the sustained generation of DNA accessibility by SWI/SNF chromatin remodelers. However, the window when SWI/SNF is acutely essential in these settings has not been identified. Here we used neuroblastoma (NB) cells to model and dissect the relationship between cell-cycle progression and SWI/SNF ATPase activity. We find that SWI/SNF inactivation impairs coordinated occupancy of non-pioneer CRC members at enhancers within 1 hour, rapidly breaking their autoregulation. By precisely timing inhibitor treatment following synchronization, we show that SWI/SNF is dispensable for survival in S and G2/M, but becomes acutely essential only during G1 phase. We furthermore developed a new approach to analyze the oscillating patterns of genome-wide DNA accessibility across the cell cycle, which revealed that SWI/SNF-dependent CRC binding sites are enriched at enhancers with peak accessibility during G1 phase, where they activate genes involved in cell-cycle progression. SWI/SNF inhibition strongly impairs G1-S transition and potentiates the ability of retinoids used clinically to induce cell-cycle exit. Similar cell-cycle effects in diverse SWI/SNF-addicted settings highlight G1-S transition as a common cause of SWI/SNF dependency. Our results illustrate that deeper knowledge of the temporal patterns of enhancer-related dependencies may aid the rational targeting of addicted cancers.

Graphical Abstract

Graphical Abstract.

Introduction

SWI/SNF subunits have been identified as context-specific dependencies in several cancer settings (1–6). In these settings, inactivation of SWI/SNF ATPase activity using the selective catalytic inhibitors BRM014 (7), FHD-286, or degrader AU-15330 (4) has anti-neoplastic effects and is well tolerated in vivo (4,5,8). As a result, the use of small molecules to impair SWI/SNF activity in these ‘SWI/SNF-addicted’ cancer settings has garnered considerable attention as a potential therapeutic strategy. However, the chromatin mechanisms and temporal windows associated with SWI/SNF dependency in each of these settings have remained uncertain, in part because these complexes regulate DNA accessibility for many activities, including transcription (9,10), sister-chromatid resolution during mitosis (11), expression of cell cycle regulators (12), mitotic bookmarking (13), DNA repair (14), DNA replication (15), and many other key pathways (16–18).

We identified neuroblastoma (NB) as an ideal model to interrogate the mechanisms underlying cancer cell addiction to SWI/SNF catalytic activity, as the primary ATPase of SWI/SNF complexes SMARCA4 (BRG1) is a dependency (3), and this cancer utilizes a core regulatory circuitry (CRC) of TFs that is conceptually similar to those seen in other SWI/SNF-addicted malignancies (1,4,19,20). NB is the most frequent extracranial pediatric solid tumor (21) and arises from cells of neural crest origin (22). Genome-wide analyses have revealed few common genetic alterations in NB, the most frequent of which is MYCN amplification (20% of all cases (23–25), rising to 50% in high-risk tumors (26)). Additionally, NB is driven by aberrant patterns of DNA accessibility (27–30). NB cells have the potential to adopt adrenergic or mesenchymal identities (31,32), which are associated with distinct CRCs (28,33). Adrenergic NB cell identity is associated with high-risk cases (31), and is governed by MYCN, ISL1, PHOX2B, HAND2, GATA3, and other TFs (30,33–36), which form an autoregulatory circuit that promotes self-renewal. The adrenergic CRC is reprogrammed by retinoids (37) used clinically as maintenance therapies, which promote cell-cycle exit and differentiation in mid-G1 phase (38).

Here we reveal that SMARCA4 activity is needed to sustain the open enhancer landscape for adrenergic CRC TFs. We find that the CRC arises from an interplay between pioneer TFs and those that require SWI/SNF catalytic activity, and inhibition of SWI/SNF causes rapid collapse of the CRC autoregulatory loop within 1 hour. We demonstrate that SWI/SNF activity is essential during G1-S transition, but dispensable for the survival of addicted cancer cells at other cell cycle phases. As a result of this G1-specific effect, we find that SWI/SNF inhibition sensitizes cells to retinoids commonly used as maintenance therapies. Our findings show that SWI/SNF activity represents a targetable site of convergence in G1 phase.

Materials and methods

Cloning

Lentiviral vectors

To generate a plasmid for expression of human SMARCA4 tagged with EGFP and the minimal auxin-induced degron (mAID), SMARCA4 (hsBRG1) was amplified from pBABE-hsBRG1, a gift from Robert Kingston (Addgene plasmid #1959; http://n2t.net/addgene:1959; RRID:Addgene_1959) and inserted into a pRRL-CAG lentiviral vector through NheI/MluI restriction sites using In-Fusion (Takara Bio #639649) according to manufacturer's instructions. Subsequently, the SMARCA4 cDNA sequence was modified to introduce mismatches against two different gRNAs while preserving amino acid sequence of the protein product using site-directed ligase-independent mutagenesis (39). In particular, the SMARCA4 cDNA sequence was wobbled to introduce mismatches against following gRNAs: 5′-GGTCCTGTTGCGGACACCGA-3′, 5′-CGGCACCTCCAAATTACAGC-3′. The MluI site was subsequently used to insert a sequence encoding EGFP-mAID obtained from synthesized DNA.

TIR1 was PCR amplified from pMK232 (CMV-OsTIR1-PURO), a gift from Masato Kanemaki (Addgene plasmid #72834; http://n2t.net/addgene:72834; RRID:Addgene_72834) and inserted into pRRL-CAG backbone through NheI and AgeI restriction sites. These sites were subsequently used to insert sequences encoding nuclear localization signal on both the N- and C- terminus of TIR1 using adaptor cloning. Sequence encoding P2A-mTagBFP was PCR amplified from pHR-pSFFV-dCas9-SunTag-P2A-BFP-dWPRE, a gift from Clifford Brangwynne (Addgene plasmid #122151; http://n2t.net/addgene:122151; RRID:Addgene_122151), and inserted into pRRL-CAG-NLS-TIR1-NLS through AgeI and EcoRV restriction sites to create pRRL-CAG-NLS-TIR1-NLS-P2A-BFP construct.

A CrisprLentiV2 transfer plasmid for CRISPR/Cas9-mediated knockout of endogenous SMARCA4 harboring SMARCA4 gRNA was cloned following protocol described earlier (40,41). Briefly, CrisprLentiV2 vector conferring G418 (geneticin) resistance was digested using BsmbI and the following gRNA sequence targeting endogenous SMARCA4 was inserted as an adaptor into the backbone: 5′-GGTCCTGTTGCGGACACCGA-3′. Plasmids for shRNA-mediated knockdown were purchased from Dharmacon (#TRCN0000015549, #TRCN0000015550, #TRCN20330, #TRCN20332, and #RHS6848).

RNAse H-EGFP was PCR amplified from pEGFP‐RNAse H1, a gift from Andrew Jackson and Martin Reijns (Addgene plasmid #108699; http://n2t.net/addgene:108699; RRID:Addgene_108699) and inserted into the pRRL-CAG-IRES-puro backbone using NheI and MluI restriction sites.

PiggyBac vectors

Vector delivering hyperactive transposase was a gift from C. Bashor (Rice University). Transposon plasmid tFucci(CA)5 was a gift from Atsushi Miyawaki (Addgene plasmid #153521; http://n2t.net/addgene:153521; RRID:Addgene_153521). For single-color controls used for flow cytometry, the plasmid tFucci(CA)5 was modified to yield piggyBac transposon plasmids that constitutively express either AzaleaB5 alone or h2-3 fluorescent protein alone by deleting degron fusions and preserving only one of the two respective fluorescent proteins.

Culturing of human cells and generation of stable cell lines

2D cell culture

Human cancer cell lines were acquired from ATCC, DSMZ, or MilliporeSigma and cultured in a humidified incubator maintained at 37ºC and 5% CO₂. Lenti-X 293T, KELLY, G-401, and SK-N-DZ cell lines were maintained in high-glucose DMEM (Gibco #11960–044), IMR-32 and RMC2C in MEM (Gibco #11095–072) and SK-N-SH, LA-N-5, THP-1, HL-60, MOLM-13, MV-4–11, LNCaP, 22Rv1, SH-SY5Y, SK-N-AS, CHP-212, SH-EP, and 92.1 in RPMI (Gibco #31800–022). All culture media was supplemented with 10% heat inactivated FBS (Corning #35–010-CV), 1× GlutaMAX (Gibco #35050–061), 1× non-essential amino acids (Gibco #11140–050), 1 mM sodium pyruvate (Gibco #11360-070), 10 mM HEPES (Gibco #15630-080), and 1× Pen/Strep (Gibco #15140–122).

