Acquired Resistance to EZH2 Inhibitor GSK343 Promotes the Differentiation of Human DLBCL Cell Lines toward an ABC-Like Phenotype

Diffuse large B-cell lymphoma (DLBCL) is the most common form of non-Hodgkin's lymphoma (NHL). Because of its aggressive nature, 30% to 40% of DLBCL will relapse or progress, with most relapses occurring in the first 2 years (1, 2), and up to 90% of patients with relapsed/refractory DLBCL (rrDLBCL) will eventually succumb to the disease (2). This remains an unmet clinical challenge and new therapeutic options are needed for this group of patients.

DLBCL is a heterogeneous disease, both phenotypically and molecularly. The cell-of-origin (COO) molecular subtypes of DLBCL include the germinal center B-cell (GCB) and activated B-cell (ABC) groups, with the latter having a worse response to therapy and overall survival (2–4). GCB versus ABC subtypes have distinct origins along the B-cell pathway; the GCB subtype arises from germinal center centroblasts, whereas the ABC subtype is derived from post-germinal center plasmablasts (5, 6). Efforts to profile gene expression and mutational differences between these subtypes have underlined their heterogeneity and differential response to chemotherapy (7–12). Although our understanding of DLBCL classification is continually updated, the most recent large-scale studies have suggested a novel taxonomy of DLBCL cases, based on stratifying patients into one of seven genetic subtypes (13, 14). Several recurrent mutations have been identified in DLBCL cases, with many affecting genes encoding histone methylsferases (HMT) and histone acetyltransferases (HAT; refs. 8, 10, 15, 16). These enzyme groups are key components of the machinery that controls epigenetic regulation of gene expression via changes in the chromatin landscape. This emphasizes the potential importance of epigenetics in contributing to a malignant DLBCL phenotype.

One of the most frequent mutations in GCB-type DLBCL affects tyrosine 641 (Y641) of the histone-lysine N-methyltransferase enhancer of zeste homolog 2 (EZH2; ref. 16). Mutations in EZH2 occur in approximately 22% and 1.7% of GCB and ABC DLBCL cases, respectively (13), and in approximately 45% of patients with DLBCL in the EZB genetic subtype (14). This activating mutation results in the upregulation of global H3K27 histone trimethylation (H3K27me³), thereby repressing genes essential for B-cell development as methylation renders chromatin less accessible to transcriptional machinery. This alteration in B-cell development leads to the expansion of B cells at the germinal center stage that can lead to malignancy (17). Overexpression of wild-type EZH2, as observed in certain cancers, has also been shown to increase H3K27me³ levels (18). Several small-molecule inhibitors of EZH2 (EZH2i) have been created with high specificity for EZH2, including GSK343, GSK126, and EPZ6438 (19–21). Promising preclinical research in DLBCL and follicular lymphoma (FL) models has shown that EZH2i strongly inhibits tumor cell proliferation both in vitro and in vivo (17, 22), with EPZ6438 (tazemetostat) currently in a Phase II clinical trial for patients with DLBCL and FL in combination with R-CHOP (NCT02889523). In Phase I trials, EPZ6438 showed clinical activity as a monotherapy in heavily pretreated, patients with refractory DLBCL (23), and encouraging clinical efficacy in combination with R-CHOP in newly diagnosed patients with DLBCL (24). In addition, EPZ6438 has shown antitumor activity in patients with solid tumors that harbor mutations in SWI/SNF subunits, leading to an oncogenic dependency on EZH2 (25).

The acquisition of resistance mechanisms to any targeted therapy has been seen in a majority of patients with cancer, thereby limiting treatment efficacy (26). This remains a clinically significant problem that has been largely unstudied in the context of EZH2 inhibitors. Our research group has previously implicated chromatin remodeling in mediating therapeutic resistance in models of acute promyelocytic leukemia (27, 28). We therefore wanted to investigate EZH2i resistance in DLBCL, aiming to better inform clinical use of EZH2i in combination with novel agents. Here, we have developed an in vitro DLBCL model of EZH2i resistance and profiled the changes in genetic and epigenetic landscapes that develop because of this resistance. We observe that chronic EZH2i treatment leads to the differentiation of DLBCL cells, which has the potential to expand the limited therapeutic opportunities available for patients with rrDLBCL.

