Point the EED inhibitor at the index fusion - a clinical-stage drug, a subunit with no paralogue, and zero leukaemia experiments
also called EED226 in KMT2A::AFF1 · MAK683 leukaemia test · paralogue-agnostic PRC2 inhibition in infant ALL
There is a class of drug that blocks Polycomb by binding a part of the machine that has no back-up copy, so the redundancy problem does not apply. One of them has already been through a 139-patient trial. Nobody has ever tested one in leukaemia.
The same thing, for a scientist
**The gap.** `(EED226 OR MAK683 OR "EED inhibitor") AND (leukemia OR MLL OR KMT2A)` returns **0** PubMed records, while the unconjoined query returns 28 covering chemistry, prostate, DLBCL, neuroblastoma, nasopharyngeal carcinoma, HIV latency and a bivalent degrader [E1]. **MAK683 has completed a first-in-human phase 1/2 in 139 patients with no leukaemia cohort** [E2]. Meanwhile EED is the one PRC2 subunit with **no paralogue**, the one whose genetic inactivation kills MLL-AF9 leukaemia where *Ezh2* inactivation does not [E3], and the best-scoring PRC2 gene in this project's DepMap panel with the lean carried by the **B-lineage** half [E4]. An allosteric EED inhibitor also covers at least one EZH2 mutant resistant to SAM-competitive drugs [E5]. **Design.** Four KMT2A::AFF1 systems (SEM, RS4;11, one infant-derived PDX, one non-rearranged B-ALL comparator), four arms: EED226 or MAK683; valemetostat (dual EZH1/EZH2); tazemetostat (EZH2-selective, the predicted-negative control); vehicle. **Primary readout is not growth** but spike-in-calibrated H3K27me3 depletion matched across arms, so that the growth comparison is at equal target engagement - the missing control in the one published head-to-head [E6]. Secondary: RNA-seq for *CDKN1C*, *TP53INP1*, *CDKN2A*, *HDAC7*; *CDKN2A* status genotyped per line, because it is the published mediator and is frequently deleted here. **Cost tier: medium.** Compounds are commercially available as probes; the PDX arm is the cost. **Timeline 6-9 months.**
Why the matched-engagement design is the whole point
The existing evidence that dual beats selective in B-ALL is a cell-line growth comparison with no stated matching of exposure or of H3K27me3 depletion between arms 6. That comparison is confounded with potency: a more potent compound outgrows a less potent one at the same nominal concentration whatever the target coverage.
This matters because the conclusion drawn from it - EZH2-selective is the wrong drug class here - is one of the two load-bearing claims of this whole beat, and it currently rests on an unmatched comparison plus mouse genetics. Matching on the mark rather than on the dose converts an inference into a measurement.
The design also makes the tazemetostat arm informative rather than decorative. If tazemetostat at a dose that achieves equal bulk H3K27me3 depletion to valemetostat still kills less, the failure is about which loci each drug reaches, not about how much mark it removes - and that is a different and more interesting result than 'EZH1 compensates', because it would mean the EZH1-dependent loci are the functionally decisive ones.
What each outcome would mean
| outcome | reading |
|---|---|
| EED inhibitor ≈ valemetostat > tazemetostat, at matched mark depletion | PRC2 catalysis is the requirement, the paralogue problem is real, and an EED-directed agent is a viable alternative with a clinical-stage compound already in hand |
| EED inhibitor > valemetostat | EED has a non-catalytic role, or allosteric inhibition reaches complexes the SAM-competitive agents do not - both would be new and both are testable further |
| all three equal at matched depletion | the paralogue argument collapses and the EZH1 story is about potency, not coverage; gene-ezh1 and drug-tazemetostat would both need rewriting |
| none works in the index fusion | PRC2 is a myeloid MLL-AF9 dependency that does not transfer to KMT2A::AFF1, which would be the single most useful negative this arm could produce and would retire the whole therapeutic line |
The last row is the one this beat most wants tested, because every mechanistic result supporting PRC2 dependence in this disease is mouse MLL-AF9, and this project's repeated finding is that KMT2A epigenetics results are myeloid until proven otherwise. The one index-fusion result is a single PDX.
