Targeting Chemotherapy to Decondensed H3K27me3-Marked Chromatin of AML Cells Enhances Leukemia Suppression

Acute myelogenous leukemia (AML) arises from transformation of early hematopoietic progenitor cells (HPC), leading to clonal expansion and differentiation arrest (1). AML is genetically heterogeneous (2–5), with distinct subclones coexisting in the same patient and creating obstacles for the development of effective molecularly targeted therapies (6–8). Thus, the frontline therapy for most patients with AML remains non-targeted intensive induction chemotherapy. This regimen is generally offered only to “fit” patients (young adult and/or individuals of high-performance status), and therefore effective therapies for elderly or unfit patients are needed.

Most conventional chemotherapeutic agents induce DNA damage and cell death by targeting tumor cells during DNA replication (9–11). However, chromatin structure influences the efficacy of DNA-damaging agents. For example, compacted chromatin is more resistant to formation of double-strand breaks (DSB; refs. 12, 13), and histone deacetylase (HDAC) inhibitors, which cause chromatin relaxation, increase sensitivity of cancer cells to ionizing radiation (14). Moreover, chemical profiling of DNA damage sites induced by cytotoxic drugs suggests that DNA damage induced by topoisomerase II inhibitors daunomycin and etoposide is not random, but preferentially targets H3K36me3- and H3K9me1-marked transcriptionally active, de-condensed regions of the genome (15, 16). By contrast, H3K27me3-marked, repressed genomic regions are relatively inaccessible to these cytotoxic agents (15, 16). In euchromatin, H3K27me3-marked nucleosome arrays represent the most compact chromatin in the genome (17, 18), suggesting that decondensing its structure might enhance the efficiency of cytotoxic agents.

Using the chromatin assembly assay (CAA) to examine the loading of chromatin proteins on nascent DNA at single-cell resolution (19–22), we discovered that, upon induction of differentiation, embryonic stem cells (ESC) and CD34⁺ HPCs exhibit a significant period when post-replicative chromatin is globally devoid of the repressive histone mark H3K27me3 (21, 22). This transient phase of “open” nascent chromatin allows the recruitment of lineage-determining transcription factors (TF) onto nascent DNA to promote differentiation (21, 22).

In the present study, we used the CAA to examine the recruitment of H3K27me3 EZH2 histone methyltransferase (HMT, a component of the polycomb repressive complex 2) and the deposition of H3K27me3 to nascent DNA of CD34⁺ AML blasts. EZH2 and H3K27me3 were readily detected on nascent DNA of AML cell lines and blasts from newly diagnosed patients. These findings provided the rationale for combining EZH2 inhibitors and cytotoxic drugs with the goal of enhancing DNA damage and apoptosis of AML blasts. We show here that “opening” the H3K27me3-marked chromatin of AML cells by EZH2 inhibition renders DNA more vulnerable to DNA-damaging agents, leading to enhanced cell death ex vivo and leukemia suppression in mice. Such effects were further improved by enhancing the targeting of nascent chromatin through enrichment of S-phase AML cells. In the p53-null KG-1 and THP-1 cell lines, treatment with EZH2 inhibitor/Doxorubicin induced a transcriptional reprogramming that was dependent, in part, on derepression of H3K27me3-marked promoters and led to increased expression of cell death and growth-inhibitory genes.

The THP-1 line was purchased from the ATCC. The KG-1 and KG-1a lines were provided by Dr. Kathrin M. Bernt (Children's Hospital of Philadelphia, Philadelphia, PA). Cells were cultured in Iscove's Modified Dulbecco's Medium (Gibco) supplemented with 10% heat-inactivated FBS (Biowest), 1% penicillin–streptomycin and 1% l-glutamine (Thermo Fisher Scientific) at 37°C, 5% CO₂. Cell lines were tested for Mycoplasma every 3 months as described previously (23). The lines carry a deletion or single nucleotide mutation in the p53 gene causing premature termination of p53 mRNA translation potentially generating truncated proteins (p53–176 and p53–226 in the THP-1 and KG-1 lines, respectively; ref. 24). However, these cell lines can be considered p53-null because they do not express detectable levels of wild-type or truncated p53 proteins.

Primary AML samples were obtained after written informed consent in accordance with the Declaration of Helsinki and the Thomas Jefferson University Institutional Review Board–approved protocol #17D.803 (PI-Palmisiano).

Molecular characteristics and clinical outcomes of the samples are described in Supplementary Table SI. Primary cells were maintained in SFEM supplemented with StemSpan CC100 and recombinant thrombopoietin (TPO), all from Stem Cell Technologies.

Doxorubicin, AraC, and bleomycin were obtained from Cayman Chemical and resuspended in sterile water. Etoposide was purchased from Sigma-Aldrich and resuspended in DMSO.

For ex vivo studies, GSK126 and UNC1999 were purchased from Cayman Chemical; EPZ-6438 was purchased from Selleckchem and palbociclib was obtained from LC Laboratories. They were resuspended in DMSO.

