Patients with acute myeloid leukemia (AML) frequently do not respond to conventional therapies. Leukemic cell survival and treatment resistance have been attributed to the overexpression of B-cell lymphoma 2 (BCL-2) and aberrant DNA hypermethylation. In a phase Ib study in elderly patients with AML, combining the BCL-2 selective inhibitor venetoclax with hypomethylating agents 5-azacitidine (5-Aza) or decitabine resulted in 67% overall response rate; however, the underlying mechanism for this activity is unknown.
We studied the consequences of combining two therapeutic agents, venetoclax and 5-Aza, in AML preclinical models and primary patient samples. We measured expression changes in the integrated stress response (ISR) and the BCL-2 family by Western blot and qPCR. Subsequently, we engineered PMAIP1 (NOXA)- and BBC3 (PUMA)-deficient AML cell lines using CRISPR-Cas9 methods to understand their respective roles in driving the venetoclax/5-Aza combinatorial activity.
In this study, we demonstrate that venetoclax and 5-Aza act synergistically to kill AML cells in vitro and display combinatorial antitumor activity in vivo. We uncover a novel nonepigenetic mechanism for 5-Aza–induced apoptosis in AML cells through transcriptional induction of the proapoptotic BH3-only protein NOXA. This induction occurred within hours of treatment and was mediated by the ISR pathway. NOXA was detected in complex with antiapoptotic proteins, suggesting that 5-Aza may be “priming” the AML cells for venetoclax-induced apoptosis. PMAIP1 knockout confirmed its major role in driving venetoclax and 5-Aza synergy.
These data provide a novel nonepigenetic mechanism of action for 5-Aza and its combinatorial activity with venetoclax through the ISR-mediated induction of PMAIP1.
The FDA recently approved the combination of venetoclax with hypomethylating agents in elderly patients with acute myeloid leukemia (AML); however, the molecular mechanism behind the combinatorial activity is unknown. We show that 5-azacitidine (5-Aza) and venetoclax provided added antitumorigenic benefit relative to either agent alone in preclinical models of AML. We characterize the 5-Aza–mediated apoptotic priming of AML cells linked to the induction of the proapoptotic protein NOXA and its binding to antiapoptotic BCL-2 family proteins. This study uncovers a nonepigenetic mechanism for the combinatorial activity between venetoclax and 5-Aza, and highlights a central role for NOXA in venetoclax-induced apoptosis.
Acute myeloid leukemia (AML) is a clonal hematologic malignancy characterized by genomic heterogeneity and epigenetic changes, including aberrant DNA methylation (1). Hypomethylating agents (HMAs), such as the cytidine analogs 5-azacitidine (5-Aza) and decitabine, demonstrate single-agent activity in 25% to 50% of patients with myelodysplastic syndrome (MDS; ref. 2) and myeloproliferative neoplasms (3), and are active in about 15% to 29% of AML (4, 5). Following cellular uptake, 5-Aza is metabolized and incorporated into both DNA and RNA to drive distinct cellular responses. Many of 5-Aza's epigenetic effects can be attributed to its incorporation into DNA to deplete DNA methyltransferase (DNMT) and drive DNA hypomethylation (6, 7). However, 80% to 90% of 5-Aza is incorporated into RNA (8–10), which may drive nonepigenetic effects (11, 12) including apoptosis (13, 14). This may account for some of 5-Aza's clinical activity because DNA hypomethylation is not predictive of clinical response to 5-Aza in myeloid malignancies (15).
Apoptosis is primarily regulated at the mitochondrial level by the B-cell lymphoma protein-2 (BCL-2) family of proteins. This collection of cell death regulators is divided into three groups that each contains at least one BCL-2 homology (BH) motif (BH1-4). The proapoptotic “BH3-only” proteins BIM, BID, PUMA, NOXA, BAD, BIK, BMF, and HRK, and the “multidomain effector” proteins BAX and BAK are activated or induced by various cell death stimuli that drive mitochondrial outer membrane permeabilization and subsequently apoptosis (16). The antiapoptotic members (BCL-2, BCL-XL, MCL-1, BCL-W, and BCL2-A1) possess BH3-binding grooves that function to constrain the “BH3-only” and multidomain effectors. Aberrant expression and/or function of BCL-2 family proteins are integral to tumorigenesis and therapeutic resistance by enabling malignant cells to evade apoptosis (16, 17). Deletion of the genes encoding the BH3-only proteins BIM (BCL2L11), PUMA (BBC3), or NOXA (PMAIP1) can drive resistance to various apoptosis stimuli (18–22) and accelerates tumorigenesis in mouse models of hematologic malignancies (23–25). Similarly, BCL-2 overexpression is a clinical feature of human hematologic malignancies (26–28) that exacerbates the malignant state (29) and drives apoptosis resistance phenotypes (30), including that induced by 5-Aza (13). Consequently, targeting apoptosis signaling, and in particular BCL-2 itself, has emerged as a tractable therapeutic approach in oncology.
