Oncofusion

start / graph / analogy-ezh2-driven-mhc-silencing-in-fus-erg-and-cbfa2t3-glis2

Analogymoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q54-immune

Another fusion family did the experiment this beat wants: in FUS::ERG and CBFA2T3::GLIS2 AML, MHC class I and II are absent at diagnosis, EZH2 drives it, and an EZH2 inhibitor restores presentation

also called polycomb-driven MHC silencing in fusion leukaemia · t(16;21) FUS::ERG immune evasion · tazemetostat restores MHC presentation · EZH2 overexpression and antigen presentation in paediatric AML

A different, rarer fusion leukaemia switches off both kinds of immune identity card before treatment even starts, and the reason is a polycomb enzyme. A licensed drug against that enzyme turns the cards back on and lets immune cells kill the leukaemia - which is the experiment nobody has done for KMT2A.

The same thing, for a scientist

The premise of this whole beat - that polycomb holds MHC genes bivalent and that releasing it restores antigen presentation - has been demonstrated **in a fusion-driven paediatric acute leukaemia**, just not this one. Buteyn et al. report that in **t(16;21) FUS::ERG AML**, MHC class I and class II receptors and costimulatory molecules are **absent at diagnosis**, mirroring the phenotype of post-transplant relapse; that this is **driven by EZH2 overexpression**, in vitro and in multiple clinical cohorts; that the same phenotype is **shared by CBFA2T3::GLIS2-driven AML** and that RUNX1::RUNX1T1 patients do significantly worse when EZH2 overexpression co-occurs; and that the FDA-approved EZH2 inhibitor **tazemetostat reverses the phenotype, re-establishes MHC presentation and elicits immune effector cell-mediated elimination** [E1]. Three of those four entities have entries in this graph - `fusion-cbfa2t3-glis2`, `fusion-runx1-runx1t1`, `drug-tazemetostat` - and none of them says anything about antigen presentation. **The analogy is structural, not loose:** same enzyme family, same mark, same readout, same drug class as `mechanism-mhc-class-i-bivalency-under-menin-inhibition`, in a paediatric fusion leukaemia with patient cohorts.

Go deeper
The mapping, term by term
KMT2A-rearranged (this beat)FUS::ERG and CBFA2T3::GLIS2 (the analogue)
MHC class I bivalency inferred, never measured in a blastMHC class I and class II absent at diagnosis, measured in patient cohorts
polycomb proposed as the repressor, from a CML screenEZH2 overexpression identified as the driver, in vitro and in multiple clinical cohorts
menin inhibitor releases the mark indirectly, by KMT2A redistributionEZH2 inhibitor releases the mark directly
T-cell killing shown in murine SCLC-OVA and in NPM1c AMLimmune effector cell-mediated elimination shown after tazemetostat
no genotype-matched immune-evasion phenotype at diagnosisphenotype is present at diagnosis and defines a high-risk subset
tazemetostat is in this graph with no immune contenttazemetostat is the therapeutic proposal

Sparbier's own data make the bridge explicit from the other end: menin inhibition phenocopies and augments EZH2 inhibition for MHC-I induction, with combined treatment 'dramatically' augmenting the capacity of an EZH2 inhibitor to induce surface MHC class I 2. So the two drug classes are already shown to converge on this readout in a non-leukaemic system, and the analogue supplies the leukaemic patient data.

What transfers, what does not, and the honest verdict

Transfers well.

  • The measurement plan. Buteyn measured class I, class II and costimulatory molecules at diagnosis across clinical cohorts. That is exactly the measurement missing for KMT2A-rearranged disease, and it is a flow panel on banked diagnostic material.
  • The comparator logic. EZH2 overexpression worsening outcome in RUNX1::RUNX1T1 patients is a within-disease dose argument, not a between-disease one.
  • The drug. Tazemetostat is approved and is in this graph already.

Does not transfer, and this is the part to say out loud.

  • The direction of the phenotype may be opposite. In FUS::ERG the blasts are MHC-null at diagnosis. In the MLL-AF9 model, class I is elevated on leukaemia-initiating cells relative to healthy marrow and is an in vivo dependency (mechanism-mhc-class-i-is-an-nk-brake-in-kmt2a-fusion-leukaemia). If KMT2A-rearranged blasts are MHC-high rather than MHC-low, the analogy inverts: there is nothing to restore, and the relevant biology is NK inhibition rather than T-cell blindness.
  • This graph's relationship with EZH2 is already contested. hyp-prc2-is-the-fusions-repressive-arm and risk-ezh2-inhibition-b-lineage-translocations pull in different directions, and the second records a specific hazard of EZH2 inhibition in B-lineage disease.
  • It is a preprint. bioRxiv, 2025, no peer review found by this beat.

