Enhancer Domains in Gastrointestinal Stromal Tumor Regulate KIT Expression and are Targetable by BET Bromodomain Inhibition

Abstract

Gastrointestinal stromal tumor (GIST) is a mesenchymal neoplasm characterized by activating mutations in the related receptor tyrosine kinases KIT and PDGFRA. GIST relies on expression of these unamplified receptor tyrosine kinase (RTK) genes through a large enhancer domain, resulting in high expression levels of the oncogene required for tumor growth. Although kinase inhibition is an effective therapy for many GIST patients, disease progression from kinase resistance mutations is common and no other effective classes of systemic therapy exist. In this study, we identify regulatory regions of the KIT enhancer essential for KIT gene expression and GIST cell viability. Given the dependence of GIST upon enhancer-driven expression of RTKs, we hypothesized that the enhancer domains could be therapeutically targeted by a BET bromodomain inhibitor (BBI). Treatment of GIST cells with BBIs led to cell cycle arrest, apoptosis, and cell death, with unique sensitivity in GIST cells arising from attenuation of the KIT enhancer domain and reduced KIT gene expression. BBI treatment in KIT-dependent GIST cells produced genome-wide changes in the H3K27ac enhancer landscape and gene expression program, which was also seen with direct KIT inhibition using a tyrosine kinase inhibitor (TKI). Combination treatment with BBI and TKI led to superior cytotoxic effects in vitro and in vivo, with BBI preventing tumor growth in TKI-resistant xenografts. Resistance to select BBI in GIST was attributable to drug efflux pumps. These results define a therapeutic vulnerability and clinical strategy for targeting oncogenic kinase dependency in GIST.

Introduction

GIST is among the most common sarcoma subtypes and the majority of cases are characterized by activating mutations in the related RTKs KIT or PDGFRA (1). GIST’s cell of origin is the interstitial cell of Cajal or its precursors, which natively rely upon signaling through wild-type KIT for survival (2). Diverse activating mutations in KIT or PDGFRA lead to ligand-independent RTK signaling associated with GIST development, and families with germline activating KIT or PDGFRA mutations are at high risk for developing GIST (3). In contrast to many other tumor types’ oncogenic transformation, GIST does not rely on amplification of KIT or PDGFRA to drive neoplastic development (4,5), relying instead on activating mutations within these natively expressed RTKs to initiate disease and spur disease progression (5).

For metastatic or unresectable GIST, inhibition of signaling through these RTKs can lead to disease control (6), and in select cases neoadjuvant or adjuvant therapy is therapeutically beneficial (7,8). Approved second- and third-line therapies also rely on kinase inhibition (9), though disease progression is common in the metastatic setting. Currently, the focus of clinical trials in GIST centers on RTK inhibition, utilizing compounds that attempt to overcome TKI-resistance mutations (1). There are no approved therapies for GIST apart from TKIs, and effective treatments with novel mechanisms of action have not yet been elaborated.

The bromodomain and extraterminal domain-containing (BET) family of proteins consists of BRD2, BRD3, BRD4 and BRDT, which bind acetylated proteins such as histones and transcription factors and function in the cell to regulate chromatin state and transcription (10). BRD4 has been well studied for its essential role in binding to large regions of chromatin modified with H3K27ac and contributing to the formation of large transcriptional regulatory domains termed super-enhancers (SEs) (11). SEs regulate expression of lineage-specific genes important for cellular function, and can be dysregulated to sustain oncogene expression and cancer development (12). Due to their dependence upon SEs, select cancer oncogenes can be preferentially targeted through the inhibition of BRD4 (13). This has generated enthusiasm for inhibiting BRD4 function with BBIs, and several such compounds have been developed that can therapeutically target BET proteins (14,15).

We previously used H3K27ac chromatin immunoprecipitation with sequencing (ChIP-seq) of GIST tumor samples and cell lines to describe the enhancer landscape of GIST (16). This analysis identified a large enhancer domain driving KIT gene expression that was unique to GIST. Based upon this finding, we hypothesized that disruption of the KIT enhancer domain represents a therapeutic vulnerability in GIST. In the present study, we show that CRISPR-mediated disruption of the KIT enhancer domain decreases GIST cell proliferation and KIT mRNA. Treatment of KIT-dependent GIST cells with BBIs causes cell cycle arrest, apoptosis and cell death by decreasing KIT gene expression. Using a CRISPR based approach, sensitivity to BBIs could be abrogated by abstracting KIT expression from its native enhancer. Both BBI and TKI treatment led to genome-wide changes in gene expression and the enhancer landscape, with BBIs eliminating BRD4 occupancy at the KIT enhancer domain. BBIs produced a phenocopy of TKI treatment through disrupting a transcriptional program dependent upon KIT signaling. Through their unique mechanisms of action independently targeting oncogenic KIT signaling, TKIs and BBIs were found to have combinatorial toxicity in vitro and in vivo. GIST cell lines were found to be resistant to select BBIs through the expression of drug efflux pumps, representing a new mechanism of resistance to this family of drugs. Taken together, these results define the mechanism of selective toxicity of BBIs in GIST, explore a new therapeutic approach for this disease exploiting combinatorial effects of BBIs and TKIs on oncogenic KIT signaling, and identify a novel mechanism of resistance to BBIs.

Materials and Methods

Cell lines and virus production.

All cell lines tested negative for mycoplasma infection on routine surveillance, with last testing performed on November 19^th 2018. Experiments using cell lines were performed within 15 passages from the initial stock. Human embryonic kidney (HEK) 293FT (ThermoFischer Cat# R70007, RRID: CVCL_6911), the GIST cell lines GIST-T1 (Cosmo Bio Cat# PMC-GIST01-COS, RRID:CVCL_4976; KIT mutation in exon 11 Δ560-578), GIST430 (RRID: CVCL_7040; KIT mutation in exon 11 Δ560-576), GIST48 (RRID: CVCL_7041; KIT mutations in exon 11 V560D and exon 17 D820A), GIST48B (RRID: CVCL_M441; KIT-independent) and GIST226 (KIT-independent) and the MaMel melanoma cell line (17) were cultured in Dulbecco’s modified Eagle’s medium containing 10% FBS, 2 mM L-glutamine, 100 mg/ml penicillin, and 100 mg/ml streptomycin. GIST882 (RRID: CVCL_7044; KIT mutation in exon 13 K642E) was grown in RPMI with identical supplementation. Non-commercial cell lines were obtained from the laboratory of Jonathan Fletcher between 2014 and 2016. KIT exons were sequenced to confirm the expected coding mutations and cell identity. Transfections were performed with Lipofectamine 2000 (Invitrogen). Lentiviral production and stable cell line generation was performed as previously described (18). Briefly, 293FT cells were cotransfected with pMD2.G (Addgene #12259), psPAX2 (Addgene #12260) and the lentiviral expression plasmid. Viral supernatant was collected at approximately 72 h and debris removed by centrifugation at 1,000g for 5 min. Cells were transduced with viral supernatant and polybrene at 8 μg/mL by spinoculation at 680g for 60 min. For growth over time assays, 15 × 10³ cells were dispensed per well in a 96 well plate and cell count performed twice per week without subculture.

ChIP-seq.