Neural crest cell and peripheral neurons

H1 human embryonic stem cells (WiCell #WA01) were maintained in laminin-coated plates (Fisher #NC1547124) in B8 media (42). H1 cells were differentiated into neural crest cells (NCCs) by seeding 5000 cells per well in a 96-well laminin-coated plate in induction media with 3 μM CHIR-99021 (Tocris Bioscience #4423) added for the first 2 days of differentiation as previously described (43,44). On day 5, differentiation into neural crest cells was confirmed by OCT4 (Santa Cruz Biotechnology #sc-5279, RRID:AB_628051) and SOX9 (MilliporeSigma #AB5535, RRID:AB_2239761) immunofluorescence (IF) staining. For differentiation into neurons, day 5 neural crest cells were detached using accutase (MilliporeSigma #SCR005) and 20 000 cells/well were plated in a 96-well laminin-coated plate. Cells were seeded in induction media which was replaced by terminal differentiation media containing 500 nM all-trans retinoic acid (Cayman Chemical #302-79-4) as described previously (43). Differentiation into neurons was assessed by TUJ1 (BioLegend #801202, RRID:AB_10063408) IF staining.

3D multicellular tumor spheroids

Multicellular tumor spheroids were grown in 96-well clear round bottom ultra-low attachment microplate culture dishes (Corning #89089–826). A 100 μl suspension of 1000 cells in culture media was added to each well. Spheroids were cultured for 3 days before treatment with DMSO or BRM014, and their growth was monitored using the Incucyte SX5 Live-Cell Analysis System (Essen Bioscience) and Incucyte SX5 Spheroid software module.

Preparation of lentiviral vectors

Lentiviral vectors were prepared by co-transfection of Lenti-X 293T cells with transfer plasmid (described in the Cloning section above) and packaging plasmids (psPAX2, pMD2.G) using polyethyleneimine hydrochloride (PEI, Polysciences #247651, MW 40000). The culture media was changed to DMEM supplemented with 2% FBS 24 h post-transfection and lentiviral particles were harvested and concentrated using a 100-kDa Amicon filter (MilliporeSigma #UFC910024) 72 h post-transfection.

Generation of cell line with SMARCA4 degron

To engineer a cell line for selective targeted protein degradation of SMARCA4 via the mAID tag (Supplementary Figure 1) (45,46), IMR-32 cells were co-spinfected (1000 g, 30 min, 32ºC) with two lentiviral vectors delivering (1) human gRNA resistant SMARCA4-EGFP-mAID and (2) NLS-TIR1-NLS-P2A-BFP. The population of EGFP and BFP double-positive cells was enriched using fluorescence-activated cell sorting (FACS). The double-positive cells were treated with 500 μM auxin for 1 h and BFP-positive EGFP-negative cells (i.e. cells responsive to auxin-induced degradation of SMARCA4-EGFP-mAID) were subsequently selected using FACS. These responsive cells were spinfected with lentiviral vector delivering a lentiCRISPRv2 for CRISPR/Cas9-mediated knockout of endogenous SMARCA4 and conferring G418 (geneticin) resistance. Cells were subsequently selected using 125 μg/ml G418 (Gibco #10131-035) and single EGFP-BFP double positive cells were seeded into 96-well plates using FACS and allowed to expand to obtain monoclonal cell lines. Knockout of endogenous SMARCA4 and auxin-induced degradation of SMARCA4-EGFP-mAID were confirmed via western blot.

Auxin-induced degradation

For all assays using the minimal auxin-induced degron, freshly prepared 500 mM auxin (indole-3-acetic acid sodium salt, MilliporeSigma #I5148) was prepared in ethanol, then diluted 1000× upon incubation with cells to yield a final concentration of 500 μM auxin. As a control, an equivalent volume of ethanol vehicle was used. Auxin-induced degradation of SMARCA4 was confirmed by western blot to be complete within 2 h of auxin treatment.

Generation of FUCCI-labeled cell lines

IMR-32 cells were transfected with vector encoding hyperactive transposase and tFucci(CA)5 in 1:5 ratio using Lipofectamine 3000 according to the manufacturer's instructions and selected with 5 μg/ml blasticidin S hydrochloride. Following complete selection, cells were further cultured with 10 μg/ml blasticidin.

BRM014 treatment

Synthesis

BRM014 was synthesized as described earlier (Supplementary Figure 2) (7) with minor changes described below. In particular, the phenyl (5-(((TBDMS)oxy)methyl)-2-fluoropyridin-4-yl)carbamate was prepared according to the published protocol, with the exception of the fluorination step, which was carried out with CsF instead of TMAF. This approach resulted in slightly lower yield but allowed us to use more accessible reagent. The 3-(difluoromethyl)isothiazol-5-amine was also prepared according to the published procedure with only minor adjustment. Borane dimethylsulfate complex was used instead of borane-THF complex. The coupling conditions used for the merging of these two fragments were optimized: longer reaction time at lower temperature was used in the LiHMDS-mediated reaction and HF-TEA was employed in the final deprotection, which resulted in doubling of the overall yield of the final compound. The final synthesized product was either crystallized or lyophilized prior to reconstitution in DMSO. Final crystallization helped remove residual color in the BRM014 and resulted in colorless needles. Both lyophilized and crystallized products were determined to have equivalent biological activity.

Validation and quality control

To assess the quality and purity of the synthesized BRM014, ¹⁹F, ¹H, and ¹³C NMR spectra were recorded on a Bruker Avance III HD 400 MHz using solvent signal as a reference (Supplementary Figure 2B). The identity of the BRM014 structure was verified by measuring ¹³C, H,C-HSQC and H,C-HMBC spectra using standard pulse programs from the library of the spectrometer. Gradient selection was used in the 2D experiments, and all signals were assigned to their respective atoms. ¹H NMR (400 MHz, DMSO-d₆) δ 11.76 and 9.25 (bs, 1H, NH), 8.05 (s, 1H, H-5), 7.77 (s, 1H, H-2), 7.09 (s, 1H, H-4′), 6.95 (t, 1H, J_H,F= 54.4 Hz, CHF₂), 5.72 (t, 1H, J_OH,CH2= 5.1 Hz, OH), 4.59 (s, 2H, J_CH2,OH= 4.4 Hz, CH₂O). ¹³C NMR (101 MHz, DMSO-d₆) δ 163.70 (H-5′), 163.64 (d, J_1,F= 230.3 Hz, C-1), 159.04 (t, J_3’,F= 26.8 Hz, C-3′), 152.25 (C=O), 148.94 (d, J_3,F= 12.5 Hz, C-3), 146.55 (d, J_5,F= 17.9 Hz, C-5), 123.04 (d, J_4,F= 3.7 Hz, C-4), 111.53 (t, J_C-F= 237.1 Hz, CHF₂), 104.03 (C-4′), 97.66 (d, J_2,F= 45.2 Hz, C-2), 58.59 (CH₂O). ¹⁹F NMR (470.4 MHz, DMSO-d₆) δ -69.98 (s). -113.82 (d, J= 54.5 Hz). Additionally, LC-MS analysis was performed using a Waters UPLC H-Class Core System, UPLC PDA detector and Mass spectrometer Waters SQD2. For all materials used in this study, the expected molecular weight of 318 Da was confirmed, and purity of > 99% was confirmed based on LC-MS measurement (Supplementary Figure 2B).

Treatments

Except where indicated otherwise, BRM014 was used at a concentration of 1 μM and compared to an equivalent volume of DMSO as a vehicle control.

Western Blotting

Immunoblotting

Total protein was isolated from cells using RIPA lysis buffer [50 mM Tris–HCl pH 7.5, 150 mM NaCl, 1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS, cOmplete ULTRA protease inhibitors (Roche #05–892-791-001)]. Harvested cells were washed 2 times in cold PBS, subsequently resuspended and incubated in cold RIPA lysis buffer for 20 min on ice. Samples were sonicated using a Diagenode Bioruptor (5 cycles of 30 seconds on/off at high power) and cleared by centrifugation at 18 000 g for 15 min at 4°C. Protein concentration in the cleared supernatant was determined using the Pierce Detergent Compatible Bradford Assay Kit (Thermo Fisher #23246) according to the manufacturer's instructions. Subsequently, 15 μg of total protein per well was loaded into Novex NuPAGE Tris-Acetate 3–8% polyacrylamide gels (Thermo Fisher #EA03785BOX) in Tris-Acetate SDS running buffer (Thermo Fisher #LA0041) and the protein bands were separated according to manufacturer's instructions. Bands from the gel were transferred to 0.45 μm PVDF membrane (MilliporeSigma #IPFL00010) overnight.

