Potent Dual BET Bromodomain-Kinase Inhibitors as Value-Added Multitargeted Chemical Probes and Cancer Therapeutics

Bromodomains (BRD) are about 110 amino acid domains that bind to and “read” acetylated lysine (KAc) residues of histones tails in a process critical for chromatin organization and gene transcription (1). BRDs regulate transcription, chromatin remodeling, gene splicing, protein scaffolding and signal transduction and, therefore, play fundamental roles in cell proliferation and division. Members of the bromodomain and extra terminal (BET) protein family (BRD2, BRD3, BRD4, and BRDT) have been implicated in a number of disease pathways and are therefore considered promising drug targets (2). The BET protein BRD4 facilitates transcriptional elongation via recruitment of the positive transcription elongation factor (P-TEFb) and displacement of negative regulators such as HEXIM1 and 7SK snRNA (3). BRD4 is overexpressed in various cancers and can undergo translocations that are a hallmark of the lethal tumor NUT midline carcinoma (4). Chemical inhibition of BRD4 exerts a broad spectrum of desirable biologic effects such as anticancer and anti-inflammatory properties (5). All known BRD4 inhibitors target the KAc recognition site, particularly through H-bonding interactions with a conserved Asn residue. Importantly, BRD4 inhibition downregulates oncogenic MYC transcription factors in several cancer cell lines. Intense efforts are underway to develop chemically diverse and highly potent BRD4 inhibitors as new cancer therapeutics (6). Currently, 8 different BRD4 inhibitors have entered phase I trials for the treatment of liquid and solid tumors (www.clinicaltrials.gov).

We and others recently reported that diverse kinase inhibitors also inhibit the KAc-binding site of BRD4 (7–9). Two of the most potent compounds, BI2536 (primary target PLK1; IC_50(BRD4) = 25 nmol/L) and TG101209 (primary target JAK2, IC_50(BRD4) = 120 nmol/L) exerted cellular modes of action consistent with inhibition of BRD4, such as downregulation of c-MYC in multiple myeloma and acute myeloid leukemia (AML) cells, accompanied by strong inhibition of cell proliferation. Combined, these findings provided compelling evidence that certain kinase inhibitor chemotypes could be further developed to target specifically cancers that depend on BRD4 functionality and aberrant kinase activity. The rationale to develop dual BRD4-kinase inhibitors is relevant as single-activity BRD4 and kinase inhibitors act synergistically in a variety of cancers. JQ1 synergizes with rapamycin, an mTOR inhibitor, to inhibit proliferation and survival of human osteosarcoma cells as well as to upregulate p21Cip1 and downregulate c-MYC (10). A quarter of AML cells undergo activating internal tandem duplication (ITD) mutations in FLT3. Combinations of JQ1 and the FLT3 inhibitors quizartinib and ponatinib are synergistically lethal in AML cells driven by FLT3-ITD (11). In metastatic breast cancer, mutations of c-MYC and members of the PI3K signaling pathway are often concurrent. In such cells, inhibition of either BET proteins or PI3K is ineffective. However, when combined, MS417 (an analogue of JQ1) and the PI3K inhibitor GDC-0941 induce cell death and tumor regression (12). BET inhibition in mantle cell lymphoma cells decreases CDK4/6 and Bruton tyrosine kinase (BTK) levels, and co-treatment with JQ1 and the BTK inhibitor ibrutinib or the CDK4/6 inhibitor palbociclib synergistically induces apoptosis (13). Combinatorial treatment of osteosarcoma cells with JQ1 and CDK inhibitors induces potent synergistic activity (14). Significant molecular synergism between BET inhibitors and the TK inhibitor lapatinib occurs in ERBB2⁺ breast cancer cells, indicating that targeting broad-acting epigenetic regulators is needed to suppress the induction of gene expression following adaptive kinome response (15). In acute T-cell lymphoblastic leukemias, BRD2 inhibition by JQ1 showed strong synergy with tyrosine kinase inhibitors in inducing apoptosis (16).

Here, we describe the development of potent dual BET-kinase inhibitors using the dianilinopyrimidine scaffold of the JAK2/FLT3 inhibitor TG101348. Lead compounds selectively inhibit BET bromodomains and a set of tyrosine kinases including the cancer targets JAK2, FLT3, RET, and ROS1. Screening across 931 cancer cell lines revealed potent growth-inhibitory potential over JQ1 and TG101348. Dual BRD4-kinase inhibitors were highly effective against cell lines and patient samples of JAK2-driven myeloproliferative neoplasm (MPN), and drug combination studies indicated synergistic lethality in cell lines with high sensitivity for these compounds. Combined, our findings indicate promising potential of these first-in-class dual BRD4-kinase inhibitors as cancer therapeutics.