Venetoclax (ABT-199) is a highly selective, orally bioavailable BCL-2 inhibitor that induces apoptosis in BCL-2–dependent tumor cells (31), and is approved by the FDA in the United States for use in patients with small lymphocytic lymphoma or chronic lymphocytic leukemia, who have tried at least one therapy (32). Further, in a phase II monotherapy study in patients with R/R AML, venetoclax has demonstrated an overall response rate (ORR) of 19% (33), providing the foundation for combinational studies in AML. Subsequent phase Ib data in treatment-naïve patients with AML indicate that combining venetoclax with 5-Aza or decitabine results in an ORR of 67% (34), compared with the historical ORR of 15% to 29% with HMA treatment alone (5, 35–37). These encouraging data have led to the initiation of randomized phase III trials to evaluate venetoclax activity in combination with 5-Aza in elderly treatment-naïve AML patients ineligible for standard induction therapy (M15-656, NCT02993523). However, the underlying mechanism for the combinational activity observed between venetoclax and 5-Aza is unknown.
Herein, we demonstrate that acute 5-Aza treatment drives combinatorial activity with venetoclax independent of epigenetic effects (6, 7). Specifically, we determine that, at pharmacologically relevant concentrations (38), 5-Aza activates the integrated stress response (ISR) pathway to induce expression of the canonical target DDIT3, as well as the BH3-only proteins NOXA (PMAIP1) and PUMA (BBC3) in human AML cell lines, priming them for apoptosis independent of DNA demethylation. We subsequently demonstrate that NOXA induction is not only required for the synergy observed between 5-Aza and venetoclax, but PMAIP1 deletion significantly affects the kinetics and depth of response to either agent alone or in combination. Together, these data provide a rationale for an ongoing randomized phase III clinical trial evaluating venetoclax activity in combination with 5-Aza (M15-656, NCT02993523), and advocate for the assessment of PMAIP1 and DDIT3 gene induction in patients treated with 5-Aza and venetoclax as potential biomarkers of response.
Materials and Methods
Cell culture and reagents
AML cell lines OCI-AML5 (RRID:CVCL_1620), SET-2 (CVCL_2187), PL-21 (CVCL_2161), SKM-1 (CVCL_0098), MOLM-13 (CVCL_2119), SKNO-1 (CVCL_2196), SHI-1 (CVCL_2191), and NOMO-1 (CVCL_1609) were purchased from DSMZ; and HEL (CVCL_2481), U-937 (CVCL_0007), MV4-11 (CVCL_0064), Kasumi-1 (CVCL_0589), KG-1 (CVCL_0374), and THP-1 (CVCL_0006) were purchased from the ATCC and cultivated for 1 to 10 passages in RPMI-1640 medium, 20 mmol/L HEPES (Gibco) supplemented with penicillin/streptomycin and 10% FBS (Invitrogen). Cells were grown at 37°C in a humidified atmosphere with 5% CO2. All cell lines were tested for authenticity by short tandem repeat profiling and mycoplasma by the AbbVie Core Cell Line Facility. Primary AML cells from peripheral blood were purchased from Discovery Life Sciences, collected with informed consent from patients, and cultured overnight in media described above, plus 30 U/mL of IL2, prior to 5-Aza treatment. AML-340 and AML-343 (stage M2) were newly diagnosed, untreated; AML-69 had previous treatment: gemtuzumab ozogamicin, and decitabine, treatment at time of collection: venetoclax; AML-3667 treatment at time of collection: mercaptopurine. Venetoclax and 5-Aza were obtained from AbbVie chemical library. ISRIB (trans-isomer) was purchased from Selleck Chemicals (S7400).
AML cell lines were seeded at 10,000 cells per well in 96-well plates and treated with 5-Aza and or venetoclax for 24 hours. For combinatorial studies, cells were treated in matrix of nine doses of venetoclax (10 μmol/L, 1:3 dilutions) and three doses of 5-Aza, 0.3, 1, and 3 μmol/L. Cell viability was subsequently determined using CellTiter-Glo reagent as described by the manufacturer's instructions (Promega Inc.). The effective concentration to induce 50% cell death (EC50) was determined by nonlinear regression algorithms using Prism 7.03 (GraphPad Software).