Detection

Western blots were subsequently blocked in 5% BLOTTO (Santa Cruz Biotechnology #NC9730946) dissolved in Tris-buffered saline with 0.1% Tween 20 detergent (TBS-T), incubated in primary antibody overnight at 4°C, washed 3 times in TBS-T, and probed with secondary antibody for 1–2 h at RT. The following antibodies were used in this study: SMARCA4 (Santa Cruz Biotechnology #sc-374197, RRID:AB_10990135), SMARCA2 (Abcam #ab15597, RRID:AB_443214), beta-actin (GeneTex #GTX629630, RRID:AB_2728646), CCND1 (Abcam #ab134175, RRID:AB_2750906), PHOX2B (Santa Cruz Biotechnology #sc-376997, RRID:AB_2813765), ISL1 (DSHB #40.2D6, RRID:AB_528315), HAND2 (Abcam #ab200040, RRID:AB_2923502), GATA3 (Thermo Fisher #MA1-028, RRID:AB_2536713), MYCN (Active Motif #61185, RRID:AB_2793543), ASCL1 (Santa Cruz Biotechnology #sc-374104, RRID:AB_10918561), EBF3 (Thermo Fisher #PA5-30985, RRID:AB_2548459), HRP-conjugated mouse anti-rabbit IgG-HRP secondary antibody (Santa Cruz Biotechnology #sc-2357, RRID:AB_628497) and m-IgGk BP-HRP secondary antibody (Santa Cruz Biotechnology #sc-516102, RRID:AB_2687626). Detection of HRP-conjugated antibodies was performed using Clarity (Bio-Rad #1705060) or Clarity Max Western ECL substrate (Bio-Rad #1705062) on a Bio-Rad ChemiDoc MP imager.

Assessment of cell growth, viability, drug synergy, and apoptosis

Proliferation and viability

Cell proliferation was assessed by crystal violet staining (VWR #548-62-9) according to the manufacturer's instructions. For crystal violet staining, cells were seeded as independent biological replicates (N = 3) for each tested condition and exposed to variable concentrations of BRM014 for 6 days before they were fixed and stained. Stained plates were scanned using an Epson 4990 scanner to obtain colored images and imaged with a Bio-Rad ChemiDoc MP imager for densitometric analysis with Bio-Rad ImageLab software.

Cell proliferation and viability were also assessed using resazurin-based indicators (Invitrogen #A13261 or #DAL1025) in cells seeded as independent biological replicates (N = 3) for each tested condition. Treatments were either 1 μM BRM014, 500 μM auxin, or equivalent volumes of vehicle controls (DMSO or ethanol, respectively). Single time-point viability measurements in IMR-32 and IMR-32 mAID cells were conducted 96 h post-treatment. All other measurements were carried out at the indicated timepoints. All viability measurements were performed according to the manufacturer's instructions. Briefly, at experimental endpoints, 10 μl (10% culture volume) of reagent was added cells in 90 μl culture medium in 96-well plate. Plates were read at 560/590 (ex/em) using a Cytation 5 or Perkin Elmer Victor plate reader.

For measurement of drug interactions, cells were exposed to drugs for 96 h (IMR-32, SK-N-DZ, SK-N-SH, LNCaP, 92.1, or THP-1) or 144 h (KELLY), passaged into media without drugs and grown for 1 week before outgrowth was assessed. Drug synergy scores were determined using the ‘synergyfinder’ R package. The following drugs were used in this study: isotretinoin (MilliporeSigma #PHR1188), all-trans retinoic acid (ATRA, MilliporeSigma #R2625), doxorubicin (MilliporeSigma #D1515), cisplatin (MilliporeSigma #232120), etoposide (MilliporeSigma #341205), topotecan (MilliporeSigma #T2705), and vincristine (MilliporeSigma #V8879).

Real-time assessment of cell proliferation and death

For monitoring cell growth and viability in real time, the adherent cell-by-cell module in the Incucyte live-cell analysis system (Essen Bioscience) was used. Cells were seeded as independent biological replicates (N = 3) for each condition and 16 images per replicate were acquired with a 10× objective every 2 h. Real-time quantification of cell death was performed by addition of Cytotox red dye (VWR #MSPP-4632E) according to the manufacturer's instructions. Confluence-normalized intensity of Cytotox red signal was determined for each acquired image relative to mean DMSO measurements. Breakpoints indicating the onset of cell death were determined in an unbiased manner from viability time courses using a segmented regression model with the ‘segmented’ R package.

Assessment of cell cycle phase

Propidium iodide (PI) staining

NB, AML, prostate cancer, uveal melanoma, and control cell lines were seeded at ∼20% confluency, treated with DMSO or 1 μM BRM014 at different timepoints, and harvested simultaneously to reach exposure corresponding to 0, 1, 2, or 3 days. Cell-cycle analysis was performed for independent biological replicates (N = 3). Cells were fixed with 70% ethanol at 4°C for 30 min, treated with RNase (Roche #11119915001) at a final concentration of 100 μg/ml at 37°C for 30 min and stained by PI solution (PBS, 0.1% Triton X-100, 50 μg/ml propidium iodide). Cells were transferred into 5 ml FACS tubes (Falcon #352235) and immediately analyzed by flow cytometry. Cell-cycle phases were determined on a Sony SH800 cell sorter at 561 nm excitation / 600 nm emission.

tFucci(CA)5 cell-cycle analysis

To analyze the cell-cycle phases using flow cytometry, IMR-32 cells expressing tFucci(CA)5 were seeded at ∼40% confluency and exposed to DMSO, 1 μM BRM014, or 10 μM ATRA and harvested. DAPI live/dead stain was added for detection of dead cells. The presence of fluorescent proteins indicating different phases of the cell cycle was analyzed using the Sony SH800 cell sorter. In particular, AzaleaB5 was monitored using 561 nm excitation / 600 nm emission, and h2-3 was monitored using 488 nm excitation / 525 nm emission.

To monitor cell-cycle changes over time, IMR-32 cells expressing tFucci(CA)5 were plated on 6-well plates as independent biological replicates for each tested condition (N = 3) and changes in their fluorescent properties were monitored using the Incucyte SX5 live-cell analysis system (Essen Bioscience). Adherent cell-by-cell classification to determine the fractions of AzaleaB5-positive, h2-3-positive, and double-positive cells was performed in Incucyte SX5 software. To synchronize cells in S phase, cells were treated with 2 mM thymidine (MilliporeSigma #T1895). After 48 h, synchronization was released by replacement of media not supplemented with thymidine. At this time, either 1 μM BRM014, 0.316 μM doxorubicin (MilliporeSigma #D1515), 0.316 μM etoposide (MilliporeSigma #341205), 10 μM nocodazole (MilliporeSigma #M1404), 10 μM cisplatin (MilliporeSigma #232120), or an equivalent volume of DMSO was added to the culture medium. Cell-cycle progression was monitored using the Incucyte SX5 system and analyzed using the ‘tidycyte’ workflow (https://github.com/hodgeslab/tidycyte).

RNA expression analysis

Isolation of RNA

Cells were seeded as independent cell culture replicates (N = 3) at ∼30% confluency, then treated with 1 μM BRM014 or vehicle (DMSO) and incubated for up to 72 h under standard cell culture conditions. Total RNA was harvested using TRIzol reagent (Invitrogen #15596026) following the manufacturer's instructions.

RT-qPCR

cDNA was generated using a high-capacity reverse transcription kit (Fisher #43–688-14). Subsequently, 1× TaqMan Fast Advanced Master Mix (Applied Biosystems #4444557), 1× PCNA probe (FAM-MGB #Hs00427214), 1× ACTB probe (VIC-MGB #Hs01060665), and 50 ng of cDNA were combined in a 20 ul volume for each sample. PCR was performed with a QuantStudio 3 Real-Time PCR system (Applied Biosystems) at 50ºC for 2 min and 95ºC for 20 s, followed by 40 cycles of 95ºC for 1 s and 60ºC for 20 s. Quantification expression was normalized to beta actin expression levels using the ΔΔCt method.

Thiol (SH)-linked alkylation for metabolic labeling of RNA (4sU RNA-seq)

4-thiouridine (4sU) labeling

IMR-32 cells engineered to express SMARCA4 degron responsive to auxin as well as parental IMR-32, KELLY and SK-N-DZ cells were seeded at ∼60% confluency. IMR-32 SMARCA4 degron cells were treated with 500 μM auxin or ethanol, and KELLY, IMR-32, and SK-N-DZ cell were treated with 1 μM BRM014 or DMSO (N = 2 independent cell culture replicates for each condition). Subsequently, cells were exposed to 500 μM 4sU (MilliporeSigma #T4509-25MG) for 2 h before the RNA was harvested.

Isolation of RNA

Total RNA was harvested from 80% confluent 2D monolayer cultures using TRIzol as described previously (47). Briefly, samples were incubated at RT for 5 min protected from light and 200 μl of chloroform was added to each tube containing 1 ml of TRIzol and the lysed cells. Samples were vortexed for 15 s, incubated at room temperature for 3 min and centrifuged at 16000 g for 15 min at 4°C. Aqueous phase was transferred to new tube and mixed with equal part of 2-propanol supplemented with 1/100 volume of 10 mM DTT (0.1 mM final concentration). Samples were vortexed, incubated at room temperature for 10 min and centrifuged at 16000 g for 20 min at 4°C. Supernatant was discarded, pellet was washed with 500 μl 75% EtOH, 5 μl of 10 mM DTT per sample, and centrifuged at 7500 g for 5 min at RT. After drying, the pellet was resuspended in water supplemented with 1 mM DTT.