ChIP-seq was performed as previously described (16). Approximately 2-10 × 10⁷ cells were incubated in 1% formaldehyde for 10 min. Following incubation, excess formaldehyde was quenched with glycine at 0.125 M. Samples were washed with PBS, and intact nuclei extracted by serial incubations in lysis buffers for 10 min at 4°C: Lysis Buffer 1 (50 mM HEPES, pH 7.5, 140 mM NaCl, 1 mM EDTA, 10% glycerol, 0.5% IGEPAL CA-630, 0.25% Triton X-100, 1x Halt protease inhibitor cocktail (Thermo Scientific)); Lysis Buffer 2 (10 mM Tris, pH 8, 200 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 1x Halt protease inhibitor cocktail). Nuclei were spun at 1,350g for 5 min between lysis steps and then resuspended in sonication buffer with SDS (50 mM HEPES, pH 7.5, 140 mM NaCl, 1 mM EDTA, 1mM EGTA, 1% Triton X-100, 0.1% Na-deoxycholate, 0.5% SDS, 1x Halt protease inhibitor) prior to Bioruptor sonication (Diagenode; high output sonication in 30 s intervals) for 20-25 min in 15 mL TPX tubes. Sonicated samples were spun 20,000g for clarification and supernatant diluted 1:5 in sonication buffer prior to overnight incubation with Dynabeads (Life Technologies) pre-bound with antibody (H3K27ac, Abcam ab4729, RRID: AB_2118291; BRD4, Bethyl Laboratories A301-985A100) per manufacturer’s recommendations. 50 μL of sonicated chromatin was saved as input control. Following overnight immunoprecipitation, samples were washed serially with sonication buffer, sonication buffer with 500 mM NaCl, LiCl buffer (20 mM Tris, pH 8, 1 mM EDTA, 25 mM LiCl, 0.5% IGEPAL CA-630, 0.5% Na-deoxycholate) and TE-NaCl buffer (50 mM Tris, pH 8, 1 mM EDTA, 50 mM NaCl). Samples were eluted by heating at 65°C for 15 min in elution buffer (50 mM Tris, pH 8, 10 mM EDTA, 1% SDS), and crosslinks reversed by incubation at 65°C for 16 h. DNA was purified by serial incubation with 0.2 mg/mL RNaseA (Thermo Scientific) then Proteinase K (Ambion), followed by extraction with phenol-chloroform and ethanol precipitation. Libraries for single-end 75 bp sequencing on an Illumina NextSeq 500 were prepared with a ThruPLEX DNA-seq Kit (Rubicon), with sample analysis and quantitation using a Qubit dsDNA HS Assay Kit (Life Technologies) and TapeStation 2200 (Agilent).

RNA-seq.

GIST-T1 cells were treated in biological triplicates for 6 or 24 h with the indicated drugs at 500 nM. Total RNA was isolated using an RNeasy Plus Kit (Qiagen). RNA was normalized to cell count, then ERCC RNA Spike-In Control (Ambion) was added to allow for normalization of gene expression changes (19). RNA concentration was measured by Nanodrop (Thermo Scientific) and quality by Bioanalyzer (Agilent). Libraries for Illumina NextSeq 500 sequencing were prepared using TruSeq Stranded mRNA Library Prep Kit (Illumina) and equimolar multiplexed libraries were sequenced with single-end 75 bp reads.

Sequencing data analysis.

Computational methods used for data analysis have been described previously (16,20) and are available from the Bradner and Lin Laboratory github pages (github.com/BradnerLab/pipeline, github.com/linlabbcm).

Sequencing read alignment and annotation.

All Chip-Seq data were aligned to the human reference genome assembly hg19, GRCh37 (https://www.ncbi.nlm.nih.gov/assembly/GCF_000001405.13/) using Bowtie2 (21) (version 2.2.1), and gene annotations by gencode annotation release 19 (gencodegenes.org/releases/19.html). Normalized read density was calculated using Bamliquidator (version 1.0) read density calculator. Aligned reads were extended by 200 bp and the density of reads per base pair was calculated. In each region, the density of reads was normalized to the total number of million mapped reads, generating read density in units of reads per million mapped reads per bp (rpm/bp).

Peak finding and enhancer classification.

Peak finding was performed using Model-based Analysis for ChIP-Seq (MACS, version 1.4.2) (22), with p-value cutoff of 1 × 10⁻⁹. As control, matched input chromatin was used to generate background signal for peak determination. Calling of typical enhancers and super-enhancers was performed using the ROSE2 (13) software package. Gene assignment for ROSE2-defined typical enhancers was performed using ROSE2 genemapper. For SE determination, an optimized stitching region was used to cluster H3K27ac peaks into contiguous enhancers at each enhancer region. Based upon gene expression data, SEs called by ROSE2 within the KIT locus were assigned to KIT.

ChIP-seq data representations.

Ranked enhancer plots were generated using ROSE2 by quantifying total H3K27ac signal at each enhancer. Transcriptional start site and H3K27ac-defined enhancer signal was merged into 50 bp windows using bamliquidator to generate meta tracks. Waterfall plots showing log₂ fold change in SE regions between treatment groups were generated by comparing compiled background-subtracted H3K27ac signal at SEs present in at least one condition. Heat map visualizations of ChIP-seq data were generated using ChAsE (23). Individual ChIP-seq track displays were generated using bamplot. Multi-sample meta tracks of H3K27ac ChIP-seq signal were generated with each individual sample displayed as a transparent track in rpm/bp and the opaque line representing the average signal across all samples.

RNA-seq alignment and analysis.

RNA-seq data, cell-count normalized where appropriate, were aligned using HISAT (24) with expression quantification using Cufflinks (25) to generate gene expression values in fragments per kilobase of transcript per million mapped reads (FPKM) units. FPKM values were normalized to ERCC spike-in, as previously described (19). Heat maps of log₂ fold change in FPKM compared to DMSO were generated using Morpheus (software.broadinstitute.org/morpheus/). Gene set enrichment analysis (GSEA) (26) was performed using C2 (curated gene sets) in the Molecular Signatures Database (software.broadinstitute.org/gsea/) or custom gene lists.

Immunoblotting.

Cells and tissue were lysed in RIPA buffer containing protease inhibitor cocktail (Roche) and centrifuged at 14,000g for 10 min to remove genomic DNA and debris. Fresh frozen tumor samples were obtained from patients following written informed consent to an Institutional Review Board (IRB) approved research protocol and undergoing surgery at the Brigham and Women’s Hospital/Dana-Farber Cancer Institute. Protein concentrations were determined using a bicinchoninic acid-based assay (Pierce Biotechnology). Protein samples were subjected to SDS-PAGE and Western blotting with the following antibodies: KIT (Cell Signaling Technology Cat# 3074S, RRID: AB_1147633; 1:1,000), phospho-KIT (Cell Signaling Technology Cat# 3073P, RRID: AB_10329035; 1:1,000), PDGFRA (Cell Signaling Technology Cat# 3174, RRID: AB_2162345; 1:1,000), HA (Cell Signaling Technology Cat# 2367S, RRID: AB_10691311; 1:1,000), ERK (Cell Signaling Technology Cat# 9107S, RRID: AB_10695739; 1:2,000), phospho-ERK (Cell Signaling Technology Cat# 4370, RRID: AB_2315112; 1:1,000), ETV1 (Abcam Cat# ab81086, RRID: AB_1640495; 1:500), HEXIM1 (Cell Signaling Technology Cat# 9064S, RRID: AB_10998620; 1:1,000) and vinculin (Santa Cruz Biotechnology Cat# sc-5573, RRID:AB_2214507; 1:1,000). ERK and vinculin were used as loading controls. Western blots were probed with anti-mouse or anti-rabbit secondary antibodies and detected using the Odyssey CLx infrared imaging system (LI-COR Biosciences). Quantification of band intensity was performed using Image Studio (LI-COR Biosciences), with values normalized to control conditions. Immunoblots shown are representative of at least three independent experiments.