Reagents and compounds for biochemical and crystallographic experiments were purchased from Sigma-Aldrich, Selleck Chemicals, and Hampton Research unless otherwise indicated. BRD4-1 was purified and crystallized as described previously (7). The synthesis of compounds 1–6 is described elsewhere (17). Human MM1.S, MV-4-11, and SAOS2 cells were purchased from ATCC and were passaged in the laboratory for fewer than 6 months after receipt or resuscitation. Human myeloproliferative syndrome–derived UKE-1 cells, which express the constitutively active JAK2-V617F mutant enzyme, were kindly provided by Dr. Ross Levine (Memorial Sloan Kettering Cancer Center, New York, NY). Human HCC-78 cells were kindly provided by Dr. Uwe Rix (Moffitt Cancer Center, Tampa, FL). UKE-1 and HCC-78 cells were authenticated by short tandem repeat (STR) DNA typing in the Moffitt Genomics core. MM1.S, UKE-1, and HCC-78 cells were maintained in RPMI-1640 medium (Life Technologies), and MV4-11 cells were maintained in IMDM media (Lonza) containing 10% FBS (Atlanta Biologicals). SAOS2 cells were maintained in McCoy 5A containing 15% FBS. All cells were grown and maintained at 37°C in a humidified atmosphere containing 5% CO₂. The following antibodies were purchased from Cell Signaling: phospho-STAT3 (Y705) (#9145), STAT3 (#9139), phospho-FLT3 (Y591) (#3466), FLT3 (#3462), cMYC (#5605), p21Cip1 (#2946), cleaved PARP (#5625), and cleaved caspase-3 (#9661). Vinculin antibody was purchased from Sigma-Aldrich (#V9131). Peroxidase-conjugated secondary antibodies were purchased from Jackson ImmunoResearch.

The binding potential of compounds against BRD4-1 was assessed by differential scanning fluorimetry (DSF) using a StepOnePlus Real-Time PCR system (Applied Biosystems). Purified BRD4-1 [4 μmol/L final concentration; 10 mmol/L HEPES (pH7.5), 100 mmol/L NaCl, and 1 mmol/L DTT] was assayed, in quadruplicates, in a 96-well plate. Inhibitors were added to a final concentration of 100 μmol/L and 2% DMSO. Protein Thermal Shift Dye (1:8000; Applied Biosystems) was used as the fluorescent probe, and fluorescence was measured using the ROX Reporter channel (620 nm). Protein stability was investigated by programing the thermocycler to increase the temperature from 25°C to 99°C using 0.2°C increments and 10-second incubations per increment. The inflection point of the transition curve/melting temperature (T_m) was calculated using the Boltzmann equation within the Protein Thermal Shift Software (v.1.1) (Applied Biosystems). (+)-JQ1 (18) and dinaciclib (8) were used as controls for strong and weak binders of BRD4-1, respectively. The ΔT_m was calculated by using DMSO control wells as reference.

Crystals of BRD4-1 were grown in the presence of 1 mmol/L ligand and 10% (v/v) DMSO from vapor diffusion hanging drops using reservoir as described previously (7), harvested in cryoprotectant [reservoir containing 25% (v/v) ethylene glycol and 0.5 mmol/L ligand], and flash-frozen in a stream of nitrogen gas. X-ray diffraction data were recorded at −180°C at the beamline 22-BM, SER-CAT, Advanced Photon Source, Argonne National Laboratories. Data were reduced and scaled with XDS (19); PHENIX (20) was employed for phasing and refinement, and model building was performed using Coot (21). All structures were solved by molecular replacement and the monomer of PDB entry 4O7A (7) as the search model. Initial models for the small-molecule ligands were generated using MarvinSketch (ChemAxon) with ligand restraints from eLBOW of the PHENIX suite. Figures were prepared using PyMOL (Schrödinger, LLC).

Human cells were seeded in 96-well plates at approximately 3,000 adherent or 20,000 suspension cells per well (0.1 mL). Adherent cells were allowed to attach overnight, and suspension cells were incubated for 1 hour before dosing. Cells were incubated with the compounds indicated in the figure legends ranging from 1 nmol/L to 10 μmol/L in the presence of vehicle (0.1% DMSO) for 72 hours, with 6 replicates per concentration. After drug treatment, 15 μL of CellTiter Blue reagent (Promega) was added to each well, followed by vigorous orbital shaking for 5 minutes and incubation for 3 hours at 37°C. Plates were placed in a Wallac EnVision 2103 Multilabel Reader (PerkinElmer), and fluorescence was determined using excitation and emission filters of 570 and 615 nm, respectively. Growth inhibition data were analyzed with the Prism6 software (GraphPad).

Development of drug resistance was assessed by culturing UKE-1 cells (1 × 10⁵/mL) in DMSO (0.1%) and inhibitors at concentrations around the respective IC₂₀ values. Cell cultures were counted every 2 to 3 days at which point they were diluted back to the starting cell density. If drug-treated cells grew at a rate of 90% or more of DMSO control, the drug concentration was doubled, and cells growing at a rate less than 90% of DMSO control were maintained at their current drug concentration.

Dose–effect analysis of drug combinations was performed essentially as described (22). Growth inhibition data upon 72-hour drug treatment with single drugs and constant 1:1 ratios of JQ1 and ruxolitinib, quizartinib, or TG101209 were determined in parallel for each cell line and the data were fit to the Hill equation (Eq. A) to account for the differences in maximum effect levels (Max) reached for each inhibitor. Rearrangement of Eq. A yields the dose (D) as a function of the fraction affected (fa; Eq. B). The combination index (CI) at a given fa value was calculated using Eq. C where (Dc)₁ and (Dc)₂ are the concentrations of each drug in the combination treatment and (Ds)₁ and (Ds)₂ are those of the single drugs.

