Both gain-of-function enhancer of zeste homolog 2 (EZH2) mutations and inactivating histone acetyltransferases mutations, such as CREBBP and EP300, have been implicated in the pathogenesis of germinal center (GC)–derived lymphomas. We hypothesized that direct inhibition of EZH2 and histone deacetyltransferase (HDAC) would be synergistic in GC-derived lymphomas.
Lymphoma cell lines (n = 21) were exposed to GSK126, an EZH2 inhibitor, and romidepsin, a pan-HDAC inhibitor. Synergy was assessed by excess over bliss. Western blot, mass spectrometry, and coimmunoprecipitation were performed. A SU-DHL-10 xenograft model was utilized to validate in vitro findings. Pretreatment RNA-sequencing of cell lines was performed. MetaVIPER analysis was used to infer protein activity.
Exposure to GSK126 and romidepsin demonstrated potent synergy in lymphoma cell lines with EZH2 dysregulation. Combination of romidepsin with other EZH2 inhibitors also demonstrated synergy suggesting a class effect of EZH2 inhibition with romidepsin. Dual inhibition of EZH2 and HDAC led to modulation of acetylation and methylation of H3K27. The synergistic effects of the combination were due to disruption of the PRC2 complex secondary to acetylation of RbAP 46/48. A common basal gene signature was shared among synergistic lymphoma cell lines and was characterized by upregulation in chromatin remodeling genes and transcriptional regulators. This finding was supported by metaVIPER analysis which also revealed that HDAC 1/2 and DNA methyltransferase were associated with EZH2 activation.
Inhibition of EZH2 and HDAC is synergistic and leads to the dissociation of PRC2 complex. Our findings support the clinical translation of the combination of EZH2 and HDAC inhibition in EZH2 dysregulated lymphomas.
Given the prevalence of enhancer of zeste homolog 2 (EZH2) mutations and histone acetyltransferase mutations in germinal center diffuse large B-cell lymphoma, the rational combination of EZH2 inhibition and histone deacetyltransferase (HDAC) inhibition was explored. Using a panel of 21 lymphoma cell lines, we demonstrate that exposure to dual inhibition of EZH2 and HDACs was synergistic in EZH2 dysregulated lymphomas. The synergistic effects of EZH2 and HDAC inhibition may be attributed to the disassembly of the PRC2 complex. In a mouse xenograft model of SU-DHL-10, the combination led to tumor growth delay and an improvement in overall survival. A basal common genetic signature among synergistic cell lines was identified using gene set enrichment analysis and metaVIPER analysis and correlates with therapeutic response. The novel combination of dual EZH2 and HDAC inhibition may serve as a future precision medicine therapeutic platform. A clinical trial to further explore this combination is in development.
Enhancer of zeste homolog 2 (EZH2) is critical in the germinal center (GC) reaction and serves as the catalytic subunit of the Polycomb Repression Complex 2 (PRC2), inducing trimethylation of histone 3 lysine 27 (H3K27me3), a marker of transcriptional repression (1). During the GC reaction, the PRC2 complex recruits histone deacetyltransferase (HDAC) 1/2 and DNA methyltransferases (DNMT) to further inhibit transcription (2, 3).
Disturbances in epigenetic pathways have been implicated in lymphomagenesis. Aberrancy of histone methyltransferases, such as EZH2, has been associated with the development of GC-derived lymphomas, including diffuse large B-cell lymphoma (DLBCL) and follicular lymphoma (4, 5). Activating mutations in EZH2 have been implicated in 22% of GC-DLBCL and 7% to 12% of follicular lymphoma (4, 5). EZH2 dysregulation has been implicated in other lymphoma subtypes, including overexpression in some subtypes of T-cell lymphoma (TCL; refs. 6–9). Given the prevalence of EZH2 dysregulation in several malignancies, EZH2 inhibitors have been developed and demonstrate superior efficacy in mutated EZH2 GC-derived lymphoma cell lines compared with wild-type EZH2 cell lines (10–12). The preclinical activity of the EZH2 inhibitors in B-cell lymphomas has been replicated in the clinic by tazemetostat, a first-in-class EZH2 inhibitor, which demonstrated an overall response rate of 38% in a phase I clinical trial (13). Notably, clinical responses were achieved irrespective of EZH2 mutational status.