A total of 1.5–2.0 × 10⁴ cells were spun onto slides using a cytocentrifuge, fixed for 10 minutes in 3.7% paraformaldehyde at room temperature and washed with PBS. Cells were permeabilized with 0.25% Triton-X/PBS and blocked with the #WESTBL-RO solution (Roche) solution for one hour. Then, cells were incubated with primary antibodies (γH2AX, 1:200 Millipore #056361; RAD51, 1:100, Santa Cruz Biotechnology #8349) overnight at 4°C. Slides were washed with 1X PBS and incubated 1 hour at room temperature with secondary antibodies (Alexa Fluor 488 goat anti-mouse or Alexa Fluor 555 goat anti-rabbit (Thermo Fisher Scientific). Samples were mounted using DAPI Fluoromount-G (Southern Biotech; #0100–20). Image acquisition was performed using a confocal microscope Nikon C2 with lasers set on λ = 405–488–562 nm. Images were analyzed using Fiji ImageJ 2.0 (Cambridge Astronomical Survey Unit) and Adobe Photoshop 2020 (Adobe Systems) software.

Neutral comet assays were performed on AML cell lines and primary samples as described previously (25). Images were captured on a Nikon Eclipse Ni (Nikon) with a × 10 objective and analyzed using CometScore software. All samples were blinded and 80–120 cells per sample were scored.

Cells were incubated at 100,000/mL with 1μL of the CellEvent Caspase 3/7 Green Detection Reagent (#C10427; Thermo Fisher Scientific) for 15 minutes at 37°C and analyzed by flow cytometry using a 488-nm excitation laser. Apoptosis was also assessed by Annexin V staining. For this assay, 100,000 cells were spun down and resuspended in 50 μL of Annexin V Binding buffer (10 mmol/L HEPES, 140 mmol/L NaCl, and 2.5 mmol/L CaCl₂, pH 7.4), containing 1 μL of Cy 5.5-Annexin V (#559935; BD Pharmigen) for 15 minutes at room temperature. Samples were analyzed by flow cytometry using a 640-nm laser.

For cell-cycle analysis, cells were incubated for 5 minutes with sodium citrate 0.1%, Triton-X 0.1%, 50 μg/mL propidium iodide. DNA content was subsequently analyzed using a BD FACS Celesta flow cytometer using a 561-nm laser.

CAA experiments were performed as described previously (21) with minor modifications. AML cells were pulse-labeled with 5 μmol/L EdU for 15 minutes. Approximately 20,000 cells per each condition were harvested and cytospun (750 rpm for 5 minutes) onto polylysine-coated slides (Polyscience), fixed with 4% formaldehyde, permeabilized in 0.25% Triton-X, and subjected to Click-iT reaction (Invitrogen) to conjugate biotin to EdU. Cells were incubated with mouse antibody to biotin and rabbit antibodies to the protein of interest (H3K27me3: Millipore #07–449; EZH2: Cell Signaling Technology # 5246; H2AK119ub: Cell Signaling Technology #8240S; RING1b: Cell Signaling Technology #5694). Proximity ligation assay (PLA) reactions were performed according to the manufacturer (Olink), as described previously (21). Following PLA, EdU was detected by immunostaining with Alexa Fluor 488–conjugated anti-biotin mouse antibody.

RNA-seq was performed on three independent biological replicates per condition. Samples from different conditions were processed together to avoid differences in batch preparation. For each sample, 2 × 10⁶ cells were used to extract total RNA and generate RNA-seq libraries. KG-1 cells were plated at 1 × 10⁶ cells/mL in complete Iscove's Modified Dulbecco's Medium (Gibco) treated with doxorubicin (0.3 μmol/L; 24 hours), GSK126 (5 μmol/L; 12 hours pretreatment plus 24 hours treatment), or the GSK126/doxorubicin combination (5 μmol/L GSK126, 12 hours pretreatment plus 24 hours treatment; 0.3 μmol/L doxorubicin, 24 hours). Total RNA was isolated with the RNeasy Plus Mini Kit (#74134, Qiagen) and used to prepare libraries using the NEBNext Ultra II RNA Library Prep Kit for Illumina (New England Biolabs Inc.), following the manufacturer's protocol. Libraries were sequenced on a NextSeq 500 instrument using 75-bp single-end chemistry. Raw FASTQ sequencing reads were mapped against the reference human genome Ensembl Version GRCh38 using additional information from the gene transfer format (.gtf) annotation from GENCODE version GRCH29 using RSEM. Total read counts and normalized transcripts per million (TPM) were obtained using RSEM's calculate-expression function. Data were deposited to SRA database (BioProjectID: PRJNA693321). Before determining differential expression, batch effects and sample heterogeneity were tested using iSeqQC (26). Differential gene expression was tested using the DESeq2 package in R/Bioconductor. Genes were considered differentially expressed (DE) if they had adjusted P ≤ 0.05 and absolute fold change ≥2. Plots were generated using R/Bioconductor. Canonical pathways modulated by drug treatments were identified by DAVID (27).

