Targeting BET proteins downregulates miR-33a to promote synergy with PIM inhibitors in CMML

Abstract

Background:

Preclinical studies in myeloid neoplasms have demonstrated efficacy of Bromodomain and Extra-Terminal protein inhibitors (BETi). However, BETi demonstrate poor single agent activity in clinical trials. Several studies suggest that combination with other anti-cancer inhibitors may enhance the efficacy of BETi.

Experimental Design:

To nominate BETi combination therapies for myeloid neoplasms, we used a chemical screen with therapies currently in clinical cancer development and validated this screen using a panel of myeloid cell line, heterotopic cell line models, and PDX models of disease. We utilized standard protein and RNA assays to determine the mechanism responsible for synergy in our disease models.

Results:

We identified PIM inhibitors (PIMi) as therapeutically synergistic with BETi in myeloid leukemia models. Mechanistically, we show that PIM kinase is increased after BETi treatment, and that PIM kinase upregulation is sufficient to induce persistence to BETi and sensitize cells to PIMi. Further, we demonstrate that miR-33a downregulation is the underlying mechanism driving PIM1 upregulation. We also show that GM-CSF hypersensitivity, a hallmark of chronic myelomonocytic leukemia (CMML), represents a molecular signature for sensitivity to combination therapy.

Conclusion:

Inhibition of PIM kinases is a potential novel strategy for overcoming BETi persistence in myeloid neoplasms. Our data supports further clinical investigation of this combination.

Introduction

Mutations in genes governing epigenetic regulation are the most common alterations in myeloid malignancies. While clinically heterogenous, mutations in epigenetic regulators such as TET2 and DNMT3a are predominantly early molecular events associated with disease initiation(1,2). In a large subset of cases, secondary mutations in genes encoding signal transduction proteins such as N/K-RAS or FLT3 result in leukemic transformation of clones harboring pre-existing epigenetic pathway mutations in both preclinical and clinical specimens(3,4). The frequency and clonal composition of mutations in genes that alter the epigenome have suggested that therapies aimed at targeting this pathway may be therapeutically attractive. However, while several therapies have been tested across a variety of myeloid neoplasms, only 5-azacitidine has been clinically demonstrated to alter the natural history of myelodysplastic syndromes. Therefore, there exists a critical need to identify therapies that capitalize on the epigenetic dysregulation that molecularly hallmarks these cancers.

The development of potent inhibitors that bind to the BET family of proteins by mimicking acetylated lysine and occupying tandem bromodomains conserved among BET proteins represents a powerful epigenetic therapeutic approach that has been clinically tested in myeloid malignancies(5). Occupation of tandem bromodomains prevents BET proteins from binding acetylated lysine resulting in widespread downregulation of gene expression, particularly those gene expression programs governed by super enhancers, such as NFkB signaling in myeloproliferative diseases(6,7). Preclinical models of myeloid malignancies have suggested that BETi may be highly effective against a variety of hematological malignancies(8–10). However, early phase BET inhibitor clinical trials have demonstrated minimal clinical efficacy with the potential for significant side effects associated with long term treatment(11). Additionally, several persistence mechanisms to BET inhibitors have been identified such as Wnt pathway upregulation(12–14), BRD4 hyperphosphorylation(15), or transcriptional reprograming leading to alternate kinome dependencies(16). Interestingly, these can be reversed with withdrawal of BET inhibitors establishing them as persistent and not resistant mechanisms (16,17). Given the alterations in both epigenetic and signaling mutations seen in myeloid malignancies and the recent preclinical and clinical efficacy observed with combining BETi and JAKi in myeloproliferative neoplasms, we hypothesized that novel combinations of BETi and kinase inhibitors may represent an effective therapeutic strategy for myeloid malignancies(18).

To test this hypothesis, we performed a broad in-house compound screen and identified PIMi as a potential synergistic combination with BETi. We validated this in vitro and in vivo utilizing human leukemia cells, primary patient material, and isogenic BETi persistent cells. Mechanistically, BETi dependent PIM1 overexpression was observed in cell lines vulnerable to the combination therapy and PIM1 overexpression alone was sufficient to both induce persistence to BETi and increase sensitivity to PIMi. We additionally demonstrate that BETi dependent PIM upregulation was secondary to downregulation of miR-33a, known post-transcriptional regulators of PIM1 via global repression of miRNA biogenesis. Last, we find that sensitivity to this combination is associated with cytokine dependent transcriptional priming of PIM1 and validate this in human specimens of chronic myelomonocytic leukemia (CMML) patients which are transcriptionally primed at the PIM1 loci and vulnerable to this therapeutic approach.

Materials and Methods

Experimental Models and Subject Details

Cell Lines

U-937 (CVCL_0007), MV4–11 (CVCL_0064), SKM-1 (CVCL_0098), OCI-AML-3 (CVCL_1844), HEL (CVCL_0001), HL-60 (CVCL_0002), THP-1 (CVCL_0006) and ML-1 (CVCL_0436) cells were cultured in RPMI with 10% fetal bovine serum (FBS). KG-1 (CVCL_0374) cells were cultured in IMDM with 20% FBS. TF-1 (CVCL_0559) cells were cultured in RPMI with 10% FBS and 2ng/mL GM-CSF. M-07e (CVCL_2106) cells were cultured in IMDM with 10% FBS and 10 ng/mL GM-CSF. U-937, KG-1, TF-1, THP-1, HL-60, and MV-4-11 cells were obtained from ATCC. SKM-1, HEL, OCI-AML-3, and M07-e cells were obtained from DSMZ. ML-1 cells were obtained from ECACC. Cell lines were not authenticated, but were routinely tested for mycoplasma by HEK Blue Detection. Cell lines were thawed from cryopreservation and used in experiments for no longer than 3 months.