AML cells were treated with 0.3, 1, 3, 10, or 30 μmol/L of 5-Aza for 24 hours. To inhibit caspase activation, cells were pretreated with 40 μmol/L Z-VAD-fmk for 1 hour prior to treatment with 5-Aza for a further 24 hours. Protein concentration was quantified using Pierce bicinchoninic acid assay (ThermoFisher Scientific), and equal amounts of cellular protein samples were separated with 4% to 12% SDS-PAGE (Invitrogen) and blotted to nitrocellulose or PVDF membranes (Invitrogen). The blots were incubated with the following primary antibodies: anti-BIM (Cell Signaling Technology; catalog No. 2933, RRID:AB_1030947), anti–BCL-2 (Abcam; catalog No. ab32124, RRID:AB_725644), anti–BCL-XL (Abcam; catalog No. ab32370, RRID:AB_725655), anti–MCL-1 (Cell Signaling Technology; catalog No. 94296, RRID:AB_2722740), anti-PUMA (Abcam; catalog No. ab9645, RRID:AB_296538), anti-NOXA (Abcam; catalog No. ab13654, RRID:AB_300536), anti-PARP (BD Biosciences; catalog No. 556494, RRID:AB_396433), anti–caspase-3 (Abcam; catalog No. ab13585, RRID:AB_300480), ani-DNMT1 (Cell Signaling Technology; catalog No. 5032, RRID:AB_10548197), anti-ATF4 (Cell Signaling Technology; catalog No. 11815, RRID:AB_2616025), anti-CHOP (Cell Signaling Technology; catalog No. 2895, RRID:AB_2089254), anti-eIF2α (Cell Signaling Technology; catalog No. 2103, RRID:AB_836874), anti–phospho-eIF2α (S51) (Cell Signaling Technology; catalog No. 3398, RRID:AB_2096481), anti–β-actin (Sigma-Aldrich; catalog No. A2228, RRID:AB_476697), and anti-GAPDH (Abcam; catalog No. ab110305, RRID:AB_10861081). The blots were imaged using the Odyssey CLx (Li-Cor) following incubation in with the IRDye secondary antibodies (LI-COR Bio, AB_10795014; AB_10796098). Approximate protein molecular size was calculated from protein size standards in Image Studio 5.0 (LI-COR).
Cells were pretreated with 40 μmol/L zVAD-fmk 1 hour prior to treatment with 0.3, 1, 3, or 10 μmol/L of 5-Aza for a further 24 hours. Cellular proteins were extracted and centrifuged in 1% CHAPs buffer. Cell lysates (250 μg) were incubated with 3 μg of biotinylated anti–BCL-2 (US Biological, catalog No. B0807-06F) or BCL-XL (R&D systems, catalog No. DYC894-2) or MCL-1 (BD Bioscience, catalog No. 624008 custom biotinylation) or IgG control (US Biological, B1750-06X) antibodies in the presence of protease inhibitors (complete tablets, Roche) overnight at 4°C. Streptavidin beads (Sigma) were added to precipitate complexes containing BCL-2, BCL-XL, or MCL-1 prior to separation by SDS-PAGE.
DNA methylation analysis
AML cells were treated with 0.3 or 3 μmol/L of 5-Aza for 24 hours or 7 days, replenished on days 0, 2, 4, and 6. For methylation analysis, 500 ng of DNA was denatured by heating the sample at 100°C. Samples were treated with 5 units of Nuclease P1 (Sigma) in reaction buffer (5 mmol/L ZnCl2, 50 mmol/L NaCl) at 50°C for 1 hour. Samples were further treated with 0.002 units of phosphodiesterase I (Sigma) at 37°C for 2 hours followed by treatment with 0.5 U of alkaline phosphatase (Invitrogen) for 1 hour at 37°C. The samples were diluted threefold to dilute out the salts and enzymes, and injected into a Acquity UPLC HSS T3 2.1-mm x 50-mm column (1.8-μm particle size; Waters catalog No. 186003538). Samples were run on a Shimadzu Nexera-X2 UHPLC coupled to an ABSciex 6500 Triple Quadrupole Mass Spectrometer. Quantification was performed using the Analyst 1.6.2 Workstation software using the intelliQuan algorithm in a multiple reaction monitoring mode. All runs included standard curves for 5-hmC, 5-mC, and 5-dC. Linear regression was performed to obtain slopes and intercepts, which were used to calculate the actual % 5-mdC and % 5-hmdC values. All standards and samples were matrix matched and treated with same dilutions.
Cell line–derived xenograft models of AML
SKM1.FP1 is a cell line derived from a subcutaneous tumor generated by the injection of 2 × 106 of SKM-1 cells (DSMZ). SKM1.FP1 and MV4-11 (ATCC) were cultured as described above. Female Fox Chase SCID Beige (RRID: IMSR_CRL:250; for SKM1.FP1; 8–10 weeks, 18–20 g) and female Fox Chase SCID (RRID: IMSR_CRL:236; for MV4-11; 8–10 weeks, 18–20 g) mice were purchased from Charles River Laboratories. Body weights upon arrival were 18 to 20 g. Food and water were provided ad libitum. Animals were on the light phase of a 12-hour light: 12-hour dark cycle. All animal studies were conducted in accordance with the guidelines established by the AbbVie Institutional Animal Care and Use Committee. In flank xenograft experiments, mice were inoculated with 2 × 106 (SKM1.FP1) or 5 × 106 (MV4-11) cells subcutaneously into the right flank. Mice were injected with a 0.1-mL inoculum of 1:1 cell mixture in culture media and Matrigel (BD Biosciences). When tumors reached approximately 225 mm3, the mice were size-matched into treatment and control groups. Mice were treated QDx14 PO 50 mg/kg with venetoclax, Q7Dx3 IV 8 mg/kg with 5-Aza, or combination of both. Venetoclax was formulated in 60% phosal 50 propylene glycol, 30% polyethylene glycol 400, and 10% ethanol, and 5-Aza was formulated with 0.9% sodium chloride. Tumor volume was measured twice per week with electronic calipers and calculated according to the formula (L x W2)/2. All treatment groups consisted of eight mice per group.