Samples from different conditions were processed together to prevent batch effects. For each condition, 3 × 10⁶ KG-1 cells were cross-linked with 1% formaldehyde for 5 minutes at room temperature, washed once in 1.25 mol/L glycine/PBS and twice with 1X cold PBS. The pellet was resuspended in chromatin immunoprecipitation (ChIP) lysis buffer following a published protocol (28). Barcoded libraries were made with the NEB ULTRA II DNA Library Prep Kit for Illumina, and sequenced on Illumina NextSeq 500, producing 75bp SE reads. ChIP-sequencing (ChIP-seq) analyses were performed as described previously in the Supplementary Methods. Data were deposited to SRA database (BioProjectID: PRJNA693).

For leukemogenesis assays, 2 × 10⁶ THP-1 Luc cells or 1.5 × 10⁶ cells from AML sample #40 were injected intravenously in 7- to 9-week-old NRG NOD-Rag1^null IL2rg^null or NRG-SGM3 mice (The Jackson Laboratory).

GSK126 (Chemgood) was dissolved in 20% Captisol adjusted to pH 4.5 with 1 N acetic acid and administered intraperitoneally every 24 hours for 7 days.

Mice were treated for five consecutive days with 5+3 chemotherapy consisting of: 3 days of intravenous coinjection of AraC (50 mg/kg; Sigma-Aldrich) and doxorubicin (1.5 mg/kg; Sigma-Aldrich), followed by intraperitoneal injection of AraC alone on days 4 and 5.

Palbociclib isethionate was purchased from LC Laboratories and mixed in the chow by Research Diets Inc. at 800 mg/kg. This dosing was based on the average daily food intake of NRG mice to deliver 150 mg/kg/d of palbociclib given at libitum for the duration of the experiment. Leukemia burden was monitored by imaging of mice injected with THP-1-luciferase cells or by flow cytometry of peripheral blood human CD45⁺ cells in mice injected with primary AML cells. Animal experiments were approved by the Institutional Animal Care and Use Committee of Thomas Jefferson University under protocol number 00012.

Data, expressed as mean ± SD of three experiments, were analyzed for statistical significance by unpaired, two-tailed Student t test or one-way ANOVA. A P value of <0.05 was considered statistically significant. Kaplan–Meier plots for mouse survival experiments were generated using the GraphPad Prism 8.0 software. Statistical significance of differences in survival was assessed by the log-rank test.

Detailed protocols for these experimental procedures are described in Supplementary Materials and Methods and Supplementary Table SII.

We used our previously developed CAA (21) to examine the accumulation of H3K27me3 on nascent DNA of single AML blasts. In this assay, DNA is labeled in living cells in a pulse-chase manner with EdU that is then chemically conjugated with biotin. The proximity of a protein to nascent (EdU positive) DNA is then examined by the PLA using antibodies to biotin and to the protein of interest. We observed that H3K27me3 accumulates rapidly on nascent DNA of AML cell lines THP-1, KG-1 (Fig. 1A and B), KG-1a (Supplementary Fig. S1A) and of CD34⁺ blasts of several AML primary samples with recurrent molecular abnormalities (e.g., FLT3-ITD/TKD, NPM1, DNMT3A, and NRAS mutations; Fig. 1C and D; Supplementary Table SI). These mutations have no direct effect on the expression/activity of H3K27 HMTs and demethylases. Instead, in CD34⁺ cells of AML sample #8, which carries a frameshift mutation of ASXL1 (G646fs*12) in approximately 40% of the cells causing decreased H3K27me3 levels (29), accumulation of H3K27me3 on nascent DNA was heterogeneous with several EdU-positive cells showing weak H3K27me3 loading (Fig. 1D).

Because the H3K27me3 mark is associated with the most condensed nucleosome arrays in the genome (17, 18, 30), the fast accumulation of H3K27me3 on nascent DNA of AML blasts indicates that these cells possess a “condensed” structure of nucleosomes at H3K27me3-marked genomic regions following DNA replication.

The level of H3K27me3 depends on the activity of EZH2 and, to a lesser extent, of EZH1 (31). We found that EZH2 is associated with nascent DNA of THP-1, KG-1, and KG-1a cell lines (Supplementary Fig. S2A) and of CD34⁺ blasts from several primary AML samples (Supplementary Fig. S2B), in line with high H3K27me3 levels on their nascent DNA.

Inhibition of EZH2 activity by GSK126 or EPZ-6438 treatment (32) of THP-1 or of primary cells of sample AML#40 markedly decreased H3K27me3 loading on nascent DNA (Supplementary Fig. S3A and S3B).

We assessed whether inhibiting EZH2 led to genome-wide changes in chromatin accessibility of THP-1 and KG-1 cells. Using the ATAC-seq assay, we identified an average of 11,823 and 16,198 new accessible sites in GSK126-treated THP-1 and KG-1 cells respectively, compared with the untreated counterparts (Supplementary Fig. S3C and S3D), consistent with enhanced chromatin accessibility. The vast majority of the newly accessible genomic sites correspond to regulatory regions, including promoters and intronic/intergenic enhancers.