Also contributing to GC lymphomagenesis is the haploinsufficiency of histone acetyltransferases (HAT). HATs control the addition of acetyl groups on histones in order to promote an open chromatin state, allowing for transcription. Mutations leading to loss of function of HATs, specifically EP300 and CREBBP, are found in 39% of GC-DLBCLs and 41% of follicular lymphomas (14), and the presence of these mutations has been reported to be associated with HDAC inhibitor sensitivity (15). Vorinostat, an HDAC inhibitor, was the first epigenetic drug to gain FDA approval in patients with relapsed/refractory TCL. Two other HDAC inhibitors have gained approval for the treatment of TCL, whereas panobinostat has been approved for the treatment of multiple myeloma. However, despite the robust link between epigenetic dysregulation in several malignancies, few diseases have demonstrated clinical benefit with single-agent epigenetic targeting therapy, including GC-derived B-cell lymphomas.
Our group and others have established a proof of principle for selective targeting of epigenetic modifiers in DLBCL. The combination of niacinamide, a sirtuin inhibitor, and pan-HDAC inhibitors, including romidepsin, is synergistic in GC-DLBCL (16). A phase I clinical study utilizing vorinostat and niacinamide in relapsed/refractory lymphoma demonstrated an overall response rate (ORR) of 24%, suggesting a potential role for combination epigenetic therapy in B-cell lymphomas. The combination of panobinostat and decitabine, a DNMT inhibitor, was found to be more synergistic in GC-DLBCL compared with activated B-cell (ABC)-DLBCL cell lines leading to a unique differential expression of various genes including SMAD1 and DNMT3A (17). Although single-agent epigenetic therapy has been disappointing in DLBCL, the aforementioned data suggest that using a platform based on a combination of epigenetic targeted agents may be a potential therapeutic method for the treatment of GC-DLBCL.
Given the frequent dysfunction of EZH2 as well as HATs in GC-derived B-cell lymphomas, we hypothesized that the rational combination of EZH2 and HDAC inhibitors would be synergistic by modulating both acetylation and methylation states, in turn, triggering apoptosis. Simultaneous mutations in EZH2 and CREBBP occur in 26 of 1,343 primary DLBCL samples (adjusted P value < 0.001), whereas co-occurrence of mutations of EZH2 and EP300 is not significant (7/1,343; refs. 18, 19). Herein, we demonstrate that GSK126, an EZH2 inhibitor, and romidepsin, a pan-HDAC inhibitor, are synergistic by disrupting the PRC2 complex, leading to modulation of histone acetylation and methylation. Sensitivity to the combination was associated with a specific gene expression signature.
Materials and Methods
Cell lines and culture
OCI-LY1, SU-DHL-2, SU-DHL-6, Pfeiffer, Farage, Toledo, Riva, HBL-1, Jeko-1, Z-138, H9, and HH were obtained from the ATCC. OCI-LY7, OCI-LY10, SU-DHL-10, and OCI-LY3 were obtained from DSMZ. PF382 and P12 were gifts from the laboratory of Adolfo Fernando. TLOM-1 and MT-1 were obtained from Kyoto University; and MT-2 was obtained from Memorial Sloan Kettering Cancer Center. All cell lines were authenticated and screened for mycoplasma using the ATCC/Promega STR Authentication Testing Kit and Lonza MycoAlert, obtained between 2008 and 2016 and revived after 2 weeks. Experiments were performed between 2015 and 2018.
Genomic DNAs from 21 lymphoma cell lines were extracted with cell culture DNA mini kit (Qiagen) and measured by NanoDrop 3300. PCR was performed by following the manufacturer's instructions. Genomic DNA was amplified by PCR with AmpliTaq Gold DNA Polymerase, PE Buffer II, and MgCl2 (Applied Biosystems) using primers designed as follows: EZH2 Y641 forward, 5′-CAGGTCTGAGGATTTACAGTGATAG-3′; EZH2 Y641 reverse, GCAGAAGTCCAGGCTGAAA-3′; EZH2 A677 forward 5′-GGCAAACCCTGAAGAACTGTA-3′; EZH2 A677 reverse 5′-GTCCATCATCACAGGACTGAAA-3′. PCR products were run on an agarose gel, purified using QIAquick PCR purification kit (28104), and sent for sequencing (Genewiz).
Cell viability assays
Cells were counted and resuspended based on their optimal density for log-phase growth. Cell viability assays were performed as previously described (17). Cells were exposed to romidepsin (Selleckchem), ACY957 (Acetylon), GSK126 (Selleckchem), EPZ011989 (Epizyme), and CPI-1205 (Selleckchem). Synergy was assessed by excess over bliss (EOB; refs. 20, 21). Sensitivity to GSK126 and romidepsin as determined by mean IC50 was correlated with EZH2 mutation/overexpression and HAT mutations, respectively, using Prism GraphPad's Student paired t test. Experiments were performed in triplicate and repeated at least twice.