Caspase-3/-7 activity time course
Note that 96-well clear-bottom black polystyrene microplates (Corning) were coated with CellTAK (Corning) as recommended by the manufacturer. For each well, 5 × 104 cells were seeded in 50 μL of RPMI media and then placed in the tissue culture incubator for 20 minutes. The IncuCyte Caspase-3/-7 Red Apoptosis Assay Reagent (Sartorius) and the compounds were dispensed into the corresponding wells and dilutions using the D300 Digital Dispenser (Tecan). The assay plate was then placed in the IncuCyte ZOOM (Sartorius) and programmed to take four images per well at a 1-hour interval for 24 hours. Data were analyzed using the IncuCyte Zoom 2017 software and plotted as the number of red objects (active Caspase-3/7+, aCasp-3/-7+) per well divided by area (in μm2) occupied by the cells, per well.
Note that 1.0 × 107 cells, for AML cell lines, and 2.5 × 105 cells per well, for primary AML cells, were treated with DMSO or 0.3, 1, or 3 μmol/L of 5-Aza, for 6 and 24 hours, at which point the cells were washed with 10 mL of cold DPBS and spun down at 4°C, 300 rcf for 5 minutes. The supernatant was removed, and the cells were lysed with 400 μL of RLT buffer containing β-mercaptoethanol. RNA was isolated using Qiagen RNeasy Plus Mini Kit (74134) for cell lines, and Micro Kit (74034) for primary samples, as recommended by manufacturer. cDNA was prepared from 3 μg of purified RNA using iScript Reverse Transcription Supermix (Bio-Rad) protocol. qPCR was performed on 2 μL of cDNA reaction, TaqMan Fast Advanced (Life Technologies, No. 4444557) in 20-μL final volume. The following PrimeTime qPCR Probes were manufactured by Integrated DNA Technologies (IDT): HEX dye probe RPLP0 (reference gene, Hs.PT.39a.22214824); 6-FAM probes: PMAIP1 (Hs.PT.58.21318159), BBC3 (Hs.PT.5839966045), DDIT3 (Hs.PT.58.39204289.g), CDKN1A (Hs.PT.47.2442322; Supplementary Table S1). The cDNA standards (fivefold dilutions) were prepared from AML5 and Kasumi-1 cells treated with SN-38 for 24 hours. qPCR was performed in 96-well PCR plates (Bio-Rad) on the Bio-Rad CFX 96. Gene expression (ΔΔCq) was quantified on the Bio-Rad CFX Manager 3.1. Target 6-FAM probes were normalized to HEX-GAPDH control gene and relative to vehicle control treated 0 μmol/L 5-Aza.
Knockout cell lines
Genomic editing for elimination of BBC3 (PUMA) and PMAIP1 (NOXA) was done using the ALT-R CRISPR-Cas9 system from IDT. All reagents were from IDT, and RNP complexes were formed as described by IDT. Briefly, the CRISPR RNA (crRNA; Supplementary Table S1) was annealed to tracRNA labeled with ATTO 550 fluorescent reagent. The annealed product was combined with Alt-R S.p. Cas9 Nuclease V3 (IDT) and subsequently introduced into the OCI-AML5, Kasumi-1, and MV4-11 cells by electroporation. The electroporation was done with a NEON Transfection System (Thermo Fisher) containing 2 × 106 cells using the following settings: OCI-AML5 (1 × 30 ms pulse @ 1,500 V), Kasumi-1 (1 × 20 ms pulse @ 1,700 V), and MV4-11 (1 × 20 ms pulse @ 1,700 V). Transfection efficiency was monitored using the ATTO 550 fluorescent reagent. At 24 hours following transfection, the top 5% ATTO 550+ cells were sorted by fluorescence-activated flow cytometry (BD FACSAria Fusion) under sterile conditions as a single cell per well of a 96-well plate and were used to isolate clonal cell lines. The expression of either NOXA or PUMA protein was monitored using Western blot analysis. Genomic DNA from clones that showed lack of expression of the intended target was amplified with PCR primers flanking the site of the target CRISPR crRNA. Primers for PCR and sequencing are shown in Supplementary Table S2. The sequencing data were analyzed using Vector NTI Express (Thermo Fisher) to confirm successful genomic editing.
Data are shown as mean and SEM of at least three independent experiments and two technical replicates. Groups were statistically compared using the Student t test or one-way and two-way ANOVA for multiple comparisons using Prism 7.03 (GraphPad Software). Asterisks on graphs denote a significant difference (*, P < 0.05), and ns for not significant (ns, P > 0.05). Spearman correlation and linear fit ± 95% confidence intervals was used for correlation of gene expression and bliss sums. The Bliss independence model was used to evaluate combinatorial activity, with positive integers indicating synergy (39). Bliss scores were calculated for each combination in the dose matrix and totaled to give a “Bliss sum” value.