A component of the polycomb repressor complex 1 (PRC1) the H2A119 ubiquitin ligase RING1B, and H2A119ub are also rapidly associated with EdU-labeled DNA in AML cell lines KG-1 and THP-1, and H2A119ub stably associates hereafter (Supplementary Fig. S4). These findings suggest that the compaction of H3K27me3-marked chromatin of AML cells is further ensured by rapidly establishing and maintaining a functional PRC1 on their nascent DNA.

Having established that inhibiting EZH2 decondenses “nascent” H3K27me3-marked chromatin in AML blasts, we investigated whether this effect increases their sensitivity to DNA-damaging agents. DNA damage was examined by counting γH2AX foci (33) and DNA breaks using the comet assay (25). Compared with untreated cells, treatment of THP-1 cells with doxorubicin caused a marked increase in the number of γH2AX foci and DNA breaks, and these numbers were further enhanced by a 16-hour pretreatment with GSK126 to de-condense H3K27me3-marked chromatin followed by GSK126/doxorubicin cotreatment (Fig. 2A). Enhanced DNA damage was also observed in EZH2-silenced, doxorubicin-treated THP-1 cells compared with cells treated with doxorubicin only (Supplementary Fig. S5A and S5B).

Higher levels of DNA damage were also detected in GSK126/doxorubicin-treated AML cell lines KG-1 (Fig. 2B) and KG-1a (Supplementary Fig. S1B), and in CD34⁺ cells of primary AML patient samples (Fig. 2C–E), as well as in THP-1 and primary AML samples cotreated with the dual EZH1/2 inhibitor UNC1999 (34) and doxorubicin (Supplementary Fig. S6A–S6D).

Notably, GSK126 treatment did not increase doxorubicin-induced DNA damage in AML sample #8 (Supplementary Fig. S7). This sample exhibits low-levels of H3K27me3-marked nascent chromatin in many cells due to the ASXL1 mutation in 40% of the cells (Fig. 1D). AML sample #8 also has a loss-of-function mutation of the STAG2 gene in >90% of the cells (Supplementary Table SI). AML cells with mutant STAG2 exhibit high sensitivity to DNA-damaging agents, including doxorubicin (35). Thus, the combination of low-H3K27me3 levels associated with the ASXL1 mutation and the high sensitivity to DNA-damaging agents associated with STAG2 mutations might explain why EZH2 inhibition did not enhance further doxorubicin-induced DNA damage in AML sample #8. Significantly higher levels of DNA damage were also observed in THP-1 cells cotreated with GSK126 and bleomycin (a drug that causes single-strand and DSBs directly) or GSK126 and Etoposide (Supplementary Fig. S8).

The homologous recombination (HR) DNA repair pathway is frequently altered in AML cells (36). Thus, we asked whether the increase in DNA damage in GSK126/doxorubicin-treated AML cells was due to defective DNA damage repair. To this end, we analyzed the number of RAD51 foci (indicative of HR-DSB repair) and observed that they were not diminished by GSK126 treatment following doxorubicin-induced DNA damage in THP-1 cells (Supplementary Fig. S9A). Likewise, GSK126 treatment of THP-1 cells did not diminish the number of 53BP1 foci (indicative of NHEJ repair) compared with treatment with doxorubicin only (Supplementary Fig. S9B).

Together, these findings strongly suggest that, at least in THP-1 cells, defective DSB repair does not play an important role in the increase in doxorubicin-induced DNA damage.

To determine whether the increased DNA damage in GSK126/doxorubicin-treated AML cells results in enhanced cell death, we assessed apoptosis by flow cytometry analysis of caspase-3/7 activation and by Annexin V staining. Consistent with the increase in DNA damage, apoptosis was markedly enhanced in GSK126/doxorubicin-treated AML cell lines and CD34⁺ blasts from several patients with AML, compared with treatment with doxorubicin only (Fig. 3A; Supplementary Fig. S10). Enhanced apoptosis was also observed in the THP-1 and KG-1 lines cotreated with the selective EZH2 inhibitor EPZ-6438 and doxorubicin and in primary AML samples cotreated with the dual EZH1/2 inhibitor UNC1999 and doxorubicin, compared with treatment with doxorubicin alone (Supplementary Fig. S11).

Apoptosis induced by the doxorubicin 0.1μmol/L/GSK126 5μmol/L cotreatment was similar to that induced by a 3-fold higher concentration of doxorubicin alone in THP-1 and AML #40 cells. In KG-1 cells, the doxorubicin 0.3 μmol/L/GSK126 10 μmol/L combination was markedly more effective than doxorubicin alone at 0.3 or 0.5 μmol/L (Fig. 3A). Compared with single drug treatment, the GSK126/doxorubicin combination increased synergistically the number of apoptotic THP-1 and KG-1 cells at multiple doses as indicated by Chou–Talalay combination indices (Fig. 3B).