Flow cytometry analysis was performed using FITC Annexin V Apoptosis Detection Kit with PI (Biolegend #640194) as previously described (16). Experiments were performed at least 3 times.
Immunoprecipitation was performed using the Pierce Co-Immunoprecipitation Kit (#26149). Columns were prepared with 20 to 40 μg of antibody. Whole protein lysate was incubated with antibody. Flow through was collected, and column was washed and eluted. Antibodies used were anti-EZH2 (Cell Signaling Technology), anti-SUZ12 (Cell Signaling Technology), anti-RbAP 46/48 (Cell Signaling Technology), anti-EED (Millipore), anti-HDAC2 (Cell Signaling Technology), and anti-DNMT3L (Novus Biologicals). Experiments were performed at least 3 times.
Western blotting was performed as previously described (16). Antibodies used were as above. Experiments were performed at least 3 times.
Mass spectrometry for acetylation of PRC2 complex
Immunoprecipitation was performed using Thermo Scientific Pierce MS-Compatible Magnetic IP Kit. Protein was incubated with EZH2 or acetylated-lysine antibody. Antibody-bound proteins were eluted and run into SDS-PAGE. The excised gel lane pieces were reduced, alkylated, and digested in Trypsin Gold (Promega) digestion buffer (Thermo Fisher Scientific). Peptides were extracted with 70% acetonitrile (Thermo Fisher Scientific). The concentrated peptide mix was reconstituted in a solution of 2% ACN and 2% formic acid for MS analysis. Peptides were eluted from the column using a Dionex Ultimate 3000 Nano LC system. Using Thermo Fusion Tribrid mass spectrometer (Thermo Scientific), eluted peptides were electrosprayed. Mass spectrometer-scanning functions and high performance liquid chromatography (HPLC) gradients were controlled by the Xcalibur data system (Thermo Fisher). Experiments were performed at least twice.
Database search and interpretation of MS/MS data
Tandem mass spectra from raw files were searched against uniprot_human_170129.fasta database using the Proteome Discoverer 2.1 (Thermo Fisher). The mouse protein database was downloaded as FASTA-formatted sequences from Uniprot protein database (January 2017). The peptide mass search tolerance was 10 ppm with a required minimum sequence length of 7 amino acids. To calculate confidence levels and FDRs, Proteome Discoverer generates a decoy database and performs the search against this concatenated database (nondecoy + decoy). Scaffold (Proteome Software, Inc.) was used to visualize and filter to <1% FDR. Spectral counts were used for estimation of relative protein abundance.
HDAC short hairpin RNA
Human HDAC2 short hairpin RNA (shRNA) plasmids were purchased from Origene (#TG312495). HEK293 cells were plated in OPTI-MEM containing shRNA or scramble using Lipofectamine 3000 (Cat#L3000075). Cells were selected with puromycin, periodically analyzed by flow cytometry and fluorescent microscopy to monitor GFP levels until a stable cell line had been generated.
MS analysis and data handling for H3K27 acetylation and methylation
Histone extraction, derivatization, and tryptic digestion were adapted from previous works (22, 23). Peptides were resuspended in 0.1% TFA for LC-MS/MS analysis.
Multiple reaction monitoring was performed on a triple quadrupole (QqQ) mass spectrometer (Thermo Fisher Scientific TSQ Quantiva) directly coupled with UltiMate 3000 Dionex nano-LC system. The following QqQ settings were used: collision gas pressure of 1.5 mTorr; Q1 peak width of 0.7 (FWHM); cycle time of 2 seconds; skimmer offset of 10 V; electrospray voltage of 2.5 kV. Modified and unmodified histone peptides monitored in the assay were selected based on previous reports (23). Raw MS files were imported and analyzed in Skyline software with Savitzky–Golay smoothing (24). Automatic peak assignments from Skyline were manually confirmed. Peptide peak areas from Skyline were used to determine the relative abundance of each histone modification. The relative abundances were determined based on the mean of three technical replicates with error bars representing the SD. Experiments were performed at least twice.
In vivo studies
Animals were maintained in accordance with an Institutional Animal Care and Use Committee–approved protocol (AC-AAAR9404). SU-DHL-10 (1 × 107) was suspended in 50% Matrigel (BD Biosciences) and 50% PBS (Gibco) and subcutaneously injected into the flanks of 5- to 7-week-old beige/SCID female mice (Taconic Farms). Mice were randomly divided into 5 cohorts (n = 9–10) upon tumor volume reaching 80 to 100 mm3 as follows: (i) Normal saline: days 1, 4, 8, 15, and 18; (ii) GSK126: 100 mg/kg days 1, 4, 8, 11, 15, and 18; (iii) romidepsin: 2 mg/kg days 1, 8, and 15; (iv) GSK126 and romidepsin; (v) pretreatment with GSK126 (days 1, 4, 8, and 15), and followed by romidepsin on days 8, 15, and 22. Dosing was selected based on prior in vivo studies (11, 25, 26). Drugs were diluted in sterile normal saline and administered via i.p. route. Weight and tumor volume were evaluated 3x/week. Statistical analysis was performed using two-way ANOVA, and overall survival (OS) was estimated using the Kaplan–Meier method (GraphPad Software, Inc.)