Thiol modification

20 ug of RNA was added to 50 μl of reaction mix [50 mM NaPO₄ pH 8, 10 mM iodoacetamide (IAA) and 50% DMSO (final concentrations)]. The reaction was incubated at 50°C for 15 min and quenched with 1 μl of 1 M DTT. RNA was precipitated by adding 5 μl NaOAc (3 M, pH 5.2) and 125 μl 100% ethanol, vortexed, then incubated for 30 min at –80°C and centrifuged at 16 000 g for 30 min. The pellet was washed with 1 ml 75% EtOH and resuspended in 20 ul of DEPC-treated water.

Library preparation and sequencing

IAA-treated RNA was subjected to polyA enrichment, fragmentation, and random-primer cDNA synthesis as previously performed (9). Sequencing of all libraries was performed on Illumina NextSeq 500 high-output or NovaSeq 6000 flow cells according to standard Illumina protocols.

Analysis of 4sU RNA-seq data

Obtained 4sU RNA-seq reads were processed by mapping to the hg38 reference human genome using hisat-3n (48), which permits mapping of sequenced reads to the genome with T > C conversions. Duplicate fragments and reads with mapping quality <10 were discarded, selecting for high-quality unique reads. Reads containing 5 or more T > C conversions were found to show >90-fold selectivity upon 4sU incubation. Reads within genes bearing conversions were counted using HTSeq and processed using DESeq2 with default parameters. Log₂ fold changes were calculated using maximum a posteriori estimation using a zero-mean normal prior (Tikhonov-Ridge regularization). FDR-corrected P values were calculated using the Benjamini–Hochberg procedure. Differential calls were made by requiring FDR-corrected P< 0.05. For analysis of expression changes across cell lines, cell-line-specific expression was normalized using the ‘limma’ R package.

Assay for transposase-accessible chromatin sequencing (ATAC-seq)

ATAC-seq library preparation and sequencing

IMR-32, KELLY, and SK-N-DZ cells were treated with 1 μM BRM014 or DMSO for 1 h and processed as described below. For cell cycle-phase specific ATAC-seq IMR-32 FUCCI-expressing cells were treated with 1 μM BRM014 or DMSO before their separation by FACS using a Sony SH800 cell sorter. AzaleaB5 was monitored using 561 nm excitation/600 nm emission, and h2-3 was monitored using 488 nm excitation / 525 nm emission and gates for individual cell cycle phases were set as indicated. Cells were collected into culture media supplemented with 1 μM BRM014 or DMSO and processed for ATAC-seq as described below.

ATAC libraries were prepared using the Omni-ATAC protocol (N = 2 independent cell culture replicates for each condition) (49). Briefly, upon harvesting of cells, nuclei were obtained by resuspending cells in 50 μl of lysis buffer [0.1% Tween-20, 0.1% NP-40 and 0.01% digitonin in RSB buffer (10 mM Tris–HCl pH 7.5, 10 mM NaCl, 3 mM MgCl₂)] and incubating on ice for 3 min. Lysis buffer was washed out by RSB buffer supplemented with 0.1% Tween-20. Subsequently, nuclei were pelleted by centrifugation for 10 min at 500 g, resuspended in 50 μl of Transposition mix [2.5 μl Tagment DNA enzyme (Illumina #20034198), 25 μl of Tagment DNA buffer (Illumina #20034198), 0.1% Tween-20 and 0.01% Digitonin] and incubated for 30 min at 37°C. DNA was purified with a MinElute PCR purification kit (Qiagen #28004), and libraries were amplified by PCR with barcoded Nextera primers (Illumina) using 2× NEBNext High-Fidelity PCR Master Mix (NEB #M0541). For sequencing, libraries were size-selected with AMPure XP beads (Beckman Coulter #NC9959336) for fragments between ∼100 and 1000 bp in length according to the manufacturer's instructions. Sequencing was performed using paired-end reads on an Illumina NextSeq 500 high-output flow cell.

ATAC-seq data analysis

Paired-end reads were processed by mapping to the hg38 reference human genome using Bowtie 2.1.0 (50). Duplicate fragments and reads with mapping quality < 10 were discarded, selecting for high-quality unique reads. Peak calling was performed by MACS 2.1.1 (51). To identify increased, decreased, and unchanged peaks, differential peak calling was performed using DESeq2 using default parameters, requiring fold changes >1.5-fold in either direction and FDR-adjusted P values < 0.10. P values for enrichment were obtained by measuring the frequency of their motif presence in increased peaks compared to unchanged peaks using the hypergeometric test. Cell cycle-dependent SMARCA4-targeted sites were identified as overlap between SMARCA4 ChIP-seq peaks and cell cycle-phase specific ATAC-seq sites accessible in at least one of the cell cycle phases in vehicle (DMSO) treated cells. Phase-specific accessibility Z scores were DMSO-centered. Motif enrichment was measured using HOMER (52) (version 4.11) to examine differential enrichment of transcription factor motifs within SMARCA4-dependent peaks, and using chromVAR (53) to evaluate differential accessibility independently of differential peak calls.

Chromatin immunoprecipitation sequencing (ChIP-seq)

Chromatin immunoprecipitation

ChIP libraries were prepared as described previously (54) from IMR-32 cells treated with 1 μM BRM014 or DMSO for 1 h or 4 h (N = 2 independent cell culture replicates for each condition). All cells for RARA ChIP were treated with 10 μM 13-cis retinoic acid along with either 1 μM BRM014 or DMSO. Single-cell suspensions were prepared before cells were fixed for 10 min in 1% formaldehyde with the exception of ASCL1 ChIP, where cells were fixed first for 45 min in 2 mM disuccinimidyl glutarate (DSG) prior to 10 min in 1% formaldehyde. Excess formaldehyde was quenched by the addition of glycine to 125 mM. Fixed cells were washed, pelleted, and snap-frozen using liquid nitrogen. Cell pellets were resuspended in shearing buffer and sonicated in a Covaris E220 sonicator to generate DNA fragments 200–1000 bp in length. Chromatin was then immunoprecipitated overnight at 4°C with SMARCA4 (Proteintech #21634–1-AP, RRID:AB_10858784), PHOX2B (Santa Cruz Biotechnology #sc-376997, RRID:AB_2813765), ISL1 cocktail (DSHB #40.2D6, #39.4D5, #39.3F7, and #40.3A4, RRIDs: AB_528315, AB_2314683, AB_1157901, and AB_528313), HAND2 (Abcam #ab200040, RRID:AB_2923502), GATA3 (Thermo Fisher #MA1-028, RRID:AB_2536713), MYCN (Active Motif #61185 RRID:AB_2793543), ASCL1 (Abcam #ab74065, RRID:AB_1859937), EBF3 (Thermo Fisher #PA5-30985, RRID:AB_2548459), H3K4me1 (Abcam #ab8895, RRID:AB_306847), or RARA (Abcam #ab41934, RRID:AB_777683) antibodies bound to Protein A (Thermo Fisher #10001D) or Protein G (Thermo Fisher #10003D) Dynabeads. Enrichment was confirmed using the following primer sets: Intergenic peak F: CCATGGGGACACGTGTATGG, Intergenic peak R: AATCCCATTAGCGCTCTCCG, CCND1 enhancer F: CACCCGCCTTGGTTTAGTGA, CCND1 enhancer R: GCAAGGGACAGCACATTGTC, FAIM2 promoter F: GCCCTGGCCTCATTACATGG, FAIM2 promoter R: CTTCACTGGCCACTCACTCA, Gene desert F: CATCCCTGGACTGATTGTC, Gene desert R: GGTTGGCCAGGTACATGTTT.

ChIP-seq library preparation

Chromatin was eluted, then digestion and de-crosslinking was performed at 65°C overnight. DNA was extracted with phenol-chloroform and precipitated with ethanol. Size selection was performed by extracting 200–400 bp DNA fragments with AMPure XP beads (Beckman Coulter #NC9959336) before PCR amplification. Subsequently, DNA was purified using AMPure XP beads. After PCR amplification, DNA was quantified by Qubit fluorometric quantitation. Sequencing was performed using paired-end reads on an Illumina NextSeq 500 high-output flow cell.

ChIP-seq data analysis

Analysis of ChIP-seq data was performed as described previously (54). Briefly, ChIP-seq reads were processed by mapping to the hg38 reference human genome using Bowtie 2.1.0 (50), rejecting reads with map qualities <10. For datasets obtained from paired-end sequencing, actual fragment sizes were used and quantification using bedtools 2.28.0 was performed. Peaks identified within 1000 bp were merged. Read coverage values in genome tracks are the mean values across both replicates from each condition and were prepared using PyGenomeTracks.