High-throughput drug treatment.

Bromodomain and kinase inhibitors were tested in an automated 384-well format, plating 1,900 cells per well and delivering 100 nL drug by robotic pin transfer (JANUS workstation, Perkin Elmer). Cell viability was determined at 72 h using the ATPlite Luminescence Assay System (Perkin Elmer). For synergy studies, GIST-T1 cells were seeded into a 384-well plate and incubated with a pairwise combination of imatinib and JQ1. Combination indices were determined using the median-effect principle of Chou and Talalay (27), and an isobologram plot generated with points representing paired values of drug assessed for synergy. The diagonal line represents a combination index of 1 to indicate drug additivity. Points below the line represent synergy, while point above the line represent antagonism. For experiments utilizing tariquidar, cells were pre-incubated for 24 h with tariquidar prior to administration of the experimental drug. All high-throughput drug treatments were performed in at least quadruplicate and on at least two independent 384-well plates. Imatinib and sunitinib were obtained from LC Laboratories, tariquidar from Selleckchem and BET bromodomain inhibitors were synthesized in the Bradner laboratory.

Cell cycle and apoptosis.

Cell cycle analysis was performed following drug treatment for 24 h (GIST-T1, GIST48B) or 72 h for slowly growing cell lines (GIST430, GIST882, GIST48). Cells were trypsinized, washed in PBS and fixed in 70% ethanol. Propidium iodide at 25 μg/mL (Life Technologies) and RNAse A at 0.2 mg/mL (ThermoFischer) were used to stain nuclear DNA. Analysis was performed on a Guava easyCyte Flow Cytometer (EMD Millipore), and single cells were assessed for nuclear content using Guava InCyte software. Apoptosis and cell death were measured following 72 h of drug treatment using Guava Nexin Reagent (EMD Millipore) per manufacturer’s recommendations. Non-apoptotic cells stain negative for Annexin V and 7-AAD, early apoptotic cells stain positive for Annexin V but negative for 7-AAD and late apoptotic and dead cells stain positive for both Annexin V and 7-AAD. Staining was assayed on a Guava easyCyte Flow Cytometer and data analyzed using Guava InCyte software.

Quantitative RT-PCR.

Cells were trypsinized and washed in PBS for RNA extraction using the RNeasy Mini Kit (Qiagen). Libraries of cDNA were made using SuperScript VILO cDNA Synthesis Kit (Invitrogen). RT-PCR was performed using SYBR Select Master Mix (Life Technologies) on a ViiA 7 Real-Time PCR System (Life Technologies). Relative mRNA levels were calculated by the ΔΔCt method using GAPDH expression as reference. All RT-PCR amplicons were verified by gel electrophoresis and DNA sequencing. PCR primers are listed in Table S1.

Cloning and CRISPR assays.

Cell lines stably expressing a human codon-optimized Streptococcus pyogenes Cas9 (Addgene #73310) or dCas9-KRAB (Addgene #89567) were generated by viral transduction. CRISPR single-guide RNAs (sgRNAs) were cloned into Lenti-sgRNA-EFS-GFP (LRG, Addgene #65656) (28), or else into a modified plasmid with GFP replaced by copGFP linked to a puromycin resistance gene by a 2A peptide (LRGP). All sgRNAs were designed using the CRISPR Design Tool (crispr.mit.edu) or CHOPCHOP (29) (chopchop.cbu.uib.no) and detailed in Table S1. Selected sgRNAs were designed to target the KIT enhancer domain or upstream coding region or kinase domain of KIT. For proliferation and RT-PCR assays, cells were infected with LRGP-derived lentivirus to target a GFP-positive rate ≥95%, and further selected with puromycin to ensure all cells express the sgRNA of interest. Cell count and GFP expression were quantified on a Guava easyCyte Flow Cytometer. For sgRNA/GFP competition assays, flow cytometry to quantify GFP-expressing cells was performed in 96-well plates. To determine indels generated following Cas9 cleavage, individual cell clones were isolated and target alleles cloned into a plasmid for Sanger sequencing, or amplified genomic DNA was directly sequenced and alleles deconvoluted using TIDE (30). For CRISPR rescue experiments, GIST-T1 cells expressing Cas9 were transduced simultaneously with a lentivirus containing the KITe1.1 sgRNA and another lentivirus containing the KIT rescue construct. KIT and HAND1 were cloned from a cDNA library generated from GIST-T1, and the KIT N822K mutation generated by primer-directed mutagenesis. ABCB1 was cloned by PCR-amplification of pHaMDRwt (Addgene #10957) with addition of restriction sites enabling ligation into a lentiviral vector.

Xenograft Models.

The PG27 patient derived xenograft was obtained from a patient undergoing clinically indicated surgery and following written informed consent to a Dana-Farber Cancer Institute IRB-approved research protocol. Cryopreserved tumor or the GIST-T1 cell line mixed 1:1 with matrigel were implanted subcutaneously into 6-week old female nude mice (NU/NU; Charles River Laboratories). GIST-T1 tested negative for mycoplasma and rodent pathogens (Charles River Laboratories). Engrafted mice were enrolled into treatment groups when tumors reached approximately 100-200 mm³ in size, as measured by calipers and determined by the tumor volume equation: volume = long diameter² × short diameter × 0.5. Mice were randomly assigned to treatment groups administered imatinib (50 mg/kg gavage daily, prepared fresh daily in 0.9% saline), JQ1 (50 mg/kg IP 5 days per week, prepared twice weekly in 5% DMSO and 10.5% cyclodextrin in 0.9% saline), GSK525762 (I-BET762; 50 mg/kg IP 5 days per week, prepared twice weekly in 5% DMSO and 20% cyclodextrin in 0.9% saline) or vehicle IP injection. No statistical methods were used to predetermine sample size, no animals were excluded from tumor volume analysis and no animals lost weight or died during drug treatment. Tumors were dissected and snap-frozen or fixed in 10% formalin 6 h after the last injection for corollary studies including RNA isolation, Western blot and H&E staining of sectioned tumors. All procedures were conducted under protocols approved by the Institutional Animal Care and Use Committee at Dana-Farber Cancer Institute.

Statistical analysis.

Center values, error bars, P-value cutoffs, number of replicates and statistical tests are identified in the corresponding figure legends. Replicates represent separate treatments of cultures from the original stock unless otherwise specified. Error bars are shown for all data points with replicates as a measure of variation within each group. Samples sizes were not predetermined, and the investigators were not blinded to allocation during experiment or outcome assessment.

Data availability:

ChIP-seq and RNA-seq data presented in this publication are available online through the GEO Publication Reference ID GSE113217. Additional sequencing data has been previously published under GSE95864 (16) and GSE106626 (31).

Results

The KIT enhancer domain is responsible for KIT gene expression and cell proliferation in KIT-dependent GIST.