Pharmacokinetic/pharmacodynamics in vivo studies
Plasma samples were collected at 0.25, 0.5, 2, 4, 8, and 24 hours after one-time infusion of GSK126 and romidepsin. Noncompartmental analysis was performed using Phoenix Winnonlin software (Certara) to define the maximum plasma concentration (Cmax), the time to maximum plasma concentration (Tmax), and the area under the plasma concentration time curve from t = 0 to the last data point (AUClast). Romidepsin and GSK126 were extracted by mixing 2:1 solution of serum/tissue homogenate in acetonitrile/methanol.
LC-MS/MS analysis was performed using Agilent 6410 triple quad mass spectrometer (Agilent Technologies). Data acquisition and peak integration were done using MassHunter software v 3.1. The assay performance was validated for mouse serum samples according to FDA guidelines (27). Intra-assay precision and accuracy for romidepsin in mouse serum were 5.55% and 105.1%, respectively, whereas the interassay precision was 5.1%. For GSK126, the intra-assay accuracy was 99.35% with a precision of 1.55%, whereas the interassay precision was 2.83%.
RNA was purified using the RNAeasy Plus Kit (QIAGEN). RNA concentration and integrity were verified using Agilent 2100 Bioanalyzer (Agilent Technologies). Libraries were generated using Illumina's TruSeq RNA sample Prep Kit v2, following the manufacturer's protocol. Note that 2 × 75 bp paired-end sequencing was performed on the HiSeq4000 sequencer. Raw RNA-Seq data were aligned to the Human reference genome (Version hg19 from UCSC) using the STAR (V 2.4.2) aligner (28). Aligned reads were quantified against the reference annotation (hg19 from UCSC) to obtain Fragments per Kilobase per million (FPKM) and raw counts using Cufflinks (v 2.2.1) and HTseq, respectively (29, 30). Differential expression was performed on raw counts with the limma package in R (31). Principal component analysis was performed on the log2-transformed FPKM expression values in R statistical software. Gene set enrichment analysis (GSEA) was performed using software from Broad Institute. Genes were ranked by the t-statistic value and used to identify significantly enriched biological pathways. Differential expression was performed, and expression profiles of synergistic (EOB ≥ 20) versus nonsynergistic (EOB < 20) cell lines were compared.
The Virtual Inference of Protein-activity by Enriched Regulon analysis (VIPER) algorithm is a computational systems biology approach to infer protein activity from gene expression profiling (32, 33). In the absence of an available cancer-type–specific regulatory network, metaVIPER (34) can be effectively used to infer protein activity.
All regulatory networks used for metaVIPER analysis were reverse engineered by ARACNe (35). Twenty-four core The Cancer Genome Atlas (TCGA) RNA-Seq–derived interactomes are publicly available in the R Bioconductor package aracne.networks (36). After standard read alignment of RNA-Seq data by STAR to the GRCh38 reference genome build and summarization of expression quantities at the gene count level, gene expression was normalized by the Variance Stabilization Transformation, as implemented in the DESeq2 package on Bioconductor (37). A gene expression signature was computed between each synergistic cell line versus the reference group of nonsynergistic cell lines using the viperSignature function in the VIPER package, followed by application of the analytic Rank-based enrichment analysis using each of the available interactomes (38, 39). Normalized enrichment scores are integrated by Stouffer's method. Pathway analysis on the differential protein activity signature was performed using GSEA with “Cancer Hallmark” and “Gene Ontology” gene sets provided in the Broad MSigDB collections (40).
A machine-learning classifier for predicting synergy with GSK126 and romidepsin using basal protein activity signatures was developed after first running VIPER on scaled gene expression signatures, resulting in protein activity profiles for each sample. The random forest method was applied iteratively with the addition of anywhere from 1 to 100 of the most differentially active proteins between synergistic and nonsynergistic cell lines. For each split in the decision trees, the minimum of the number of proteins made available for classification of 5 was used. The random Forest algorithm was run with 1,000 iterations of 3-fold cross-validation to estimate the ROC.