Blast cells of AML sample #40 are exclusively CD34⁺ and a significant portion of these cells consists of the more primitive CD34+CD38low subset, providing the opportunity to test whether this stem cell-enriched subset is also vulnerable to cotreatment with GSK126/doxorubicin. Both subsets exhibited a similar amount of H3K27me3-marked nascent chromatin and a comparable increase in apoptosis upon GSK126/doxorubicin cotreatment (Fig. 3C). Enhanced apoptosis susceptibility by GSK126/doxorubicin co-treatment was also observed in the CD34+CD38high and the CD34+CD38low subsets of two additional AML samples (Supplementary Fig. S12). Notice that AML sample #15 has a homozygous p53-null mutation in approximately 90% of the cells (Supplementary Table SI). Enhanced apoptosis was also observed in EZH2-silenced, doxorubicin-treated THP-1 cells compared with cells treated with doxorubicin only (Supplementary Fig. S5C).

We also examined the effect of the GSK126/doxorubicin combination on colony formation of AML cells. Single treatments with doxorubicin or GSK126 caused moderate suppression of AML colonies; however, the GSK126/doxorubicin cotreatment almost completely eradicated clonogenic AML cells in the tested samples (Supplementary Fig. S13), suggesting that this cell subset may be particularly vulnerable to enhanced DNA damage promoted by chromatin decondensation. Notably, the GSK126/doxorubicin combination was at least as effective as single doxorubicin treatments used at 2–3 fold higher concentrations.

Doxorubicin induces DNA damage during DNA replication by inhibiting DNA topoisomerase II and by preventing religation of DNA breaks (10). Consistent with this mode of action, inhibiting S-phase entry by treating THP-1 cells with the CDK4/6 inhibitor palbociclib (Supplementary Fig. S14A) completely suppressed doxorubicin-induced DNA damage and apoptosis (Supplementary Fig. S14B and S14C). On the basis of these findings, we hypothesized that our strategy of combining an EZH2 inhibitor with an S-phase-selective cytotoxic agent (doxorubicin) would further increase its efficiency when directed to replicating cells. Thus, THP-1 cells were first treated with palbociclib to induce G₁ arrest, and then allowed to re-enter the cell cycle after palbociclib washout. This treatment significantly increased the number of S-phase cells (∼54% at 8–10 hours after washout) compared with the unsynchronized counterpart (S-phase: ∼24%; Fig. 4A). Likewise, GSK126 treatment of S-phase-enriched THP-1 cells markedly increased their chromatin accessibility compared with GSK126-treated unsynchronized cells, as revealed by the detection of approximately 30,000 new ATAC-seq peaks in S-phase–enriched THP-1 cells (Fig. 4B). Consistent with these effects, GSK126/doxorubicin cotreatment of S-phase–enriched THP-1 cells strongly increased the number of DNA damage foci and of apoptotic cells, compared with GSK126/doxorubicin-treated unsynchronized THP-1 cells (Fig. 4C).

The increased DNA damage and apoptosis of GSK126/doxorubicin-treated AML cells is likely to correlate with changes in gene expression that may reflect the cumulative effects induced by doxorubicin and GSK126 treatment, a new transcription program modulated by the GSK126/doxorubicin combination, or both.

To investigate this, we performed RNA-seq analyses of untreated, doxorubicin- or GSK126-treated (24 hours) KG-1 cells, or of cells pretreated with GSK126 (16 hours) and then cotreated with GSK126/doxorubicin for 24 hours. The p53-null KG-1 line was chosen because the GSK126/doxorubicin treatment was highly effective in inducing DNA damage and apoptosis (detected after 24 and 48-hour treatment, respectively), compared with single treatments (Fig. 2B; Fig. 3A). Of note, no apoptosis is detected in cells treated with GSK126/doxorubicin for 24 hours. Moreover, AML with p53 mutations is associated with chemoresistance and disease relapse (37), suggesting that findings in the p53-null KG-1 line may be translationally significant.

Compared with untreated cells, the GSK126/doxorubicin co-treatment was much more effective in modulating gene expression than treatment with doxorubicin or GSK126 alone (1,593 genes up-/downregulated by GSK126/doxorubicin cotreatment versus 80 and 81 up-/downregulated by GSK126 or doxorubicin treatments alone; FDR <5%; Fig. 5A). Of the genes differentially expressed in GSK126/doxorubicin-treated versus untreated cells, 1,508 were specific for the GSK126/doxorubicin cotreatment. Using DAVID (26) to perform functional analysis of these genes, cytokine-mediated signaling (P = 4.5 × 10^–20), type I and IFNγ response (P = 7.1 × 10^–16 and P = 1.8 × 10^–8, respectively) were the three top-ranked pathways. Figure 5B shows the heatmap of selected genes included in these pathways that were upregulated by the GSK126/doxorubicin cotreatment; qPCR analysis confirmed the selective GSK126/doxorubicin-induced upregulation of these genes, a subset of which is shown in Fig. 5C. These pathways include genes with the potential to induce apoptosis (e.g., STAT1, IRF1, IRF9, ISG15, and ISG20; refs. 38–40). Several genes typically regulated by doxorubicin via a p53-dependent mechanism [e.g., BBC3 (PUMA), CDKN1A (p21), TP53INP2, and TP53I11)] were also markedly overexpressed in GSK126/doxorubicin-treated p53-null KG-1 cells (Fig. 5B and C). Increased expression of some of the genes activated by GSK126/doxorubicin cotreatment was also confirmed by western blot, concomitant with a marked increase in γH2AX levels (Fig. 5D).