Patient Samples

Bone Marrow samples were obtained from 12 patients. Patient samples were collected after obtaining written informed consent. The use of human materials was approved by the Institutional Review Board of the Moffitt Cancer Center Scientific Review Committee and the University of South Florida Institutional Review Board in accordance with the Declaration of Helsinki. Patient characteristics for patient samples used in PDX studies and Colony Forming Assays are shown in Supplemental Table S3.

Heterotopic Cell Line Models and CMML PDX

All animal studies were approved by the Moffitt Cancer Center Institutional Animal Care and Use Committee.

U-937, SKM-1 or SKM-1 P1–14 cells were resuspended in cold 0.9% sterile saline and mixed with Matrigel Matrix to a final protein concentration of 7mg/mL. 3×10⁵ U-937 or 1×10⁶ SKM-1 and SKM-1 P1–14 cells were injected into the right flank of NGS-S(RRID:IMSR_JAX:013062) mice (male, 20–30g, 10–30 weeks old) and allowed to reach between 100mm³ and 150mm³ before beginning treatment. Tumors were measured at least twice a week by caliper and tumor volume was calculated using the formula; Tumor volume = width × width × length × 0.52. Mice were randomized into 4 groups: vehicle, INCB057643, INCB053914 or Combination. INCB057643 was administered once a day at 10mg/kg, 5 days a week by oral gavage. INCB053914 was administered twice a day at 30mg/kg, 5 days a week by oral gavage. Both compounds were dissolved in 5% N,N-dimethylacetimide/95% 0.5% methylcellulose.

For CMML PDX experiments, frozen BMMCs were first thawed and treated with DNAse I for 15 minutes to create a single cell suspension. Cells were washed once and resuspended in 0.9% sterile saline and injected via tail vein into NSG-S mice (male or female, 20–30g, 10–30 weeks old) sub-lethally irradiated the day prior. At least 2 million cells were injected into each mouse and treatment started between 2–3 weeks after injection. Mice were randomized into the same groups as the heterotopic cell line models. Treatment lasted 2 weeks and all mice were euthanized shortly after the end of treatment. The spleen, peripheral blood and femurs were taken for analysis. One femur and a portion of the spleen were fixed in formalin and used for IHC. Another portion of the spleen, peripheral blood and bone marrow were further processed by creating a single cell suspension and lysing red blood cells with ACK lysis buffer. Cells were then washed with PBS and stained with zombie violet viability dye (Biolegend Cat#: 423114) before fixation in 1.6% formaldehyde and storage at 4°C.

Methods

Viability Assays

For the drug screen, cells were plated with compounds in 384 well plates and viability was assessed after 72hrs using Cell-Titre Glo (Promega) according to the manufacturer’s instructions. For all other viability assays, cells were plated with compounds in 96 well plates and viability was assessed after 72hrs using CCK8 following the manufacturer’s instructions. synergy was calculated using Zero Interaction Potential (ZIP) via SynergyFinder(19)

Persistent Cell Lines

U-937 and TF-1 cells were grown in medium containing 500nM INCB054329 and SKM-1 cells were grown in medium containing 300nM INCB054329. Persistence was determined by significantly increased IC₅₀ by CCK8 and steady growth in medium containing INCB054329.

Colony Forming Assays

Frozen BMMCs were thawed and prepared in a similar manner to PDX experiments. Cells were then resuspended in IMDM + 2% FBS at a concentration of 200,000 cells per mL. 300μL of cell suspension and 3μL of each compound were added to 3mL Methocult 4034 (StemCell Cat#: 04034) and mixed by vortexing for 1 minute. 1.1 mL of cell mixture was plated in StemCell smart dishes, incubated for 14 days and read on StemVision (Stem Cell Technologies) for the final colony count.

Western Blotting

All cells were lysed using RIPA lysis buffer and protein quantified using BCA. SDS–polyacrylamide gel electrophoresis was performed using 7.5, 10 or 12.5% bis-tris gels and protein was transferred to PVDF membranes using a wet transfer system (Bio-Rad). Membranes were blocked with 5% milk in TBS-T and incubated overnight with primary antibody in either milk or BSA at manufacturer recommended concentrations. Blots were washed multiple times in TBS-T before addition of HRP-conjugated secondary antibody diluted in 5% milk and incubated for 1hr at room temperature. Antibodies used: BRD2(RRID:AB_10835146), BRD4(RRID:AB_2687578), FLAG-TAG(RRID:AB_2572291), PIM1(RRID:AB_2799461), PIM2(RRID:AB_2163921), PIM3(RRID:AB_1904094), BRD3(RRID:AB_1907250), Actin(RRID:AB_476744), Vinculin(RRID:AB_477629).

RNA Extraction

Total RNA from cultured cells was extracted using either Quick-RNA Miniprep (Zymo Research) for gene expression or miRNeasy/miRNeasy advanced (Qiagen) for miRNA detection.

q-RT-PCR

RNA was converted into cDNA using iScript^® Reverse Transcription Supermix for RTqPCR (Bio-Rad). qRT-PCR reactions were performed for in triplicate using probes designed and ordered from IDT (PIM1, PIM2, TBP) or off the shelf TaqMan assays (SREBF2, pri-33a, pre-33a) (ThermoFisher Scientific). For miRNA, cDNA was generated using the TaqMan Advanced miRNA cDNA Synthesis Kit. miRNA qRT-PCR reactions were also performed in triplicate using off the shelf TaqMan Advanced miRNA Assays for miR-33a, miR-33b, miR-16, and miR-26b and miR-93(Endogenous Control miR)(ThermoFisher Scientific). Primers: PIM1 forward: GCTTCGGCTCGGTCTACTCA, reverse: AGTGCCATTAGGCAGCTCTC; PIM2 forward: GGACACCGCCTCACAGATCG, reverse: TGACTGAGTCTGACAAGGGGG; TBP forward: CACGAACCACGGCACTGATT, reverse: TTTTCTTGCTGCCAGTCTGGAC.