WSU-DLCL2 (RRID:CVCL_1902), SU-DHL-6 (RRID:CVCL_2206), and OCI-LY7 (RRID:CVCL_1881) were purchased from the ATCC. Human DLBCL cells were maintained in RPMI-1640 with 10% FBS and antibiotics at 37°C with 5% CO₂ as well as their GSK343-resistant clones (see below). All cell lines were passaged twice between thawing and use in experiments, and routinely tested to confirm absence of Mycoplasma contamination using e-Myco VALiD Mycoplasma PCR Detection Kit (FroggaBio). Culture media, FBS and antibiotics were purchased from Wisent. Drug information is listed in Supplementary Table S1. Diluted compounds were kept at −80°C.

The same protocol was followed for WSU-DLCL2, SU-DHL-6, and OCI-LY7 cells. Cells were seeded at 0.3×10⁶ cells per mL and maintained in 0.5 μmol/L GSK343 (19) for 14 days. Cells were seeded back at 0.3×10⁶ cells per mL when they reached confluence. Doses of GSK343 were slowly increased to reach a concentration equal or higher to twice the estimated GI₅₀. Targeted concentration values of 4, 5, and 6 μmol/L were reached for WSU-DLCL2, SU-DHL-6, and OCI-LY7, respectively. The polyclonal population obtained from that process was separated into monoclonal populations by sequential dilution. All resistant clones were maintained in full media supplemented every second day with GSK343. The resistant cell lines, WSU-R4, SU-DHL-6-R5, and LY7-R6, were named according to their parental counterpart and the GSK343 concentration they are maintained in. When required, the resistant cells were grown without the presence of GSK343 for 7 days and labeled as wash-off.

Western blotting was performed as previously reported (28). Detailed antibody information is listed in Supplementary Table S2. See Supplementary Materials and Methods for protein extraction methods.

Differential gene expression analysis for all cell lines was performed as previously reported (29); see Supplementary Materials and Methods for details. For functional enrichment analyses, goseq v1.30.0 was used. Background was matched to the differentially expressed (DE) gene list using the genefinder command of the genefilter v1.60.0 package. A weight of 20 and the Manhattan method was used to recruit at least 10 background genes for each DE gene. Then the weighting function of goseq was applied using gene length as the bias against which to normalize. The analysis was conducted through gene ontology (GO) and Transfac and Jaspar. The latter required the creation of a many-to-many library from Enrichr libraries (https://amp.pharm.mssm.edu/Enrichr). Transformations were made using bash commands and dplyr v0.7.4 package. Dot and bar plots were designed using ggplot2 v2.2.1 package (RRID:SCR_014601). Gene set enrichment analysis (GSEA, RRID:SCR_003199) for DE genes, ranked by log fold change, was performed using the R package fgsea, with the Molecular Signatures Database (MSigDB) v7.4. Plots were generated using the built-in plotting function.

Lymphoma/leukemia molecular profiling project (LLMPP) score was calculated from our RNA-seq data to classify our different models of human lymphoid malignancies in molecular terms (30). The predictive model genes are listed at: http://llmpp.nih.gov/DLBCLpredictor/LLMPP_Model.txt. The data were first normalized to gene length as reads or fragments per kilobase of transcript per million mapped reads (RPKM or FPKM). The log₂ normalized data were then manipulated to fit micro-array data: Mean was adjusted to equal 0 and standard deviation to equal 1. The adjusted data were then used to calculate the LLMPP score as the sum of the products of adjusted data and the LLMPP coefficient for each gene present in both our assay and in the LLMPP list. Data were plotted showing the 10^th and 90^th percentile as marks for the ABC and GCB signature. All data manipulations were performed in the R statistical computing environment v3.4.3 (http://www.R-project.org) using DESeq2 v1.18.1, data.table_1.10.4–3 and dplyr_0.7.4 packages (RRID:SCR_000154).

RNA was extracted from cells using Ribozol RNA Extraction Reagent (Amresco). cDNA was generated from 1 μg total RNA using the iScript cDNA Synthesis Kit (Bio-Rad Laboratories). qPCR was performed as previously described (31). Primers are listed in Supplementary Table S3.