Orthotopic mouse model

Tumor implantation and harvest

All animal studies were performed according to protocols approved by the Institutional Animal Care and Use Committee of Baylor College of Medicine (Houston, TX) in compliance with ethical conduct in the care and use of animals. The orthotopic NB mouse model was established as described earlier (55). Briefly, 10⁶ IMR-32 cells were injected in the sub-renal capsule of six‐week‐old female NCr nude mice (CrTac:NCr-Foxn1nu, Taconic, NY, USA). Three weeks after implantation, tumor engraftment was validated and tumor size was determined via magnetic resonance imaging (MRI, echo time (TE) = 80 ms, repetition time (TR) = 3030 ms, slice thickness = 1.2 mm, field of view = 80 mm, number of slices = 18, matrix = 256 × 250, no. of signal averages = 2, dwell time = 25 μs, scan time ∼ 3.5 min, 1.0 T permanent MRI scanner, M2 system, Aspect Technologies, Israel). Mice were assigned into two groups with equivalent distribution of tumor sizes (N = 11 animals per group). One group was assigned to receive 20 mg/kg/day of BRM014 in 10% DMSO and 1% methylcellulose (MilliporeSigma #M0430-100G) and the second control group received vehicle (10% DMSO and 1% methylcellulose) through oral gavage once daily for 14 days. During treatment, mice were weighed twice per week and tumor growth was monitored using small-animal MRI once per week. 16 h after last treatment, mice were euthanized, kidneys with tumors and contralateral (non-injected) kidneys were resected, and weighed to infer tumor weight by subtracting the weight of non-injected kidneys.

Immunohistochemistry (IHC) and histology

Kidneys containing tumors were fixed using 10% formalin for 24 h and subsequently dehydrated in 70% ethanol for 48 h. Samples were embedded in paraffin and 4-μm sections were prepared. Sections were stained with hematoxylin and eosin and IHC staining was performed using the following antibodies: KI67 (BioCare #901–325-042921, RRID:AB_272118), and ImmPRESS Goat anti-rabbit (Vector Laboratories #MP-7451, RRID:AB_2631198). Imaging of sections was performed using a Nikon ECLIPSE Ti-2 inverted microscope and quantification of KI67 IHC staining was determined using QuPath (version 0.3.0) (56).

Analysis of genome-wide CRISPR screens

DepMap analysis

To compare dependency on SMARCA4 and SMARCA2 across lineages, Chronos scores from CRISPR KO dataset generated for 1057 cell lines were downloaded from the DepMap portal and analyzed (57). Expression levels of SMARCA4 and SMARCA2 across different cell lines were obtained from RNA-seq datasets provided by the DepMap portal (57).

Analysis of patient samples

Analysis of survival in human NB tumors

Gene expression profiles of 498 primary NB tumor samples and clinical outcomes of the patients were obtained from the SEQC cohort (58). Patients were stratified into quartiles based on their SMARCA4 or SMARCA2 expression levels and survival analysis was performed using the ‘survminer’ and ‘survfit’ R packages.

Analysis of cell-cycle markers in SWI/SNF-addicted human cancers

Gene expression profiles of NB, AML, castration-resistant prostate cancer tumor samples, were obtained from the AML TARGET (59), NB TARGET (60), SU2C-PCF (61), and NB SEQC (58) cohorts. Additionally, FPKM expression values for TCGA-LAML, TCGA-PRAD, and TCGA-UVM were downloaded from the TCGA GDC data portal. To focus only on high-grade prostate tumors rather than low- or medium-grade tumors, TCGA-PRAD samples were filtered to include only samples with Gleason score ≥ 8.

Gene set enrichment was conducted using the ‘GSVA’ R package (version 1.42.0) in R version 4.1.2. GSVA is an unsupervised, non-parametric method that provides scaled sample-wise gene set enrichment scores. GSVA was run for each sample cohort separately, resulting in gene-set by sample matrices of GSVA enrichment scores. A G2–G1 score was calculated as the difference in GSVA score for the ‘REACTOME MITOTIC G2-G2/M PHASES’ and ‘REACTOME G1 PHASE’ gene sets. Points with Cook's distance larger than 4/N were considered to be outliers and removed, and the association between the G2–G1 score and SMARCA4 expression was quantified using a linear regression model. To assess the relationship between CCND1 and SMARCA4 expression, tumors within each cohort were assigned as SMARCA4-high or -low based on a statistical threshold using standard approaches (62), and P values were reported using T tests.

Statistical analysis

All statistical tests were performed as two-sided tests using GraphPad Prism 9.0 software (GraphPad Software, San Diego, CA, USA), or R (3.6.1 and 4.2.1). For multiple hypothesis testing, P values were adjusted using the Benjamini-Hochberg false discovery rate (FDR) correction procedure to obtain P_adj values. P values smaller than the 64-bit double precision machine epsilon (2⁻⁵² = 2.22e-16) are reported as P< 2.2e-16.

Code availability

All analyses were performed using publicly or commercially available software. Scripts used for genome-wide analyses are available on Zenodo (63) and at the following link:

https://github.com/hodgeslab/workflows.

Results

SMARCA4 is a targetable tumor-specific dependency of adrenergic NB cells

To evaluate sensitivity to SWI/SNF catalytic inhibition, we treated a panel of NB cell lines with BRM014, a small-molecule ATPase inhibitor of SMARCA4 and SMARCA2 (BRM) (7). BRM014 treatment induced morphological changes and dose-dependent cell death in NB cell lines previously annotated to have adrenergic character (35,64), with sub-micromolar IC50 regardless of MYCN amplification or 1p36 deletion (65), a marker of advanced disease (Figure 1A, B, Supplementary Figure 3A). Morphological changes induced by sub-lethal treatment included axonal extension and were consistent with neuronal differentiation (Figure 1B). We also observed a dramatic reduction in 3D spheroid growth (Supplementary Figure 3B), confirming the persistence of SWI/SNF dependency when cells are removed from 2D culture conditions. Similar sensitivity was not observed in NB cells annotated to have mesenchymal identity (35,64) (Figure 1A, Supplementary Figure 3A), demonstrating that SWI/SNF addiction is a specific feature of adrenergic NB. No loss of viability was observed in control cell lines (Figure 1A, Supplementary Figure 3A), and similar effects were observed using the orthogonal SMARCA2/4 inhibitor FHD-286 and degrader AU-15330 (Supplementary Figure 3C–E), confirming a common, on-target mechanism of action and ruling out non-specific toxicity. Additionally, no significant dependency on SWI/SNF catalytic activity was detected in cultured human neural crest cells or neural crest-derived peripheral neurons (Figure 3C, D), confirming that the requirement of SWI/SNF catalytic activity is a selective vulnerability of transformed NB cells and not shared with the neural crest lineage.

Figure 1.

SMARCA4 is a targetable, tumor-specific vulnerability of adrenergic neuroblastoma (NB). (A) Crystal violet staining of adrenergic (ADRN) and mesenchymal (MES) NB or control cell lines treated with serial titration of SWI/SNF inhibitor BRM014. Orange squares indicate MYCN amplification or 1p36 deletion. Quantified in Supplementary Figure 3A. (B) Representative images of NB cells treated with lethal (1 μM) and sub-lethal (10–100 nM) BRM014 or vehicle control (DMSO). (C) Preparation of neural crest cells (NCCs) and NCC-derived peripheral neurons from human embryonic stem cells (hESCs). (D) Cell growth over time upon treatment with BRM014 or DMSO. Error bars: mean ± 95% confidence interval (N = 3 independent samples), abbreviations used as in (C). (E) Representative images of IMR-32 cells upon expression of shRNA mediating SMARCA4 or SMARCA2 knock-down, or non-targeting control (NTC). (F) Western blot analysis of SMARCA4 and SMARCA2 levels after SMARCA2 knock-down compared to NTC. (G) Western blot analysis of SMARCA4 and SMARCA2 levels in IMR-32 SMARCA4 degron-containing cell line (SMARCA4-EGFP-mAID) after 2 h of vehicle (EtOH) or auxin treatment. (H) Quantification of IMR-32 (WT) and SMARCA4-EGFP-mAID (mAID) cell line viability upon treatment with BRM014, auxin or vehicle controls for 96 h. Error bars: mean ± SE. (I) Experimental design of animal study. (J) Small animal MRI imaging and kidneys extracted from animals comparing BRM014- and vehicle-treated animals (N = 11 per group). (K) Tumor growth time course measured by small-animal MRI. Error bars: mean ± SE. (L) Comparison of tumor weights of BRM014 and vehicle control treated animals. Error bars: mean ± SE. (M) Representative H&E staining of tumor sections comparing BRM014- and vehicle-treated animals.

Figure 3.

G1-S transition represents a unique window of sensitivity to SWI/SNF inhibition. (A) Decreased gene sets upon BRM014 treatment ranked by significance in NB cells. Cell cycle regulation-related gene sets are highlighted. (B) Depletion of S-phase genes upon SMARCA4 inactivation measured by gene set enrichment analysis. (C) Flow cytometry cell cycle analysis of cells upon time course treatment with BRM014 or DMSO (N = 4). (D) Quantification of cells in G1 phase from in vivo NB mouse model following BRM014 or vehicle control treatment. (E) FUCCI fluorescent cell cycle reporter system. (F) Representative images of FUCCI-labeled IMR-32 cells following treatment with BRM014 or DMSO. (G) Representative trajectory from real-time monitoring of FUCCI-labeled IMR-32 cells upon release from synchronization and treatment with drugs (N ≥ 4 independent replicates). Closed triangles indicate local signal maxima. Abbreviations: doxorubicin (Doxo); etoposide (Etop); nocodazole (Nocod). (H) Representative trajectory from real-time monitoring of cell cycle progression and cell death in synchronized and unsynchronized cells (N ≥ 4 independent replicates). Closed triangles indicate local signal maxima and open triangles indicate onset of cell death. (I) Timing of cell death onset following BRM014 treatment initiated at distinct phases of the cell cycle (N = 3 independent replicates). (J) Summary and interpretation: SWI/SNF activity is essential in G1 but dispensable for survival in other cell-cycle phases.