The increased expression of selected genes induced by the GSK126/doxorubicin cotreatment may depend on demethylation of H3K27me3-marked promoters, allowing chromatin “opening” and binding of TFs expressed/activated in GSK126/doxorubicin-treated AML cells.

Thus, we performed an anti-H3K27me3 ChIP-seq experiment in KG-1 cells under the same conditions used for the RNA-seq experiment. Approximately 27% of the genes exclusively upregulated by the GSK126/doxorubicin treatment show strong H3K27me3 peaks at their promoters in untreated cells; this percentage is higher (35%), among genes with ≥ 4-fold increased expression. Both percentages are higher than expected by chance (Fisher's exact test P < 2.2 × 10^–16). Selected genes that lost the H3K27me3 mark in GSK126- and GSK126/doxorubicin-treated AML cells are labeled in red in Fig. 5B and C and shown in Fig. 5E.

The remaining genes lack the H3K27me3 mark in their promoters in untreated cells but are transcriptionally activated only after GSK126/doxorubicin treatment. This suggests that they became targets of TFs induced/activated in GSK126/doxorubicin-treated KG-1 cells.

Among the H3K27me3-marked genes activated by GSK126/doxorubicin treatment, CDKN1A is particularly interesting because its expression may be responsible for the slow cycling/late S-phase arrest of GSK126/doxorubicin-treated KG-1 cells, which contrasts with the rapid G₂ arrest of KG-1 cells treated with doxorubicin only (Supplementary Fig. S15). The increase in the fraction of late S-phase cells induced by GSK126/doxorubicin treatment may enhance the cell population targeted by doxorubicin, possibly augmenting DNA damage (41).

Similar studies were also performed using the p53-null THP-1 cell line. In contrast with the KG-1 line in which changes in gene expression were mostly induced by the GSK126/doxorubicin co-treatment (Fig. 5A), the RNA-seq analysis of untreated or drug-treated (GSK126, doxorubicin, or both) THP-1 cells revealed that doxorubicin treatment alone modulated the expression of 765 genes compared with the 1,217 genes regulated by the GSK126/doxorubicin cotreatment; of interest, EZH2 inhibition alone modulated the expression of 416 genes (Supplementary Fig. S16A). Of the genes differentially expressed in GSK126/doxorubicin-treated versus untreated cells, 599 were specific for the GSK126/doxorubicin cotreatment (FDR <5%). However, some of the genes upregulated by doxorubicin alone exhibited increased expression upon GSK126/doxorubicin cotreatment of THP-1 cells. The heatmap of a subset of genes exhibiting selective or enhanced expression in GSK126/doxorubicin-treated cells is shown in Supplementary Fig. S16B. Several of these genes are involved in biological processes such as growth suppression/apoptosis (CDKN1A, DUSP4, GADD45A, INKA2, RASL10A, SMAD9, and TNFRSF11A), and stress response (ATF3, CLU, ETV7, FOSB, FOXD3, and SGK1) that may explain the increased apoptosis of GSK126/doxorubicin-treated THP-1 cells.

To assess whether changes in expression induced by the GSK126/doxorubicin cotreatment were associated with loss of H3K27me3 marking in the promoter of GSK126/doxorubicin-modulated genes, anti-H3K27me3 ChIP-seq experiments of THP-1 cells were performed under the same conditions used for the RNA-seq experiments. Approximately 32% of the genes exclusively upregulated by the GSK126/doxorubicin treatment exhibit loss of the distinct H3K27me3 peaks at their promoters that are present in untreated cells; this percentage is similar to that (27%) observed in doxorubicin/GSK126-treated KG-1 cells. The H3K27me3 mark was also lost in the promoter of some genes that, compared with treatment with doxorubicin alone, exhibited a further increase in expression upon GSK126/doxorubicin cotreatment (e.g., the ATF3 gene). Examples of genes that lost the H3K27me3 mark are labeled in Supplementary Fig. S16B (red) and shown in Supplementary Fig. S16C.

Together, these data strongly suggest that loss of H3K27me3 marking is involved in the transcriptional reprogramming induced by GSK126/doxorubicin cotreatment of AML cells.

Frontline therapy for de novo AML typically consists of a remission-inducing protocol that combines AraC (7 days) with doxorubicin (3 days). We tested whether inhibiting EZH2 enhances the anti-leukemia effect of this therapy using preclinical models of human AML. To this end, NOD-Rag1null IL2rgnull (NRG) mice injected with luciferase-expressing THP-1 cells were treated with vehicle (control), GSK126, doxorubicin/AraC (DA 5+3; ref. 42), or DA + GSK126 (Fig. 6A). The therapeutic effect was assessed by monitoring leukemia burden by bioluminescence imaging of leukemic mice 1- and 2-weeks after treatment and by determining their median survival. The DA/GSK126 combination was more effective than DA or GSK126 alone in suppressing leukemia burden (Fig. 6B) and in prolonging survival (median survival: 61 days vs. 55 and 54 days, respectively; Fig. 6C).