We have previously demonstrated the H3K27ac landscape in GIST, which we used to define a conserved enhancer and transcriptional program in this disease (16). The enhancer program was found to be unique to GIST in comparison to other cell types that express KIT, with a conserved set of transcription factors (TFs) driving the gene expression program in GIST tumors and cell lines. GIST does not rely on KIT amplification and was found to possess large enhancer domains that drive high levels of KIT gene expression. We reasoned that genomic disruption of this SE by CRISPR would affect KIT gene expression and cell viability in KIT-dependent GIST cell lines.

In comparison to the KIT-independent GIST cell line GIST48B, which has lost expression of KIT through in vitro selection, all KIT-dependent GIST cell lines displayed a conserved enhancer landscape at the KIT locus, with a SE upstream of the KIT TSS (Fig. S1A). Evaluating ATAC-seq (Assay for Transposase-Accessible Chromatin using sequencing) data to identify sites of DNA accessibility in KIT-dependent and KIT-independent GIST cell lines, complimentary findings to the H3K27ac signal were identified at the KIT locus (Fig. S1B). The TFs ETV1 (32), FOXF1 (31) and HAND1 (16) have previously been implicated in GIST biology, and ChIP-seq signal of these proteins is wholly enriched within the KIT SE (Fig. S1C). Thus, the KIT SE signifies a highly regulated region of the GIST genome with considerable complementarity in H3K27ac ChIP-seq and ATAC-seq signal, and contains exclusive regions of binding within the KIT locus for TFs essential to GIST biology (Fig. 1A).

Figure 1.

The KIT enhancer domain regulates KIT gene expression and cell viability in GIST. A, Overlap of H3K27ac (red, n=4), ATAC (gray, n=3) and TF (green, n=3) signal at the KIT SE in KIT-dependent GIST cells shown in Fig. S1. Blue lines and text indicates the location of sgRNAs targeting regions of ATAC and TF peak signal. Markers labeled A-D indicate primer locations and directionality. B, Relative cell count in GIST-T1 cells expressing either Cas9 (black) or dCas9-KRAB (red) following 10 days of infection with the indicated sgRNAs targeting the KIT SE or luciferase (Luc) as control; mean ± s.e.m., n=4. Relative KIT mRNA in GIST-T1 cells expressing dCas9-KRAB and the indicated sgRNA following 10 days of infection is shown on the right axis (blue); mean ± s.e.m., n=3. C, Relative day 10 cell count in GIST-T1/Cas9 cells expressing sgRNAs targeting Luc or RPS19 as controls or co-treatment with se1.1 and se5.2 sgRNAs flanking the KIT SE; mean ± s.e.m., n=4. D, Amplification of genomic DNA from KIT-independent GIST48B cells expressing the indicated sgRNAs, demonstrating deletion of the KIT SE in a subset of cells treated with se1.1 and se5.2 sgRNAs. E, Relative day 10 cell count in KIT-dependent GIST430/Cas9 cells expressing sgRNAs targeting Luc, RPS19, or se1.1 and se5.2 sgRNAs alone or in combination; mean ± s.e.m., n=4. F, Relative day 10 cell count in KIT-independent GIST48B/Cas9 cells expressing sgRNAs targeting Luc, RPS19, or se1.1 and se5.2 sgRNAs alone or in combination; mean ± s.e.m., n=4. Data in panels B-C and E-F were analyzed by one-way ANOVA with Dunnett’s multiple comparisons test (compared to Luc; *, P<0.05; **, P<0.01; ***, P<0.001). G, Relative KIT mRNA in GIST-T1/Cas9 and GIST430/Cas9 cells expressing Luc or se1.1 and se5.2 sgRNAs in combination ~96 h following infection; mean ± s.e.m., n=3. Data were analyzed by unpaired t-test (compared to Luc; *, P<0.05).

To evaluate for functional consequence of disruption of the KIT SE, we designed sgRNAs within the ATAC and TF signal peaks and expressed them together with Cas9 to produce double strand brakes or dCas9-KRAB to incite local enhancer repression (33). In the KIT-dependent cell line GIST-T1, which bears an activating KIT mutation, both Cas9 and dCas9-KRAB expression together with a subset of SE-directed sgRNAs led to decreased cell proliferation (Fig. 1B, left axis), with double strand breaks having more deleterious effects on proliferation than enhancer repression at most sites. Enhancer repression with dCas9-KRAB also led to a decrease in KIT mRNA at several of these sites (Fig. 1B, right axis). By co-expressing sgRNAs flanking the KIT SE, effects on GIST cell proliferation were more marked, and deletion of the entire SE could be seen in a subset of cells (Fig. 1C-D). Similar toxic effects of KIT SE manipulation were seen in the KIT-dependent cell line GIST430, but not the KIT-independent cell line GIST48B, as expected (Fig. 1E-F). The more modest effects on cell count seen in GIST430 may be related to the far slower proliferation rate of these cells compared to GIST-T1 and GIST48B (Fig. S1D). In GIST-T1 and GIST430, co-expression of the SE-flanking guides led to a significant decrease in KIT mRNA shortly following infection (Fig. 1G), but no change in expression of the neighboring PDGFRA gene (Fig. S1E). Taken together, these data demonstrate the essential and highly regulated function of the KIT SE in KIT gene expression and GIST biology.

BBI treatment leads to cell cycle arrest, apoptosis, cell death and reduced KIT gene expression in KIT-dependent GIST

In light of GIST’s dependence upon enhancer-driven gene expression of a singular oncogene, we hypothesized that KIT gene expression could be therapeutically targeted by disrupting the oncogene-associated SE with BBIs. In support of this hypothesis, the KIT-dependent cell line GIST-T1was more susceptible to the prototypical BBI, JQ1, with a 9-fold lower IC₅₀ and greater maximal effect, compared to the KIT-independent cell line GIST48B (Fig. 2A). GIST48B has lost expression and dependency upon KIT and, together with a comparable proliferative rate to GIST-T1 (Fig. S1D), represents an ideal control. JQ1 induced G0/G1 arrest (Fig. 2B and Fig. S2A-B) and provoked apoptosis and cell death in GIST-T1 and other KIT-mutant and KIT-dependent cell lines with slower proliferative rates (Fig. 2C, Fig. S2C-F and Materials and Methods); the TKI imatinib was used as comparator. Cell cycle alterations and apoptosis were not observed with doses of JQ1 at or below 50 nM under these treatment conditions, in contrast to imatinib, with higher doses of JQ1 causing uniform G0/G1 cell cycle arrest but variable effects on apoptosis between KIT-dependent GIST cell lines.

Figure 2.

Sensitivity of KIT-dependent GIST to BBIs. A, Cellular viability in KIT-dependent and KIT-independent GIST cell lines in response to imatinib and JQ1; mean ± s.e.m., n=8. B, Cell cycle analysis by propidium iodide staining in response to 500 nM imatinib or JQ1 at 24 h; mean value shown, n=6. C, Early apoptosis or late apoptosis and cell death, as indicated by 7-AAD and annexin V staining, of GIST cell lines in response to 500 nM imatinib or JQ1 at 72 h; mean ± s.e.m., n=4. D, Relative KIT mRNA levels by RT-PCR in four KIT-dependent GIST cell lines in response to the indicated doses of JQ1 at 6 and 24 h; mean ± s.e.m., n=3. E, Western blot for designated proteins in five GIST cell lines with the indicated KIT activating mutations. Cells were treated with imatinib at 500 nM for 24 h or JQ1 at 50 or 500 nM for 6 or 24 h. Values inset in Western blots represent average band signal intensity across three independent experiments with experimental samples processed in parallel and normalized to the control condition. Data were analyzed by one-way ANOVA, where indicated, with Dunnett’s multiple comparisons test (compared to DMSO; *, P<0.05; **, P<0.01; ***, P<0.001).