| design | Four-arm comparison in KMT2A::AFF1 systems (SEM, RS4;11, an infant-derived PDX, and a non-rearranged B-ALL comparator): (1) EED-directed allosteric inhibitor (EED226 as probe, MAK683 as clinical-stage compound), (2) valemetostat (dual EZH1/EZH2), (3) tazemetostat (EZH2-selective; predicted-negative control), (4) vehicle. Doses titrated to MATCHED spike-in-calibrated H3K27me3 depletion before growth is compared. |
|---|---|
| readout | Primary: growth and survival at matched H3K27me3 depletion. Secondary: spike-in-calibrated H3K27me3 CUT&RUN (bulk and locus-resolved); RNA-seq for CDKN1C, TP53INP1, CDKN2A, CDKN2B, HDAC7; differentiation immunophenotype; apoptosis. |
| controls | CDKN2A deletion status genotyped per line, because it is the published mediator and is frequently deleted in this disease; a non-rearranged B-ALL comparator for fusion specificity; matched target engagement rather than matched dose; normal cord-blood CD34+ progenitors as the toxicity arm, since the published therapeutic-window claim was made against immortalised lines rather than primary HSPCs. |
| cost tier | med |
| timeline | 6-9 months |
| most valuable outcome | a clean negative. Every mechanistic result supporting PRC2 dependence in this disease is mouse MLL-AF9; if no PRC2-directed agent works in an index-fusion system at matched engagement, the whole therapeutic line retires and the graph gains a bounded negative in place of an assumption. |
| why not done | (EED226 OR MAK683 OR "EED inhibitor") AND (leukemia OR MLL OR KMT2A) returns 0 records; the MAK683 phase 1/2 enrolled seven tumour types and no leukaemia cohort. |
tests
[ "hyp-prc2-is-the-fusions-repressive-arm" ]
- Every arm of this experiment removes H3K27me3 and raises H3K27ac, which is the reported route to increased chromosomal translocation in B cells with high AID activity or DNA-repair deficiency. A B-lineage experiment that scores growth and ignores genomic integrity would miss the hazard it is generating - the translocation-capture readout belongs in the design, not in a later study.
- A positive result here creates pressure to take a marrow-suppressive drug into infants on the strength of cell lines and one PDX. The toxicity arm using primary progenitors is therefore not optional, because the existing therapeutic-window estimate comes from immortalised lines and the clinical experience contradicts it.
- If the EED arm works, the obvious next step is combination with a menin inhibitor - which hyp-prc2-is-the-fusions-repressive-arm predicts is redundant. Designing the combination before the mechanism experiment (exp-menin-inhibitor-h3k27me3-timecourse-index-fusion) reports would risk a predictably uninformative trial.
- Does an infant-derived KMT2A::AFF1 PDX with adequate characterisation exist in an accessible biobank? Every index-fusion pharmacology result in this beat rests on adult or unstated-age material, and the lineage lesson of this project is that provenance decides the answer.
- Is EED's contribution in leukaemia allosteric or structural? An EED-pocket inhibitor blocks the H3K27me3-reading stimulation loop but leaves the subunit in the complex, so it will not phenocopy an Eed null if the requirement is scaffolding - and the bivalent EED-targeted degrader reported in the literature is the reagent that would separate them.
- Should the toxicity arm use cord-blood CD34+ progenitors rather than a cell line? The published therapeutic-window claim for PRC2 suppression was measured against immortalised non-transformed lines, and the clinical dose-limiting toxicities are cytopenias - so the existing window estimate is probably optimistic and this experiment is the cheapest place to correct it.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
PubMed returns 28 records for (EED226 OR MAK683 OR "EED inhibitor") - covering medicinal chemistry, prostate cancer, DLBCL, neuroblastoma, nasopharyngeal carcinoma, HIV latency reversal, acute kidney injury, a bivalent EED-targeted PRC2 degrader and a first-in-human trial - and 0 records for the same query conjoined with (leukemia OR MLL OR KMT2A).