The effect of the chemotherapy/GSK126 combination was also assessed in NRG-SGM3 mice injected with AML sample #40. In this model, mice were treated with vehicle, GSK126 alone, doxorubicin/AraC alone, or the GSK126/doxorubicin/AraC combination when peripheral blood leukemic cells, detected by anti-CD45 flow cytometry, were approximately 10%–35%. Doxorubicin was given at a dose of 1.0 mg/kg/3 days and AraC was given at 33.3 mg/kg/5 days, which is one third lower of that used in the THP-1 model. Leukemia burden was monitored by assessing the percentage of CD45⁺ leukemic cells three times at one week-interval, starting 10 days after termination of the therapy.

Vehicle- and GSK126-treated mice exhibited a similar increase in leukemia burden over the course of the experiment (Fig. 6D). As expected, treatment with chemotherapy alone suppressed AML growth. However, 24 days after therapy termination peripheral blood leukemic cells were more numerous than at the start of the treatment in 6/6 mice (Fig. 6D). By contrast, the combined GSK126/AraC/doxorubicin treatment was clearly more effective with a significant improvement over treatment with chemotherapy alone in suppressing AML growth at each time of analysis (Fig. 6D, bottom). In particular, 24 days after therapy termination only 1/8 mice had a leukemia burden (peripheral blood CD45⁺ cells) higher than at the start of the treatment (Fig. 6D). Notice that this AML sample is from a high-risk patient who had a rapidly fatal disease relapse after a short-lived complete remission (Supplementary Table SI).

Potentiation of chemotherapy-induced DNA damage and apoptosis by decondensing H3K27me3-marked chromatin was significantly enhanced in S-phase–enriched THP-1 cells (Fig. 4). We extended this strategy to our in vivo model of NRG mice injected with THP-1-luciferase cells (Fig. 7A). We observed that, compared with the control group, leukemic mice temporarily given a palbociclib-containing diet (72 hours), and analyzed 48 hours later, exhibited an increased number of bone marrow S-phase THP-1 cells (29.8% vs. 13.7%, respectively; Fig. 7B), consistent with the effect seen in the ex vivo experiment (Fig. 4). Decondensing the H3K27me3-marked chromatin of S-phase–enriched AML cells, followed by doxorubicin/AraC cotreatment was more effective than pretreatment with palbociclib followed by the doxorubicin/AraC combination in suppressing THP-1 leukemia burden (Fig. 7C) and in prolonging survival (median survival 81 days vs. 75 days; Fig. 7D). Comparison of the Fig. 6 and Fig. 7 survival experiments strongly suggest that the GSK126/chemotherapy combination was more effective when given to mice pretreated short-term with palbociclib to increase the fraction of S-phase AML cells.

We show here that blast cells from most patients with AML exhibit rapid association of the H3K27me3 repressive mark with nascent DNA. This chromatin structure protects from cytotoxic agent-induced DNA damage, which typically occurs during DNA replication. In an effort to improve the current protocol of AML remission induction, we designed a novel therapy based on the sequential treatment with an EZH2 inhibitor to decondense the H3K27me3-marked chromatin followed by chemotherapy. This approach significantly increased Doxorubicin-induced DNA damage and apoptosis ex vivo, and leukemia suppression in mice. Such effects were further improved by enhancing the targeting of chemotherapy to newly replicating de-condensed DNA after short-term CDK4/6 inhibition.

The structure of chromatin and chromatin-based epigenetic mechanisms are thought to play critical roles in cancer etiology/maintenance and thus are considered very attractive therapeutic targets (43–45). Significant effort has focused on developing inhibitors of histone-modifying enzymes, with the expectation that changing the structure of chromatin will affect overall transcription and thus will be beneficial in cancer treatment. Although some success has been achieved, for example, in the treatment of cutaneous T-cell lymphoma (CTCL) with HDAC inhibitors (46), EZH2 inhibitors have shown limited clinical success (47). To some extent, this is reflected in our findings that EZH2 inhibition by GSK126 treatment induced only a modest decrease in AML colony formation and a modest increase in the survival of mice injected with THP-1 cells.

Chromatin structure is also correlated to therapy response (48, 49). In particular, increased chromatin accessibility is associated with response to HDAC inhibitors in CTCL (46), and appears to pre-determine glucocorticoid sensitivity in acute lymphoblastic leukemia (50). Moreover, high expression of PRC2 genes correlates with low chromatin accessibility and anthracycline resistance in breast cancer (49).

Chromatin structure is likely to influence the efficacy of anthracyclines, as these drugs induce DNA damage preferentially at de-condensed genomic regions (15, 16), and promote histone eviction by incorporating more efficiently into open rather than compact chromatin (16).

However, this knowledge has not been exploited yet to design more effective therapies with currently used cytotoxic agents. Our approach is based on the novel concept of using EZH2 inhibitors in AML, and possibly in other malignancies, not as direct therapeutic agents but as modifiers of chromatin structure to enhance the effects of DNA-damaging agents.