To assess the effect of JQ1 on KIT gene expression, we treated four KIT-dependent GIST cell lines with JQ1 and evaluated KIT mRNA levels by RT-PCR. JQ1 produced significant and, in more sensitive cell lines, sustained reductions in KIT mRNA levels (Fig. 2D). This reduction in KIT mRNA by JQ1 translated into variable reduction in KIT protein levels and effectors of the KIT signaling pathway, with alterations in KIT, phospho-KIT, phospho-ERK and ETV1 seen in all KIT-dependent cell lines at early time points (Fig. 2E). For GIST882, higher doses of JQ1 were able to achieve sustained reductions in KIT mRNA and protein at 24 h, though in GIST48 only acute treatments led to such changes (Fig. S3A-C), perhaps owing in part to the low rate of cell cycle progression in this cell line facilitating adaptation to enhancer disruption (Fig. S2B). By comparison, imatinib treatment induced biphasic changes in normalized KIT mRNA and variable changes in KIT protein levels and maturation (Fig. S3D-E). While receptor signaling promotes KIT protein turnover (34), only three of the four GIST cell lines exhibited KIT protein accumulation with KIT inhibition; imatinib treatment in GIST-T1 decreased KIT protein levels, seen both here and in prior reports (35). This may owe to context-dependent effects of imatinib treatment, which has also been shown to cause lysosomal degradation of KIT protein in a leukemia cell line (36). Biphasic changes in KIT mRNA with imatinib treatment have previously been observed (37), and may result from a temporal balance of trophic KIT signaling loss and later transcriptional de-repression with reduction of negative regulatory proteins (e.g. Sprouty and DUSP gene families). These context-dependent mechanisms of KIT expression changes in GIST have yet to be fully described.

Taken together, these data demonstrate that KIT-dependent GIST cell lines, which rely on a large enhancer domain to drive KIT gene expression, are susceptible to BET bromodomain inhibition. BBIs reduce KIT gene expression and cause alterations in the levels of KIT protein and mediators of KIT signal transduction, resulting in cell cycle arrest, apoptosis and cell death in GIST.

The KIT enhancer domain accounts for selective toxicity of BBIs in GIST

To isolate the mechanism of selective toxicity of BBIs in KIT-dependent GIST, we utilized CRISPR/Cas9 as a means to disconnect KIT expression from its native enhancers. First, we used an sgRNA/GFP competition assay in GIST-T1 and GIST48B cells expressing Cas9 to identify sgRNAs capable of eliminating endogenous KIT signaling without off-target toxicity (Fig. 3A). All KIT sgRNAs, targeting upstream coding regions or the kinase domain, reduced cellular viability in GIST-T1 but not GIST48B. sgRNAs targeting the kinase domain in exon 12 showed significant toxicity, consistent with the kinase domain’s intolerance of amino acid changes resulting from DNA repair following Cas9 cleavage (28). The most active KIT sgRNA, KITe1.1, spans the translational start site into the 5’ UTR. Cas9 cleavage at this site results in indels that fully eliminate productive KIT gene expression (Fig. 3B).

Figure 3.

The KIT enhancer domain accounts for BBI sensitivity. A, CRISPR/Cas9 GFP-competition assay over time with sgRNAs targeting KIT in the KIT-independent cell line GIST48B (grayscale) and KIT-dependent cell line GIST-T1 (red); mean ± s.e.m., n=3. B, Schema of the KITe1.1 sgRNA relative to the KIT start codon, predicted Cas9 cleavage site (red arrowhead) and PAM (red text). Example indels sequenced from clones isolated following Cas9 cleavage and rescue with KIT cDNA are listed. C, Western blot for the designated proteins in the GIST-T1 parental cell line (left column) or GIST-T1 KITe1.1 sgRNA/Cas9 cell lines rescued with the indicated HA-tagged KIT constructs driven by viral promoter; the KIT exon 11 mutant construct (GIST-T1/KIT^Δe11) is in the middle column and compound KIT exon 11 and 17 mutant construct (GIST-T1/KIT^Δe11/N822K) is in the right column. Cellular viability effects of imatinib (D) or JQ1 (E) in the parental GIST-T1 cell line and KIT-rescue cell lines; mean ± s.e.m., n=8. The shaded yellow area represents the 95% confidence interval for GIST48B JQ1 treatment shown in Fig. 2A.

We next co-expressed the KITe1.1 sgRNA together with Cas9 and a rescue KIT HA-tagged expression vector. As the rescue cDNA construct is driven by viral promoter, the lack of the 5’ UTR prevents targeting by KITe1.1 sgRNAs and permits trophic KIT expression and signaling detached from native enhancer regulation. Compared to the parental GIST-T1 cell line, the rescue cell lines with an identical activating exon 11 mutation (GIST-T1/KIT^Δe11) or compound exon 11 and 17 mutations (GIST-T1/KIT^Δe11/N822K) driven by viral promoter were resistant to JQ1, with preservation of KIT protein levels and downstream effectors of the KIT pathway (Fig. 3C). The exon 17 N822K mutation lies within the activation loop and confers imatinib-resistance, as indicated by persistence of phospho-KIT, phospho-ERK and ETV1 in the presence of imatinib. In the parental cell line, protein levels of KIT were dramatically reduced by JQ1, resulting in decreased phospho-KIT, phospho-ERK and destabilization of ETV1. By contrast, KIT protein levels and downstream signaling effectors were preserved in the rescue cell lines upon JQ1 exposure. In each cell line, greater effects on ETV1 destabilization were seen with combination JQ1 and imatinib than either drug alone. The increase in HEXIM1 protein with JQ1 treatment indicates similar global drug target engagement between cell lines.

The rescue cell lines showed expected reduction in viability with imatinib treatment based on KIT mutation (Fig. 3D), in agreement with reductions in phospho-KIT seen by Western blot upon imatinib exposure. However, both of the KIT enhancer-independent rescue cell lines were less vulnerable to JQ1 compared to the parental cell line (Fig. 3E). The yellow curve in Fig. 3E indicates the 95% confidence interval of viability for GIST48B, which closely matches the curves for the rescue cell lines. Comparable results were seen in GIST430, with the GIST430/KIT^Δe11 rescue cells exhibiting analogous susceptibility to imatinib but relative resistance to JQ1 (Fig. S4A-B). Compared to JQ1 treatment in the parental cell line, the rescue KIT^Δe11 cell lines had an IC₅₀ 7.7-fold (GIST430) and 4.6-fold (GIST-T1) higher with reduced maximal effect; this resistance arises solely from abstracting KIT expression from its enhancer domain. To determine if overexpression of a TF that binds within the KIT SE and is responsible for KIT gene expression (16) could generate similar BBI resistance, we cloned and overexpressed HAND1. However, HAND1 expression alone was neither sufficient to inculcate JQ1 resistance in GIST-T1 nor evoke KIT expression from a KIT-independent GIST cell line (Fig. S4C-E). Together with the above results, these data demonstrate that the selective toxicity of BBIs in KIT-dependent GIST arises from effects at KIT enhancers, which decreases KIT gene expression, lowering KIT protein levels and trophic stimulation through the KIT signaling pathway.

