EZH2 Inhibition and Myeloid Risk: A Feature of On-Target Biology

Abstract

In March 2026, Ipsen voluntarily withdrew the EZH2 inhibitor tazemetostat (Tazverik) from all markets and indications following results from the confirmatory Phase Ib/III SYMPHONY-1 trial demonstrating hematologic second primary malignancies (SPMs) in 5.7% of treated patients, compared with none in the control arm. The most frequently reported SPMs were myelodysplastic syndrome (MDS) and acute myeloid leukemia (AML). Although this finding may be interpreted as an idiosyncratic liability of a single agent, the biology of EZH2 suggests a broader, target-driven risk.

EZH2 functions in a context-dependent manner as both an oncogene and a tumor suppressor. Gain-of-function mutations drive transformation in germinal center B-cell lymphomas, whereas loss-of-function alterations impair hematopoietic stem cell differentiation and are recurrent in myeloid malignancies, including MDS and AML. In selected solid tumor contexts, such as specific subtypes of medulloblastoma, EZH2 loss can likewise promote tumorigenesis. These observations raise concern that chronic pharmacologic inhibition of EZH2 may phenocopy loss-of-function states that initiate myeloid neoplasia.

We argue that the withdrawal of tazemetostat exposes a fundamental vulnerability in the development of epigenetic therapies: target validation has focused on tumor-intrinsic effects while underweighting consequences in normal stem cell compartments. This issue is particularly relevant for ongoing Phase III programs in prostate cancer, where prolonged exposure may amplify latency-dependent risks. We propose changes to preclinical safety assessment, trial design, and pharmacovigilance to mitigate predictable, target-mediated toxicities.

Introduction

On March 9, 2026, Ipsen announced the voluntary global withdrawal of tazemetostat (Tazverik) across all indications including relapsed/refractory follicular lymphoma (FL) and epithelioid sarcoma citing emerging data from the confirmatory Phase Ib/III SYMPHONY-1 trial of tazemetostat plus lenalidomide (revlimid) and rituximab (R²) versus R² alone (1). The trigger for the withdrawal was unambiguous and serious. As of the March 6, 2026, it was reported that 18 of 318 patients (5.7%) treated with tazemetostat developed hematologic second primary malignancies (SPMs), predominantly myelodysplastic syndrome (MDS) and acute myeloid leukemia (AML), with additional cases of B-cell acute lymphoblastic leukemia and clonal cytopenia of undetermined significance, compared with zero events in the control arm (2). Three patients died and 14 had unresolved disease. Critically, SPMs emerged as early as 7.5 months, clustered after one to three years of exposure (median treatment duration 15.8 months), and occurred in some patients after treatment had stopped (2). At accelerated approval in 2020, SPMs had been a labeled risk, but at an estimated incidence of roughly 1.7% (2,3); this confirmatory trial revealed a rate more than threefold higher.

Several caveats temper interpretation of this signal. The per-arm distribution of prior genotoxic exposure, including alkylating agents and topoisomerase-II inhibitors which are known contributors to therapy-related myeloid neoplasia has not been reported for SYMPHONY-1. Therefore, the contribution of prior therapy cannot be disentangled from any drug-specific effects in SYMPHONY-1. Notably, no SPMs were reported in the control arm, whereas background rates of therapy-related MDS/AML in relapsed/refractory FL have been estimated at approximately 1–2% (4); a zero-event comparator is therefore lower than might be anticipated and may reflect between-arm imbalances in prior therapy, age, or follow-up rather than a true absence of risk. For the same reasons, the apparent difference from the ~1.7% incidence recognized at accelerated approval derives from distinct studies, populations, regimens, and durations of follow-up and should be regarded as a descriptive observation rather than a controlled, quantitative comparison of relative risk. These uncertainties should temper any causal attribution to EZH2 inhibition alone, and we join others in urging timely, transparent publication of the complete SYMPHONY-1 safety dataset.

Tazemetostat was the first EZH2 inhibitor approved for clinical use, so the signal reverberates across the entire enhancer of zeste homolog 2 (EZH2)/Polycomb repressive complex 2 (PRC2) pipeline. The simplistic interpretation is an idiosyncratic failure of one molecule, or one combination is reassuring but, we argue, mistaken. The hematologic toxicity that led to tazemetostat’s withdrawal is better understood as an on-target, on-pathway, off-tissue consequence of inhibiting a developmental chromatin regulator whose normal function in the blood is tumor suppression. If the interpretation that this episode is not an isolated effect of one drug targeting this pathway is correct, then this episode is a warning for the solid-tumor drug development programs, above all in prostate cancer, now betting on chronic EZH2 inhibition.