PubMed esearch counts and title check for EED-directed inhibitor queries, run 2026-09-20 · NCBI PubMed 2026What this evidence cannot show
The symbol set is two compound names and one phrase; EEDi-5285, EEDi-5273, the bivalent degrader by name, unnamed chemical-series papers and any study framed as 'PRC2 inhibitor' without naming EED are not covered. Per Rule 64 the complex form was run and recovers no leukaemia EED-inhibitor study. This is an absence of published study under these terms - an absence of testing, not a negative result.
Evidence that argues against this (1)
- The functional equivalent of this experiment has been done and is positive: genetic Eed inactivation kills MLL-AF9 leukaemia and combined pharmacologic EZH2/EZH1 inhibition has a strong anti-leukaemic effect. So the absence is of a chemical modality, not of the biological question, and the experiment proposed here is a translation step rather than a discovery step. Danis E et al. 2015
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
MAK683, a first-in-class selective allosteric EED inhibitor, completed a first-in-human phase 1/2 dose-escalation study in 139 patients with advanced malignancies across seven tumour types with no leukaemia cohort; 15 patients had 22 dose-limiting toxicities including thrombocytopenia and febrile neutropenia, overall response rate was 5.8% with activity in DLBCL and epithelioid sarcoma, and pharmacodynamics were measured as H3K27me3 change from baseline. NCT02900651.
A first-in-human phase 1/2 dose-escalation study of MAK683 (EED inhibitor) in patients with advanced malignancies. · Ribrag V, Iglesias L, De Braud F, Ma B, Yokota T, Zander T, et al. · Eur J Cancer 2025What this evidence cannot show
Abstract read from the eutils record; full text not read. Cited for the existence of a clinical-stage EED-directed compound and for its toxicity profile, not for efficacy - a 5.8% overall response rate across seven solid and lymphoid tumour types is a weak clinical signal, and the dose-limiting toxicities are haematological, which is the least favourable profile for a leukaemia extension.
Evidence that argues against this (1)
- The haematological dose-limiting toxicities are the direct expression of PRC2 being required for haematopoietic stem cell maintenance, so in leukaemia the target organ and the toxic organ coincide. A 5.8% ORR in solid tumours combined with marrow-limited dosing is a weak prior for this experiment succeeding clinically even if it succeeds preclinically. Yamagishi M et al. 2026
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Inactivation of the essential PRC2 gene Eed ablates PRC2 function and is incompatible with MLL-AF9 leukaemia growth, whereas Ezh2-null leukaemias maintain self-renewal to the third round of transplantation with locus-specific H3K27me3 persisting through suggested Ezh1 compensation.
Polycomb repressive complex 2 is required for MLL-AF9 leukemia. · Neff T, Sinha AU, Kluk MJ, Zhu N, Khattab MH, Stein L, et al. · Proc Natl Acad Sci U S A 2012What this evidence cannot show
Abstract read from the eutils record; full text not read. Mouse MLL-AF9, myeloid, no B-lineage or infant arm - the transferability of this result to KMT2A::AFF1 is the thing the proposed experiment is meant to test rather than assume. Genetic ablation removes scaffolding and non-catalytic functions with the mark, so it is an upper bound on what an allosteric catalytic inhibitor could achieve.