Global enhancer changes induced by KIT and bromodomain inhibition in GIST

To characterize global effects on chromatin arising from bromodomain and KIT inhibition, we employed H3K27ac and BRD4 ChIP-seq in cells treated with imatinib or JQ1. In the KIT-independent GIST48B cell line, there were minimal changes in H3K27ac signal at the transcriptional start site (TSSs) or enhancers of all enhancer-associated genes (Fig. 4A). In this cell line, SEs defined by H3K27ac signal were also largely unaffected by drug treatment (Fig. 4B). In contrast, the KIT-dependent GIST-T1 cell line showed global decreases in H3K27ac signal at TSSs and enhancers of enhancer-associated genes from drug treatment, and most dramatically with JQ1 (Fig. 4C). SEs defined by H3K27ac were also considerably affected by drug treatment in this cell line, with imatinib driving the largest log₂ fold-change in H3K27ac signal lost and gained (Fig. 4D). Many SE-associated genes were similarly affected by imatinib and JQ1 in GIST-T1. These results demonstrate the authoritarian role of oncogenic KIT signaling in shaping enhancer activity and nuclear organization in GIST.

Figure 4.

Global chromatin changes following imatinib and JQ1 treatment in GIST. A, H3K27ac ChIP-seq meta tracks of TSSs (top panel) and all H3K27ac-defined enhancer peaks (bottom panel) in GIST48B. B, Waterfall plots showing log₂ fold change of SE regions in GIST48B H3K27ac ChIP-seq signal following the indicated 24 h drug treatment at 500 nM. C-D, H3K27ac ChIP-seq meta tracks at TSSs (top panel) and all H3K27ac-defined enhancer peaks (bottom panel) and log₂ fold change of SE regions in GIST-T1 following 24 h treatment with the indicated drug. E-F, BRD4 ChIP-seq meta tracks at TSSs (top panel) and BRD4-defined enhancers (bottom panel) and log₂ fold change in BRD4-defined SE regions in GIST-T1 following 24 h drug treatment. Changes in H3K27ac (red, G) and BRD4 (blue, H) ChIP-seq signal at the KIT locus in GIST-T1 following the indicated 24 h drug treatment.

Comparable enhancer and SE alterations in GIST-T1 were also seen using BRD4 ChIP-seq (Fig. 4E-F), with larger changes seen with JQ1 treatment, as expected with the direct targeting of BRD4. Ranked enhancer plots of GIST-T1 H3K27ac and BRD4 ChIP-seq show similar enrichment in GIST-associated genes (Fig. S5A-B). In contrast, GIST-48B loses enhancers in many of the KIT-dependent GIST-associated genes, particularly at SE-associated genes (Fig. S5C). This is reflected by the loss or reduction in ChIP-seq signal at all GIST-T1 enhancers in this cell line (Fig. S5D).

As the above results indicate that the KIT enhancer domain is responsible for BBI selectivity in GIST, we evaluated this locus for changes in H3K27ac and BRD4 signal following treatment with imatinib and JQ1 (Fig. 4G-H). Compared to DMSO and imatinib treatment, JQ1 reduced H3K27ac signal and wholly eliminated BRD4 at the KIT locus. Similar findings were observed in GIST430 (Fig. S5E). These results demonstrate the genome-wide epigenetic effects on H3K27ac and BRD4 occupancy arising from treatment with imatinib and JQ1, and further implicate the elimination of BRD4 signal at the KIT enhancer as the substrate for BBI toxicity in GIST.

Gene expression changes induced by bromodomain inhibition phenocopy direct KIT inhibition

To evaluate for global transcriptional changes associated with imatinib and JQ1 exposure, we treated GIST-T1 cells for 6 or 24 hours with each drug alone or in combination. Cell count and ERCC spike-in normalized RNA was evaluated by RNA-seq. Compared to vehicle control, both imatinib and JQ1 caused widespread reductions in gene expression, with a disproportionate effect on GIST-associated SE genes (Fig. 5A). Among expressed RTKs, KIT was the most reduced by JQ1 and combination treatment (Fig. 5A, middle panel), and the expression of several cell stress- and apoptosis-associated genes were increased (Fig. 5A, bottom panel). Both imatinib and JQ1 treatment, alone or in combination, disproportionally affected SEs compared to typical enhancers (TEs) at all time points (Fig. 5B).

Figure 5.

Gene expression alterations following imatinib and JQ1 treatment in GIST. A, Heat map showing expression of genes that are up or down regulated by imatinib (IM) and/or JQ1 after 6 or 24 h treatment at 500 nM. Genes are ordered by maximal fold change with combination imatinib/JQ1 treatment. Red lettering indicates SE-associated genes. The top panel displays the most downregulated genes, the middle panel RTKs and the bottom panel select genes associated with cell stress. B, Box plots showing log₂ fold change in expression of genes with TEs or SEs (Welch’s t-test; *, P<2 × 10⁻⁵). C, GSEA comparing gene expression changes in the top 366 SE-associated genes between JQ1 and imatinib treatment at 24 h. D, GSEA showing analogous loss of expression of RTK signaling-associated genes following treatment with imatinib, JQ1 or the drug combination in the 229 gene set. NES = normalized enrichment score, FDR = false discovery rate.

Using gene set enrichment analysis (GSEA) to compare the effects of JQ1 and imatinib at SE-associated genes, each inhibitor had preferential effects on a different subset of these highly regulated genes (Fig. 5C). While imatinib had greater effects on negative regulators of KIT signaling (e.g. SPRY4, DUSP6), JQ1 preferentially affected KIT expression, among other GIST-associated genes including ETV1 and HAND1.

The enhancers at several genes showed a differential effect upon treatment with imatinib and JQ1, and a close correlation was found between these data and changes in expression by RNA-seq (Fig. S6A-H). SPRY2, a negative regulator of KIT signaling, has an enhancer and associated RNA expression that are more modestly affected by JQ1. However, direct suppression of KIT signaling with imatinib led to more rapid enhancer loss and decrease in SPRY2 gene expression (Fig. S6E-F). In contrast, both the enhancer and expression values of FGFR2 were modestly affected by imatinib, but more dramatically attenuated by JQ1 (Fig. S6G-H). Evaluating for transcriptional effects at genes neighboring the KIT locus, KDR is not expressed in any GIST cell line, while PDGFRA is expressed in only a subset of GIST cell lines (Fig. S7A). In GIST-T1, which expresses PDGFRA at levels >10-fold lower than KIT, JQ1 treatment had significantly less impact on PDGFRA expression in comparison to KIT (Fig. S7B-C). GIST882, which has the highest protein expression of PDGFRA among GIST cell lines (16), showed reductions in PDGFRA protein from treatment with either imatinib or JQ1 (Fig. S7D). Finally, dependency screening in GIST-T1 (38) reveals that only KIT expression is necessary for cell proliferation among these three genes (Fig. S7E-G). These results, together with findings that KIT-mutant GIST tumors do not universally express PDGFRA (16), suggest that PDGFRA and KDR expression are noncore for KIT-mutant GIST.