We recognize that the SYMPHONY-1 signal may instead reflect factors specific to a single agent, combination, or disease context, and this explanation cannot be excluded. Our argument is not that such possibilities are naive, but that the mechanistic rationale for a potential class effect rooted in the context-dependent tumor-suppressor function of EZH2 and PRC2 warrants that the signal be evaluated across the class rather than assumed to be confined to one molecule. Both possibilities are compatible with the current data and distinguishing between them will require the longer-term, cross-agent safety information that is not yet publicly available.

EZH2 is context-dependent, not monolithically oncogenic

EZH2 is the catalytic subunit of PRC2, depositing the repressive H3K27me3 mark to silence lineage and developmental genes and to maintain bivalent promoters that poise stem and progenitor cells for differentiation (5,6). The paradigm that EZH2 may be oncogenic and which justified the development of tazemetostat is based on a specific, well-defined context. Recurrent gain-of-function mutations at EZH2 Tyr641, present in roughly 22% of germinal-center B-cell diffuse large B-cell lymphomas and 7% of FL, renders the enzyme hyperactive and creates a therapeutic dependency (7). With that perspective, chronically inhibiting EZH2 is mechanistically-rational and clinically-relevant.

But limiting one’s view to EZH2 as an oncogene does not consider its normal physiologic functions. Across tissues, EZH2 behaves as either an oncogene or a tumor suppressor depending on cellular context, including through non-canonical, PRC2-independent activities (8). Nowhere is the tumor-suppressive function more consequential than in the blood (Fig. 1).

Figure 1.

The context-dependent biology of EZH2 explains the tazemetostat second-malignancy signal. Left panel: in germinal-center B cells, gain-of-function EZH2 (Tyr641) hyperactivates H3K27me3-mediated silencing, creating an oncogenic dependency that EZH2 inhibition exploits. Right panel: in hematopoietic stem/progenitor cells, EZH2 acts as a tumor suppressor; recurrent loss-of-function mutations drive MDS/AML, and chronic pharmacologic inhibition phenocopies this state, expanding clonal hematopoiesis toward therapy-related myeloid neoplasms over a latency of months to years. Center: the same drug action is therapeutic in the tumor and oncogenic in the marrow—an on-target, off-tissue toxicity. Bottom: implication for chronic dosing in long-survival solid-tumor settings (prostate cancer). Figure 1 was created and refined using FigureLabs (figurelabs.ai). The authors reviewed and approved the final figure and take responsibility for its scientific accuracy.

In myeloid hematopoiesis, EZH2 is a tumor suppressor - the crux

Two landmark 2010 studies independently identified recurrent inactivating EZH2 mutations in MDS, MDS/myeloproliferative neoplasm overlap syndromes, and myelofibrosis, where these mutations are associated with an adverse prognosis (9,10). Functional genetics then established causality showing that conditional Ezh2 deletion in mice augments hematopoietic stem-cell repopulating capacity and induces an MDS/myeloproliferative disease that is markedly accelerated by concurrent Tet2 loss (11) Ezh2 loss also cooperates with RUNX1 mutation to drive MDS (12). The consensus that emerged, which was reinforced across subsequent series, is that EZH2 acts as a bona fide tumor suppressor in the myeloid compartment with loss-of-function lesions sitting squarely within the genetic architecture of myeloid neoplasia (13).

This biology helps to reframe the SYMPHONY-1 long-term follow-up results. A small molecule used to chronically suppress EZH2’s catalytic activity appears to reproduce the effects of EZH2 inactivating mutations in the blood compartment that initiate MDS and AML. Chronic dosing is, in effect, recapitulating EZH2 loss of function mutations. Interpreted through this lens, the excess of MDS/AML with chronic tazemetostat seems very biologically-based (Fig. 2).