A different dose–effect analysis was performed for a separate study in which UKE-1 cells treated with non-constant ratios of JQ1 and ruxolitinib were counted over a period of 6 days. Growth rate constants (k_obs) were determined for each drug treatment using Eq. D, where y₀ is the number of cells at t = 0 (2 × 10⁵ cells/mL plated). Cell numbers were determined at days 2, 3, 4, and 6 by trypan blue exclusion from 2 separate experiments.

Fa values were calculated from e^kobs values for each drug/combination relative to DMSO control. Dose–effect analysis was performed using CompuSyn (http://www.combosyn.com/index.html).

Peripheral blood was obtained from JAK2-V617F–positive MPN patients, consented through the Moffitt Cancer Center Total Cancer Care protocol (MCC 14690/Liberty IRB #12.11.0023) and approved by the Moffitt Cancer Center Scientific Review committee. Blood was treated with HetaSep (STEMCELL Technologies, Inc.) to remove the majority of red blood cells. Peripheral blood mononuclear cells (PBMC) were isolated by ficoll separation. PBMCs (1 × 10⁵ to 4 × 10⁵) were then plated in 1 mL of methylcellulose medium containing rhSCF, rhIL-3, and rhGM-CSF (MethoCult #H4534; STEMCELL Technologies, Inc.). All drug-treated samples contained 0.1% DMSO as the final concentration. Erythroid colonies were counted after 14 days.

Compounds 2, 3, and 6 were screened across 931 cell lines using curve fitting and IC₅₀ estimation essentially as described (23). All cell lines were sourced from commercial vendors. Cells were grown in RPMI or DMEM/F12 medium supplemented with 5% FBS and penicillin/streptavidin and maintained at 37°C in a humidified atmosphere at 5% CO₂. Cell lines were propagated in these 2 media to minimize the potential effect of varying the media on sensitivity to therapeutic compounds in our assay and to facilitate high-throughput screening. To exclude cross-contaminated or synonymous lines, a panel of 92 SNPs was profiled for each cell line (Sequenom) and a pairwise comparison score calculated. In addition, STR analysis (AmpFlSTR Identifiler, Applied Biosystems) was conducted and was matched to an existing STR profile. More information on the cell lines screened, including their SNP and STR profiles, is available on the Genomics of Drug Sensitivity in Cancer project website (www.cancerRxgene.org). Cells were seeded at variable density (∼10%–25% confluence) to ensure optimal proliferation during the assay. For adherent cell lines, after overnight incubation, cells were treated with 9 concentrations of each compound (3-fold dilutions series) using liquid handling robotics and then returned to the incubator for assay after 96 hours. For suspension cell lines, cells were treated with compound immediately following plating and were also returned to the incubator for 96 hours. Cells were then stained with 55 μg/mL resazurin (Sigma) prepared in glutathione-free media for 4 hours. Quantitation of fluorescent signal intensity was performed using a fluorescent plate reader at excitation and emission wavelengths of 535/595 nm. All screening plates were subjected to stringent quality control measures (including coefficient of variations of untreated control wells <20%). Effects on cell viability were measured and a curve-fitting algorithm was applied to this raw dataset to derive a multiparameter description of drug response, including the half maximal inhibitory concentration (IC₅₀). Data for JQ1 (733 cell lines overlapping with compound 3) and TG101348 (885 cell lines overlapping with compound 3) were collected in separate experiments using the same experimental conditions and procedures with the exception that cells were exposed to drug for 72 hours. Gene mutation and expression status of drug targets and biomarkers for the screened cell lines were obtained from online databases of the GDSC (www.cancerrxgene.org), Cosmic (cancer.sanger.ac.uk/cosmic), and cBioPortal (www.cbioportal.org).

Cells were seeded in 6-well plates (2 × 10⁵ adherent or 1 × 10⁶ suspension cells per well) and incubated for 6 hours with increasing inhibitor concentrations. Cells were harvested by centrifugation at 300 × g for 5 minutes and resuspended in CelLytic M Cell Lysis Reagent (Sigma-Aldrich) containing Halt Protease Inhibitor Cocktail and Halt Phosphatase Inhibitor Cocktail (Thermo Scientific) and 5 mmol/L EDTA at 4°C. Protein concentrations were determined with Bio-Rad Protein Assay Reagent, and samples were diluted with 1:3 volume 4× SDS sample buffer and heated at 95°C for 5 minutes. Samples were subjected to 10% or 12.5% SDS-PAGE and transferred to polyvinylidene difuoride (PVDF) or nitrocellulose membranes. Western blot analyses were developed with the appropriate pairs of primary and secondary antibodies, and signals were visualized using HyGLO Chemiluminescent reagent (Denville Scientific).

MM1.S cells were treated with 0.5 μmol/L compound or 0.1% vehicle (DMSO) for 24 hours. Cells were harvested and spun down at 4°C, washed with ice-cold PBS, and fixed on ice for at least 30 minutes with 70% ethanol. Cells were washed again with ice-cold PBS, filtered with a cell strainer to achieve a single-cell suspension, and stained with 1 μg/mL DAPI (BD Biosciences #564907) at a cell density of 1 × 10⁶ to 2 × 10⁶ cells/mL for 1 to 2 hours. Sample analysis was performed on a FACSCanto II (BD Biosciences) with DIVA 8 software, and histograms were generated using FlowJo v9 cytometry analysis software (Tree Star, Inc.).