Probing the molecular signatures database for gene sets most similar to this RNA-seq dataset, the greatest enrichment was found in a curated gene set downregulated in an EGFR-dependent cell line treated with an EGFR inhibitor (39). This signature produced the highest NES in the imatinib and combination treatment groups, and was among the top gene sets enriched in the JQ1 treatment groups (Fig. 5D). Taken together with the above results, these data demonstrate that, in KIT-dependent GIST, the selective toxicity from bromodomain inhibition arises from suppression of KIT enhancers, leading to decreased KIT gene expression, decreased KIT signaling and creating a phenocopy of direct KIT inhibition. Further, these data underscore the global epigenetic and transcriptional changes resulting from inhibition of a driver oncogene distal to the nucleus (e.g. inhibition of KIT with imatinib), and how these alterations overlap with direct bromodomain inhibition.

Combinatorial effects of bromodomain and kinase inhibition in GIST

We next sought to extend these findings of selective BBI toxicity in KIT-dependent GIST to a panel of additional bromodomain inhibitors, many of which are under clinical development. TKIs demonstrated expected selectivity in cell lines, with KIT-independent GIST exhibiting insensitivity and the KIT-dependent GIST48 line, with compound KIT exon 11 juxtamembrane and exon 17 kinase domain mutations, showing relative resistance (Fig. 6A, top panel and Fig. S8A). The KIT-amplified and KIT-dependent MaMel melanoma cell line, bearing an exon 9 mutation in KIT, was similarly susceptible to TKIs as sensitive GIST cell lines. With notable exceptions detailed below, KIT-dependent GIST cell lines were more sensitive to the eight evaluated BBIs than KIT-independent GIST cell lines or the MaMel cell line (Fig. 6A, bottom panel and Fig. S8B). IC₅₀ values from drug treatments are listed in Table S2.

Figure 6.

Selective toxicity in KIT-dependent GIST to bromodomain inhibition and synergy with imatinib in vitro and in vivo. A, Heat map of mean IC₅₀ of TKIs and BBIs in KIT-dependent GIST cell lines, KIT-independent GIST cell lines and a KIT-dependent melanoma cell line; n=8. Black outlines in MT1 and RG6146 rows indicate KIT-dependent GIST cell lines >4-fold less sensitive to drug compared to GIST-T1 and GIST430. B, Isobologram plot where points represent paired values of JQ1 and imatinib assessed for synergy. The diagonal line designates the combination index indicative of additivity, with points below the line demonstrating synergy and those above antagonism; n=8 per data point. C, Late apoptosis and cell death, as indicated by 7-AAD and annexin V staining, of GIST cell lines in response to 100 nM imatinib and/or JQ1; mean ± s.e.m., n=4. D, Tumor volume of GIST-T1 cell line xenografts in response to treatment with vehicle (n=5), JQ1 (50 mg/kg IP, 5 days per week; n=6), imatinib (50 mg/kg gavage, daily; n=7) or imatinib and JQ1 combination (n=7) over a 21-day treatment period. E, Tumor volume of PG27 TKI-resistant patient-derived xenograft in response to vehicle, JQ1 or GSK525762 (50 mg/kg IP, 5 days per week; n=5 per group). Tumor volume data were analyzed by two-way ANOVA with Tukey’s multiple comparisons test (compared to vehicle or the indicated group; *, P<0.05; **, P<0.01; ***, P<0.001). F, Mitoses per 10 high power fields (HPF) in the PG27 tumors treated with the indicated drugs. G, Relative KIT mRNA levels in PG27 tumors by RT-PCR normalized to vehicle control. Tumors were harvested 6 h after the last administered treatment with vehicle (n=5), JQ1 (n=4) or GSK525762 (n=5). Data were analyzed by one-way ANOVA with Tukey’s post-hoc test (compared to vehicle or DMSO control; *, P<0.05; **, P<0.01; ***, P<0.001).

Given the proposed independent mechanisms of action of BBIs in GIST, which converges upon the KIT signaling pathway without direct kinase inhibition, we hypothesized a synergistic toxicity arising from the combination of imatinib and JQ1. To test this, we utilized a high-throughput platform to compare the individual and combined effects of imatinib and JQ1 on the viability of GIST-T1 cells. Using the median-effect principle (27), we performed isobologram analysis, which demonstrated a synergistic effect arising from the drug combination (Fig. 6B). In all KIT-dependent cell lines, combination treatment with imatinib and JQ1 showed greater effects on viability and apoptosis than either drug alone (Fig. 6C, Fig. S8C-F), while the KIT-independent cell line GIST48B remained resistant to individual and combination treatments. These results support a complimentary effect of combined kinase and bromodomain inhibition in GIST, which independently target the KIT signaling pathway.

To evaluate combinatorial effects of imatinib and JQ1 in vivo, we utilized the GIST-T1 cell line xenograft model. Treatment of tumor bearing mice for 21 days with 50 mg/kg imatinib daily by gavage and/or 50 mg/kg JQ1 five days per week IP led to significant reductions in tumor growth, with the combination treatment group having significantly less tumor burden than other treatment or control groups (Fig. 6D) without overt toxicity (Fig. S9A). Drug treatment was administered below the maximal tolerated dose to enable studies of the treatment combination. To assess the ability of JQ1 and a BBI under clinical development, GSK525762, to control tumor growth in a TKI-resistant tumor, we utilized the PG27 patient-derived xenograft (PDX) model (16,40). Treatment of PG27-tumor bearing mice with JQ1 or GSK525762 for 21 days at 50 mg/kg five days per week IP led to significant reductions in tumor growth (Fig. 6E) without weight loss from treatment (Fig. S9B). Analysis of the tumors after treatment showed a significant reduction in mitotic rate and KIT mRNA in the BBI treated mice (Fig. 6F-G). Treatment-related effects were seen by Western blot (Fig. S9C-F), with an increase in the ratio of KIT to phospho-KIT signal without changes in total KIT protein levels, and histology (Fig. S9G). The findings of stable total KIT protein levels despite reductions in KIT mRNA and KIT signaling may represent adaptive changes following 21 days of drug treatment, whereby decreased KIT signaling decreases receptor turnover (34).These results demonstrate the efficacy of BBIs alone and in combination with TKIs in vivo, and support suppression of KIT gene expression as a primary mechanism of BBI vulnerability in GIST.