Figure 2:

Proposed mechanism linking chronic EZH2 inhibition to clonal hematopoiesis and therapy-related myeloid neoplasia. In normal hematopoiesis, the polycomb repressive complex 2 (PRC2), composed of EZH2, EED, and SUZ12, catalyzes trimethylation of histone H3 lysine 27 (H3K27me3). This represses stemness- and proliferation-associated genes in hematopoietic stem and progenitor cells (HSPCs), maintaining a balance between self-renewal and differentiation. Chronic pharmacologic inhibition of EZH2 (e.g., with tazemetostat) reduces H3K27me3-mediated gene repression, resulting in activation of self-renewal and proliferation programs. In the setting of pre-existing clonal hematopoiesis of indeterminate potential (CHIP)— particularly clones harboring mutations in genes such as DNMT3A, TET2, or ASXL1—this altered epigenetic state may confer a selective growth advantage and lead to expansion of mutant hematopoietic clones. Over time, sustained clonal expansion and impaired differentiation promote clonal evolution and acquisition of additional leukemogenic events, culminating in therapy-related myelodysplastic syndrome (MDS) or acute myeloid leukemia (AML). These effects may emerge after months or years of exposure and are likely to persist after discontinuation of EZH2 inhibitor therapy, highlighting a potential mechanism for treatment-associated myeloid neoplasia. Figure 2 was created and refined using FigureLabs (figurelabs.ai). The authors reviewed and approved the final figure and take responsibility for its scientific accuracy.

Clonal hematopoiesis as the permissive substrate

The appearance of clonal cytopenia of undetermined significance among the SYMPHONY-1 SPMs is also revealing. Aging patients frequently harbor clonal hematopoiesis of indeterminate potential (CHIP), commonly driven by mutations in DNMT3A, TET2, ASXL1, all of which cooperate with EZH2 loss (11,12). Hence, therapies that impose differentiation blockade or selective pressure can expand these pre-malignant clones (14). EZH2 inhibition is a plausible example of such pressure, and the kinetics fit. There is a latency of months to years, and persistence of the phenotype after drug discontinuation suggest there has been a durable epigenetic and clonal “hit” (3).

A complementary mechanism may explain the lymphoid SPMs. EZH2 inhibition can increase activation-induced cytidine deaminase (AID)–dependent chromosomal translocations in B cells when combined with PI3Kδ inhibition or in DNA-repair–deficient backgrounds, even as monotherapy appears genetically silent in the same assays (15). SYMPHONY-1 paired tazemetostat with lenalidomide, which itself associated with second malignancies. Therefore, combination-specific genotoxicity must be considered alongside the dominant stem-cell mechanism.

Trial-design and regulatory lessons for epigenetic agents

Three features of this episode should reshape how the field generates evidence for chromatin-targeted drugs. First, latency mismatch: SPMs began at 7.5 months and clustered at one to three years, a timeframe that conventional early-phase safety windows and short median follow-ups often miss (2). Therefore, surrogate-endpoint accelerated approvals of agents targeting chromatin regulators should demand confirmatory trials that are appropriately powered and with long-term safety follow-up to detect delayed, stem-cell-mediated toxicity. Second, carry over risk toxicity that manifests after drug cessation defies the assumption that stopping the drug stops the danger. This effect provides additional support for long-term safety follow-up in trials with these agents. Third, there is the issues of combination confounding when a developmental-regulator inhibitor is co-administered with a genotoxic or immunomodulatory partner. Every attempt should be made to determine the side effects and risks attributable to each agent prospectively, rather than deconstructed post hoc (2,15).

This risk must also be placed in context. Therapy-related MDS/AML is not unique to EZH2 inhibitors. It is an established risk of PARP inhibitors (16), radioligand therapies such as 177Lu-PSMA, and various DNA-damaging chemotherapeutics. For each of these classes, the field has accepted a defined risk of secondary myeloid neoplasia when justified by a commensurate magnitude of clinical benefit, and EZH2 inhibitors should be evaluated within, not apart from, this same benefit–risk framework. The concern for toxicities such as MDS/AML is heightened in settings such as newly diagnosed metastatic prostate cancer. In this disease state, patients are often exposed to multiple genotoxic agents sequentially. Exposure to an EZH2 inhibitor, a PARP inhibitor, a radiopharmaceutical, and DNA-damaging chemotherapy nearly all of which most prostate cancer patients receive may compound the risk of therapy-related myeloid neoplasia. Thus, the cumulative genotoxic burden should be weighed explicitly in trial design and in clinical decision-making.