For cell viability assays: cell death was assessed by flow cytometry using amine reactive LIVE/DEAD Fixable Aqua Dead Cell Stain Kit (Invitrogen) on a suspension of one million cells. For antibody-based assays: Following viability staining, cells were washed in cold PBS and incubated for 30 minutes on ice in blocking solution (1% human FcBlock, BD Biosciences) diluted in FACS buffer (PBS, 5% FBS, 0.1% sodium azide). Staining solution was then added to each sample (see Supplementary Table S2 for antibodies used) and incubated for 30 minutes on ice protected from light. Data were acquired on a LSRFortessa (BD Biosciences) after compensation calculation using DIVA software. Analysis was done using FlowJo v10 (BD Biosciences; RRID:SCR_008520).

A total of 1×10⁶ cells were lysed for RNA extraction using TRIzol (Ambion by Life Technologies). cDNA was synthesized from 1 μg RNA using the High-Capacity cDNA Reverse Transcription Kit (Life Technologies). A PCR reaction was performed using XBP1 primers that encompass the splice sites under the following conditions: 94°C for 3 minutes, 30 cycles of 94°C for 30 seconds, 58°C for 30 seconds, 72°C for 3 minutes, and 72°C for 3 minutes. The PCR fragments were resolved on a 2.5% agarose-1,000 (Invitrogen) gel. Primer sequences are listed in Supplementary Table S3.

Chromatin immunoprecipitation (ChIP) analyses were performed as previously reported (28); see Supplementary Materials and Methods for details.

Use of primary DLBCL samples for this project was approved by the Research Ethics Board of the Lady Davis Institute protocols (11–047 and 12–052) and is in accordance with the declaration of Helsinki. Informed written consent was obtained from each patient. One-millimeter cores of primary diagnostic DLBCL formalin-fixed paraffin-embedded tissue were used to construct a tissue micro-array. Patients were all treated with curative intent with R-CHOP-like regimens. 51 cores (1 per patient) were stained for SLAMF7 (see Supplementary Table S2 for details) at a 1:100 dilution, followed by detection with ImmPACT DAB substrate (Vector Laboratories). Hematoxylin-counterstained samples were then mounted with coverslips. Staining intensity was determined by H score using QuPath software v0.1.2 (RRID:SCR_018257). Threshold values of 0.2, 0.3, and 0.45 DAB OD were used for scores of 1, 2, and 3, respectively. COO calls were made by clinical pathologists at the Jewish General Hospital (Montréal, Canada) using Hans algorithm. Correlation analyses were performed using Prism 6 software.

Prism 6 software (RRID:SCR_005375) or R was used to perform statistical analysis. The significance of differences between groups was determined using the Mann–Whitney test, one-way ANOVA, or two-way ANOVA (with multiple comparison tests) as appropriate. P values of <0.05 were considered significant.

The data generated in this study are available within the article and its Supplementary Files. The sequencing data from this study are publicly available in SRA at PRJNA770766.

To model resistance mechanisms, we developed EZH2i-resistant cell sub-lines. The parental WSU-DLCL2 cell line, which contains the Y641 EZH2 mutation, is sensitive to EZH2 inhibition as shown by strongly reduced proliferation rates in the presence of drug (Fig. 1A). Cells were continuously exposed to increasing doses of GSK343, until reaching a concentration equal or higher to twice the estimated GI₅₀. Through this process, a resistant cell sub-population emerged. Monoclonal populations were isolated from this cell pool, and we chose to move forward with one of these as our resistant cell line, named WSU-R4. These resistant cells can survive and proliferate in the presence of EZH2i (Fig. 1A). Importantly, WSU-R4 proliferation rates remain the same despite drug wash-off for 7 days (Fig. 1A). As expected, treatment with GSK343 in the parental cell line led to a significant reduction in H3K27me³, alongside a moderate decrease in EZH2 protein levels (Fig. 1B). The H3K27me³ level was similarly decreased in WSU-R4–resistant cells in the presence of drug, without a concurrent decrease in EZH2, and H3K27me³ levels were restored upon wash-off of GSK343 (Fig. 1B). This indicates that the acquired resistance to the growth inhibitory effects of EZH2i is independent of H3K27me³. Furthermore, these cells maintain their resistance to EZH2i in response to two other inhibitors—GSK126 (Fig. 1C) and EPZ6438 (Fig. 1D)—suggesting the establishment of a common mechanism of cross-resistance to multiple inhibitors targeting EZH2 due to GSK343 treatment alone.