ChIP-PCR

U-937, SKM-1, or MV411 cells were serum starved overnight. The next day, cells were stimulated with 10ng/mL GM-CSF for 15mins and immediately fixed with 1% formaldehyde for 10mins. Formaldehyde was quenched with glycine and cells were washed 2X with cold PBS before being snap frozen on dry ice and stored at −80°C. Fixed cells were then prepared using the SimpleChIP Magnetic Bead Kit (Cell Signaling Technologies) according to manufacturer’s recommendations. DNA was sheared using a Qsonica Q800R3 with the following settings: 50% amplitude, 30sec pulse, 5min shearing time. DNA was sheared 5min, spun down, and sheared an additional 5min. Antibodies for STAT5(RRID:AB_2737403), RNA PolII(RRID:AB_2732926), and IGG Negative Control (Included with kit) were incubated overnight before continuing with the protocol according to manufacturer’s recommendations. Primers for PIM1 promoter and enhancer were custom designed and ordered from IDT. PIM1 Enhancer forward: CTTCAACTGCTGTGCTGGTTC, Reverse: CGGGGTTAGTAGTGCAAGGC; PIM1 Promoter forward: GCAGCATCTGGCATCACAAC, Reverse: AAAGCACCAGCGAATCCTGA.

RNA-seq and GSEA

U-937 cells were treated with DMSO, JQ1 and INCB054329 for 24hrs in quadruplicate. Persistent cells treated with INCB054329 were also included in quadruplicate. RNA was extracted and screened for quality on an Agilent BioAnalyzer(RRID:SCR_019715). The samples were then processed for RNA-sequencing using the Nugen Universal RNA-seq kit(NuGEN). Briefly, 100 ng of RNA was used to generate cDNA and a strand-specific library following the manufacturer’s protocol. Quality control steps including BioAnalyzer library assessment and quantitative RT-PCR for library quantification were performed. Two libraries failed QC and were excluded. The libraries were then sequenced the Illumina NextSeq 500 v2 sequencer with two high-output 75-base paired-end runs in order to generate approximately 25 to 30 million read pairs per sample. Sequencing data was mapped to hg38 using STAR “Spliced Transcripts Alignment to a Reference”(20). Raw data was cleaned by removing any genes with less than 10 reads or present in less than half of the samples before running differential analysis through DESeq2(RRID:SCR_015687)(21). Normalized counts were run through GSEA 4.1.0(RRID:SCR_003199) with default parameters except permutation type, which was set to gene_set(22).

Transduction of Cells with PIM1

SKM-1 cells were transduced with a Flag-Tagged, 34kDa isoform of PIM1 in a pCDH-CMV-MCS-EF1α-GreenPuro Cloning and Expression Lentivector(System Biosciences) via the Spinfection method. Briefly, cells were resuspended in Opti-MEM and plated into 6-well plates along with fresh virus, Lipofectamine-2000 and polybrene. Cells were centrifuged for 90mins at 2200rpm in a 37°C centrifuge, and incubated at 37°C for 1 hour, after which 1 mL of normal growth medium was added and cells were incubated overnight. Cells were then centrifuged and resuspended in normal growth medium. After 1 week, cells were single cell sorted for GFP positivity. Single cell clones were profiled for successful transduction by western blotting for Flag-tag.

Electroporation

The Neon Transfection System(ThermoFisher Scientific) with 100μL tips was used to deliver siRNA or miRNA mimics. Cells were first washed with PBS and resuspended in R buffer at a concentration of 5×10⁷ cells per mL. siRNA or miRNA mimics were added to a final concentration of 5μM, mixed thoroughly and cells were electroporated with the following settings: 1400V, 10 pulse width, 3 pulses. Cells were then added to 10mL RPMI with 10% FBS and incubated for either 48 (miRNA mimics) or 72 (siRNA) hours before collection for qPCR and western blotting. For experiments with miRNA mimics, INCB054329 was added 24hrs after electroporation.

miRNA Array

RNA was extracted from U-937 cells treated with DMSO, INBC054329 or JQ1 for 24hrs using miRNeasy Advanced kit (Qiagen). Thermo GeneChip miRNA 4.0 arrays were processed and hybridized according to the manufacturer’s protocol (ThermoFisher Scientific, Waltham, MA). Briefly, 500ng of RNA was processed using the FlashTag Biotin HSR RNA Labeling Kit and following poly-adenylation and ligation of a biotinylated RNA tag, the product was hybridized to GeneChip miRNA 4.0 arrays at 48C for 16 hours at 60 RPM using the GeneChip Hybridization Oven 645. The hybridized miRNA arrays were then washed and stained using the GeneChip Fluidics Station 450, followed by scanning on the Thermo GeneChip Scanner 3000 7G. Data were reviewed for quality control and analysis was performed using the GeneChip Transcriptome Analysis Console v4.0.

Competition Assays

SKM-1 P1–1 or P1–14 were plated with SKM-1 WT cells in a 1:10 ratio. The initial mixture of cells was checked before any treatment started. Cells were plated with 150nM INCB054329 and incubated for 5 days. Each day, 1mL of cell suspension was taken out for analysis of GFP positive cells and replaced with fresh medium with INCB054329 or DMSO.