Dissection of the underlying molecular mechanisms requires identification of the ATPase subunit responsible for SWI/SNF’s dependency. We therefore sought to dissect the dependency on the SWI/SNF ATPases SMARCA4 and SMARCA2. We analyzed genome-wide CRISPR knock-out screens spanning 1057 cell lines from the DepMap portal and found that NB cells (N = 31) show signs of a lineage-specific addiction to SMARCA4 (P= 7.1e-8), but not SMARCA2 (P> 0.05), along with high SMARCA4 expression (Supplementary Figure 4A–C). Moreover, analysis of human NB tumors revealed that unlike SMARCA2, high SMARCA4 is linked to worse patient outcomes (Supplementary Figure 4D–G). We compared the dependency on SMARCA4 and SMARCA2 by performing shRNA-mediated knockdown of these ATPase subunits in IMR-32 cells. Unlike SMARCA2 knockdown, which did not alter proliferation, SMARCA4 knock-down resulted in loss of cell viability and a viable cell line could not be derived following selection (Figure 1E, F). We therefore engineered IMR-32 cells for auxin-inducible targeted protein degradation (Figure 1G, H), which confirmed that loss of SMARCA4 alone was sufficient to induce loss of viability.

To assess whether SWI/SNF catalytic activity was targetable in vivo, we established a MYCN-amplified orthotopic model of NB by sub-renal injection of IMR-32 cells in nude mice (43) and monitored the formation of tumors by small-animal MRI (Figure 1I). Following tumor establishment, we applied a daily treatment of 20 mg/kg BRM014 or vehicle control for 14 days via oral gavage. BRM014 treatment resulted in significant slowing of tumor growth (ANOVA, P< 0.0001) without detectable change in body weight over the 14-day treatment duration (Figure 1J–M, Supplementary Figure 5). Our results reveal that SMARCA4 is a pharmacologically targetable, tumor-selective vulnerability of adrenergic NB cells in vitro and in vivo.

SWI/SNF sustains chromatin occupancy and expression of CRC members

Highly proliferative MYCN-amplified human NB tumors are associated with poor outcomes (23) and exhibit uniformly high SMARCA4 expression (Supplementary Figure 4G). We therefore focused on MYCN-amplified NB cell lines to establish the chromatin mechanisms related to SMARCA4. We compared ATAC-seq profiles of multiple adrenergic cell lines following 1-h treatment with BRM014 or DMSO, performed TF motif profiling to determine which motifs were altered in response to BRM014 treatment, and cross-referenced motif accessibility changes to measurements of dependency observed in genome-wide CRISPR knock-out screens of NB cells in DepMap (Figure 2A). We found that SWI/SNF inhibition causes profoundly reduced DNA accessibility at sites bearing the motifs of adrenergic CRC TFs PHOX2B, HAND2, MYCN, GATA3, ISL1, and ASCL1 (Figure 2A), which together represent a collection of the most essential genes in NB cells. Interestingly, we also observed decreased accessibility at sites bearing the motifs of EBF3 (Figure 2A), which is a TF involved in neurogenesis (66), a dependency of NB cells, and a potential new member of the adrenergic CRC.

Figure 2.

SMARCA4 sustains enhancer binding and expression of the adrenergic core regulatory circuitry (CRC). (A) DNA accessibility changes of CRC transcription factor (TF) binding motifs induced by BRM014 treatment plotted against sensitivity of NB cells to depletion of individual CRC TFs. (B) Histograms of CRC TFs occupancy changes upon 1 h or 4 h BRM014 treatment. N values for each factor's histogram correspond to the Venn diagram intersections in Supplementary Figure 6. (C) Similarity heatmap of CRC TFs occupancy changes upon 1 h of BRM014 treatment. (D) Enrichment of genomic features at SMARCA4-bound sites characterized by unchanged or decreased binding of individual CRC TFs. (E) Quantification of decreased CRC TFs at SMARCA4-targeted sites. (F) Browser tracks showing changes of MYCN and GATA3 binding at PHOX2B locus upon 1 h of BRM014 treatment. (G) Western blot time course of PHOX2B expression upon BRM014 treatment across MYCN-amplified NB cell lines. (H) Correlation between SMARCA4 and PHOX2B expression in human patient NB samples. (I) Model of SMARCA4 regulation of adrenergic NB CRC.

To determine whether the chromatin binding of CRC TFs is dependent on SWI/SNF catalytic activity, we prepared ChIP-seq libraries at 1 and 4 h following BRM014 treatment and compared them to DMSO. Of the targets tested, HAND2, MYCN, ASCL1, ISL1, PHOX2B, and EBF3 displayed nearly uniform reduced binding to chromatin when subjected to BRM014 treatment at sites co-occupied by both SMARCA4 and each individual TF (Figure 2B, Supplementary Figure 6A). The strong reductions of chromatin occupancy occurred prior to reductions of protein levels (Supplementary Figure 6B), thereby confirming that SWI/SNF catalytic activity is essential for post-translational chromatin binding of these TFs and ruling out protein levels as the cause of reduced chromatin binding.

Interestingly, unlike other CRC TFs, GATA3 preserved its interaction with chromatin even after 4 h of sustained SWI/SNF catalytic inhibition (Figure 2B), in agreement with its role as a pioneer TF (67,68). Metrics of genome-wide changes revealed that, aside from GATA3, other CRC TFs displayed a highly similar pattern of genome-wide changes, indicating a high degree of coordination (Figure 2C). While unchanged sites were associated with promoter regions and marked with H3K4me3, decreased sites for each TF were significantly enriched at intronic and intergenic regions as well the enhancer mark H3K4me1 (Figure 2D), and 77.5% of sites with at least one reduced TF experienced reduced binding of two or more CRC TFs (Figure 2E). This result reveals that SMARCA4 sustains adrenergic identity by acting as a common dependency for multiple adrenergic CRC TFs.

In several malignancies, CRC TFs form autoregulatory loops that promote each other's expression. We therefore assessed whether sustained SWI/SNF inhibition could break the adrenergic autoregulatory loop. We observed several sites with rapid loss of binding of SMARCA4-dependent CRC TFs at the PHOX2B locus within 1 h of BRM014 treatment, as exemplified by MYCN (Figure 2F). Sustained inhibition led to steady downregulation of PHOX2B protein levels across NB cell lines (Figure 2G). These results demonstrate that SWI/SNF is involved in both the regulation of chromatin occupancy and expression of CRC members. Interestingly, GATA3 binding at the PHOX2B locus was preserved (Figure 2F); the subsequent loss of PHOX2B expression therefore indicates that pioneer activity alone is insufficient to sustain the expression of this CRC member. The positive correlation between the expression levels of SMARCA4 and PHOX2B observed in NB patient samples across two different cohorts (58,60) validates the operation of this regulatory circuit in human tumors (Figure 2H). Our findings reveal that SWI/SNF-dependent CRCs arise from an interplay between SWI/SNF-independent pioneer TFs and those that require SWI/SNF activity. Despite the presence of these pioneer TFs however, sustained inhibition of SWI/SNF is sufficient to break key expression feedback loops (Figure 2I).

SWI/SNF inactivation induces arrest and cell death selectively in G1 phase

To analyze acute transcriptional effects of SWI/SNF inactivation, we performed 4sU metabolic labeling of RNA, which permits genome-wide detection of nascent transcripts (Supplementary Figure 7A–C) (47,69). 4sU RNA-seq revealed strong loss of S phase transcripts across three adrenergic NB cell lines upon BRM014 treatment (Figure 3A, B, Supplementary Figure 7D–G, Supplementary Datasets 1–2). Similar transcriptional changes were observed across all three cell lines, as well as upon targeted degradation of SMARCA4 alone (Supplementary Figure 7D–G, Supplementary Datasets 1–2). To ascertain the impact of these transcriptional changes on cell-cycle progression, we conducted cell-cycle analysis by flow cytometry across MYCN-amplified and non-amplified adrenergic NB cell lines and control cells. Remarkably, BRM014 but not DMSO resulted in G1 cell-cycle arrest in all tested NB cells but not control cells (Figure 3C, Supplementary Figure 8A). Similar results were obtained with FHD-286, AU-15330, or with auxin-induced targeted protein degradation of SMARCA4 alone (Supplementary Figure 8B-D). Treatment of orthotopic mouse tumors with BRM014 similarly caused a reduction in the proportion of cells expressing KI67, a marker associated with G2/M phase (Wilcox test, P< 0.001, Figure 3D). Additionally, low SMARCA4 expression is linked to an increase in G1 phase signature and reduction in the G2 phase signature in both the SEQC (58) and TARGET (60) NB patient cohorts (Supplementary Figure 8E), altogether demonstrating similar mechanisms operating in cell culture as well as murine and human tumors.