We subsequently produced two other EZH2i-resistant DLBCL cell lines using the same procedure. Parental SU-DHL-6 and OCI-LY7 cell lines are both sensitive to EZH2i and yielded resistant sub-populations upon continued GSK343 exposure (Supplementary Fig. S1A and S1B). Notably, inhibitor-induced proliferation arrest is independent of EZH2 mutation status; although SU-DHL-6 cell lines also harbor the Y641 EZH2 mutant, OCI-LY7 express only wild-type EZH2. Upon drug wash-off, however, both resistant sub-populations increase proliferation rates, indicating that the full growth inhibition remains dependent on the drug.

Epigenetic modifiers function to remodel chromatin primarily by targeting the methylation and acetylation of histones and DNA. We wanted to confirm that the development of resistance to EZH2i (inhibition of an epigenetic writer) did not alter cross-sensitivity to inhibitors of other epigenetic regulatory processes. Accordingly, no difference was observed between the sensitivity of WSU-DLCL2 and WSU-R4 cell lines to treatment with vorinostat, an inhibitor of histone acetylation (Fig. 2A). Similar results were observed for SU-DHL-6 and OCI-LY7 EZH2i-resistant lines (Supplementary Fig. S2A and S2B). In addition, cell viability remained largely consistent between WSU-DLCL2 and WSU-R4 cells upon treatment with increasing concentrations of azacitidine, an analogue of cytosine that incorporates into DNA (Fig. 2B). Furthermore, we saw no difference in cell viability between parental and EZH2i-resistant WSU-DLCL2 cells in the presence of varying doses of JQ1, a potent inhibitor of BET family proteins, including BRD2, BRD3 and BRD4 (Fig. 2C). Our SU-DHL-6 and OCI-LY7 EZH2i-sensitive and resistant cells similarly displayed no difference in their sensitivity to azacitadine or JQ1 (Supplementary Fig. S2C–S2F). Together these data suggest that the development of EZH2i resistance does not alter sensitivity to other inhibitors of epigenetic modifiers. Interestingly, we did observe that the development of EZH2i resistance is associated with changes in sensitivity to the chemotherapy drug doxorubicin. WSU-R4 cells are significantly more sensitive to treatment with mid-range concentrations of doxorubicin (0.5–1 μmol/L), compared with WSU-DLCL2 cells, as measured by increased cell death in the presence of drug (Fig. 2D).

As EZH2 inhibitors are believed to work mainly by modulating gene expression, we wanted to better understand the changes occurring in the transcriptional landscape as a response to acquired EZH2i resistance. We, therefore, used RNA-sequencing (RNA-seq) to compare the transcriptomes of parental and resistant WSU-DLCL2 cell lines, treated with GSK343 or not. Principal component analysis (PCA) revealed that when separated by the two most prominent transcriptional variables of the WSU-DLCL2 RNA-seq, the conditions under review were distinct. These conditions were: Untreated WSU-DLCL2 (parental), WSU-DLCL2 treated for 7 days with GSK343, WSU-R4 (resistant) and WSU-R4 wash-off (Fig. 3A). Interestingly, the transcriptomes of WSU-R4 cells in the presence of drug and after 7 days of wash-off were largely similar, helping to confirm that our resistant sub-line does not depend on the presence of drug for their resistance to EZH2 inhibition (Fig. 3A and B). Similar PCA plots were created for RNA-seq completed on SU-DHL-6 and OCI-LY7 parental and resistant cell lines (Supplementary Fig. S3A and S3B), although in these models we observed a bigger difference in the transcriptomes of resistant and wash-off conditions, which supports our conclusion that complete inhibition of proliferation in these resistant cell lines remains dependent on the presence of GSK343 (Supplementary Fig. S1A and S1B).