Flow Cytometry

For SKM-1 competition assays, live cells were washed with FACS buffer and stained with DAPI before running on a FACSCanto (BD Biosciences). For GM-CSF stimulation experiments, cells were starved overnight, incubated with zombie violet for 20mins, washed, and stimulated with varying concentrations of GM-CSF for 15 minutes. Immediately after stimulation, paraformaldehyde was added to a final concentration of 1.6% and cells were fixed for 10 mins. Cells were then washed with PBS and permeabilized using 2mL of ice-cold 95% methanol. After washing off methanol, cells were stored in FACS buffer until analysis. On the day of analysis, cells were stained with pSTAT5 antibody (BD Biosciences) for 15 minutes, washed and run on a FACSCanto. For PDX experiments, cells were resuspended in 50μL FACS buffer with 2μL of both human and mouse FCR blocking antibody and incubated for 10mins. An antibody cocktail comprised of hCD45, mCD45, hCD3, hCD33, and hCD34(BD Biosciences) was added to each tube, incubated for 15 minutes and washed with FACS buffer. Cells were run on an LSRII (BD Biosciences). Data was analyzed in FlowJo(RRID:SCR_008520).

Immunohistochemistry

Slides were stained using a Ventana Discovery XT automated system (Ventana Medical Systems). Briefly, slides were deparaffinized with Discovery Wash solution and heat-induced antigen retrieval method was used in Ribo CC. Rabbit primary antibodies for hCD45(#ab10558, Abcam), PIM1(#PA5–22315, Invitrogen) and PIM2 (#710504, Invitrogen) were used in Dako antibody diluent (Carpenteria) and incubated for 60, 32 and 32 min respectively. Slides were then stained with anti-rabbit secondary (Ventana). Detection was performed using the Ventana ChromoMap kit and slides were counterstained Hematoxylin. Slides were then dehydrated and coverslipped as per normal laboratory protocol.

Microscopy Analysis

Serial slide sections stained for CD45, PIM1, and PIM2 we scanned with a Leica Aperio AT2 digital Pathology Slide Scanner (Leica Biosystems, Vista, CA) with a 20x/0.7NA objective lens. SVS image files were imported into Visiopharm version 2022.02 (Visiopharm A/S, Denmark) where the Tissuealign tool was used to co-register images for the 3 IHC biomarkers. After alignment, the software’s manual drawing tool was used to select Regions of Interest (ROIs) on each aligned image set and a simple intensity threshold segmentation was applied to the ROIs in order to label each co-registered image pixel into staining categories. All thresholds and settings for pixel labeling were identical for each image analyzed within each experiment.

Statistical Analysis

Statistical analyses and graphical representations were performed using GraphPad Prism 9.0 (RRID:SCR_002798). For comparisons involving only 2 sets of data, unpaired t-tests were used. For data involving more than 2 comparison, one-way ANOVA was used. For experiments involving tumor size measurements over time, or the SKM-1 co-culture experiment, we first measured the area under the curve (AUC) of each individual sample to get as single data point describing that sample over time. Then, depending on the groups being compared, either unpaired t-test or one-way ANOVA was used to test significance.

Data Availability

Raw data corresponding to the drug screen in Fig. 1A is attached as Supplemental Table 1. Raw RNA-seq data corresponding to Fig. 2C and 4D and miRNA array data corresponding to Supplemental Fig. 4F is included as Supplemental Data. ChIP-seq, ATAC-seq and DIP-seq data were previously deposited in the Gene Expression Omnibus (GEO) database under GSE159886. For original data, please contact Eric.Padron@moffitt.org.

Figure 1. BETi and PIMi are synergistic in cell line models of CMML:

(A)Results of kinase screen performed in U-937 and SKM-1 cells. Top: Ratio of base drug +/− experimental compounds for all targets. Bottom: Targets filtered by previously published research. (B) ZIP synergy plots generated by SynergyFinder of U-937, SKM-1 and TF1 cell lines; red indicates synergy, green indicates antagonism. Cell lines were treated with 7 increasing doses of both BETi and PIMi for 72hrs(3 biological replicates, 6 technical replicates). (C) IC₅₀ of parental cell lines and their persistent counterparts treated with either BETi(i) or PIMi(ii) for 72hrs (2 biological, 6 technical replicates). (D)Tumor size calculations of mice subcutaneously injected with either U-937 or SKM-1 cells and treated with BETi(INCB057643), PIMi(INCB053914) or combo (U-937: Vehicle and BET n=10, PIMi n=8 Combo n=9; SKM-1: Vehicle, BETi, PIMi n=9, Combo n=8). Mice were treated for 2 weeks unless tumors showed signs of ulceration.

Figure 2. PIM kinases are increased after BET inhibition:

(A) Western blot of cells lines treated with BETi for 24hrs. + indicates treatment, * indicates persistent cell lines. U-937 and TF1 cells were incubated with 500nM of either BETi, SKM1 cells were incubated with 100nM. Each PIM kinase was run on a separate gel due to similar sizes and combined to produce the figure. (B) qPCR of cell lines treated with BETi for 24hrs (2 biological replicates, 6 technical replicates). (C) RNA-seq analysis of BETi treated (500nM) U-937 cells showing the top 20 up and downregulated genes. U5P indicates the persistent cell line. Red = upregulated, Blue = downregulated. (D) Graphic detailing the method for generating the data in Fig. 2E. Figure created in Biorender. (E) Correlation plot of PIMi IC₅₀ and PIM kinase changes of cells treated with BETi for 24hrs. (F) Western blot of BET family proteins in cells treated with siRNA against each individual BET protein. BET proteins were run on separate gels due to similar sizes and combined to produce the figure. Actin was used instead of vinculin because vinculin is a similar size to BRD2 and BRD3. (G) Western blot of PIM1 in cells treated with siRNA against BET proteins

Figure 4. miR-33a is downregulated after BET inhibition and is responsible for PIM1 upregulation:

(A) Western blots of PIM1 in cells treated with AGO2 or Dicer inhibitors. (B) Flow chart of process for selecting miRNAs for further analysis. (C) qPCR of 4 candidate miRNAs in SKM-1 cells treated with BETi 500nM U937, 100nM SKM1) for 2–16hrs. (D) GSEA enrichment plot for miR-33a/miR-33b targets in U-937 cells treated with BETifor 24hrs. (E) qPCR of cells treated with both miR-33a mimic and BETi(500nM U937, 100nM SKM1). (F) Western blot of cells treated with miRNA mimic and BETi(500nM U937, 100nM SKM1).