The half maximal inhibitory concentration (IC₅₀) of each compound against BETs was determined by Reaction Biology Corp. using a chemiluminescent alpha screen binding assay. Briefly, donor beads coated with streptavidin were incubated with biotinylated histone H4 peptide (residues 1–21) containing KAc (K5/8/12/16Ac). In the absence of inhibitor, His-tagged BRD binds to KAc-histone H4 peptide, thereby recruiting acceptor beads coated with a nickel chelator. Binding potential is assessed by detecting light emission (520 to 620 nm) from acceptor beads following laser excitation (680 nm) of a photosensitizer within the donor beads which converts ambient oxygen to singlet oxygen. Binding potential for BRD4-1 and profiling across 32 human bromodomains was performed by Discoverx Corp. The amount of BRD captured on an immobilized ligand in the presence or absence of compound was measured using a quantitative real-time PCR (qPCR) method that detects the associated DNA label tagged to the bromodomain. The results are reported as:

Profiling of compounds 3 and 5 was performed at a single concentration of 2 μmol/L.

Inhibitory activity of compounds against JAK2, FLT3, RET, ROS1, and other kinases was determined in dose response by Reaction Biology Corp using a ³³P-ATP radiolabeled assay (10 doses from 0.5 nmol/L to 10 μmol/L). ATP concentration was 10 μmol/L and staurosporine served as a positive control. Residual enzymatic activity (in % of DMSO controls) was determined in duplicate. Profiling of compounds 3 and 5 against a panel of 365 kinases was performed by Reaction Biology at a single concentration of 0.1 μmol/L in duplicate.

Atomic coordinates and structure factors for complexes of BRD4-1 with compounds 1–5 have been deposited in the Protein Data Bank (PDB) under accession codes 5F5Z, 5F60, 5F61, 5F62, and 5F63.

BRDs and kinases are functionally and structurally unrelated, and the respective KAc- and ATP-binding sites are uniquely different in architecture. TG101209, a close analogue of TG101348 (fedratinib), inhibits JAK2 and the first bromodomain of BRD4 (BRD4-1) with IC₅₀ values of 0.5 and 130 nmol/L, respectively (Table 1). The functional groups required for binding to the hinge region of the ATP site in JAK2 (Fig. 1A) directly interact with the side chain of Asn140 in the KAc site of BRD4-1 (7), a conserved residue among bromodomains which is critical for the binding of acetylated proteins (Fig. 1B). To increase binding potential for BRD4 while maintaining high activity against JAK2 and FLT3, we designed analogues of TG101209 to explore the KAc site in several regions, of which the ‘WPF shelf’ (W81, P82, F83), the ‘KL flank’ (K91, L92), and the ‘ZA-channel’ (structurally conserved water molecules) are prominent features (Fig. 1C). For structure–activity relationship (SAR) studies, the binding potential of new compounds was initially assessed by DSF using a standard curve of known BRD4 inhibitors covering a range of IC₅₀ values between 0.025 and 25 μmol/L (7). Inhibitory activity was determined by alpha screen assay for BRD4 and other BETs, and a ³³P-labeled assay was employed to determine activity against kinases. High-resolution co-crystal structures of compounds with human BRD4-1 were determined throughout to guide the design of high-affinity inhibitors (Supplementary Fig. S1 and Supplementary Table S1). The chemical synthesis of this series of compounds is described elsewhere (17). Here, we characterized a select set of compounds that emerged as promising leads for the development of dual BET-kinase inhibitors as cancer drugs.

Early into the SAR studies, we realized that reversal of the sulfonamide functionality in the 3′-position of the A-ring (∼SO₂NH-R to ∼NHSO₂-R) along with the introduction of halogen substituents in either aniline ring significantly increased binding potential for BRD4 (Table 1). Compound 1, an isomer of TG101209, assumes an extended conformation in the KAc site, stabilized by H-bonding interactions of the sulfonamide group with Lys91 and Asp88 below the KL flank (Fig. 1D and E). The binding potential of compound 1 toward BRD4 and kinases JAK2 and FLT3 was similar to TG101209. Introducing a fluorine in the 3′-position of the B-ring (compound 3) increased binding activity for BRD4 4-fold (IC₅₀ = 34 nmol/L). As the binding partners and the conformations of compounds 1 and 3 in the KAc site are identical, the substantial increase in binding activity of compound 3 is likely a result of the fluorine acting as an electron sink, rendering the −NH groups flanking the pyrimidine core more acidic for H-bonding with Asn140 and Pro82. Similarly, introducing a chlorine in the 4′-position of the A-ring (compound 2) also resulted in a 4-fold increase in activity for BRD4 but rendered the compound 10-fold less active against JAK2 and FLT3 (Table 1). It appears that chlorine in 4′-position of the A-ring favors a different inhibitor conformation in the KAc site, the tert-butyl group now neatly positioned in a hydrophobic subsite above the WPF shelf (Fig. 1E). Notably, the chlorophenyl group of JQ1 occupies the same subsite in BRD4, and superimposition of the respective co-crystal structures revealed that compound 2 adopts a conformation mimicking the binding mode of JQ1 (Supplementary Fig. S2). To further diversify the parent compound, several other modifications were introduced, of which compounds 4 and 5 were among the most potent BRD4 inhibitors described to date, with IC₅₀ values of 14 and 21 nmol/L, respectively, similar to JQ1 (IC₅₀ = 21 nmol/L). As expected, the 4′ Cl containing compounds 4 and 5 were less active against kinases and assumed conformations in the KAc site identical to compound 2 (Fig. 1E). Therefore, compounds 2, 4, and 5 can be considered equipotent inhibitors of BRD4 and JAK2/FLT3 (IC₅₀ 10–30 nmol/L), whereas compound 3 is about 10-fold more potent against these kinases. Notably, compounds 2–5 were equally active against the first and second bromodomains of BRD4 and showed only slightly reduced activity against BRDT-1, whereas JQ1 was 7-fold less active against BRDT-1 (Supplementary Table S2). Compound 6, which carries a cyclohexyl moiety fused to the pyrimidine ring of compound 1 (Supplementary Fig. S3), was about 10-fold less active against BRD4 and greatly reduced kinase activity; it served as a negative control in cellular assays.