To begin profiling the transcriptional changes that correlate with EZH2i resistance, GO term analysis was performed on a set of genes that were shown to have an altered expression between WSU-DLCL2 and WSU-R4 cells (Fig. 3B, red circle). Of particular importance, we observed that EZH2i-resistant cells had a significant enrichment in genes associated with cell differentiation and development (Fig. 3C). Encouragingly, equivalent analysis of the SU-DHL-6 and OCI-LY7 RNA-seq datasets also revealed that resistance to EZH2i is associated with a significant enrichment in cell differentiation in both models (Supplementary Fig. S3C and S3D), as well as cell activation (Supplementary Fig. S3C), and cellular development (Supplementary Fig. S3D). We further wanted to elucidate potential differences in transcription factor binding profiles between EZH2i-sensitive and -resistant groups. To this end, we analyzed our RNA-seq data through TRANSFAC and JASPAR libraries. From this, we noted a significant enrichment of the E2F4-binding site profile in our WSU-R4–resistant cell line (Fig. 3D) and SU-DHL-6-R5–resistant cells (Supplementary Fig. S3E). The E2F4 transcription factor plays a crucial role in cell-cycle progression (32), and it is known that primary B cells from E2F4 knockout mice are less able to differentiate (33). In our OCI-LY7 data, however, we did not find a significant enrichment of any transcription factor–binding site profiles (Supplementary Fig. S3F). Taken together, we interpret the upregulation of these processes in EZH2i-resistant cells as differentiation along the B-cell lineage, likely toward a more ABC-like state.

To provide support to the conclusions drawn from our RNA-seq, we chose to pursue validation of an ABC-like cell status, focusing on WSU-DLCL2 and WSU-R4 cell lines. First, analysis of our RNA-seq revealed a decrease in the GCB-associated genes BCL6 and MME (CD10) in WSU-R4 cells compared with WSU-DLCL2, with a parallel increase in the ABC-associated genes PIM1 and ENTPD1 (CD39; Fig. 4A). Second, the expression of the B-cell surface marker CD20 is known to decrease upon differentiation of B cells to plasma cells, and more recently has been shown to have reduced expression in ABC DLBCL samples compared with GCB (34). As such, we confirmed by flow cytometry that the surface expression of CD20 decreases significantly upon development of EZH2i resistance (Fig. 4B). Using GSEA, we further confirmed a strong positive upregulation of a plasma cell gene signature in our WSU-R4 cells (Fig. 4C). Plasma cells have high levels of proteotoxic stress, and elevated unfolded protein response (UPR) signaling, a gene signature that we also show as being positively enriched in WSU-R4 cells (Fig. 4D). This state of heightened stress makes plasma cells more sensitive to treatment with proteasome inhibitors (such as bortezomib) or drugs that increase proteotoxic stress (such as lenalidomide; ref. 35). Accordingly, we observed that our WSU-R4 cells are more sensitive than WSU-DLCL2 cells to treatment with bortezomib (Fig. 4E), as well as having an increased sensitivity to lenalidomide (Fig. 4F). Mechanistically, the transcription factor XBP1 is differentially spliced by IRE1 in response to ER/UPR stress and is known to be essential for plasma cell differentiation and immunoglobulin production (36). We observed an increase in relative expression of the spliced (activated) form of XPB1 in WSU-R4 cells as compared with WSU-DLCL2 (Fig. 4G). In parallel, we detected an increased expression of phospho-IRE1 in our EZH2i-resistant WSU-R4 cells (Fig. 4H); this phosphorylation event on Ser724 is associated with an increase in the RNase activity of IRE1 (37). In addition, we showed that our WSU-R4 cells have increased levels of the NFκB p65 subunit (Fig. 4I), indicating increased activity of NFκB signaling, which is required for B-cell pro-survival and maturation signals (38). Surprisingly, we saw no significant change in intracellular IgM production in our WSU-R4 cells compared with WSU-DLCL2, but a modest increase upon drug wash-off (Fig. 4J). Finally, CD79B, which encodes a subunit of the B-cell receptor, is frequently mutated in ABC-DLBCLs, often leading to an amplification of expression (13). Furthermore, a higher CD79B expression has been shown to mediate resistance to ibrutinib—a selective inhibitor of Bruton's tyrosine kinase that sits downstream of the B-cell receptor (39). Encouragingly, our WSU-R4 cells have significantly higher surface expression of CD79B compared with WSU-DLCL2 cells (Fig. 4K), as well as increased resistance to ibrutinib-induced cell death (Fig. 4L). Together, these data provide strong evidence that WSU-R4 EZH2i-resistant cells have differentiated from a GCB-like to a plasmablast (ABC)-like state.