Results

Clinically relevant kinase inhibitor screen identifies BET and PIM inhibitors as synergistic in models of myeloid malignancies.

In order to test potential synergies between BET inhibitors and inhibitors in clinical development, we utilized an in-house targeted chemical screen of 300 compounds which are FDA approved or in clinical cancer development (Supplemental Table 1)(23). U-937 and SKM-1 cells were incubated with the IC₂₀ (U-937:155nM, SKM-1: 30nM) of the BET inhibitor INCB054329 and two doses (0.5μM and 2.5μM) of each library compound. Cell viability was evaluated 72hrs post-treatment using CellTitre-Glo. Combinations with a drug - base/drug + base ratio greater than 2 were chosen for further consideration as previously described(23). As expected, known synergies with JAK, HDAC, CDK, MEK, and PI3K inhibitors were found supporting the validity of our chemical screen to identify clinically relevant BETi combinations(24–33). After previously published interactions were filtered out (Supplemental Table 2), the only combination with a drug - base/drug + base ratio greater than 2 was with SGI-1776, a pan-PIM inhibitor (Fig. 1A). To validate therapeutic synergy between these BET and PIM inhibitors, we repeated the experiment in three human myeloid cell lines (U-937, TF1 and SKM-1) with 7 doses of INCB054329 and either pan-PIM inhibitors SGI-1776 or INCB053914. In all lines, and in both BET/PIM inhibitor combinations, in vitro synergy was observed consistent with our initial compound screen (Fig. 1B). AZD1208, another PIM inhibitor included in the screen, did not achieve the threshold. However, given that the only compound that met our threshold was a PIM inhibitor we performed formal synergy analysis and were able to demonstrate that AZD1208 was indeed synergistic with BET inhibitors similarly to that observed with INCB054329 and SGI-1776 (Supplemental Figure 1A). Importantly, synergy was evident in the low dose PIMi chemical screen and most models when testing low doses of both BETi and PIMi.

Rapid persistence to BETi has been shown to occur through various mechanisms in both leukemia and solid tumors, likely responsible for its limited clinically efficacy as a single agent(12–16). To determine whether PIMi could overcome persistence to BETi, we generated 3 BET persistent human leukemia cell lines. We then compared the IC₅₀ of PIMi to that of the parental cell lines tested. Persistence was achieved by daily treatment of cell lines with 500nM INCB054329 or 300nM INCB054329 for SKM-1 cells (Fig. 1Ci). These cells were chronically exposed to INCB054329 to maintain persistence. Consistent with the notion that these cells demonstrate BET inhibitor persistence, drug withdrawal for 1 week led to resensitization to BETi (Supplemental Figure 1B). At 60 days, all three cell lines demonstrated an increase in PIMi sensitivity compared to their parental counterparts, particularly in the human monocytic leukemia cell line U-937 (Fig. 1Cii). To determine whether the observed in vitro synergy was present in vivo, heterotopic tumors were established in NSG-S mice(34) with either U-937 (n=10/group) or SKM-1 cells (n=10/group). After tumors reached between 100 and 150mm³, drug treatment was started with 10 mpk (milligrams per kilogram) INCB057643 and 30 mpk INCB053914 via oral gavage either as single agent or in combination and continued for 2 weeks, with tumor measurements occurring twice per week and at endpoint. These experiments identified a statistically significant decrease in tumor volume utilizing both cell line models with combination treatment suggesting that this combination strategy may be synergistic in vivo (Figure 1D).

PIM kinases are upregulated in response to BETi in a subset of leukemia cell lines and correlate to PIMi sensitivity

PIM proteins are serine/threonine kinases with a short half-life that do not require post-translational modifications for their activation and therefore, their activity is primarily transcriptionally and post-translationally mediated(35,36). Given the profound effects that BET inhibitors exert on the transcriptome, we first sought to examine the effect of BET inhibition on RNA and protein expression of PIM kinases. Interestingly, PIM kinase protein and RNA expression of cells treated with BETi after 24 hrs revealed a significant increase in expression of PIM kinases. This increase was highest in BETi persistent cells where significant increases in PIM1 and PIM2 were observed (Fig. 2A–B). Notably, PIM kinases have been previously implicated in drug resistance in AML(37–39). Further, time course studies demonstrated that PIM mRNA upregulation occurs as early as 8hrs (Supplemental Fig. 2A). Because PIM1 was expressed in both short-term treated and persistent cell lines, we prioritized this PIM isoform for further investigation in subsequent experiments. Differential gene expression analysis of RNA-seq data from U-937 cells identified that PIM1 was among the top 20 upregulated genes in BETi treated cells compared to DMSO control (Enrichment score = −3.298) and that a gene set previously reported to be enriched in PIM overexpressing myeloid cells was also upregulated in our BETi treated cells (Fig. 2C, Supplemental Figure 2B)(40). We next confirmed the increased PIM levels after BETi in multiple myeloid leukemia cell lines. Four of nine cell lines demonstrated increased PIM kinase protein levels at 24 hours (Supplemental Fig. 2C). While PIM upregulation was heterogeneous, the BETi dependent increases in PIM levels correlated to increased synergy with BETi and PIMi in vitro (Fig. 2D and E). Since current inhibitors in clinical development are pan-BET inhibitors, including those tested here, we sought to investigate which BET proteins were most associated with PIM upregulation. We individually genetically depleted BRD2, 3 and 4 in U-937 and SKM-1 cells and found that only BRD4 knockdown resulted in significant upregulation of PIM1 levels (Fig 2F and G) consistent with the known expression of BRD4 in the hematopoietic compartment(8,41). Collectively, these data suggest that BET inhibition leads to increased PIM expression in a subset of cell lines that is associated with synergy between BET and PIM inhibitors.