GSK126 and romidepsin synergize in EZH2 dysregulated lymphomas
To understand the effects of EZH2 inhibition and HDAC inhibition in cell lines with or without EZH2 dysfunction and HAT mutations, a panel of 21 lymphoma cell lines was exposed to GSK126, an EZH2 inhibitor, and romidepsin, a pan-HDAC inhibitor, as single agents. Both B-cell lymphoma and TCL were selected in order to establish a range of drug sensitivity and mutational status. EZH2 mutational status was confirmed via PCR, whereas EZH2 overexpression and HAT mutational status was established from literature including Cancer Cell Line Encyclopedia (Broad Institute). The concentration–effect relationship of 21 cell lines was established over varying time exposures and increasing concentrations to determine the IC50 to GSK126 and romidepsin (Fig. 1). Lymphoma cell lines with an activating mutation in EZH2 were more sensitive to GSK126 as compared to wild-type EZH2 (P = 0.02) as rank ordered by the IC50 at 144 hours (Fig. 1A and C). In regards to cell lines with EZH2 overexpression, there was no clear association with increased sensitivity to GSK126 as compared with wild-type (P = 0.52). Published literature suggests that HAT mutations predict sensitivity to HDAC inhibitors (14, 15). However, only a trend toward romidepsin sensitivity and the presence of EP300 or CREBBP mutation was observed (P = 0.05; Fig. 1B and D).
To investigate the dual effects of EZH2 inhibition and HDAC inhibition (Fig. 2A), lymphoma cell lines were simultaneously exposed to GSK126 and romidepsin over 72 hours. Low drug concentrations (IC20–40) were selected in order prevent untoward toxicity that may be seen with high concentration when combined. Coexposure to GSK126 and romidepsin demonstrated potent synergy with the highest EOB value reaching 61.7 (Fig. 2B; Supplementary Fig. S1). Cell lines harboring EZH2 mutations demonstrated the highest level of synergy. Drug schedule with pretreatment of GSK126 was evaluated but did not affect synergy (Supplementary Fig. S2A and S2B). Combination of romidepsin with other EZH2 inhibitors including EPZ011989 and CPI-1205 also demonstrated synergy, suggesting that the combination of EZH2 inhibition and romidepsin is a class effect of EZH2 (Supplementary. Fig S3).
To confirm induction of apoptosis, 4 GC-DLBCL cell lines with different EZH2 mutational status were simultaneously treated with GSK126 and romidepsin for 24 to 48 hours and evaluated by flow cytometry (Fig. 2C and D; Supplementary Fig. S4). A time point prior to the maximum EOB value was selected in order to capture the events prior to complete cellular demise (24 hours for Pfeiffer; 48 hours for OCI-LY7, SU-DHL-10, and SU-DHL-6). Increased apoptosis was observed with the combination as compared with single-agent exposure. Apoptosis was also confirmed by decreased pro–caspase 3 and increased PARP cleavage following exposure to the combination as measured by immunoblot (Fig. 2E). In addition, as compared with single-agent treatment, the level of p21 was increased after exposure to GSK126 and romidepsin (Fig. 2E).
Coexposure to GSK126 and romidepsin leads to enhanced acetylation and hypomethylation of H3K27 as well as dissociation of the PRC2 complex
To understand the effects of dual epigenetic targeting on both acetylation and methylation of histone, 4 GC-DLBCL cell lines were exposed to control, GSK126, romidepsin, or the combination. Treatment with GSK126 and romidepsin led to increased acetylation and decreased trimethylation of H3K27 as compared with single agents as detected by histone extraction and immunoblot (Fig. 3A). These findings were validated by mass spectrometry (Fig. 3B–E).
Protein levels of EZH2 and other members of PRC2 complex (SUZ12, EED, and RbAp 46/48) were significantly decreased after dual treatment with GSK126 and romidepsin compared with single-agent exposure (Fig. 3F). Coimmunoprecipitation pull-down with EZH2 demonstrated dissociation of the PRC2 complex after simultaneous exposure to GSK126 and romidepsin. Specifically, exposure to romidepsin alone or in combination with GSK126 led to dissociation of EZH2 from EED, RbAp 46/48, and AEBP2 as compared with control, suggesting that romidepsin directly contributes to the breakdown of the PRC2 complex (Fig. 3G). In addition, HDAC2 and DNMT3L were also found to disassemble from the EZH2–PRC2 complex after combination therapy. Mass spectrometry confirmed disappearance of members of the PRC2 complex from EZH2 (Fig. 3H and I). With this in mind, we hypothesized that romidepsin may be responsible for the disruption of the PRC2 complex through direct acetylation of one or more subunits of the complex. To evaluate this hypothesis, SU-DHL-10 cells were treated with romidepsin, and immunoprecipitation using acetyl-lysine antibodies was performed. Based on mass spectrometry analysis, a 2-fold increase estimated by spectral counts of RbAp 46/48 (RBBP4) was observed after exposure to romidepsin as compared with control (FDR < 1.0%; Fig. 3J). Taken together, this suggests that the disruption of the PRC2 complex was secondary to direct acetylation of RbAp 46/48, which is responsible for PRC2 complex recruitment to nucleosomes (41).