Given this conclusion, we opted to characterize the gene expression data of each of our cell lines according to the COO hierarchical clustering algorithm established by the LLMPP. This algorithm determines whether a given sample is GCB versus ABC, based on the expression profile of a set of 27 previously described genes (30). As expected, we confirmed that parental WSU-DLCL2 are GCB-like and found that the WSU-R4 sub-population has strong expression of markers of an ABC-like phenotype (Fig. 4M). Remarkably, the SU-DHL-6 cell lines follow this trend, with the EZH2i-resistant cells developing an ABC-like transcriptional profile whereas the parental and drug-sensitive lines are GCB/GCB-like (Fig. 4M). Interestingly, although our OCI-LY7 EZH2i-resistant cells also shift toward an ABC-like profile, treatment of the parental cells with GSK343 is sufficient to push them toward a more differentiated state (Fig. 4M). These data suggest that chronic exposure to EZH2i results in the differentiation of DLBCL cells.

To begin to illuminate the mechanism behind resistant cell plasmablast differentiation, we mined our RNA-seq dataset for promising targets. One such target is SLAMF7 (signaling lymphocytic activation molecule F7). This cell surface protein is heavily expressed on multiple myeloma (MM) cells (40), and treatment with the anti-SLAMF7 antibody elotuzumab has shown potent efficacy against MM in the clinic (41, 42). Our sequencing data show that SLAMF7 is upregulated in resistant cells, with qPCR confirming this at the mRNA level (Fig. 5A). Flow cytometry (Fig. 5B) and Western blotting (Fig. 5C) validate that SLAMF7 protein levels are increased in EZH2i-resistant cells and remain high even after drug wash-off for 7 days.

High-resolution ChIP tiling of the SLAMF7 locus shows high levels of H3K27me³ histone methylation in WSU-DLCL2 cells (Fig. 5D). This is indicative of a chromatin landscape that is transcriptionally non-permissive, which explains the comparatively low-expression levels of SLAMF7 in parental cells (Fig. 5B and C). As expected, treatment with GSK343 in sensitive cells decreases this histone methylation mark, whereas there was no change in H3K27 acetylation levels (H3K27ac; Fig. 5D). SLAMF7 expression, therefore, remains low in the parental group upon EZH2 inhibition. In the resistant cells, however, there was both a decrease in H3K27me³, and an increase in H3K27ac (Fig. 5D), an epigenetic mark that is linked to transcriptional activation. The net result is the switch from a non-permissive to permissive chromatin landscape to allow transcription of the SLAMF7 locus. In agreement with this model, flow cytometry profiling of a selection of human DLBCL cell lines showed us that the expression of SLAMF7 is higher in ABC cell lines as compared with GCB cell lines (Fig. 5E). In addition, we performed SLAMF7 IHC staining on GCB versus non-GCB DLBCL patient tissue (Fig. 5F). Overall, expression was variable between patients; H scores ranged from 7 to 203 (Supplementary Fig. S4A). When stratifying by subtype, we observed that SLAMF7 expression was significantly higher in non-GCB samples compared with GCB patient samples (Fig. 5G), but only in patients that were responsive to treatment. Curiously, the differential expression of SLAMF7 between subtypes was lost in patients with relapsed disease (Fig. 5H). Furthermore, we observed no significant correlation between SLAMF7 and patient survival, either overall or when stratified by subtype (Supplementary Fig. S4B–S4F), although there is a trending negative correlation between SLAMF7 expression and overall survival in treatment-responsive patients (Supplementary Fig. S4C).

The development of acquired mechanisms of resistance is inevitable for many targeted therapies. In this study, we chose to investigate the establishment of such mechanisms against a group of epigenetic modulatory therapies—EZH2 inhibitors—specifically in DLBCL. Through the establishment and characterization of an in vitro resistance model, we have shown that chronic exposure to EZH2i causes differentiation along the B-cell pathway toward a plasmablast-like state, concomitant with an upregulation of SLAMF7.