We postulated that decondensing the H3K27me3-marked chromatin of AML blasts by inhibiting EZH2 activity would render DNA more accessible to DNA-damaging agents, causing enhanced cell death. We also hypothesized that enriching S-phase AML cells with open chromatin and targeting them with chemotherapy would further enhance DNA damage and cell death. Both hypotheses were validated by the ex vivo and in vivo studies reported here using AML cell lines and primary patients' samples.

Our approach is applicable to most cases of AML, with the possible exception of those (approximately 10%) with gene mutations impairing the activity of the PRC2 complex (e.g., loss-of-function mutations/deletions of EZH2, ASXL1, or BCOR). However, these mutations are often heterozygous and might be present in a minority of AML cells, raising the possibility that high levels of H3K27me3 may be also present in AML samples with these mutations.

EZH2 inhibition by GSK126 treatment alone induced a slight decrease in AML colony formation and a modest increase in the survival of mice injected with THP-1 AML cells. These findings are consistent with results of a study showing that EZH2 expression/activity is required for AML colony formation and disease maintenance in mice (51). However, AML suppression induced by EZH2 gene deletion was more effective than inhibition of EZH2 activity (51), suggesting the involvement of enzyme-independent mechanisms.

In our studies, the effect of EZH2 inhibition was initially designed as mostly auxiliary, because it promotes enhanced DNA damage and apoptosis by rendering the chromatin structure of AML cells more accessible to DNA-damaging agents. However, chromatin decondensation induced by EZH2 inhibition can also allow the reactivation of repressed genes if TFs expressed in GSK126/doxorubicin-treated AML cells bind to gene promoters not accessible in untreated cells due to H3K27me3 marking. In the KG-1 and THP-1 cell lines, this mechanism might be responsible for the increased expression of several cell death-promoting and growth-inhibitory genes. GSK126/doxorubicin treatment of AML cells also induced the expression of genes lacking the H3K27me3 mark in their promoter; we propose that in this case, H3K27me3 loss might have an indirect role, through reactivation and DNA binding of TFs that are repressed in untreated AML cells due to condensed chromatin at their promoters.

In summary, our approach of targeting cytotoxic agents to S-phase cells with decondensed post-replicative chromatin has the potential to achieve a more effective killing of AML cells than presently possible due to the intrinsic “chemoresistance” caused by their highly compacted chromatin structure. Our strategy could be especially beneficial in limiting chemotherapy side effects in elderly patients with AML or those with significant comorbidities. For these patients, reduced doses of cytotoxic agents might achieve AML suppression comparable with that of standard chemotherapy, if combined with EZH2 inhibitors to enhance DNA accessibility of cytotoxic agents.

Most promisingly, our protocols for treating AML-bearing mice have the potential to be adopted in the clinic, because CDK4/6 and EZH2 inhibitors are currently approved and/or under evaluation for anticancer therapies.

B. Calabretta reports a patent for Methods for enhancing DNA damage and apoptosis of leukemic cells pending. P. Porazzi reports a patent for method for enhancing DNA damage and apoptosis of leukemic cells pending to Bruno Calabretta, Alexander Mazo, Svetlana Petruk, and Porazzi Patrizia. S. Petruk reports a patent for methods for enhancing DNA damage and apoptosis of leukemic cells pending. P. Porcu reports other support from Viracta Therapeutics, Innate Pharma, Daiichi, and Kyowa outside the submitted work. C.M. Eischen reports other support from AbbVie outside the submitted work. A. Mazo reports a patent for methods for enhancing DNA damage and apoptosis of leukemic cells pending. No disclosures were reported by the other authors.

P. Porazzi: Conceptualization, data curation, formal analysis, validation, investigation, writing–original draft, writing–review and editing. S. Petruk: Formal analysis, investigation, methodology. L. Pagliaroli: Formal analysis, methodology. M. De Dominici: Investigation, methodology. D. Deming II: Investigation, methodology. M.V. Puccetti: Formal analysis, investigation, methodology. S. Kushinsky: Formal analysis, investigation. G. Kumar: Software, formal analysis. V. Minieri: Data curation. E. Barbieri: Data curation, methodology. S. Deliard: Software, formal analysis. A. Grande: Resources. M. Trizzino: Resources, formal analysis. A. Gardini: Resources, data curation, formal analysis. E. Canaani: Methodology. N. Palmisiano: Resources. P. Porcu: Resources. A. Ertel: Software, formal analysis. P. Fortina: Resources. C.M. Eischen: Resources, data curation, supervision. A. Mazo: Conceptualization, resources, data curation, supervision, funding acquisition. B. Calabretta: Conceptualization, resources, supervision, funding acquisition, writing–review and editing.

The authors thank the support of the Flow Cytometry Core Facility and Bioimaging Shared Resource of the Sidney Kimmel Cancer Center at Thomas Jefferson University, which are supported by the NCI Core grant (NCI 5 P30 CA-5603). This work was supported by NIH grants R01-HL127895 (to B. Calabretta and A. Mazo), R01CA257251 (to B. Calabretta), R01GM075141 (to A. Mazo), R01CA226432 (to C.M. Eischen), and the Herbert A. Rosenthal Fund. D. Deming II was supported by training grant T32CA236736.

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.