To assess inhibitory potential against BRDs outside the BET family, compounds 3 and 5 were profiled against a panel of 32 human BRDs (Fig. 2A). Both compounds showed high selectivity for BETs (Family II). Outside the BET family, only compound 5 showed weak binding potential for TAF1 (Family VII) and EP300 and CREBBP (both Family III). Profiling against a panel of 365 kinases confirmed that compound 3 displayed significantly higher activity against the kinome than compound 5 (Fig. 2B). The primary targets were tyrosine kinases (TK) particularly JAK2, FLT3, and RET with IC₅₀ values between 0.9 and 1.1 nmol/L. Other TKs significantly inhibited by compound 3 and causally implicated in cancer (ref. 24; cancer.sanger.ac.uk/census) were NTRK3 (IC₅₀ = 5 nmol/L), ROS1 (IC₅₀ = 11 nmol/L), PDGFRb (IC₅₀ = 16 nmol/L), and FGFR1 (IC₅₀ = 43 nmol/L). The most significant differences in inhibitory potential were noted for ULK3, ULK1, and ERN2/IRE2, which were potently inhibited by compound 3 but appeared to be insensitive towards compound 5. Statistical evaluation of kinase selectivity using the Gini coefficient (25) yielded selectivity scores for compounds 3 and 5 similar to those of ruxolitinib, quizartinib, and fedratinib using published kinome profiling data (ref. 26; Supplementary Fig. S4).

On-target inhibition of dual BRD4-kinase inhibitors was assessed in the multiple myeloma and AML cell lines MM1.S and MV-4-11, both of which are highly sensitive to JQ1 accompanied by downregulation of c-MYC levels (9, 27). Growth of MM1.S cells was potently inhibited by compounds 1–5 with IC₅₀ values between 0.07 and 0.38 μmol/L (Table 1, Fig. 3A). Growth-inhibitory activity of the most potent BRD4 inhibitors 4 and 5 was similar to that of JQ1 (IC₅₀ = 0.1 μmol/L) and greatly improved over JAK2 inhibitor ruxolitinib (IC₅₀ > 10 μmol/L) or FLT3 inhibitor quizartinib (IC₅₀ > 10 μmol/L), both of which lack binding potential for BRD4. Inhibition of BRD4 by JQ1 has been reported to induce G₁ arrest in MM1.S cells (27). Flow cytometric analysis of cells treated with dual BRD4-kinase inhibitors revealed a substantial increase of cells in G₁ arrest similar to JQ1 (Fig. 3B; Supplementary Fig. S5). Single kinase inhibitors ruxolitinib and quizartinib only slightly affected MM1.S cell cycle, whereas the parent compound TG101209 showed a moderate increase of cells in G₁. It appears that the growth inhibition and the ability to induce G₁ arrest are directly correlated. Compounds 1–5 showed a dose-dependent reduction of c-MYC and p-STAT3 levels, reflecting the concomitant inhibition of BRD4 and JAK2, respectively (Fig. 3C). Notably, p-STAT3 levels were detectable only upon stimulation with human IL6, which did not alter STAT3 or c-MYC levels (Supplementary Fig. S6). The pSTAT3 and c-MYC reduction potential of compounds was in general agreement with the inhibitory activities against JAK2 and BRD4, the weaker kinase inhibitors 2 and 5 being less effective at reducing pSTAT3 levels than compounds 1 and 3. MYC levels were affected strongest by compounds 3 and 5, slightly improved over JQ1, compound 3 being most active in attenuating both c-MYC and p-STAT3 levels. As expected, control compound 6 showed only moderate c-MYC reduction potential without affecting p-STAT3 levels, and ruxolitinib and JQ1 exclusively reduced p-STAT3 or c-MYC levels, respectively. Similar results were obtained for the FLT3-driven cell line MV-4-11 (28) which upon treatment with compounds 1–5 showed dose-dependent reduction of p-FLT3 and c-MYC levels (Supplementary Fig. S7). Compounds were also probed for their ability to affect levels of p21Cip1, previously reported to increase in certain cell lines upon treatment with JQ1 (10). In MM1.S, upregulation of p21Cip1 and downregulation of c-MYC levels caused by BRD4 inhibition were time-dependent (Supplementary Fig. S8). Increase in p21Cip1 levels was paralleled by a decrease of c-MYC levels for compounds 1–6, compounds 3 and 5 showing activities similar to JQ1 (Fig. 3C). Onset of apoptosis, assessed by increased levels of c-PARP, was pronounced for dual BRD4-kinase inhibitors, whereas the single activity inhibitors JQ1, ruxolitinib and compound 6, were ineffective.