Chronic or acute 5-Aza treatment synergizes with venetoclax in preclinical models of AML
5-Aza is a known DNA hypomethylation agent (40). To determine its hypomethylating potential in vitro, we treated AML cell lines with various doses of 5-Aza, replenished every 48 hours, over a total period of 7 days. No difference in methylation was observed within the first 24 hours of treatment of Kasumi-1 or SET-2 cells (Supplementary Fig. S1A). However, 5-Aza treatment for 7 days at the 300 nmol/L dose was able to decrease global DNA methylation in four AML cell lines tested (Fig. 1A). When these 5-Aza pretreated cells were then exposed to venetoclax for an additional 24 hours, OCI-AML5, Kasumi-1, and MV4-11 cell lines were found to be more sensitive to venetoclax compared with the DMSO pretreated cells (Fig. 1B). In contrast, the SHI-1 and SET-2 cell lines were not sensitized to venetoclax (Supplementary Fig. S1B). In humans, 5-Aza has an elimination half-life of approximately 4 hours, reaching maximum plasma concentrations of approximately 11 μmol/L, with intravenous dosing of 75 mg/m2 (38). To capture these acute clinical parameters in vitro in the context of venetoclax cotreatment, we measured the kinetics and number of cells with active caspase-3/-7 (aCasp-3/-7+) immediately following the addition of the two agents to OCI-AML5 and MV4-11. The combination of 5-Aza and venetoclax resulted in a rapid induction in the number of aCasp-3/-7+ cells that was greater than either agent alone (Fig. 1C). To further explore this observation, we evaluated the acute combinatorial activity of venetoclax and 5-Aza in a panel of 14 AML cell lines treated for 24 hours at various doses of either agent. In this setting, 5-Aza significantly sensitized the AML cell lines to venetoclax-induced cell death (Fig. 1D; Supplementary Fig. S1C) and showed synergistic activity in 12 out of 14 AML cell lines, as measured by the Bliss independence model (ref. 39; Fig. 1E). To further validate this observation, we generated cell line–derived mouse xenografts of SKM-1 and MV4-11 and treated the mice with either 5-Aza (once per week for 3 weeks), venetoclax (daily for 14 days), or the combination of 5-Aza and venetoclax (Fig. 1F). In both xenograft models, the tumor growth inhibition caused by the combination of the two agents was increased as compared with the either agent alone, consistent with the reduction of cell survival in vitro.
5-Aza upregulates NOXA and PUMA in AML
To identify a molecular mechanism of 5-Aza in promoting apoptosis in combination with venetoclax, we characterized the protein expression levels of both proapoptotic and antiapoptotic BCL-2 family members by Western blot following 5-Aza treatment for 24 hours. Through its previously reported target-binding activity (7), 5-Aza completely depleted the DNMT1 protein. This was associated with a concomitant dose- and time-dependent increase in NOXA and PUMA protein expression (Fig. 2A; Supplementary Fig. S2). To examine whether the upregulation of PUMA and NOXA was a cause or consequence of cell death, we pretreated the cells with the pan-caspase inhibitor Z-VAD-fmk to abrogate the induction of apoptosis. In the presence of Z-VAD-fmk and 5-Aza, a robust upregulation of NOXA and PUMA was detected (Fig. 2B). Further, qPCR indicated that 5-Aza was able to significantly increase both PMAIP1 (NOXA) and BBC3 (PUMA) transcripts in a dose-dependent manner as early as 6 hours after treatment, with longer treatment periods of up to 24 hours being associated with a further increase in BBC3 expression in the OCI-AML5, Kasumi-1, and MV4-11 cell lines (Fig. 2C). Interestingly, synergy between venetoclax and 5-Aza significantly correlated with the 5-Aza induction of PMAIP1 and BBC3 transcripts in a broader AML cell line panel, and was independent of TP53 status (Fig. 2D; refs. 41, 42). Similarly, primary AML patient cells treated with increasing doses of 5-Aza resulted in an upregulation of BBC3 (3/4 patient samples) and PMAIP1 (4/4 patient samples) after 6 hours, with elevated BBC3 expression maintained at 24 hours after treatment in 2 of 4 patient specimens (Fig. 2E). Chronic 5-Aza treatment did not have any significant effect on the gene expression or methylation, as the promoter regions for both PMAIP1 and BBC3 genes were unmethylated (Supplementary Fig. S3A) and their transcript levels were unaltered (Supplementary Fig. S3B) in a sample of AML cell lines. The rapid induction of these two transcripts at 6 hours, coupled with the low baseline methylation of either PMAIP1 or BBC3 and absence of methylation changes following 5-Aza treatment, collectively indicated a nonepigenetic mechanism for transcriptional induction.