It is highly unlikely that there will be one common mechanism underpinning the emergence of drug resistance between cell lines. This creates a problem when trying to target resistance mechanisms clinically. It has been shown that the development of resistance to EZH2i occurs via two broad mechanisms (43); (i) increased signaling through the pro-survival PI3K, MAPK, and IGF-1R pathways, and (ii) the development of secondary EZH2 mutations that interfere with drug-target binding. The specific pathways and mutations implicated, however, are dependent on the choice of drug and cell line treated. Our RNA-seq data show that WSU-DLCL2, SU-DHL-6, and OCI-LY7 cell lines develop a pro-differentiation transcriptome upon continued exposure to GSK343. Interestingly, we have observed that EZH2 mutational status may impact transcriptomic changes in response to EZH2 inhibition. In our parental cell lines with mutant EZH2, we observe a gradual change in differentiation in the presence of GSK343, which is furthered as these cells develop resistance to EZH2i. The OCI-LY7 parental cells, however, which express wild-type EZH2, differentiate much further toward an ABC-like state upon treatment with GSK343. Furthermore, the maturation state of the OCI-LY7 cells does not change between GSK343-sensitive and -resistant cells (Fig. 4M). It may be that the presence of GSK343 is sufficient to induce differentiation in cells that harbor wild-type EZH2, whereas the adequate reduction of EZH2 activity in cells with an activating mutation requires developed resistance. Our data also suggest that E2F4 may play a role in the differential response of EZH2 wild-type and mutant cells to EZH2i, although likely not in the development of resistance, as we observed enrichment in its binding profile only in our mutant EZH2 models. There is precedent in the literature that EZH2 mutation status can correlate with response to EZH2i, as recent evidence shows that mutant EZH2 NHL cell lines treated with EPZ6438 show a cytotoxic response, compared with a cytostatic response in wild-type EZH2 cells (44). Whether the Y641 mutation is causative or correlative in influencing the changes we see in our model systems remains unknown.

Our studies show that this altered transcription profile between parental and resistant cell lines pushes the WSU-R4, SU-DHL-6-R5, and OCI-LY7-R6 EZH2i-resistant cells toward a plasmablast-like state, displaying altered protein expression profiles, signaling pathway activity and drug sensitivity. Although patients with ABC-DLBCL have a poorer prognosis compared with those with GCB-DLBCL, this mechanism of escape from EZH2i may be exploited therapeutically by taking advantage of the fact that plasma cells exist under higher levels of physiological stress. We have shown that our EZH2i-resistant WSU-R4 cells are more sensitive to treatment with bortezomib (Fig. 4E), a proteasome inhibitor, or lenalidomide (Fig. 4F), which disrupts the substrate specificity of E3 ubiquitin ligase complexes (45). In support of this, our group has also previously shown that U937 cells with acquired resistance to vorinostat develop an activation of the UPR, exhibit marks of ER stress and have increased sensitivity toward bortezomib (46). Excitingly, EPZ6438 is currently in a Phase III clinical trial alongside rituximab and lenalidomide in patients with FL (NCT04224493). Combination therapy of an EZH2 inhibitor with drugs targeting the proteasome may, therefore, prove to be a clinically effective treatment strategy, although reduced expression of CD20 following EZH2i (Fig. 4B) may limit the use of its combination with rituximab.

Aside from differences in sensitivity to proteasome inhibitors, we demonstrate how epigenetic regulation can change chemotherapeutic response in vitro. Our WSU-R4 cells become increasingly sensitive to treatment with doxorubicin when compared with parental cells (Fig. 2D). This finding becomes even more striking when compared with the lack of difference in sensitivity to other classes of drugs that target epigenetic modulators. Detailed in vivo modeling is certainly needed to determine whether this relationship between EZH2i resistance and sensitivity to doxorubicin exists beyond in vitro culture. In addition, such experiments should be expanded to examine the potential for altered responses of EZH2i-resistant DLBCL cells to other chemotherapeutic drugs. Importantly, altered sensitivity to doxorubicin may have direct clinical relevance to patients with FL who relapse following treatment with a combination of EPZ6438, lenalidomide and rituximab (NCT04224493).

In addition, we have shown that WSU-R4, and some established ABC-DLBCL cell lines, have increased expression of SLAMF7. Although we have traced this change in expression to alterations in the chromatin landscape surrounding the SLAMF7 locus, it is still unknown whether changes in SLAMF7 directly influence DLBCL cell differentiation. To help answer this question, future efforts should be directed at illuminating the mechanistic details of how EZH2i resistance leads to the increased ability to acetylate chromatin. Continued investigation is also required to determine whether similar switches from non-permissive to permissive epigenetic marks are occurring at other gene loci aside from SLAMF7. It is worth noting that the development of EZH2i resistance in our WSU-R4 cells was not completely reversible, as removal of GSK343 did not lead to reversion to drug sensitivity. Similarly, our group has previously shown that resistance to HDAC inhibitors in U937 cells was not completely reversible upon HDACi wash-off (47). How the development of resistance to therapies targeting epigenetic regulators leads to irreversible changes in gene expression remains unknown.