To dissect the influence of SWI/SNF inhibition on cell-cycle progression in real time, we employed a fluorescence ubiquitination-based cell cycle indicator (FUCCI) reporter (Figure 3E, Supplementary Figure 9A), which leverages endogenous cell-cycle machinery to report on cell-cycle state in living cells (70). Like propidium iodide staining, FUCCI labeling also confirmed G1 accumulation upon BRM014 treatment compared to DMSO (Figure 3F, Supplementary Figure 9B, C). We synchronized FUCCI-labeled IMR-32 cells in S phase using thymidine (Supplementary Figure 9D) then 48 h later, changed into thymidine-free media supplemented with DMSO, BRM014, or other drugs. DMSO-treated cells readily oscillated between cell-cycle phases, while cells treated with BRM014 successfully completed S, G2, and M phases upon release, but arrested in G1 phase (Figure 3G). In contrast to BRM014, treatment with DNA damage-inducing agents (doxorubicin, etoposide, and cisplatin) or nocodazole resulted in successful completion of S phase and arrest in G2/M (Figure 3G). The divergent effects of these drugs demonstrate that SWI/SNF inhibition acts in a manner distinct from DNA damage or G2/M arrest. Additionally, cell death induced by SWI/SNF inactivation was not rescued by overexpression of RNAse H (Supplementary Figure 9E), excluding these effects arising from R-loop formation due to transcription-replication conflict.

Real time analysis of cell death using Cytotox red revealed that the onset of SWI/SNF inhibition-induced cell death was delayed for ∼23 h after release from synchronization into media containing BRM014, when cells were in mid G1 phase (Figure 3H). To test whether the delay was due to the synchronized entry of cells to G1 phase or to delayed drug onset, we separately treated an unsynchronized cell population, where a fraction of cells is always in G1 phase. In unsynchronized cells, cell death onset was detected within 2 h after BRM014 treatment, ruling out delayed drug onset (Figure 3H). Similar results were observed following treatment with FHD-286 or AU-15330 (Supplementary Figure 9F–I), thereby validating an on-target mechanism of action and ruling out non-catalytic roles of SWI/SNF ATPases being essential towards survival outside of G1 phase.

To test whether the dependency in G1 phase was caused by inhibition of processes temporally restricted to G1 phase or instead reflected a delayed consequence of defects that occurred during previous phases of the cell cycle, we treated cells with BRM014 at different timepoints after release from synchronization. Cell death onset was maintained regardless of whether BRM014 was supplied immediately, or 6 h (following completion of S phase), 12 h (following completion of mitosis), or 18 hours (early G1) after release from synchronization (Figure 3I, Supplementary Figure 9J). These results exclude the possibility that loss of viability occurs due to defects occurring during previous phases of the cell cycle. We therefore reveal a role for SWI/SNF that is essential in G1 phase and necessary for G1-S transition, but dispensable for survival in other phases (Figure 3J). To our knowledge, this is the first report that SWI/SNF ATPase activity is essential only during G1 phase.

SWI/SNF-dependent CRC enhancers are G1-specific and needed for cell-cycle progression

We combined FACS-based separation of FUCCI-labeled cells with ATAC-seq to directly map the temporal variation of genome-wide accessibility at SMARCA4-bound sites across the cell cycle (Figure 4A, Supplementary Figure 10A). Of 14436 SMARCA4-bound sites identified via ChIP-seq, 13289 (∼92%) displayed overlap with DNA accessibility peaks in at least one cell-cycle phase. We refer to these sites as ‘SMARCA4-targeted’ sites. Our analysis revealed cyclic patterns of DNA accessibility at SMARCA4-targeted sites across the cell cycle (Figure 4B). Unbiased k-means clustering of SMARCA4-targeted sites revealed the existence of 3 clusters with distinct temporal patterns (Figure 4C). Sites in cluster 1 (N = 4740 sites) displayed the highest accessibility in G1 phase, while cluster 2 (N = 4511) peaked during S phase, and cluster 3 (N = 4167) exhibited elevated accessibility in G2. Pooling of all cell-cycle ATAC-seq datasets revealed that each cluster was associated with distinct overall responses to BRM014: Cluster 1 showed the strongest reduction of DNA accessibility in response to SWI/SNF catalytic inhibition (mean log₂fc = –0.53), followed by cluster 3 (mean log₂fc = –0.32). Both clusters 1 (P< 2.2e-16) and 3 (P< 2.2e-16) were significantly reduced compared to cluster 2 (mean log₂fc = -0.17, Figure 4D). Annotation of each cluster's genomic regions revealed that cluster 1 was strongly enriched in intergenic regions, consistent with enhancers, while cluster 2 was enriched in promoter regions, and cluster 3 displayed mixed enrichment of enhancer and promoter regions (Supplementary Figure 10B). DNA accessibility at SMARCA4-regulated sites thus oscillates in a cell cycle-dependent manner and sites with the highest dependency on SWI/SNF activity are enhancers with peak accessibility in G1 phase.

Figure 4.

Temporal patterns of DNA accessibility reveal CRC losses are enriched at G1-specific enhancers regulated by SMARCA4. (A) FACS strategy and gating to analyze genome-wide accessibility of cells in individual cell cycle phases. (B) Example browser tracks of cell cycle-phase dependent site sensitive to SWI/SNF inactivation. (C) Classification of genome-wide SMARCA4 sites based on their cell cycle-phase dependent accessibility changes, and their response to SWI/SNF inactivation. (D) DNA accessibility changes upon BRM014 treatment at SMARCA4-bound clusters shown in panel (C). Box plot centers indicate median values. (E) Enrichment of sites with reduced CRC TF occupancy within clusters shown in panel (C). (F) Heatmap of cell cycle phase-specific accessibility changes of CCND1 enhancer, and browser tracks showing reduced accessibility in late G1 phase at CCND1 enhancer. Inset shows rapid loss of CRC TF binding at CCND1 enhancer within 1 h of BRM014 treatment. (G) Western blot analysis of CCND1 expression changes upon BRM014 treatment across NB cell lines. (H) Expression of CCND1 in NB patient tumors with low SMARCA4 expression compared to those with high SMARCA4 expression. Box plot centers indicate median values.

Finally, we investigated the association between cell-cycle phase-specific clusters and the changes in CRC ChIP-seq peaks following catalytic inhibition of SWI/SNF. Compared to all sites, cluster 1 sites showed significant enrichment with loss of HAND2 (hypergeometric P= 1.6e-38), MYCN (P= 6.0e-20), ISL1 (P= 8.9e-16), PHOX2B (P= 7.6e-3), or EBF3 (P= 0.039, Figure 4E). Analysis of the cyclic patterns of DNA accessibility over the cell cycle therefore revealed that SWI/SNF-dependent CRC binding sites are enriched at enhancers activated specifically during G1 phase.

Although we found 4740 such sites, one noteworthy example of a cluster 1 site is the enhancer region of cyclin D1 (CCND1), a critical regulatory factor involved in G1-S transition (71), whose transcription was decreased across NB cell lines following BRM014 treatment (Supplementary Figure 7F). The CCND1 enhancer normally has peak accessibility in G1 phase but is strongly downregulated by BRM014 (Figure 4F). The loss of accessibility at the CCND1 enhancer coincides with loss of SMARCA4-dependent CRC TFs within 1 hour (Figure 4F) and a reduction in CCND1 protein levels following SWI/SNF inhibition across a panel of adrenergic NB cells (Figure 4G). In agreement, human NB tumors in TARGET (P= 1.8e-9) and SEQC (P= 1.8e-6) cohorts with low SMARCA4 expression have significantly lower levels of CCND1 than tumors with high SMARCA4 (Figure 4H), consistent with the same mechanisms operating in human tumors.

SWI/SNF inhibition potentiates cell-cycle exit by retinoic acid

During G1 phase of the cell cycle, cell fate is determined by the competing influences of replication-promoting cues (72) and differentiation cues. We hypothesized that the disruption of CRC activity that promotes G1-S transition by SWI/SNF inhibition could render cells more responsive to differentiation cues. One of the most widely employed differentiation agents for NB is retinoic acid (73), which induces G1 arrest (Figure 5A) and promotes differentiation by re-shaping the CRC landscape of NB cells (37). Depletion of CRC activity by SMARCA4 inhibition could therefore confer a kinetic advantage for retinoic acid-mediated differentiation (Figure 5B).

Figure 5.