We investigated the effect of compounds on the JAK2-driven UKE-1 cell line which is a model for MPNs (29). Aberrant JAK activation is a fundamental feature of many hematologic malignancies, including pediatric and Down syndrome–associated precursor B-cell acute lymphoblastic leukemia (B-ALL; ref. 30), Hodgkin lymphoma (31), and Philadelphia chromosome–negative MPNs, including polycythemia vera, essential thrombocytosis, and primary myelofibrosis (32). Chromosomal translocations or somatic alteration (e.g., V617F) result in overactivation of the STAT proteins and other JAK2 effector pathways and is the reason behind the high transforming potential associated with constitutively active JAK2 (33). UKE-1 cells were sensitive for ruxolitinib (IC₅₀ = 0.17 μmol/L) and JQ1 (IC₅₀ = 0.21 μmol/L), but insensitive to quizartinib (IC₅₀ = 9.5 μmol/L; Fig. 4A). Parent compound TG101209 showed good activity (IC₅₀ = 0.32 μmol/L), whereas most dual activity inhibitors showed increased activity, compound 3 being the most active (IC₅₀ = 0.08 μmol/L). Growth of the erythroleukemia cell line HEL and a JAK2-V617F–transformed BaF3 cell line was inhibited by compound 3 with IC₅₀ values of 0.37 and 0.04 μmol/L, respectively, essentially identical to ruxolitinib (Supplementary Fig. S9). Dual BRD4-kinase inhibitors reduced phospho-STAT3 and c-MYC levels similar to those observed in MM1.S cells, whereas JQ1 and ruxolitinib exclusively reduced c-MYC and pSTAT3 levels, respectively (Fig. 4B).

To assess the ability of MPN cells to develop resistance toward single and dual activity inhibitors, UKE-1 cells were exposed to drug over a period of 2 months, starting with drug concentrations equaling the respective IC₂₀ values (see Materials and Methods for details). While cells readily developed resistance for TG101209 and ruxolitinib, they remained highly sensitive to compound 3 (Fig. 4C), suggesting that the concomitant inhibition of BRD4 and JAK2 could not be escaped. A similar observation was previously reported for JAK2 inhibitor–resistant HEL cells that did not readily develop resistance to BRD4 inhibitors (34). Selection of JAK2-V617F–driven cells in escalating doses of JAK2 inhibitors leads to cells that are cross-resistant to type I JAK2 kinase inhibitors (35). Such cells do not contain JAK2 mutations that induce permanent drug resistance (a phenomenon that mimics clinical observations), as cells can become resensitized to the drug after culturing in the absence of drug (35). Remarkably, UKE-1 cells growing persistently in 1 μmol/L ruxolitinib (UKE-R cells) and that are cross-resistant to other JAK2 inhibitors (35, 36) continued to grow in the presence of single inhibitors but could not proliferate when single activity kinase inhibitors were combined with JQ1. However, compound 3 alone completely suppressed UKE-R cell growth (Fig. 4D). To determine whether UKE-1 cells were more sensitive to the combination of JAK2 and BRD4 inhibition, cell growth was monitored over 6 days with JQ1 or ruxolitinib alone and in different combination ratios (Fig. 4E; Supplementary Fig. S10). While cells treated with single inhibitors continued to proliferate, albeit at a slower rate, the drug combinations completely suppressed cell growth. Dose–effect analysis of the growth rate constants according to Chou (22) indicates synergistic activity of ruxolitinib and JQ1 in these cells. Finally, we assessed the efficacy of dual BRD4-JAK2 inhibitors against erythroid colony formation in primary cells from JAK2-V617F–positive MPN patients. A characteristic feature of primary myeloid progenitor cells of MPN patients is their ability to form erythropoietin-independent erythroid colonies in methylcellulose, which is widely used to test MPN therapeutics. Compound 3 was highly effective against colony formation in 2 patient samples with IC₅₀ values < 50 nmol/L, superior to ruxolitinib and TG101209 (Fig. 4F). This was particularly evident for one patient sample that showed reduced sensitivity to ruxolitinib. Taken together, the data on JAK2-driven cell lines and ex-vivo patient samples indicate that dual BRD4-kinase inhibitors exhibit promising activity against MPNs.