Verdict: a strong analogy whose main value is the measurement it licenses, not the therapy it suggests. Measure MHC class I and II at diagnosis in KMT2A-rearranged material. If it is low, the FUS::ERG programme transfers wholesale. If it is high, this beat's NK entry becomes the load-bearing one and the menin-inhibitor immune rationale weakens.

Facts and figures
domainengineering
transferable experimentflow panel for MHC class I, MHC class II and costimulatory molecules on banked diagnostic KMT2A-rearranged material, against a genotype-matched comparator - the measurement that decides whether this analogy holds or inverts
mapping
[
  {
    "bio": "MHC class I bivalency in KMT2A-rearranged blasts is inferred from a CRISPR screen in a CML line and has never been measured in a leukaemic blast",
    "analog": "MHC class I and II absence at diagnosis in FUS::ERG AML is measured directly in multiple clinical cohorts"
  },
  {
    "bio": "polycomb is proposed as the repressor that a menin inhibitor releases indirectly, by redistributing KMT2A",
    "analog": "EZH2 overexpression is identified as the driver and an EZH2 inhibitor releases it directly"
  },
  {
    "bio": "T-cell killing after menin inhibition is shown in murine SCLC-OVA and in NPM1c AML, not in the index genotype",
    "analog": "immune effector cell-mediated elimination is shown after tazemetostat in the index genotype of that disease"
  },
  {
    "bio": "no immune-evasion phenotype is documented at diagnosis in KMT2A-rearranged leukaemia",
    "analog": "the phenotype is present at diagnosis and defines a targetable high-risk subset"
  },
  {
    "bio": "in the MLL-AF9 model class I is ELEVATED on leukaemia-initiating cells and is an NK-dependent in vivo dependency",
    "analog": "in FUS::ERG the blasts are MHC-null - so the analogy may invert, and that is the decisive unknown"
  }
]
generated hypotheses
[
  "hyp-menin-inhibitor-immunotherapy-ordering-is-antigen-dependent"
]
graph entries this touches that carry no immune content
[
  "fusion-cbfa2t3-glis2",
  "fusion-runx1-runx1t1",
  "drug-tazemetostat",
  "gene-ezh2"
]
nup98 bounded absence
{
  "query": "NUP98 AND (\"MHC class I\"[tiab] OR \"antigen presentation\"[tiab] OR \"immune evasion\"[tiab] OR \"immune escape\"[tiab])",
  "count": 3,
  "on_topic_after_title_check": 0,
  "date": "2026-09-20"
}
How this could bite us back
  • If the analogy inverts - if KMT2A-rearranged blasts are MHC-high rather than MHC-low - then a whole therapeutic programme built on restoring antigen presentation is being aimed at a phenotype this disease does not have.
  • EZH2 inhibition in B-lineage disease carries its own documented hazard in this graph (`risk-ezh2-inhibition-b-lineage-translocations`), so borrowing the FUS::ERG therapy into infant B-ALL is not a neutral import.
  • Menin inhibition needs an intact polycomb system to act against; combining it with thorough polycomb disruption removes its own substrate, which the source paper shows directly.
  • The analogue is a preprint. Building a measurement plan on it is defensible; building a trial on it is not.
Still open
  1. Are MHC class I and class II low, normal or high on KMT2A-rearranged blasts at diagnosis, relative to a matched comparator? This one measurement decides whether the FUS::ERG programme transfers or inverts, and it can be done on banked material with a flow panel.
  2. Does EZH2 overexpression stratify outcome in KMT2A-rearranged disease as it does in RUNX1::RUNX1T1? The expression data exist in published cohorts and the analysis has not been run.
  3. If KMT2A-rearranged blasts turn out to be MHC-high, is the correct analogy not FUS::ERG at all but the NK-evasion literature - and should the therapeutic direction be NK engagement rather than antigen-presentation restoration?
Every source behind this entry (3)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.preprint

    In t(16;21) FUS::ERG-positive paediatric AML, major histocompatibility complex class I and class II receptors and costimulatory molecules are absent at diagnosis, mirroring the phenotype and outcomes of post-transplant relapse; this is driven by overexpression of EZH2, shown in vitro and in multiple clinical cohorts; the phenotype is shared by CBFA2T3::GLIS2-driven AML, and RUNX1::RUNX1T1 patients have significantly worse outcomes when EZH2 overexpression co-occurs; and the FDA-approved EZH2 inhibitor tazemetostat reverses the phenotype, re-establishes MHC presentation and elicits immune effector cell-mediated elimination.