These observations argue for embedding 1) pre-specified SPM/therapy-related myeloid neoplasm endpoints with defined stopping rules, into registration trials of PRC2-axis agents, 2) baseline clonal-hematopoiesis genotyping as a stratification variable, and 3) serial molecular monitoring as a built-in safety readout, rather than an afterthought.

We wish to be explicit about the current limitations of these proposals. We envision embedding panel-based next-generation sequencing of peripheral blood for canonical myeloid clonal-hematopoiesis genes (e.g., DNMT3A, TET2, ASXL1, TP53, PPM1D, and spliceosome components) (17) in trials with EZH2 inhibitors or other agents with a clear risk of myelotoxicity based on target function. Because mandatory pre-treatment sequencing would extend screening timelines and could delay treatment for patients who require prompt initiation, baseline clonal-hematopoiesis status is best captured as a prospectively collected exploratory biomarker rather than as a randomization stratification factor or eligibility criterion. Critically, no evidence-based variant-allele-fraction threshold currently exists to exclude a patient at baseline or to trigger discontinuation on serial monitoring. We propose defining such thresholds in prospective studies. Moreover, at present, there is no evidence that discontinuing an EZH2 inhibitor in response to emergent molecular findings reduces the subsequent risk of MDS/AML. Therefore, routine serial monitoring cannot yet be recommended as a requirement for all trial participants or all patients treated in standard practice due to the cost, its burden, and the risk of prematurely withdrawing therapy from a patient who is deriving clinical benefit. We therefore frame these measures not as an EZH2-specific mandate but of relevance to broad classes of agents that may lead to therapy-related myeloid neoplasia and whose clinical utility remains to be established.

Clinical implications: counseling and surveillance

If previously approved EZH2-pathway agents return, or if other investigational EZH2 pathway agents advance, clinical practice must adapt. Informed consent should explicitly state that there may be a delayed risk of new blood cancers and convey that this risk may persist after stopping therapy. This is a risk that is quite distinct from acute, reversible toxicities. A pragmatic surveillance framework would include baseline complete blood count and molecular screening for CHIP/clonal cytopenia, standardized longitudinal monitoring, and pre-defined triggers for referral to hematology for bone marrow evaluation (14). Patients with baseline CHIP, prior genotoxic exposure, or DNA-repair deficiency may warrant intensified monitoring (14,15). Further, careful selection of drug combinations with EZH2 inhibitors and longer-term safety follow-up pre-clinical studies may reveal the enhanced toxicity of certain combinations. Finally, benefit–risk of side effects such as hematologic toxicity is context-dependent: a risk of leukemia long-term may be acceptable in a patient with advanced, treatment-resistant cancer with few therapeutic options. However, testing of these agents in earlier settings with longer survival timelines would require careful thought and justification.

The available data on other PRC2-axis agents should be interpreted with these caveats in mind. Mevrometostat, valemetostat, and tulmimetostat (18–20) have now been evaluated in several hundred patients. To date, no comparable secondary-malignancy signal has been publicly reported for these agents. This experience is reassuring but not yet definitive. In the phase 1/2 mevrometostat experience, median follow-up was approximately 9.6 months as of the September 2, 2024 data cut (21) substantially shorter than the interval over which therapy-related MDS/AML typically emerge. This is important since most SYMPHONY-1 cases arose after one to three years of exposure. With longer follow-up, we will have a greater sense of the safety data patients exposed to other PRC2 inhibitors. Continued vigilance is clearly warranted. Although the sponsor’s specific long-term monitoring plan is not public, we believe prospective surveillance for delayed secondary malignancies is warranted and may become a regulatory requirement.

Risk is also unlikely to be governed by duration of exposure alone. Disease context (hematologic-malignancy populations often carry substantial prior alkylator and topoisomerase-II exposure and, in SYMPHONY-1, concomitant lenalidomide), prior and concomitant genotoxic therapy, and pharmacologic differences among the agents. Indeed, there are now new PRC2 inhibitors with different EZH1/EZH2 selectivity (dual EZH1/EZH2 inhibition for valemetostat versus more EZH2-selective inhibition for tazemetostat and mevrometostat), potency, pharmacokinetics, and dosing. All of these factors are possible mediators of risk, though no published data currently link any specific pharmacologic property to differential second-malignancy risk.