PIM1 overexpression is sufficient to induce persistence to BETi and sensitivity to PIMi

We next sought to determine whether increases in PIM1 alone could drive persistence to BET inhibition as well as contribute to the observed synergy seen in vitro. To test this, single cell PIM overexpressing SKM-1 clones were derived by transducing a GFP expressing lentiviral vector encoding PIM1 (Fig 3A, Supplemental Figure 3A). All four SKM-1 clones engineered to overexpress PIM1 were both persistent to BETi (Figure 3Bi), and significantly more sensitive to PIMi (Fig. 3Bii). Moreover, PIM1 levels correlated with persistence to BET inhibition (R²=0.9925, p=.0037), indicating that PIM1 overexpression is sufficient for BET inhibitor persistence and sensitization to PIM inhibition in vitro (Fig 3C). Of note, although all PIM1 overexpressing clones were more sensitive to PIM inhibition, there was no correlation between levels of PIM1 expression and PIMi sensitivity (Supplemental Fig. 3B). Additionally, we performed in vitro competition assay by co-culturing SKM-1 cells with two isogenic PIM1 overexpressing clones in the presence BETi or vehicle control. After 5 days of treatment with BETi, there was a statistically significant increase in PIM overexpressing isogenic cells indicating that PIM1 overexpressing cells were selected in the presence of their parental counterparts (Figure 3D). To determine whether PIM1 overexpression leads to BET inhibitor persistence and PIM sensitivity in vivo, heterotopic SKM-1 xenograft models were generated of P1–14 SKM-1 PIM overexpressing clones and isogenic controls. As in the above in vivo experiments, flank tumors were allowed to grow until 100–150 mm³ and treatment was initiated for two weeks. These experiments demonstrated statistically significant decreases in tumor volume in PIM over expressing SKM-1 clones after PIM inhibition compared to parental cells suggesting that PIM overexpression is sufficient for PIM inhibitor sensitivity in vivo (Fig. 3E).

Figure 3. PIM1 overexpression is sufficient to induce BETi resistance and PIMi sensitivity:

(A) Western blot of SKM-1 cells transduced with Flag-Tagged PIM1. (B) IC₅₀ curves of SKM-1 cells treated with BETi (INCB054329)(i) and PIMi(INCB053914)(ii). Cells were incubated with drug for 72hrs (2 biological replicates, 6 technical replicates). (C) Correlation of PIM1 expression and BETi IC₅₀ for WT and PIM1 overexpressing SKM-1 cells. (D) Competition assay with SKM-1 P1–1 and SKM-1 P1–14 cells cultured with WT cells at a 1:10 ratio and treated with BETi for 5 days. Flow cytometry was used to determine GFP positivity (E) Tumor volumes of mice with subcutaneously implanted SKM-1 P1–14 cells treated with PIMi(INCB053914). N=4 mice per group, INCB053914 N=3. Mice were euthanized when tumors reached 2cm in size.

BETi decreases miR-33a expression leading to increased PIM1 levels

BETi exert profound effects on the transcriptome but are generally thought to downregulate transcriptional activity(6,42). Therefore, to resolve the paradoxical increase in PIM levels after treatment we first explored BET inhibitor dependent miRNA depletion hypothesizing that depletion of miRNAs that target PIM may lead to the observed increases in PIM levels. BET inhibitors can augment miRNAs via inhibition of miRNA biogenesis at super enhancer regions and/or via direct transcriptional repression of precursor RNA species(43,44). We treated both U-937 and SKM-1 cells with either an Argonaute RISC Catalytic Component 2 (AGO2) inhibitor (Acriflavin) or a Dicer inhibitor (Poly-l-lysine), two central components of miRNA biogenesis, and measured protein PIM1 levels. Indeed, treatment with either AGO or Dicer inhibitors was sufficient to increase PIM1 levels across both cell lines suggesting that inhibition of miRNA biogenesis can recapitulate BET inhibitor induced PIM1 upregulation (Fig. 4A). To narrow down putative repressed miRNAs that may be negative regulators of PIM1, we used the computational approach outlined in Figure 4B. Briefly, miRNAs were identified by cross referencing putative PIM1 binding miRNA from the microRNA Data Integration Portal (miRDIP), miRNA with super enhancers from Suzuki et al. and published PIM1 interacting miRNA(45–49). This led to the identification of 4 putative miRNAs whose expression was evaluated after BET inhibitor treatment. Of these, miR-33a was the only miRNA with a significant time dependent decrease after treatment with two BETi (Fig. 4C). Although this reduction in miR-33a was modest, previous literature supports modest changes in miRNA leading to large impacts on gene expression of its targets(50). This was consistent with GSEA analysis of whole transcriptome RNA-sequencing performed in U-937 cells that demonstrated increased expression of genes targeted by miR-33a(Fig. 4D). To determine if miR-33a depletion was necessary for BET dependent PIM upregulation, we electroporated a miR-33a mimic into both U-937 and SKM-1 cells treated with either BETi or DMSO for 24hrs and collected pellets for both RNA and protein after 48hrs (Fig. 4E). Evaluation of PIM1 protein levels demonstrated that cells with miR-33a overexpression were protected from BETi dependent PIM upregulation (Fig. 4F). These data suggest that reduced levels of miR-33a after BET inhibition leads to an increase in PIM1 expression.