5-Aza primes the antiapoptotic proteins with NOXA and PUMA
To determine the functional consequence of NOXA and PUMA upregulation, we measured the binding of these two proteins to the antiapoptotic proteins BCL-2, BCL-XL, and MCL-1 in AML cell lines after 5-Aza treatment, by immunoprecipitation. The amount of NOXA bound to BCL-2 in Kasumi-1 and MV4-11 cells, and NOXA bound to MCL-1 in Kasumi-1, OCI-AML5, and MV4-11 cells increased in a dose-dependent fashion following 5-Aza treatment (Fig. 3). Although to a lesser degree than MCL-1, PUMA binding to BCL-2 and BCL-XL was also increased. These results collectively indicate that 5-Aza enhances the amount of BH3-only proteins found in complex with the antiapoptotic BCL-2 family members.
The ISR induces NOXA following 5-Aza acute treatment
5-Aza has been shown to also inhibit protein synthesis through RNA incorporation (11). The ISR pathway is induced by various stress signals, including proteotoxic stress. During conditions of severe stress, the ISR's main effector, activating transcription factor 4 (ATF4), can tip the cell toward death through transcriptional induction of proapoptotic targets (43). As the transcriptional upregulation of both BBC3 and PMAIP1 appeared to occur in a p53-independent manner (Fig. 2D), we postulated that the transcription factor ATF4 may play a role in the induction of these transcripts. ATF4 has been reported to bind to the PMAIP1 promoter to induce its transcription (44, 45). The protein levels of ATF4 increased significantly when the AML cell lines were exposed to 5-Aza for 24 hours (Fig. 4A). This was indicative of ISR pathway activation and was confirmed by upstream phosphorylation of serine 51 (S51) on the alpha subunit of the eukaryotic Initiation Factor 2 (eIF2α; Fig. 4A). In addition, we observed significant induction of a transcriptional target of ATF4, CCAAT-enhancer–binding protein homologous protein (CHOP). In agreement with the upregulation of CHOP at the protein level (Fig. 4A, Supplementary Fig. S4A), its transcript, DDIT3, was also induced in a dose-dependent manner by 5-Aza after 6-hour treatment of AML cell lines (Fig. 4B) and primary AML patient samples ex vivo (Fig. 4C), although the level of DDIT3 induction was reduced after 24-hour 5-Aza treatment (Fig. 4C). The level of DDIT3 induction also correlated significantly with the synergy observed between venetoclax and 5-Aza combination in 14 AML cell lines (Fig. 4D). Finally, to explore whether the ISR pathway may be responsible for the induction of PMAIP1 and BBC3, we used ISRIB, a small molecule that acts as a potent inhibitor of this pathway by reversing the effect of eIF2α phosphorylation (46). OCI-AML5 cells were treated for 6 hours with ISRIB, 5-Aza, or ISRIB and 5-Aza in combination. As expected, 5-Aza induced the PMAIP1, BBC3, and DDIT3 transcripts; however, in the presence of ISRIB, the 5-Aza–mediated induction of all three transcripts was significantly suppressed (Fig. 4E, Supplementary Fig. S4B). ISRIB partially rescued AML cells from 5-Aza–induced death (Supplementary Fig. S4C). ISRIB also induced significant resistance to the venetoclax+5-Aza combinatorial cell killing (Fig. 4F), enhancing the venetoclax AUC (Fig. 4G) and reducing the synergy between these two agents (Supplementary Fig. S4D).
Deletion of PMAIP1 abrogates 5-Aza and venetoclax combinatorial activity
The 5-Aza–mediated induction of NOXA and PUMA, as well as their subsequent binding to the antiapoptotic BCL-2 family members, suggested a role in priming the AML cells for apoptosis. To understand the importance of NOXA and PUMA in this context, CRISPR-Cas9 technology was employed to delete PMAIP1 (Fig. 5A) and BBC3, respectively (Supplementary Fig. S5A). Following deletion of the PMAIP1 gene, no consistent changes were observed in the expression of other BCL-2 family members across cell lines (Fig. 5A). The OCI-AML5, Kasumi-1, and MV4-11 PMAIP1−/− cell lines were each more resistant to venetoclax-induced cell death compared with their respective parental cell lines. The Kasumi-1 and MV4-11 PMAIP1−/− cell lines were also more resistant to 5-Aza–induced cell death (Fig. 5B). In agreement with the cell viability data, the kinetics of induction and number of aCasp-3/-7+ cells in PMAIP1-deficient cell lines were substantially reduced in response to either venetoclax or 5-Aza treatment (Fig. 5C). Because venetoclax and 5-Aza induced broad synergistic cell killing in AML cell lines, we assessed the impact of PMAIP1 deletion on this combination. PMAIP1 deletion reduced the magnitude of cell death resulting from combined venetoclax/5-Aza treatment in MV4-11, OCI-AML5, and Kasumi-1 cells (Fig. 6A), which was associated with a loss of venetoclax potency (EC50) in the presence of increasing concentrations of 5-Aza (Fig. 6B) and the abrogation of synergy (Fig. 6C).