Evidence that argues against this (1)
- An Eed null and an EED-pocket inhibitor are not the same perturbation: the inhibitor leaves EED in the complex and blocks the H3K27me3-reading allosteric loop, which removes stimulation rather than the subunit. If EED's contribution in leukaemia is structural rather than allosteric, the drug will not phenocopy the genetics and the experiment will return a false negative for the biology. Qi W et al. 2017
- E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
EED is the best-scoring PRC2 subunit across 11 KMT2A-rearranged DepMap leukaemia lines against 43 other leukaemia lines - index mean -0.458, comparator -0.340, difference -0.117, rank 824 of 17,728 - with the difference carried by the B-lineage half of the index arm (-0.224) rather than the myeloid half (-0.058).
analyses/prc2-dependency-lineage-split - DepMap (Chronos, via Open Targets) PRC2 panel scored with the index arm split by lineage · Q41-prc2-arm · Oncofusion knowledge graph, internal analysis 2026What this evidence cannot show
A difference of -0.117 is modest and below this project's -0.5 selectivity convention; the lymphoid half is 4 lines and no infant KMT2A::AFF1 line exists in DepMap. Raw effect sizes only, no z, no permutation null, so this is a consistency check against the mouse genetics rather than a hit. Cellosaurus fusion annotation may be incomplete. DepMap release version not stated by the Open Targets API.
Evidence that argues against this (1)
- The lymphoid lean is the opposite of this project's usual finding and rests on four cell lines, so the most likely explanation is noise. It is recorded because it agrees in direction with the two index-disease pharmacology results, and because a suppressed observation cannot later be checked - but it should not be treated as a reason to expect the experiment to succeed. Ito J et al. 2023
- E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
EED226 binds the H3K27me3-binding pocket of EED, induces a conformational change causing loss of PRC2 activity, blocks H3K27 methylation of PRC2 target genes and induces regression of human lymphoma xenografts comparably to SAM-competitive inhibitors, and also effectively inhibits PRC2 containing an EZH2 protein mutated to resist SAM-competitive inhibitors.
An allosteric PRC2 inhibitor targeting the H3K27me3 binding pocket of EED. · Qi W, Zhao K, Gu J, Huang Y, Wang Y, Zhang H, et al. · Nat Chem Biol 2017What this evidence cannot show
Abstract read from the eutils record; full text not read. Efficacy shown in human lymphoma xenografts; no leukaemia, no KMT2A-rearranged system, no B-ALL. The resistance-mutant coverage is an enzyme or cell observation about a specific mutant, not a demonstration of clinical cross-resistance breadth.
Evidence that argues against this (1)
- Allosteric EED targeting does not escape every PRC2 resistance route: compensatory DNA methylation is reported as an escape from dual EZH1/2 inhibition and is upstream of any PRC2 binding site, so changing which subunit is bound does not address it - and t(4;11) infant ALL is among the most promoter-hypermethylated leukaemia subtypes. Yamagishi M et al. 2026
- E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The only published head-to-head of a dual EZH1/EZH2 inhibitor against an EZH2-selective inhibitor in a leukaemia context reports that DS-3201 inhibited B-ALL cell line growth more significantly and strongly than EPZ-6438 and that DS-3201 inhibited an MLL-AF4 patient-derived xenograft - but it is a growth comparison with no stated matching of exposure or H3K27me3 depletion between arms, and EPZ-6438 was not tested in the PDX.
Dual inhibition of EZH1/2 induces cell cycle arrest of B cell acute lymphoblastic leukemia cells through upregulation of CDKN1C and TP53INP1. · Ito J, Yamagata K, Shinohara H, Shima Y, Katsumoto T, Aikawa Y, et al. · Int J Hematol 2023What this evidence cannot show
Abstract read from the eutils record; full text not read - the full text may contain matched pharmacodynamic data the abstract omits, and this experiment should be scoped against it before being run. One PDX, no n, no dose, no patient age.
Evidence that argues against this (1)
- If the full text does show matched H3K27me3 depletion between the DS-3201 and EPZ-6438 arms, the central design justification for this experiment weakens considerably and it should be rescoped to the EED arm alone. Reading that paper is a prerequisite, not an optional step. Honma D et al. 2017
matched-engagement four-arm comparison in index-fusion systems
the clinical-stage EED inhibitor has never been given in any leukaemia
tazemetostat is the predicted-negative control arm, run at matched H3K27me3 depletion