Last, to explore whether BETi directly and specifically impact miR-33a we profiled transcript levels of SREBF2 after BET inhibition as miR-33a is intronically located between exons 19 and 20 of this gene (Supplemental Fig. 4A). This analysis demonstrated a transient increase in SREBF2 transcript expression after 4hr treatment suggesting that BET inhibitors do not directly impact primary miR-33a transcription in leukemia cells (Supplemental Fig. 4B). This was observed both with primers probing the intronic region between exons 19 and 20 as well as primers measuring total SREBF2 (Supplemental Fig. 3B). Importantly, no other promoters were identified in our CMML ChIP-seq data near SREBF2 that would transcribe miRNA-33a independently (Supplemental Fig. 4D). While the rapid turnover of miRNA precursor species precludes precise measurements of their relative abundances after treatment, we attempted to profile the range of miR-33a precursors after BET inhibitor treatment at different time points. Indeed, mature miRNA isoforms (i.e. 3p and 5p) were consistently depleted upon BET inhibitor treatment, but pre-miR-33a did not significantly decrease congruent with the postulated role of BET inhibitor repression of miRNA biogenesis(44)( Supplemental Fig. 4C and E). Moreover, broad miRNA expression profiling in U-937 cells demonstrated a global statistical down regulation of miRNAs in 2 replicates suggesting that miR-33a downregulation may occur through impairment of miRNA biogenesis (Supplemental Figure 4F).

Upregulation of the GM-CSF/STAT5 axis is associated with sensitivity to combination therapy

The GM-CSF/STAT5 axis is the canonical upstream signal required for PIM transcription of all isoforms (51–54). Given the above mechanism of synergy with respect to PIM upregulation and our previous studies demonstrating that GM-CSF hypersensitivity is a universal feature of the monocytic leukemia CMML, we hypothesized that the subset of monocytic leukemia cell lines which upregulated PIM after BET inhibition could be identified a priori by their respective pre-treatment GM-CSF sensitivity. We posited that leukemia cells, which were responsive to GM-CSF stimulation at low doses would be primed to upregulate PIM upon BETi dependent downregulation of its inhibitory miRNAs. To explore this possibility, we measured pSTAT5 in the presence or absence of low dose (0.1ng/ml) granulocyte macrophage colony stimulating factor (GM-CSF). Consistent with our hypothesis, cells that exhibited BET dependent PIM protein upregulation demonstrated greater pSTAT5 activation after only 0.1ng/mL of GM-CSF stimulation compared to cells that did not upregulate PIM kinase (Fig. 5A). In the context of PIM1 specifically, pSTAT5 enrichment was accompanied by STAT5 occupation at the PIM1 downstream enhancer in PIM1 upregulating leukemia cells at a far greater magnitude compared to those cell lines that did not upregulate PIM1 after BETi (Fig. 5B). Additionally, GM-CSF stimulation led to enrichment of RNA PolII at both the PIM1 enhancer and promoter in PIM1 upregulating cell lines but not in a leukemia cell line that did not upregulate PIM1 (Fig. 5C). Collectively our data suggests that GM-CSF hypersensitive myeloid leukemia cells are transcriptionally primed at the PIM1 loci, upregulate PIM kinases after BETi, and are associated with synergy to BET and PIM inhibition.

Figure 5. Upregulation of the GM-CSF/STAT5 axis is associated with sensitivity to combination therapy.

(A) Flow cytometry analysis of pSTAT5 levels after stimulation with 0.1ng/mL GM-CSF in 11 myeloid cells lines with corresponding changes in PIM protein levels after treatment with a BETi. Different PIM protein are denoted by a black circle (PIM1) or black square (PIM2) above each bar. (B) ChIP-PCR of STAT5 levels in U-937 and MV-4–11 cells at the PIM1 promoter and enhancer after stimulation with GM-CSF(10ng/mL) and treatment with BETi (U937 500nM, MV-4–11 100nM)(2 biological replicates, 6 technical replicates). (C) ChIP-PCR of RNA PolII at the PIM1 promoter and enhancer in U-937, SKM-1 and MV-4–11 cells after stimulation with 10ng/mL GM-CSF)(2 biological replicates, 6 technical replicates). (D) ChIP-seq data from 16 unique CMML patients at the PIM1 locus.

Chronic Myelomonocytic Leukemia (CMML) is a rare hematologic malignancy classified as a Myelodysplastic/Myeloproliferative overlap syndrome by the World Health Organization(55). Clinically and pathologically, this disease is characterized by bone marrow dysplasia, peripheral monocytosis, cytopenias, and a propensity for transformation to Acute Myeloid Leukemia, all of which contribute to a poor overall survival(55). Molecularly, CMML is hallmarked by GM-CSF hypersensitivity in a mutational and subtype independent manner (56,57). To determine whether this molecular feature was associated with transcriptionally primed PIM, we leveraged our previously published multi-omic epigenetic dataset of 16 CMML patients that enabled us to probe chromatin accessibility and histone marks at the PIM1 loci (58). Both when viewed in aggregate (Fig. 5D) or as individual patients (Supplemental Fig. 5) the PIM1 promoter and enhancer demonstrated epigenetic marks consistent with transcriptional activity supporting the notion that CMML may represent a subtype of leukemia enriched for sensitivity to BET and PIM inhibition.

Combination therapy with BET and PIM inhibition is a viable therapeutic strategy in primary CMML cells in vitro and in vivo