Compounds 2, 3, and 6 were screened for growth-inhibitory activity across a large cell line panel comprising 931 liquid and solid tumor cell lines (Fig. 5). Compound 3 (mean IC₅₀ = 0.51 μmol/L) was significantly more potent than compound 2 (mean IC₅₀ = 1.2 μmol/L), indicating that its higher kinase activity contributes to increased cell kill potential (Fig. 5A). The 2 datasets showed high correlation with Pearson coefficient of 0.86, reflecting the same mechanism of action of these 2 compounds (Fig. 5E). As expected, compound 6 was substantially less active due to its reduced binding potential for BRD4 and kinases (mean IC₅₀ = 56 μmol/L). Comparison with cell line screening data for JQ1 and TG101348, separately collected under similar experimental conditions, revealed significantly weaker growth-inhibitory activity with mean IC₅₀ values of 10 μmol/L for both compounds against 733 and 855 overlapping cell lines, respectively. The Pearson coefficients of the JQ1 and TG101348 datasets with that of compound 3 were 0.46 and 0.47, respectively (Fig. 5F and G). By tissue type, the cell lines most sensitive to compounds 2 and 3 were bone and blood (mean IC₅₀ = 0.18 and 0.25 μmol/L, respectively), and the least sensitive were skin and pancreas (mean IC₅₀ = 0.88 and 1.1 μmol/L, respectively; Fig. 5B). The distribution of cell line sensitivity grouped by tissue type and relative to the respective overall IC₅₀ values was similar for compound 3 and TG101348 but differed slightly from JQ1 (Fig. 5C and D; Supplementary Fig. S11). JQ1 showed increased activity for breast and nervous system cell lines but was significantly less active against aerodigestive tract, bone, and lung cancer.

To identify potential gene–drug sensitivity correlations, the resources and databases of www.cancerrxgene.org, cancer.sanger.ac.uk/cosmic, and www.cbioportal.org were used to assign the mutational and overexpression/amplification status of relevant target proteins in the cell lines tested. Enhanced sensitivity for dual BRD4-kinase inhibitors was statistically significant in cell lines overexpressing wild-type or mutant FLT3, RET, PDGFRb, and FGFR1 (Fig. 6A). Altered c-MYC status did not affect drug sensitivity, but cell lines overexpressing BRD4 were significantly more sensitive. Among blood cancer cell lines, those overexpressing FLT3 showed significantly increased sensitivity for the most potent FLT3 inhibitors, compound 3 and TG101348, exclusively (Fig. 6B). MYCN is an established target of BET inhibition in neuroblastoma (4), and consequently cell lines with altered MYCN status showed significantly increased sensitivity only for the most potent BRD4 inhibitors, compounds 2, 3, and JQ1 (Fig. 6C). Unexpectedly, bone cancer cell lines harboring the EWS-FLI1 fusion protein characteristic of Ewing sarcoma were significantly more sensitive for dual BRD4-kinase inhibitors than for JQ1 (Fig. 6D), suggesting the involvement of a kinase in cell killing.

The synergistic lethality of JQ1 and ruxolitinib in UKE-1 cells (Fig. 4E) and the increased growth-inhibitory activity of compounds 2 and 3 over that of JQ1 (Fig. 5A) suggested positive combination effects of BRD4 and kinase inhibition in certain cell lines. We therefore performed drug combination studies of JQ1 with the JAK2/FLT3 inhibitors ruxolitinib, quizartinib, and TG101209 in cell lines with different sensitivity for JQ1. MM1.S and UKE-1 cells as well as the lung cancer cell line HCC-78 and the bone cancer cell line SAOS-2 were exposed to drugs alone and 1:1 combinations with JQ1 for 72 hours (Fig. 7). HCC-78 was chosen because of its dependency on ROS1 (37), which is strongly inhibited by compound 3in vitro (IC₅₀ = 11 nmol/L). SAOS-2 cells were included because they overexpress several kinases potentially targeted by compound 3 including FLT3 (T382S), ROS1, and NTRK3. Ruxolitinib alone was effective only in UKE-1 cells, and combinations of JQ1 with ruxolitinib, TG101209, or quizartinib were synergistic (Fig. 7A), corroborating the results of Fig. 4E. For MM1.S cells, the effect of JQ1 kinase inhibitor combinations was additive (Fig. 7B). Both HCC-78 and SAOS-2 cells showed limited sensitivity for JQ1, as the maximum fraction affected was only approximately 0.5 (Fig. 7C and D). However, both TG101209 and quizartinib showed strong synergism with JQ1 in these cells. For the cell lines tested here, the dose–response properties of compound 3 were similar to that of JQ1 in combination with TG101209, resulting in a complete suppression of cell viability.

Several chemical warheads have been reported as acetyl-lysine mimetics for the development of BRD inhibitors (5). Following the discovery that certain kinase inhibitors also inhibit the KAc site of BRD4 (7, 9), we focused our attention on the optimization of the dianilinopyrimidine-based JAK2 inhibitors fedratinib and TG101209, which are moderately active against BRD4. Dianilinopyrimidine is a privileged scaffold for kinase inhibitor design, and several such drugs are in clinical trials including the ALK inhibitors ceritinib (38) and brigatinib (AP26113; ref. 39), the BTK inhibitor spebrutinib (AVL-292; ref. 40), the EGFR inhibitor rociletinib (41), and the Syk inhibitor fostamatinib (42). Our data demonstrate that the dianilinopyrimidine scaffold is also suited to potently and selectively inhibit BET proteins (Figs. 1, 2). Current lead compounds have been designed as near equipotent inhibitors of BRD4 and a set of cancer-relevant tyrosine kinases including JAK2, FLT3, RET, and ROS1. These compounds potently inhibited the growth of diverse cancer cell lines particularly those of blood, bone, nervous system, and lung cancer. The increased growth-inhibitory activity of dual BRD4-kinase inhibitors over JQ1 suggested synergistic lethality caused by the concomitant inhibition of BRD4 and kinase(s) in certain cell lines. Drug combination studies revealed synergism of BRD4 and JAK2 inhibition in UKE-1 MPN cells (Figs. 4E and 7A), a finding in line with the recently reported synergistic lethality of BRD4 and JAK2 inhibitors against cultured and patient-derived secondary AML blast progenitor cells (43). Synergism of JQ1 and kinase inhibitors in cell killing was also observed in solid tumor cell lines that were sensitive to our inhibitors (Fig. 7C and D). These findings indicate the potential of dual BRD4-kinase inhibitors against cancer cells that depend on aberrant TK activity and BRD4 functionality.