The highest-stakes test case: prostate cancer

Nowhere are these considerations more relevant and consequential than in prostate cancer. EZH2 is overexpressed in most castration-resistant prostate cancers and has long been implicated as a driver of progression and a marker of poor outcome (22,23). The contemporary therapeutic hypothesis is compelling: EZH2 inhibition may counteract lineage plasticity, or differentiation change, and androgen-receptor–independent resistance mechanisms, which may resensitize tumors to AR-directed therapy (24). Based on this rationale, Pfizer’s mevrometostat is in a Phase 3 combination study with enzalutamide in men with metastatic androgen receptor pathway modulation-resistant (APMR) prostate cancer (previously known as castration-resistant prostate cancer) (MEVPRO-1) and in men with metastatic androgen receptor pathway modulation-naïve (APMN) prostate cancer (previously known as hormone-naïve prostate cancer; terminology standardized by the Prostate Cancer Working Group 4 [PCWG4]) (MEVPRO-3) (25). Additionally, ORIC-994, an inhibitor of EED, which is another member of the PRC2 complex is currently in clinical trials in men with metastatic APMR (NCT05413421, (25). The prostate cancer populations where mevrometostat and ORIC-994 are being tested are vastly larger than the rare diseases in which tazemetostat gained approval. Moreover, many of these prostate cancer patients, particularly those with APMN, may live for years on continuous AR-directed therapy (25).

This long-term treatment scenario is precisely the situation that led to the increased risk of hematologic SPMs with tazemetostat. If hematologic SPMs require months to years of exposure and can arise after discontinuation (2), then chronic EZH2 inhibition in men with advanced prostate cancer with extended survival expectancy might lead to even greater cumulative stem-cell exposure, enhancing the risk of clonal evolution. The prostate biology adds a further caution about optimal selection of patients. Emerging data show that EZH2 loss does not simply reverse neuroendocrine lineage plasticity but can rewire transcription-factor programs and redistribute the AR cistrome. However, full emergence of neuroendocrine differentiation in tumors may lead to poor response (24,26). Context dependence operates inside the tumor as well as inside the marrow.

A path forward

The tazemetostat withdrawal is not the end of EZH2 as a target; it is a course correction with four critical imperatives. First, elevate the normal developmental biology of the target to a formal pillar of preclinical safety. Before chronic-dosing trials, an agenťs effect on stem- and progenitor-cell homeostasis especially hematopoietic should be characterized as rigorously as its anti-tumor activity. Second, treat hematologic risk as a class-level anticipated risk and not an afterthought. Whether the dual EZH1/EZH2 inhibitor valemetostat (20), with its distinct mechanism but shared role in hematopoietic stem-cell maintenance, mitigates or aggravates this liability is an open question, as is whether EZH2 degraders that remove scaffolding functions shift the efficacy–toxicity balance (27). Harmonized, biomarker-resolved pharmacovigilance across tazemetostat, valemetostat, mevrometostat, and emerging degraders is needed. Third, target downstream effectors where possible. Lineage-specific PRC2 targets and context-restricted modalities may decouple anti-tumor effect from stem-cell toxicity (24). Fourth, recalibrate benefit–risk thresholds and explore intermittent or adaptive dosing schedules pre-clinically that could preserve on-tumor effects while reducing cumulative hematopoietic exposure in people.

Target validation in tumor cells and cancer control with EZH2 inhibitors tells us EZH2 is a target that can be blocked. The marrow has revealed what the cost of hitting that target is. For a field that continues to test chronic EZH2 administration strategies, this lesson could not be more timely.

Statement of Translational Relevance.

The withdrawal of tazemetostat due to excess secondary hematologic malignancies should be recognized not as an isolated safety event, but as a predictable, target-mediated consequence of inhibiting a chromatin regulator with tumor-suppressive roles in normal hematopoiesis. As EZH2 inhibitors advance into Phase III trials in metastatic prostate cancer, where patients may receive continuous therapy for years, failure to account for target biology in normal stem cell compartments represents a critical gap in drug development. This demands immediate changes: routine baseline and longitudinal monitoring for clonal hematopoiesis, trial designs incorporating latency-matched follow-up, and explicit informed consent addressing delayed, potentially irreversible myeloid toxicity. These risks must be weighed rigorously in benefit–risk assessments, particularly in earlier disease settings, where tolerance for late-onset, treatment-induced malignancy is appropriately low.