HDAC2 plays a critical role in the synergy between GSK126 and romidepsin
Based on the finding that HDAC2 dissociated from PRC2 complex after dual inhibition of EZH2 and HDACs (Fig. 3G), direct targeting of HDAC2 using a selective HDAC 1/2 inhibitor, ACY957 (42), was combined with GSK126 and was found to be synergistic (Fig. 4B). HDAC2 shRNA constructs were developed in order to confirm the role of HDAC2 inhibition in the synergy between GSK126 and romidepsin. Increased acetylation of H3K27 was found in HDAC shRNA HEK 293T cells, mimicking the effects of romidepsin, which was further enhanced by treatment with GSK126 (Fig. 4C). Decreased methylation of H3K27 was more pronounced in HDAC2 shRNA cells treated with GSK126, mirroring the effects of GSK126 and romidepsin exposure. Single-agent GSK126 exposure in HEK 293T cells did not significantly change the status of acetylation or methylation of H3K27.
GSK126 and romidepsin lead to improved OS and tumor growth delay in an in vivo mouse xenograft model
A SU-DHL-10 mouse xenograft model was selected due to the fact that SU-DHL-10 represents a GC-DLBCL cell line that harbors an EZH2-activating mutation as well as HAT mutations (CREBBP and EP300). Mice were exposed to control, GSK126, romidepsin, or the combination as detailed in Fig. 5A. The combination was well tolerated in mice with no appreciable change in weight (Fig. 5B). Compared with single-agent exposure, dual therapy with GSK126 and romidepsin led to significant tumor growth delay (P < 0.05) and increase in OS (P < 0.0001; Fig. 5C and D). Moreover, pretreatment with GSK126 for 1 week did not improve tumor growth kinetics as compared with simultaneous exposure (Supplementary Fig. S2C and S2D).
Pharmacokinetic analysis of both serum and tumor samples was performed after a single exposure to GSK126 and romidepsin at various time points. The median Cmax of GSK126 was 1657.5 ± 413.6 ng/mL which translates to 3.15 μmol/L (in vitro IC50 of GSK126 in SU-DHL-10 is 0.7 μmol/L), whereas romidepsin was 98.24 ± 62.50 ng/mL or 0.18 μmol/L (in vitro IC50 of romidepsin in SU-DHL-10 is 2.59 nmol/L; Fig. 5E and F). The serum AUC0–last of GSK126 and romidepsin were 2828.57 (h*ng/mL) and 5.51 (h*ng/mL), respectively. The intratumor concentration of GSK126 increased over time, whereas the romidepsin concentration was below the level of detection. A similar observation was observed in prior work performed by our group during which the intratumor levels of alisertib increased over time, whereas intratumor levels of romidepsin were below the level of detection after combination therapy (25).
Synergistic cell lines share a common basal gene expression and protein activity profile
Differential gene expression profiling was performed on pretreatment lymphoma cell lines to determine their basal expression pattern and correlated to synergy (n = 21). Cell lines with EOB ≥ 20 after treatment with GSK126 and romidepsin were defined as synergistic. There were a total of 69 genes identified (FDR < 0.2) that were differentially expressed in the synergistic cell lines compared with nonsynergistic cell lines, suggesting that a common basal gene expression profile is shared among the synergistic cell lines (Fig. 6A; Supplementary Fig. S5). Pathway analysis determined by GSEA revealed synergistic cell lines are characterized by upregulation in chromatin remodeling genes and transcriptional regulators such as HDAC9 and HCFC1 as well as pathways implicated in epigenetic regulation (Fig. 6A and B). Moreover, of the 69 genes that were found to be differentially expressed in synergistic cell lines compared with nonsynergistic cell lines, 34 genes have been identified to be altered in more than 1.0% of primary patient DLBCL samples as confirmed by TCGA database and cBioPortal (Supplementary Fig. S6; ref. 18).
metaVIPER was used to identify proteins whose activity predicts, and potentially mediates, sensitivity to dual EZH2-HDAC inhibition in lymphoma cell lines. We computed a differential protein activity signature between cell lines that demonstrate synergy by EOB and those that did not, and subsequently performed pathway analysis on this signature. Synergistic cell lines were markedly enriched in pathways involving cell-cycle control, DNA replication, and chromatin remodeling (Fig. 6C). This finding is similar to what was observed using GSEA at the RNA expression level. Downregulated pathways include inflammatory pathways as well as differentiation/developmental genes (Fig. 6D).