Despite strong initial responses, cancer cells frequently develop resistance to kinase inhibitors by acquiring active site mutations or exploiting the intrinsic redundancy of kinase signaling pathways (44). This shift to alternative kinase signaling nodes occurs through a process termed “adaptive kinome reprogramming” resulting in transcriptional upregulation and activation of compensatory kinases and their adaptor proteins. The failure of single-agent targeted therapy to prevent adaptive kinome reprogramming is seen in almost all cancer types, including in those driven by chromosomal translocations or somatic mutations. For example, gain-of-function somatic and internal tandem duplication mutations of FLT3 are the most common denominators in AML (28). Although FLT3 inhibitors can induce short-term remission in clinical trials, the onset of resistant clones remains a significant challenge (45). Similarly, although JAK2 inhibitor therapy of MPNs can reduce disease burden, it does not reverse the disease by eliminating the malignant MPN clone (46). Overcoming these problems will require attacking cancer cells at multiple levels, either by drug combinations or single drugs targeting multiple proteins. For example, it has been demonstrated that combined targeting of the JAK2 and Bcl-2/Bcl-xL pathways via administration of multiple JAK2 and Bcl-2/Bcl-xL inhibitors overcomes acquired resistance to single-agent JAK2 inhibitor treatment (47). Resistance to BET inhibitors has recently been reported in leukemia as a consequence of increased Wnt/β-catenin signaling (48, 49). As indicated by their high efficacy against MPN cells and patient samples (Fig. 4), dual BRD4-kinase inhibitors may be particularly useful in the treatment of cancers with evolved resistance to single activity TK or BET inhibitors.

H.R. Lawrence has Ownership Interest (including patents) as an inventor and has provided licensing fees. N.J. Lawrence has Ownership Interest (including patents) as an inventor on a patent and has provided licensing fees. E. Schonbrunn has Ownership Interest (including patents) as an inventor and has provided licensing fees. The IP described in patent applications concerning the dual BET-kinase inhibitors was licensed from the Moffitt Cancer Center to Aptose Biosciences. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S.W. Ember, M. Ayaz, C.C. Lynch, N.J. Lawrence, E. Schönbrunn

Development of methodology: S.W. Ember, M. Ayaz, M. Tauro, J.-Y. Zhu, H.R. Lawrence, N.J. Lawrence, E. Schönbrunn

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S.W. Ember, N. Berndt, M. Ayaz, M. Tauro, P.J. Cranfill, P. Greninger, C.C. Lynch, C.H. Benes, G.W. Reuther, N.J. Lawrence, E. Schönbrunn

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.W. Ember, N. Berndt, M. Ayaz, M. Tauro, J.-Y. Zhu, P.J. Cranfill, C.C. Lynch, H.R. Lawrence, G.W. Reuther, N.J. Lawrence, E. Schönbrunn

Writing, review, and/or revision of the manuscript: S.W. Ember, N. Berndt, M. Ayaz, J.-Y. Zhu, C.C. Lynch, H.R. Lawrence, G.W. Reuther, N.J. Lawrence, E. Schönbrunn

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S.W. Ember, Q.T. Lambert, J.-Y. Zhu, P.J. Cranfill, P. Greninger, C.C. Lynch, H.R. Lawrence, E. Schönbrunn

Study supervision: S.W. Ember, C.C. Lynch, N.J. Lawrence, E. Schönbrunn

Other (synthesis and characterization/analysis of key compounds used in this study): S. Gunawan

Other (directed the synthesis of compounds/analogs, interpretation of data): H.R. Lawrence

We thank the Southeast Regional Collaborative Access Team (SER-CAT, University of Georgia) for assistance with Synchrotron data collection at Argonne National Laboratory. We also thank Mathew J. Garnett and Ultan McDermott (Wellcome Trust Sanger Institute, Hinxton, UK) for assistance with cell line screening setup.

This work was supported in part by the Chemical Biology Core at the Moffitt Cancer Center (NIH/NCI: P30-CA076292; H.R. Lawrence, E. Schönbrunn), the National Institute of Child Health and Human Development (NIH/NICHD: HHSN275201300017C; E. Schönbrunn), and a Moffitt Team Science Award (G.W. Reuther, N.J. Lawrence, and E. Schönbrunn). C.H. Benes and P. Greninger were supported by a grant from the Wellcome Trust (102696).