Lastly, we hypothesized that PUMA might also play a role in priming the cells for apoptosis. The gene encoding PUMA, BBC3, was deleted in the parental and the PMAIP1−/− cell lines, OCI-AML5 and MV4-11. The crRNA was designed to target both BH3-containing isoforms of PUMA, alpha and beta. Western blot analysis indicated complete absence of PUMA at the protein level (Supplementary Fig. S5A). Unlike the PMAIP1−/− cell lines, the BBC3−/− OCI-AML-3 and MV4-11 cell lines responded similarly to venetoclax and/or 5-Aza as the parental cell lines (Supplementary Fig. S5B and S5C). In addition, BBC3 deletion in the PMAIP1−/− AML cell lines did not further affect the response of PMAIP1−/− cells to the venetoclax/5-Aza combination as measured by cell viability (Supplementary Fig. S5B). Taken together, our findings show that, although both NOXA and PUMA are induced by 5-Aza treatment of AML cell lines, only NOXA is critical in sensitizing the AML cells to 5-Aza/venetoclax-induced apoptosis.
Venetoclax has been approved by the FDA for use in combination with 5-Aza in the treatment of elderly patients with AML (aged ≥ 75 years) who are ineligible for induction therapy based on the significant improvement in ORRs (34). Herein, we utilize preclinical models of AML to provide mechanistic insights into the clinical activity observed between 5-Aza and venetoclax. At physiologically relevant concentrations (38), we demonstrate that 5-Aza combines with venetoclax in AML cell lines via nonepigenetic mechanisms requiring the induction of NOXA via the ISR pathway.
Treatment of AML cell lines with the combination of venetoclax and 5-Aza induced the rapid induction of apoptosis that was greater than either 5-Aza or venetoclax as single agents, both in terms of rate and magnitude of response. This corresponded with broad synergistic cell killing observed across a panel of AML cell lines that was also reflected in vivo, where the cotreatment of MV4-11 and SKM-1 xenograft models with 5-Aza and venetoclax inhibited tumor growth superior to either agent alone. The rapid kinetics (<24 hours) of apoptosis associated with the combinatorial activity between venetoclax and 5-Aza indicate that although DNMT1 expression is eliminated, the resulting cell death was independent of changes in DNA methylation, which required a week of continuous treatment. Similarly, acute cell death was previously reported in cells of patients with AML treated ex vivo for 24 hours with the venetoclax/5-Aza combination (47), aligning with the rapid blast clearance observed within 24 to 72 hours after treatment in patients with AML treated with this combination therapy (48).
In the acute treatment setting, 5-Aza did not affect the expression of antiapoptotic proteins, but it significantly induced the expression of the proapoptotic BH3-only proteins PUMA and NOXA, independent of their respective gene methylation status. Several mechanisms beyond hypomethylation have been proposed to be responsible for 5-Aza's activity (12–14), including DNA damage (11). However, although PMAIP1 and BBC3 are both p53 response genes (21, 22, 49, 50), they were both induced independent of TP53 status following 5-Aza treatment, and were not associated with changes in p53 phosphorylation status (data not shown). We subsequently hypothesized that the induction of these BH3-only proteins was not a result of DNA incorporation by 5-Aza (7, 40), but rather a consequence of RNA incorporation (8–10) and inhibition of protein synthesis (11), which may result in activation of ISR. The regulatory region of PMAIP1 contains the binding sites for over 40 different transcription factors and coactivators including those mediated by DNA damage, hypoxia, epigenetic regulation, metabolic stress, proteasome inhibition, autophagy, and ISR. One of these transcription factors and ISR component, ATF4, induces PMAIP1 transcription and enhances the expression of NOXA protein in a p53-independent manner, following treatment of cells with proteotoxic agents (44, 45). Exposure of AML cell lines to 5-Aza resulted in the activation of the ISR pathway, as evident by CHOP and ATF4 induction coupled with phosphorylation of eIF2α (S51). Subsequent pharmacologic inhibition of ISR prevented the 5-Aza–mediated induction in PMAIP1, BBC3, and DDIT3. Importantly, the degree of PMAIP1, BBC3, or DDIT3 expression induced following 5-Aza treatment correlated with the degree of synergy observed between venetoclax and 5-Aza in AML cell lines in vitro. A recent study identified that the venetoclax and 5-Aza combination killed leukemic stem cells from patients with AML by decreasing amino acid uptake and mitochondrial respiration (47). Interestingly, amino acid deprivation activates the ISR through general control nonderepressible-2 kinase signaling (43), offering a potential mechanism for the ISR activation observed in our study.