Resistance to BBIs in GIST is mediated by multidrug resistance pumps

Two KIT-dependent GIST cell lines, GIST882 and GIST48, were uniquely resistant to the BBI RG6146 (41) and the bivalent BBI MT1 (42), with an IC₅₀ >4-fold higher than other KIT-dependent GIST cell lines (Fig. 6A, bottom panel, emphasized in black border). Hypothesizing a selective mechanism of resistance to these drugs, and drawing from prior studies finding expression of efflux pumps in metastatic GIST (43), we probed these cell lines for an expression pattern of multidrug resistance (MDR) pumps aligned with the cell lines exhibiting selective BBI resistance. By RT-PCR ABCB1, which encodes P-glycoprotein (Pgp), among candidate MDR genes was found to be most highly expressed in GIST882 and GIST48 relative to other cell lines (Fig. 7A). Western blot confirmed this expression pattern (Fig. 7B). To establish the causative role of Pgp expression in BBI resistance, we co-treated cells with the selective Pgp inhibitor tariquidar (44). Co-incubation of GIST882 and GIST48 cells with tariquidar and RG6146 led to restoration of BBI potency in these resistant cell lines (Fig. 7C). This change in IC₅₀ was not seen in cell lines that lack Pgp expression, and tariquidar did not change the IC₅₀ of BBIs not subject to Pgp activity (Fig. 7D). To further validate Pgp activity as a mechanism of BBI resistance, we stably expressed Pgp in the GIST-T1 cell line, which lacks endogenous ABCB1 expression, or GFP as control (Fig. 7E). Expression of Pgp led to a significant decrease in potency of both RG6146 and MT1, with restoration of drug sensitivity upon co-incubation with tariquidar (Fig. 7F-G). In comparison to other BBI compounds, Pgp expression in GIST-T1 cells led to unique and significant loss of MT1 and RG6146 potency, which was reversible with tariquidar (Fig. 7H). As MDR expression has been described in clinical GIST samples, we compared enhancers at the ABCB1 locus between GIST tumors and cell lines. Both GIST tumors and the two cell lines expressing ABCB1 exhibit similar enhancer peaks at the ABCB1 locus, in contrast to cell lines that lack ABCB1 expression (Fig. 7I). Pgp expression was further confirmed in primary tumor samples by Western blot (Fig. 7J). These data demonstrate that BBIs can be substrates of MDR efflux pumps, and raises the possibility of existing or acquired resistance to specific BBIs through expression of ABCB1.

Figure 7.

Resistance to BBIs in GIST is mediated by multidrug resistance pumps. A, Relative mRNA level of candidate MDR genes in GIST cell lines. Data were normalized to GIST-T1; mean ± s.e.m., n=3. B, Western blot for Pgp or vinculin (Vinc) as loading control in GIST cell lines. C, Cellular viability effects of RG6146 with or without the addition of Pgp inhibitor tariquidar; mean ± s.e.m., n=8. D, Fold reduction in IC₅₀ for the indicated drugs and cell lines with the addition of tariquidar; mean ± s.e.m., n=8. E, Expression of Pgp in GIST-T1 cells following viral transduction with ABCB1 or GFP as control. F-G, Cellular viability effects of RG6146 and MT1 in GIST-T1 cells expressing Pgp or GFP as control, with or without the addition of tariquidar; mean ± s.e.m., n=8. H, Fold reduction in IC₅₀ for the indicated drug and cell line with the addition of tariquidar; mean ± s.e.m., n=8. I, Meta tracks at the ABCB1 locus in GIST tumors (n=10), GIST cell lines expressing ABCB1 (n=2) and GIST cell lines lacking ABCB1 expression (n=3). J, Western blot of GIST tumors probing for Pgp or vinculin as loading control. Data were analyzed by two-way ANOVA, where appropriate, with Tukey’s post-hoc test (compared to JQ1 control; **, P<0.01; ***, P<0.001).

Discussion

GIST represents one of the few sarcomas where we have both an understanding of oncogenesis and available targeted therapies (1). Though the advent of kinase inhibitors represents a revolutionary improvement in the clinical management of GIST, resistance to these therapies invariably develops and there are no other approved treatments for this disease (45). We have previously used H3K27ac ChIP-seq to characterize the conserved enhancer landscape in GIST, which was notable for a large enhancer domain regulating KIT gene expression conserved in vivo and in vitro (16). In GIST, KIT gene expression is principally regulated through its enhancer domain, as translocation or amplification are not mechanisms of oncogenesis seen in this disease (46).

Using a genetic approach, we found that disruption of the KIT enhancer led to reductions in KIT gene expression and compromised GIST cellular proliferation. These findings led to the hypothesis that disrupting the KIT enhancer domain with BET bromodomain inhibitors represents a therapeutic vulnerability in this disease. BBIs were found to cause cell cycle arrest and induce apoptosis, with associated decrease in KIT mRNA levels, KIT protein and mediators of KIT signal transduction. Using a CRISPR-based approach, the enhancer domain regulating KIT expression was identified as the underpinning of BBI vulnerability. BBI-sensitive GIST cell lines could be rendered resistant by abstracting KIT gene expression from the native enhancer through driving expression with a viral promoter.

Both kinase inhibition and bromodomain inhibition led to genome-wide changes in enhancers and gene expression in KIT-dependent GIST cells. The degree of suppression of gene expression at TEs and SEs was comparable between each type of inhibitor, and each inhibitor had preferential effects on a subset of SEs. Direct inhibition of KIT with imatinib led to more alterations in the H3K27ac-defined SE landscape than did JQ1. That alteration of an oncogene distal to the nucleus results in such dramatic effects on chromatin organization has been previously demonstrated with oncogenic Ras (47), and underscores the importance of KIT signaling in entraining genome regulation in GIST.

Using GSEA, we found that bromodomain inhibition in GIST cells produced a change in expression profile that overlapped with kinase inhibition, further demonstrating epigenetic inhibition of KIT signaling as a primary means of cellular toxicity. Bromodomain inhibition has previously been shown to prevent kinase inhibitor-induced feedback activation of RTKs (48), and more recently to repress KIT transcription by suppressing an acute-myeloid leukemia-specific downstream enhancer (49). Our work adds support to the hypothesis that bromodomain inhibitors can therapeutically target the enhancers of RTKs, and further that diseases with primary dependency upon SE-driven, unamplified RTKs may be most susceptible.

In agreement with our proposed mechanism of action of BET bromodomain inhibitors in GIST, multiple BBIs were consistently found to have greater activity in KIT-dependent GIST cell lines as compared with KIT-independent cell lines or a melanoma cell line dependent upon amplified mutant KIT. With the independent and complimentary mechanisms of bromodomain inhibitor and kinase inhibitor activity, we identified a combinatorial toxic effect between imatinib and JQ1 in vitro and in vivo. A PDX derived from a TKI-resistant GIST tumor was susceptible to BBIs, with a reduction in tumor mitotic rate and KIT gene expression leading to a decrease in tumor growth. These results establish the pre-clinical utility of BET bromodomain inhibition in this disease.

Select KIT-dependent GIST cell lines were found to be resistant to the BBIs MT1 and RG6146, and using pharmacologic and genetic methods this resistance was attributed to the expression of Pgp. In addition to prior studies on bromodomain inhibitor resistance mechanisms demonstrating utilization of alternative enhancers, upregulation of alternative signaling pathways or phosphorylation of BRD4 (50), this work shows MDR efflux pumps as a novel mechanism of pre-existing or potentially acquired resistance to BBIs. Together with the previously reported expression of MDR proteins in GIST (43) and in our clinical samples, these data argue for the preferential development of BBIs not subject to MDR pumps in GIST and other malignancies where drug efflux is an established means of drug resistance.

These results anchor future translational investigation using BBIs in patients with GIST. BBIs act to suppress KIT gene expression independent of kinase mutation, which may open a therapeutic option for patients bearing tumors with primary TKI resistance mutations, or patients with disease that has elaborated TKI resistance following treatment with available lines of kinase inhibitors. Given the therapeutic combinatorial effects of kinase and bromodomain inhibition, and that even in advanced GIST patients continue to benefit from TKI therapy, combined treatment with TKI and BBI represents a promising translational opportunity.

Statement of Significance:

Expression and activity of mutant KIT is essential for driving the majority of GIST neoplasms, which can be therapeutically targeted using BET bromodomain inhibitors.