SWI/SNF inhibition synergizes with retinoic acid to promote cell-cycle exit in G1. (A) Flow cytometry cell cycle analysis of NB cells upon treatment with 13-cis retinoic acid (13-cis RA), BRM014, or vehicle control (DMSO). (B) Model by which SWI/SNF inhibitors sensitize cells to retinoic acid-mediated cell cycle exit. (C) Metagene plots of MYCN and retinoic acid receptor alpha (RARA) chromatin binding at SMARCA4 sites in the presence of BRM014 or DMSO. (D) Example ChIP-seq browser track demonstrating preserved RARA binding despite loss of MYCN binding. (E) Experimental design. (F) Outgrowth of NB cell lines following cross-titration of BRM014 and 13-cis RA. (G) Heatmap of synergy scores for BRM014 and 13-cis RA cross-titration across NB cell lines measured by Loewe additivity (N ≥ 3 independent replicates). (H) Isobolograms of BRM014 and 13-cis RA treatments demonstrating synergy (N ≥ 3 independent replicates).

To ascertain whether SWI/SNF inhibition interferes with retinoic acid signaling, we investigated the impact of SWI/SNF inhibition on the chromatin occupancy of retinoic acid receptor alpha (RARA). In contrast to CRC factors like MYCN, whose binding is rapidly decreased at SMARCA4 sites upon BRM014 treatment within 1 hour, RARA chromatin occupancy remained preserved upon acute inhibition (Figure 5C, D) despite significant overlap in SMARCA4 and RARA peaks in NB cells (Supplementary Figure 11A). RARA thus operates independently of acute SWI/SNF catalytic activity.

We cross-titrated BRM014 and 13-cis retinoic acid (13-cis RA, isotretinoin), a compound used extensively for maintenance therapy for NB, across several MYCN-amplified adrenergic NB cell lines, and assessed the ability of cells to grow out upon drug release (Figure 5E). When jointly administered, BRM014 strongly sensitized NB cells to differentiation by 13-cis RA, as assessed by Loewe additivity of drug synergy and isobologram analysis (Figure 5F–H). Similar results were obtained using all-trans retinoic acid (ATRA, Supplementary Figure 11B), or when FHD-286 or AU-15330 was applied instead of BRM014 (Supplementary Figure 11C). Drug synergy was also observed in SK-N-SH cells, a line of NB cells that lacks MYCN amplification, indicating that this synergy may also influence non-MYCN-amplified adrenergic NB cancer cells (Supplementary Figure 11D-E). Drug synergy was not observed with other drugs known to primarily target S or G2/M (Supplementary Figure 11F), revealing the specificity of sensitization during G1 phase. We conclude that impaired replication cues induced by SWI/SNF inhibition can potentiate signals and interventions that act during G1 to promote differentiation.

Retinoid-induced differentiation in G1 is enhanced by SWI/SNF inhibition in diverse SWI/SNF-addicted cells

In addition to NB, SWI/SNF ATPase activity has emerged as a crucial oncogenic dependency in several other cancer types, such as acute myeloid leukemia (AML) (2), prostate cancer (PRAD) (4), and uveal melanoma (UVM) (5). Consistent with our findings in NB, SWI/SNF plays a pivotal role in establishing accessibility for autoregulatory circuits driven by distinct TFs within each of these cancer contexts (1,4,5). To evaluate the broader applicability of our findings beyond NB, we examined cell-cycle changes following SWI/SNF inactivation in cell lines derived from the SWI/SNF-addicted malignancies described above. Remarkably, all SWI/SNF-addicted cell lines tested exhibited G1 cell-cycle arrest following treatment with 1 μM BRM014 (Figure 6A, Supplementary Figure 12A, B). G1 arrest moreover coincided with a reduction in CCND1 protein levels. These changes were not observed in control cell lines where SWI/SNF activity is dispensable (Figure 6B).

Figure 6.

SWI/SNF inhibition potentiates retinoic acid-induced growth arrest in diverse addicted cancers. (A) Cell cycle analysis of SWI/SNF-addicted cells using propidium iodide (PI) staining and flow cytometry following treatment with BRM014 or DMSO. (B) Western blot analysis of CCND1 upon BRM014 treatment in acute myeloid leukemia (AML), uveal melanoma (UVM), and prostate cancer (PRAD) cell lines compared to control cell lines. (C) Correlation between SMARCA4 expression and G1 or G2 cell cycle signatures in human patient samples determined by gene set variation analysis (GSVA). Linear regression of each dataset is presented. (D) Expression of CCND1 in SWI/SNF-addicted tumors with low SMARCA4 expression compared to those with high SMARCA4 expression. Error bars: median ± SE. (E) Heatmap of synergy scores for BRM014 and all-trans retinoic acid (ATRA) cross-titration measured by Loewe additivity.

In agreement with our findings in vitro, we analyzed data from cohorts of human AML, PRAD, and UVM tumors and found that low SMARCA4 expression in these settings is associated with increased G1 and reduced G2 signatures (Figure 6C) as well as reduced expression of CCND1 (Figure 6D). These results underscore the general significance of SWI/SNF activity in replication commitment across SWI/SNF-addicted cancer contexts. To evaluate the interaction between retinoids and SWI/SNF inhibition in these cancer settings, we cross-titrated BRM014 and ATRA in THP-1 (AML), LNCaP (PRAD), and 92.1 (UVM) cells, and assessed their ability to grow out following drug release (Figure 6E). BRM014 strongly sensitized these cells to retinoid-induced differentiation, confirming the generality of our findings in diverse SWI/SNF-dependent settings.

Discussion

Here, we explore how the dependency of cells on SWI/SNF catalytic activity varies over the cell cycle. The mechanisms underlying addiction to SWI/SNF that we uncover are distinct from those described in other contexts, such as regulation of genomic integrity (14,74), R-loop formation (15), mitotic bookmarking (13), or sister chromatid resolution (11). In NB, the dependency is temporarily restricted to G1 phase and arises from the co-opting of SMARCA4 activity to promote DNA accessibility for autoregulatory, tumor-specific circuitry that operates in G1, where it is essential for replication commitment.

We show that DNA accessibility at most SWI/SNF-regulated chromatin sites oscillates in characteristic temporal patterns over the course of the cell cycle. Our approach uncovers that SWI/SNF-dependent CRC enhancers display widespread activation specifically in G1 phase, when they are required to express genes needed for cell-cycle progression. Because impairment of this circuitry is not deleterious in S or G2/M, this implies that a primary function of the CRC is to provide an alternative pathway for constitutive replication commitment (i.e. in the absence of mitogens). In settings where constitutive SWI/SNF-dependent CRCs promote replication, SWI/SNF inhibition impairs the ability of this network to express factors needed for replication commitment.

The ATPase activity of SWI/SNF complexes represents a targetable oncogenic dependency that has been explored against other cancer types, including AML (1,8), prostate cancer (4), and uveal melanoma (2,4,5), each of which is associated with a distinct CRC. Despite little overlap in the specific TFs making up these CRCs, we find that SWI/SNF inhibition remarkably promotes G1 accumulation, impairs CCND1 expression, and potentiates retinoic acid-mediated growth arrest across cells from each of these SWI/SNF-addicted cancers (Figure 6). Because SWI/SNF inhibition prevents cell-cycle progression in G1, it may generally potentiate retinoid therapies, which are currently used clinically against NB (75) and AML (76). As a result, broader consideration of SWI/SNF inhibitors, not only in acute settings, but also as part of maintenance therapies is warranted. Our observations provide insight into the temporal utilization of SWI/SNF activity, as well as a compelling rationale to further explore the temporal basis for enhancer activation and the interactions of SWI/SNF inhibitors with drugs that act in G1 phase of the cell cycle.

Data availability

The SEQC cohort was analyzed from Gene Expression Omnibus (GEO) accessions GSE47792, GSE49710, and GSE49711 (58). SU2C-PCF (61) cohort data were obtained from cBioPortal. AML TARGET and NB TARGET cohort data were obtained from the TARGET consortium. TCGA-LAML, TCGA-PRAD, and TCGA-UVM were obtained from the TCGA GDC data portal. CCND1 enhancer-promoter annotation was confirmed using HiChIP data from GEO accession GSE136208 (35). All high-throughput sequencing data generated for this project have been deposited in the GEO database with SuperSeries accession number GSE190709. Scripts used for genome-wide analyses are available on Zenodo (63) and at the following link: https://github.com/hodgeslab/workflows.

Supplementary data

Supplementary Data are available at NAR Online.

Funding

V Foundation [V2018-003 to H.C.H.]; Gabrielle's Angel Foundation, the Mark Foundation for Cancer Research [20–024-ASP to H.C.H.]; Helis Family Medical Research Foundation, National Institutes of Health (NIH) [R01CA272769 to H.C.H, R35GM137996 to H.C.H., R01CA268380 to W.W.]; Cancer Prevention and Research Institute of Texas (CPRIT) [RR170036 to H.C.H.]; U.S. Department of Defense (DOD) [PC210079 to W.W.]; Czech Academy of Sciences [RVO: 61388963 to R.N.]; Cytometry and Cell Sorting Core at BCM (CPRIT) [RP180672]; NIH [P30CA125123, S10RR024574]; MD Anderson Next-Generation Sequencing Core (CPRIT) [RP120348]. Funding for open access charge: NIH [R01CA272769].

Conflict of Interest Statement

None declared.