The capacity of the antiapoptotic BCL-2 family members to sequester or buffer elevations in their proapoptotic counterparts through direct protein–protein interactions functions to thwart the execution of apoptosis (16). In this context, MCL-1 functions as a resistance factor to venetoclax, operating in part as a sink to sequester BH3-only proteins released from BCL-2 upon venetoclax binding (51). 5-Aza enhanced the association of NOXA with MCL-1 in a dose-dependent fashion, potentially neutralizing the antiapoptotic capacity of this protein similar to one of several mechanisms ascribed to the activity of bortezomib in combination with venetoclax in multiple myeloma (52). Although full-length NOXA binds MCL-1 with a Kd of 3.4 nmol/L, it also possesses a dissociation constant of 250 nmol/L for BCL-2 (53). Reflecting these data, 5-Aza treatment also consistently enhanced the interactions between NOXA and BCL-2 in AML cell lines, providing an additional mechanism by which this hypomethylating agent may potentially prime AML cells for apoptosis induction by venetoclax. 5-Aza–mediated PMAIP1 induction correlated with venetoclax/5-Aza synergy in AML cell lines in vitro; the failure to induce PMAIP1 associated with the poor combinatorial activity observed between these two agents in THP-1 and SET-2 cells.
To further understand the impact of NOXA or PUMA on the combinatorial activity between venetoclax and 5-Aza, we utilized gene editing technology to delete PMAIP1 and/or BBC3, respectively, in OCI-AML-5, Kasumi-1, and MV4-11 cells. PMAIP1 deletion, but not BBC3 deletion, significantly inhibited the synergy between venetoclax and 5-Aza, additionally reducing the kinetics and magnitude of caspase-3/-7 activation when compared with the parental cell lines. BBC3 deletion did not inhibit apoptosis induced by either venetoclax or 5-Aza and did not further enhance the resistance to venetoclax activity mediated by PMAIP1 deletion. Although these data contrast with DNA damage–induced apoptosis, where PUMA drives most of the apoptosis with partial contributions from NOXA (54), our observations that PMAIP deletion restricts apoptosis induced by venetoclax single-agent treatment align with recent studies demonstrating that elevated PMAIP1 expression is responsible for enhanced sensitivity to venetoclax in preclinical models of diffuse large B-cell lymphoma (55) and neuroblastoma (56). What is more, loss of PMAIP1 inhibited 5-Aza–induced cell death in two of three AML cell lines assessed, complementing data that demonstrate NOXA peptides can discriminate the clinical responses of patients with AML, MDS, and MDS/PMN treated with 5-Aza (13).
Collectively, these data indicate that NOXA expression is a key determinant of venetoclax activity, both as a monotherapy and in the combination setting with 5-Aza. In addition, this work demonstrates that, at pharmacologically relevant concentrations, 5-Aza induces NOXA through the ISR pathway to sensitize AML cells to venetoclax in preclinical models of this malignancy. These data provide a rationale behind the ongoing randomized phase III clinical trial evaluating the activity of venetoclax in combination with 5-Aza in treatment-naïve subjects with AML who are ineligible for induction therapy (M15-656, NCT02993523). Consequently, understanding the association of NOXA and CHOP expression in patients with AML in relation to venetoclax clinical activity is of significant interest.
Disclosure of Potential Conflicts of Interest
S. Jin is an employee of AbbVie. D. Cojocari is an employee of AbbVie. J.J. Purkal is an employee of AbbVie. R. Popovic is an employee of AbbVie. N.N. Talaty is an employee of and holds ownership interest (including patents) in AbbVie. Y. Xiao is an employee of AbbVie. E.R. Boghaert is an employee of and holds ownership interest (including patents) in AbbVie. J.D. Leverson is an employee of and holds ownership interest (including patents) in AbbVie. D.C. Phillips is an employee of, and holds ownership interest in AbbVie, and is listed as a coinventor on a patent regarding a diagnostic test to predict sensitivity of non-Hodgkin's lymphoma (NHL) patients to BCL-2 Antagonists and ABT-199 (Venetoclax) that is owned by AbbVie. No potential conflicts of interest were disclosed by the other authors.
Conception and design: S. Jin, D. Cojocari, R. Popovic, N.N. Talaty, L.R. Solomon, E.R. Boghaert, J.D. Leverson, D.C. Phillips
Development of methodology: S. Jin, D. Cojocari, J.J. Purkal, R. Popovic, N.N. Talaty, Y. Xiao, L.R. Solomon, D.C. Phillips
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Jin, D. Cojocari, J.J. Purkal, N.N. Talaty, Y. Xiao, E.R. Boghaert
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Jin, D. Cojocari, J.J. Purkal, R. Popovic, N.N. Talaty, Y. Xiao, L.R. Solomon, E.R. Boghaert, J.D. Leverson, D.C. Phillips
Writing, review, and/or revision of the manuscript: S. Jin, D. Cojocari, J.J. Purkal, R. Popovic, N.N. Talaty, Y. Xiao, E.R. Boghaert, J.D. Leverson, D.C. Phillips
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Jin, D. Cojocari, N.N. Talaty, D.C. Phillips
Study supervision: E.R. Boghaert, D.C. Phillips
The design, study conduct, and financial support for this research were provided by AbbVie Inc. AbbVie Inc. participated in the interpretation of data, review, and approval of this publication.
We would like to acknowledge Loren M. Lasko from AbbVie Oncology Discovery for his technical expertise in single-cell sorting of the knockout cells.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.