Promisingly, the results of this study implicate a previously untapped combination of clinically targetable axes. SLAMF7 is a surface homotypic receptor that can regulate immune cell functions via recruitment of various SH2 domain-containing proteins to its cytoplasmic switch motifs (48). Increasing amounts of evidence suggest that SLAMF7, as well as other members of the SLAM family, play crucial roles in mediating tumor-immune crosstalk in the tumor microenvironment. For instance, recent literature has highlighted a role for SLAMF7 expression on tumor-associated macrophages in augmenting CD8⁺ T-cell exhaustion in a model of clear cell renal cell carcinoma (49). In addition, the development of elotuzumab, a successful monoclonal antibody therapy targeting SLAMF7, has shown encouraging clinical response in patients with MM via indirect activation of NK cell–mediated killing (50). Its efficacy in DLBCL, though, remains untested. As chronic EZH2i treatment induces an upregulation in SLAMF7 alongside B-cell differentiation, our resistant cell lines may be uniquely primed for targeting by elotuzumab. Furthermore, our patient data demonstrate that SLAMF7 expression is upregulated in non-GCB compared with GCB DLBCL samples (Fig. 5G), although this may be limited to patients who are successful R-CHOP responders. This strengthens the argument that DLBCLs arising from further along the B-cell differentiation pathway may be more susceptible to therapies against SLAMF7. We will continue to verify SLAMF7 as a molecular target in DLBCL, with a specific focus on assessing the impact of a therapeutic “double hit” via EZH2 and SLAMF7 dual targeting.

Collectively, this study characterizes a previously undocumented response to EZH2 inhibitor therapy in DLBCL cell lines. On the basis of our results, we implicate a novel therapeutic strategy that warrants further investigation in the context of relapsed/refractory DLBCL.

F. Pettersson reports personal fees from Bristol Myers Squibb outside the submitted work; and reports employment with Bristol Myers Squibb, Canada. N.A. Johnson reports other support from Epizyme outside the submitted work. J.D. Licht reports grants from Celgene and Epizyme outside the submitted work; as well as reports a patent for WO2015143424A3 pending to Glaxo Smith Kline. W.H. Miller Jr reports personal fees from Merck, Bristol Myers Squibb, Roche, Novartis, Amgen, and GSK outside the submitted work. No disclosures were reported by the other authors.

S.E.J. Preston: Conceptualization, investigation, writing–original draft, project administration, writing–review and editing. A. Emond: Investigation, methodology. F. Pettersson: Investigation, methodology. D. Dupéré-Richer: Formal analysis, investigation. M.J. Abraham: Investigation. A. Riva: Formal analysis, investigation. M. Kinal: Investigation. R.N. Rys: Investigation. N.A. Johnson: Investigation, writing–review and editing. K.K. Mann: Conceptualization, supervision, writing–review and editing. S.V. del Rincon: Conceptualization, supervision, funding acquisition, writing–review and editing. J.D. Licht: Conceptualization, supervision, funding acquisition, writing–review and editing. W.H. Miller Jr: Conceptualization, supervision, funding acquisition, writing–review and editing.

We kindly thank Dr. J. Nichol for her role in project management. We appreciate the assistance of Dr. K.K. Oros in troubleshooting our LLMPP analysis. We further thank Christian Young (Lady Davis Institute Flow Core Facility) for his continued help and guidance. This research is funded by grants from the NIH Research Project Grant Program (R01 CA180745; to J.D. Licht), Leukemia and Lymphoma Society Specialized Center of Excellence (to J.D. Licht), Leukemia and Lymphoma Society Special Fellow Award (to D. Dupéré-Richer), Samuel Waxman Cancer Research Foundation (to W.H. Miller Jr and J.D. Licht) and a Cole Foundation Transition Award (to S.V. del Rincon). S.E.J. Preston and M.J. Abraham were supported by a fellowship from the Fonds de recherche du Québec—Santé.

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.