To determine if the synergy between BET and PIM inhibitors is evident in primary CMML samples we performed clonogenicity assays with bone marrow mononuclear cells (BMMCs) from 10 unique CMML patients (Supplemental Table 3), in duplicate, treated with 100nM INCB054329 and 500nM INCB053914 or the combination. While single agent INCB054329 and INCB053914 demonstrated significant reductions of clonogenicity, combination therapy was significantly more efficient at reducing clonogenicity compared to all groups. (Fig. 6A and B). To determine whether combination therapy was a viable therapeutic approach in vivo, we generated CMML patient derived xenografts (PDX) as previously described (Supplemental Table 3) (59). After engraftment was established in each model, mice were randomized (3–5 mice per group) and treated with BET inhibitor, PIM inhibitor, or the combination for 2 weeks using the same doses as heterotopic cell line xenograft experiments (Fig. 6C). Initially, mice treated with the maximum tolerated dose of BETi and PIMi rapidly lost weight and had unacceptable toxicity (data not shown). However, given that in vitro synergy optimally occurred at lower doses of both inhibitors and nanomolar levels of BETi were sufficient to induce PIM upregulation (Supplemental Fig. 6), PDX experiments were repeated with low dose BET and PIM inhibition. After 2 weeks of treatment, all mice were simultaneously euthanized in order to compare engraftment, consistent with other preclinical myeloid studies on drug efficacy(60,61). Mice treated with either low-dose INCB057643 or INCB053914 alone showed a variable response to treatment, similar to in vitro experiments, while the combination was consistently able to reduce leukemic engraftment as evidenced by a reduced percentage of human CD45+ cells in the bone marrow (60–62) by both flow (Fig. 6D) (BETi vs Combo mean rank diff. = 13.49. PIMi vs Combo mean rank diff. = 11.30) and IHC (Fig. 6E and F). Finally, we profiled PIM expression in our PDX models to determine whether the postulated mechanism of synergy occurred in primary patient samples. Immunohistochemistry (IHC) was performed on spleen sections using rabbit anti-PIM1 and anti-PIM2. Our anti-PIM1 and anti-PIM2 antibodies cross-react with human and mouse tissues. To overcome this limitation we computationally overlaid PIM IHC with that of human CD45 to quantitate human specific PIM expression(see methods)(Fig. 6G). This analysis demonstrated that PIM upregulation occurred after BET inhibitor treatment in vivo in primary samples.

Figure 6. PDX models of CMML recapitulate in vitro data.

(A) Representative images from 3 patient sample CFAs. (B) Quantification of colony counts of CFA data, n=10 unique patients. (C) Graphical representation of PDX experiment timeline. (D) Flow cytometry analysis of hCD45 content in bone marrow of mice from 4 PDX experiments with 4 unique patients. Mice were treated with either BETi(INCB057643), PIMi(INCB053914) or combination. Significance determined using Kruskal-Wallis. (E) Representative images of bone marrow and spleen slides stained with hCD45. (F) Quantification of hCD45 in bone marrow and spleen IHC slides from PDX experiments. (G) Left: Representative image of a PDX spleen stained with hCD45, PIM1 and PIM2. Slides were stained with individual markers and overlaid using a computational program described in methods. Blue color represents area of hCD45 and PIM1 colocalization. Right: Quantification of the colocalization of hCD45/PIM1 and hCD45/PIM2 in spleen samples taken from PDX experiments.

Discussion

Despite advances in the molecular pathobiology and genetics of myeloid malignancies, no targeted therapeutics have demonstrated an impact on overall survival or augment natural history. This is especially evident in CMML where there are no CMML-specific approved therapies and the vast majority of patients will succumb to disease within 5 years (63). To address this therapeutic gap, we utilized a targeted chemical screen and identified BETi and PIMi as a synergistic combination in preclinical models in vitro and in vivo. While this synergy has not been previously reported, it is consistent with recent studies suggesting that PIM kinase upregulation may be associated with disease progression and resistance to cytotoxic therapies in AML(37–39). It is also consistent with the notion that BET inhibitor and kinase inhibitor combination therapy may be an attractive therapeutic strategy in hematologic malignancies(64–68).

Our study identified that PIM protein and RNA levels were paradoxically upregulated after BET inhibitor treatment in multiple cell lines and that PIM1 upregulation was necessary for sensitivity to PIM inhibition. Although no correlation was observed between PIM1 protein levels and PIMi sensitivity, all PIM1 overexpressing cells were significantly more sensitive to PIM inhibition than their parental control perhaps due to a threshold effect that was met in all of our PIM1 overexpression cell lines. That PIM kinase upregulation was sufficient to induce this phenotype, without upstream activation, is consistent with its known mechanism of phosphorylation. Unlike many serine threonine kinases which require a secondary phosphorylation event in order to become active, PIM kinases are constitutively active after translation(69). While we fully investigated the mechanism of BET inhibitor dependent PIM1 upregulation which was increased in all BET inhibitor persistent cells generated and our PDX models, our data provide a mechanism of synergy likely consistent in the context of other PIM isoforms given their overlapping function and known role in chemotherapy resistance(35).

Given the profound downregulation of transcriptional activity and the paradoxical increase in PIM kinase RNA in our leukemia models, we hypothesized BET inhibitors may downregulate a post-transcriptional repressor of PIM. Indeed miR-33a, a known regulator of PIM kinase, was downregulated and necessary for the observed BETi dependent PIM upregulation(45,70,71). Further, our data strongly suggests that BETi dependent impairments in miRNA biogenesis, and not direct transcriptional repression of mir-33a precursors, underlies the mechanism of miR-33a downregulation making this proposed combination therapy mechanistically novel.

Last, we demonstrated that GM-CSF hypersensitive cells were transcriptionally primed at the PIM1 loci and were associated with sensitivity to combination therapy. We therefore utilized our CMML PDX models, which recapitulate many features of the human condition (59), to credential this therapy in a randomized murine clinical trial and identified statistically superior repression of human leukemia engraftment with combination therapy across all models. Our data provides strong evidence to clinically test low dose BET and PIM inhibition in CMML, which we predict would be most responsive to this combination based on its hypersensitivity to GM-CSF(56,59). Future clinical studies are necessary to identify the biologically effective dose of BET inhibitor that upregulates PIM kinase in humans so that responses are enriched and toxicity minimized.

Translational Relevance.

Acquired cellular persistence to BET inhibitors (BETi) poses a significant challenge to their clinical efficacy in myeloid neoplasms. This study identified BETi-dependent PIM upregulation as a targetable mechanism of persistence and nominated BET and PIM inhibitor combination therapy for a subset of myeloid neoplasms characterized by GM-CSF hypersensitivity.