Oncofusion

start / graph / hyp-prc2-is-the-fusions-repressive-arm

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q41-prc2-arm

PRC2 is the fusion's repressive arm, recruited through menin, and the index disease's version of it runs to HDAC7

also called menin-EZH2 axis hypothesis · PRC2 as fusion effector · the fusion has two outputs

The usual story is that this leukaemia's fusion protein switches genes ON. This hypothesis says it also switches genes OFF, using the Polycomb machinery, and that it reaches that machinery through the same menin anchor the whole field is already drugging - which would make menin inhibitors partly Polycomb drugs without anyone having noticed.

The same thing, for a scientist

**Statement.** In KMT2A-rearranged leukaemia the fusion has two chromatin outputs, not one: an activating output through DOT1L/SEC at *HOXA*/*MEIS1*, and a **repressive output through PRC2** at differentiation genes, and the repressive arm is licensed by **menin**. The evidence is an architecture repeated across two lineages: in mouse MLL-AF9, menin binds the *Ezh2* promoter and drives its expression, and EZH2 then represses C/EBPalpha targets [E1]; in human t(4;11) infant pro-B ALL, EZH2 acts with KMT2A::AFF1 to silence *HDAC7*, and a menin inhibitor restores it [E2]. **Prediction 1: a menin inhibitor is partly a PRC2 inhibitor.** EZH2 protein and H3K27me3 at the relevant loci should fall after menin inhibition, on a timescale set by mark dilution rather than by transcription. **Prediction 2: the index-disease effector is HDAC7 and its loss is what the repressive arm buys.** **Prediction 3: adding a PRC2 inhibitor to a menin inhibitor is redundant, not synergistic**, at the loci that run through this axis. **The hypothesis is genuinely at risk.** A competing mechanism for the same intervention exists and predicts the opposite for PRC2 occupancy: menin inhibition phenocopies polycomb disruption at bivalent genes because a menin-*independent* KMT2A/B function opposes polycomb there, and releasing KMT2A from active genes shifts the balance [E3]. And the DepMap profile shows **neither** catalytic paralogue is a dependency in this disease, so if the repressive arm matters it must matter through the complex rather than through EZH2 as a protein [E4]. Prior probability 0.35: two independent systems with the same architecture, one of them the index disease, against a directly competing mechanism and no measurement of H3K27me3 after menin inhibition anywhere.

Go deeper
Why this is worth stating as a hypothesis rather than a mechanism

Because the load-bearing step has never been measured. Both supporting papers establish an association between EZH2 and a silenced gene and then perturb menin, not PRC2:

  • 1 depletes menin and observes EZH2 expression, EZH2 binding and H3K27 methylation fall at C/EBPalpha targets. That is the axis, measured, in mouse myeloid cells.
  • 2 names EZH2 and KMT2A::AFF1 as the silencers of HDAC7 and then tests MI-538 plus chidamide. There is no EZH2 inhibitor arm, no reported EZH2 ChIP at HDAC7, and no reported H3K27me3 at HDAC7 in the abstract.

So the index-disease half of the architecture is currently one association and one menin-inhibitor phenotype. If the EZH2 step at HDAC7 is a correlate rather than a cause, the hypothesis collapses to 'menin inhibition de-represses HDAC7', which is already in the graph and is not a Polycomb finding at all.

The hypothesis earns its place because if it is right it changes three things: menin inhibitors acquire a free pharmacodynamic (H3K27me3 / EZH2 level), PRC2 becomes a downstream node of the menin-interface hypothesis rather than an independent target, and the PRC2-plus-menin combination that looks obvious becomes predicted-redundant.

What would kill it

Listed in data.falsifiers and restated here because the strongest falsifier is cheap:

  1. H3K27me3 and EZH2 occupancy at HDAC7 do not fall after menin inhibition in an index-fusion line. One CUT&RUN timecourse. If the mark does not move, menin is not licensing PRC2 there.
  2. A dual EZH1/EZH2 inhibitor does not restore HDAC7. If PRC2 catalysis is the silencer, blocking it should phenocopy the menin inhibitor at that locus. If it does not, EZH2 is a bystander.
  3. Menin inhibition leaves EZH2 protein unchanged in human cells. The menin-to-Ezh2 transcriptional link is a single mouse result; if it does not reproduce in human KMT2A::AFF1 cells, Route 1 has no mechanism.
  4. The bivalency mechanism accounts for the whole effect. If after menin inhibition H3K4me3 rises at bivalent genes with PRC2 occupancy intact 3, and HDAC7 behaves like a bivalent gene rather than a PRC2-silenced one, this hypothesis is superseded rather than refuted - and that is the more likely outcome than either being simply wrong.
Facts and figures
statementThe KMT2A fusion has a repressive output as well as an activating one, delivered by PRC2 and licensed by menin; in the index disease (t(4;11) infant pro-B ALL) the effector of that repressive output is HDAC7, as C/EBPalpha targets are in MLL-AF9.
originliterature-gap
prior probability0.35
predictions
[
  "EZH2 protein falls after menin inhibition in a human KMT2A::AFF1 line",
  "H3K27me3 and EZH2 occupancy fall at HDAC7 after menin inhibition, on a mark-dilution timescale rather than a transcriptional one",
  "A dual EZH1/EZH2 inhibitor restores HDAC7 expression without a menin inhibitor",
  "Adding a PRC2 inhibitor to a menin inhibitor is redundant rather than synergistic at loci running through this axis",
  "The repressive output is visible as a set of PRC2-silenced differentiation genes that come up under BOTH menin inhibition and dual EZH1/2 inhibition - the intersection is the axis"
]
experiments
[
  "exp-menin-inhibitor-h3k27me3-timecourse-index-fusion",
  "exp-eed-allosteric-inhibition-in-kmt2a-aff1"
]
analyses
[
  "ana-prc2-dependency-lineage-split",
  "ana-prc2-arm-enumeration"
]
falsifiers
[
  "H3K27me3 and EZH2 occupancy at HDAC7 do not fall after menin inhibition in an index-fusion line",
  "A dual EZH1/EZH2 inhibitor does not restore HDAC7 expression in an index-fusion line",
  "Menin inhibition leaves EZH2 protein level unchanged in human KMT2A::AFF1 cells, making the menin-to-EZH2 transcriptional link a mouse-only observation",
  "The genes de-repressed by menin inhibition and by dual EZH1/2 inhibition do not overlap above chance",
  "HDAC7 behaves as a bivalent gene (H3K4me3 rising, PRC2 occupancy intact) after menin inhibition, in which case the bivalency-balance mechanism supersedes this one"
]
what it would change if true
[
  "menin inhibitors acquire a free pharmacodynamic biomarker (EZH2 level, or H3K27me3 at a defined locus) that no trial currently measures",
  "PRC2 becomes a downstream node of the menin-interface hypothesis rather than an independent target, which lowers rather than raises the case for a PRC2 inhibitor in this disease",
  "the obvious menin-plus-PRC2-inhibitor combination is predicted redundant before anyone runs it"
]
How this could bite us back
  • If this hypothesis is true it ARGUES AGAINST a PRC2 inhibitor in this disease rather than for one, because the menin inhibitors already in the clinic would be hitting the same axis with a fusion-specific interface and better selectivity. A hypothesis that opens a target is common; this one mostly closes one.
  • It also predicts that the PRC2-plus-menin combination someone will eventually propose is redundant at the relevant loci, so a negative combination result would be the expected outcome rather than a surprise - and should not be read as evidence against either agent.
  • If the competing bivalency mechanism dominates instead, menin inhibition activates bivalent genes broadly in normal cells as well as blasts, including in pluripotent cells. That is a developmental exposure in infants that no response-rate endpoint can see.
Still open
  1. Has anyone measured EZH2 protein or H3K27me3 in a patient or a model before and after menin inhibition? Across revumenib, ziftomenib, bleximenib and MI-538 this beat found no such measurement, which is surprising given that the menin-EZH2 link has been in print since 2013.
  2. If the repressive arm is real, is it the same arm in every fusion partner? The effector differs between MLL-AF9 (C/EBPalpha) and KMT2A::AFF1 (HDAC7), which suggests the arm is lineage-tuned - and the graph's partner-outcome arm would predict that KMT2A::MLLT3 in a B-lineage patient uses the lymphoid effector rather than the myeloid one.
  3. Does the fusion recruit PRC2 physically anywhere, or only transcriptionally through menin? No physical menin-PRC2 or fusion-PRC2 complex is reported in any paper this beat read, and a proximity-labelling experiment in an index-fusion line would settle it - this graph already has a proximity-interactome arm that could absorb the question.
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Menin binds the Ezh2 promoter and promotes its expression; EZH2 interacts with C/EBPalpha and represses its targets; menin depletion reduces MLL binding at the Ezh2 locus, EZH2 expression, and EZH2 binding and repressive H3K27 methylation at C/EBPalpha targets, inducing pro-differentiation C/EBPalpha target genes in MLL-AF9 leukaemia.

    PMID 23349306DOI
    What this evidence cannot show

    Abstract read from the eutils record; full text not read. Mouse MLL-AF9 with acute menin depletion; the effector C/EBPalpha is myeloid. A transcriptional, indirect link - no physical menin-PRC2 complex is reported. Whether EZH1 substitutes when menin-driven EZH2 falls is not addressed, which matters because DepMap says neither paralogue alone is required.