Differential protein activity on 48 TCGA DLBCL primary patient samples was inferred using a pan-TCGA reference to compute gene expression signatures followed by interrogation with metaVIPER. Eighty-one percent of DLBCL tumors demonstrate significantly increased EZH2 activity (Bonferroni P value < 0.01), in spite of only a few of the tumors harboring mutations in EZH2. Unbiased cosegregation analysis between EZH2 and a set of 400 “druggable” proteins demonstrated that the aberrant activity of several proteins is strongly associated with EZH2 activation, including HDAC 1/2 and DNMT (Fig. 6E), further supporting dual targeting of EZH2 and HDACs in DLBCL. Taken together, interrogation of protein activity as a means to identify essential pathways that are common among synergistic cell lines describes a cellular state that is characterized by a (1) high level of proliferation; (2) transcriptional silencing through chromatin remodeling/condensation; (3) halt in cellular differentiation; and lastly (4) suppression of inflammatory response. Interestingly, TGFβ signaling, which promotes T-regulatory cell function, is found to be more enriched in nonsynergistic cell lines.
Many groups have demonstrated that gene expression profiles can be used to develop robust classifiers to predict drug sensitivity, but are difficult to validate in new datasets due to the inherent noise of RNA expression measurements and the risk of false discovery (33). In contrast, VIPER inference of protein activity is highly reproducible and biologically relevant. We developed a random forest classifier from the basal protein activity profiles of this diverse set of lymphoma cell lines to predict synergy between GSK126 and romidepsin. This classifier demonstrated good ROC on 3-fold cross-validation, with an AUC of 0.89 and an accuracy rate of 0.83 for predicting synergy (Supplementary Fig. S7). The classifier plateaued in performance with the inclusion of only 8 proteins (NDUFA13, CREBRF, MRPL12, KAT2B, ASF1B, BMPR2, POLRSI, and IL65T), consistent with the ability of VIPER to identify biologically relevant proteins. Interestingly, decreased activity of KAT2B, an important HAT protein, was one of the most prominent features in the classifier for predicting synergistic activity of GSK126 and romidepsin.
Epigenetic alterations have been implicated as drivers of lymphomagenesis, with EZH2 dysregulation and HAT inactivating mutations being central to the pathogenesis of GC-DLBCL. Given the prominence of EZH2 dysregulation in lymphoma, selective EZH2 inhibitors have been developed and have shown single-agent activity in early clinical studies (13, 43). Individually, mutations in EZH2 and HAT produce a repressed transcriptional state, and together, the PRC2 complex recruits HDAC 1/2, leading to additional transcriptional repression. In this context, dual inhibition of EZH2 and HDACs may serve as a rational therapeutic platform in lymphomas harboring epigenetic derangements (Fig. 2A). We describe that the combination of GSK126 and romidepsin was synergistic in EZH2 dysregulated lymphoma cell lines secondary to disassembly of the PRC2 complex due to acetylation of RbAP 46/48. This in turn caused attenuation of H3K27 methylation, increased acetylation, and upregulation of p21, which together triggered apoptosis.
Acetylation of tumor suppressors and oncogenes has been well described (16, 44). EZH2 has been shown to be directly acetylated by P300/CBP-associated factor (PCAF) and deacetylated by SIRT1 in lung adenocarcinoma models, with acetylation of EZH2 having no effects on EZH2′s ability to interact with other members of the PRC2 complex (45). Acetylation of EZH2 itself was not identified in our studies, however, we demonstrate that exposure to GSK126 and romidepsin leads to acetylation of RbAP 46/48, in turn, causing instability of the PRC2 complex, preventing EZH2 from catalyzing trimethylation, and leading to an open chromatin state.
Xenograft experiments demonstrated improvement in OS and tumor growth delay favoring the combination arm. Interestingly, intratumor concentrations of romidepsin were below the level of detection after cotreatment with GSK126 and romidepsin, which we have observed in prior combination studies (25). However, despite the undetectable intratumor concentration of romidepsin, intratumor concentrations of GSK126 increased over time, with the combination arm demonstrating potent synergy compared with single-agent therapy as manifested by increased OS and delayed tumor growth kinetics. Although complete tumor shrinkage was not observed in our xenograft studies, SU-DHL-10 has a very high proliferative rate owing to the fact that it harbors translocations of both MYC and BCL2 classifying it as a double hit lymphoma (46). Double hit lymphomas are most frequently of GC origin and are notoriously clinically challenging as patients often relapse after first-line therapy and salvage chemotherapy (47). Thus, our data may suggest a role of dual inhibition of EZH2 and HDACs for the treatment of double hit lymphomas. Given there has been limited success in identifying targeted therapy for double hit lymphomas, this warrants further investigation.