    EZH2-driven immune evasion at disease presentation defines a targetable high-risk subset of acute leukemia exemplified by t(16;21) FUS::ERG AML. · Buteyn NJ, LaMantia SJ, Burke CG, Sartori VJ, Deering-Gardner E, DeBruine ZJ, et al. · bioRxiv 2025
    PMID 38798454DOI
    What this evidence cannot show

    A bioRxiv preprint, not peer reviewed as of 2026-09-20; this graph's preprint discipline applies and the confidence tier reflects that rather than the quality of the design. Cohort sizes are not given in the abstract. t(16;21) FUS::ERG is very rare, so 'multiple clinical cohorts' will be small. 'MHC absent at diagnosis' is a strong claim that the abstract does not quantify - absent by flow, by transcript, or both, and against what comparator, is not stated. No KMT2A-rearranged arm. Read: abstract only; PubMed record and DOI fetched 2026-09-20.

    Evidence that argues against this (1)
    • The direction of the phenotype may be the opposite in this graph's index family. In the MLL-AF9 model a classical class I molecule is elevated on leukaemia-initiating cells relative to healthy bone marrow and is a top in vivo dependency, and HLA-A is reported elevated on human AML blasts relative to normal haematopoietic stem cells. If KMT2A-rearranged blasts are MHC-high rather than MHC-low, the FUS::ERG programme does not transfer and the EZH2-inhibitor rationale does not either. Ghosh S et al. 2025
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Menin inhibition functionally phenocopies EZH2 inhibition: treatment of MHC class I-low human SCLC and neuroblastoma lines with VTP50469 increased IFN-gamma-induced cell-surface MHC-I with a degree of induction mirroring the EZH2 inhibitor EPZ-011989; MHC-I is not fully derepressed after EED deletion and additional menin or LEDGF depletion augments it; combination of menin and EZH2 inhibition dramatically augmented the capacity of EZH2 inhibition to induce cell-surface MHC-I; and EZH2-mutant DLBCL showed similar sensitivity to VTP50469 as to EZH2 inhibition with combination therapy substantially enhancing tumour killing.

    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., Dawson MA · Nat Cell Biol 2023
    PMID 36635503DOI
    What this evidence cannot show

    None of this is in leukaemia except the DLBCL viability arm, and none in a KMT2A-rearranged cell. The phenocopy is at the level of a surface readout in MHC-I-low solid tumour lines, which are chosen precisely because polycomb silences their class I - a system built to show this effect. Read: full text via Europe PMC fullTextXML (PMC7614190), 2026-09-20.

    Evidence that argues against this (1)
    • The same paper also shows the two drugs are not interchangeable in the opposite direction: disrupting both PRC1 and PRC2 largely abrogated the effect of menin inhibition on MHC-I even though those cells had not reached maximum MHC-I, which means the menin inhibitor needs an intact polycomb system to act against. A combination that removes polycomb thoroughly would remove the menin inhibitor's own substrate. Sparbier CE et al. 2023
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    As of 2026-09-20 the analogous literature is thin and asymmetric: `NUP98 AND ("MHC class I"[tiab] OR "antigen presentation"[tiab] OR "immune evasion"[tiab] OR "immune escape"[tiab])` returns 3 records, all title-checking to unrelated topics (two versions of a SARS-CoV-2 ORF6 study and a 2006 nuclear-pore paper), so NUP98-rearranged leukaemia has no documented antigen-presentation phenotype; while `(CBFB-MYH11 OR "inv(16)" OR RUNX1-RUNX1T1) AND (the same four phrases)` returns 5 records of which the FUS::ERG preprint is the only one reporting an antigen-presentation phenotype in a core-binding-factor fusion context.

    What this evidence cannot show

    Bounded absence for NUP98 specifically. Name list: NUP98 alone - not NUP98-NSD1, NUP98-KDM5A, nucleoporin 98, or the individual partner symbols, and rule 64 warns that a component-keyed absence can be structurally wrong. Four phrases were used and none of the symbol forms (HLA-A, B2M, TAP1). The CBF query uses hyphenated fusion strings, and `inv(16)` is a parenthesised token of exactly the class the brief warns can be mis-parsed - its querytranslation was read back and expanded to "inv 16"[All Fields], which is correct behaviour and not the der(n) failure mode.

    Evidence that argues against this (1)
    • This graph has a NUP98 beat with its own findings, and a zero from four phrases is not a survey of that literature. The correct reading is that no NUP98 antigen-presentation phenotype is reachable from these phrases, not that none exists - and PMID 37078511 and others in the wider NUP98-plus-immunotherapy set were not examined here. Q54-immune et al. 2026
Connected to