    Evidence that argues against this (1)
    • Menin inhibition can produce polycomb-like de-repression without touching PRC2 at all: a menin-independent KMT2A/B function maintains H3K4me3 and opposes polycomb at bivalent genes, and releasing KMT2A from active genes after menin targeting shifts that balance. The two mechanisms predict opposite things about PRC2 occupancy and identical things about gene expression. Sparbier CE et al. 2023
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In t(4;11) infant B-ALL, HDAC7 is epigenetically silenced by EZH2 together with the KMT2A::AFF1 fusion protein; the menin inhibitor MI-538 restores HDAC7 expression, an effect enhanced by chidamide, driving leukaemic pro-B cells toward differentiation, impairing aberrant proliferation in an HDAC7-dependent manner and increasing glucocorticoid sensitivity of PDX cells by repressing RUNX2.

    HDAC7 induction combined with standard-of-care chemotherapy provides a therapeutic advantage in t(4;11) infant B-cell acute lymphoblastic leukemia. · de Barrios O, Ocón-Gabarró I, Gusi-Vives M, Collazo O, Meler A, Romecín PA, et al. · Biomark Res 2025
    PMID 40722046DOI
    What this evidence cannot show

    Abstract read from the eutils record; full text not read. **The EZH2 step is the load-bearing one and is the least measured**: the abstract does not report EZH2 ChIP or H3K27me3 at HDAC7, and no PRC2 inhibitor arm was run. The intervention tested is menin inhibition plus a class I HDAC inhibitor. A multiomics attribution of silencing to EZH2 is weaker than a chromatin measurement at the locus.

    Evidence that argues against this (1)
    • Rule 65 applies to the effector gene itself: HDAC9 carries HDAC7 as a synonym and its protein name reads 'Histone deacetylase 7B', and in this disease the two point in opposite therapeutic directions - HDAC7 low and to be induced, HDAC9 high and to be repressed. Any downstream use of this hypothesis must assert HDAC7 = Q8WUI4, 952 aa. UniProt 2026
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    A whole-genome CRISPR screen on a bivalent MHC class I readout identifies MTF2-PRC2.1, PCGF1-PRC1.1 and Menin-KMT2A/B as bivalency maintainers; menin loss or pharmacological inhibition phenocopies polycomb disruption and de-represses bivalent genes, and a menin-independent KMT2A/B function maintains H3K4me3 and opposes polycomb at bivalent genes, with release of KMT2A from active genes after menin targeting shifting the balance toward activation.

    Targeting Menin disrupts the KMT2A/B and polycomb balance to paradoxically activate bivalent genes. · Sparbier CE, Gillespie A, Gomez J, Kumari N, Motazedian A, Chan KL, et al. · Nat Cell Biol 2023
    PMID 36635503DOI
    What this evidence cannot show

    Abstract read from the eutils record; full text not read. Cancer cell lines and pluripotent stem cells, not a KMT2A-rearranged leukaemia, and the KMT2A/B concerned is wild-type and SET-competent. A fusion lacking the SET domain cannot perform the H3K4me3-maintaining function described, so if this mechanism operates in a fusion-driven blast the redistributing species may be the untranslocated wild-type allele.

    Evidence that argues against this (1)
    • That the fusion lacks the SET domain is also an argument that this mechanism may simply not apply in the index disease, in which case it is not a competitor to this hypothesis at all but a description of normal cells. Deciding which requires knowing whether wild-type KMT2A redistributes in a fusion-driven blast under a menin inhibitor, which nobody has measured. Janssens DH et al. 2021
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    In 11 KMT2A-rearranged DepMap leukaemia lines neither EZH2 (-0.201, difference +0.004) nor EZH1 (-0.100, +0.002) is a dependency, while EED is (-0.458, difference -0.117, rank 824 of 17,728) and the EED difference is carried by the B-lineage half of the index arm (-0.224) rather than the myeloid half (-0.058).

    What this evidence cannot show

    Single-gene knockouts cannot see the EZH1/EZH2 redundant pair, so the catalytic nulls are uninformative about PRC2 catalysis. The EED lean is modest (-0.117) 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 effects, no z, no permutation null.

    Evidence that argues against this (1)
    • If the hypothesis is right, EZH2 PROTEIN LEVEL is the menin-controlled variable, and a CRISPR knockout that removes EZH2 entirely tests something different from a menin-driven reduction in EZH2 with EZH1 intact. The DepMap null is therefore weak evidence against the hypothesis and is recorded here as a constraint on how the hypothesis can be stated, not as a refutation. Neff T et al. 2012
Connected to