With the use of next-generation sequencing, individualized approaches to cancer therapy may arise based on unique gene expression patterns and mutational profiles that collectively contribute to a specific molecular phenotype. In an effort to identify a gene expression profile that may select patients that would benefit from dual EZH2 and HDAC inhibition, pretreatment RNA sequencing on a panel of lymphoma cell lines was performed. Cell lines demonstrating synergy to combined epigenetic therapy share a common basal genetic signature with enrichment in chromatin remodeling and gene silencing pathways, with identification of 69 genes that are expressed in a similar pattern. Using metaVIPER, enrichment of chromatin modification and epigenetic pathways was verified, but it also identified enrichment of DNA repair/synthesis and cell-cycle regulation pathways as well as downregulation of immune/inflammatory pathways in synergistic cell lines as compared with nonsynergistic cell lines. Interestingly, recent evidence suggests that EZH2 and DNMT1 inhibit tumor cell production of T helper 1 type cytokines CXCL9 and CXCL10 as well as infiltration by effector T cells, all of which can be reversed by inhibition of EZH2 and DNMT (48). Correlative studies to characterize the tumor T-cell infiltrate in the context of pre- and on-treatment biopsies after treatment with EZH2 inhibitor in conjunction with HDAC inhibitor would further assist in understanding these observations. Therefore, a phase I/II clinical trial investigating this novel combination with extensive biological correlatives is in development.
In line with the shift toward precision medicine, recent genomic analysis of primary DLBCL patient samples has led to two new proposed DLBCL classification systems, including an “EZB signature” characterized by EZH2 mutations and BCL2 translocations (49), and a “cluster 3” subgroup identified by BCL2 mutations in conjunction with KMT2D, CREBBP, and EZH2 dysregulation (50). The identification of a DLBCL molecular subtype, predominately of GC origin, characterized by EZH2 mutations and BCL2 abnormalities in conjunction with the data presented here suggests that the addition of a BCL2 inhibitor to the combined inhibition of EZH2 and HDACs may be synergistic. This ultimately requires further investigation.
Our findings provide the biological rationale and lay the groundwork for a future clinical trial of targeted epigenetic therapy in GC-DLBCL. The combination of dual EZH2 and HDAC inhibition may potentially serve as a precision medicine therapeutic platform in lymphomas derived from the GC and those harboring an epigenetically repressed transcriptional state.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: J.K. Lue, S.A. Prabhu, E.I. Chen, S. Cremers, N.L. Kelleher, J.E. Amengual
Development of methodology: J.K. Lue, S.A. Prabhu, Y. Liu, C. Qiao, R. Nandakumar, J.E. Amengual
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.K. Lue, S.A. Prabhu, Y. Gonzalez, N. Abshiru, J.M. Camarillo, S. Mehta, E.I. Chen, C. Qiao, R. Nandakumar, S. Cremers, J.E. Amengual
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.K. Lue, S.A. Prabhu, Y. Liu, A. Verma, P.S. Mundi, J.M. Camarillo, E.I. Chen, R. Nandakumar, S. Cremers, O. Elemento, J.E. Amengual
Writing, review, and/or revision of the manuscript: J.K. Lue, S.A. Prabhu, Y. Liu, P.S. Mundi, J.M. Camarillo, E.I. Chen, R. Nandakumar, N.L. Kelleher, J.E. Amengual
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.K. Lue, Y. Gonzalez, J.E. Amengual
Study supervision: N.L. Kelleher, J.E. Amengual
Others (laboratory technician): Y. Gonzalez
This work was supported by the American Society of Clinical Oncology (ASCO) Young Investigator Award (J.K. Lue), SWOG HOPE Foundation SEED Fund (J.E. Amengual), the National Resource for Translational and Developmental Proteomics under Grant P41 GM108569 from the National Institute of General Medical Sciences, NIH (N.L. Kelleher), the Sherman Fairchild Foundation (N.L. Kelleher), and the Proteomics Shared Resource of the Herbert Irving Comprehensive Cancer Center/Columbia University Irving Medical Center (P30CA013696). We would like to also acknowledge the Lymphoma Research Fund of Columbia University for its generous support.