MechanismmoderateQ3-mechanism-core

Polycomb in KMT2A-r leukemia — core PRC2 required, EZH2 and BMI1 dispensable, and CBX8 acting outside its repressive role

also called PRC1 PRC2 in MLL-r · CBX8 non-canonical · EZH2 dispensable · BMI1 dispensable MLL-AF9 · EED requirement

Polycomb proteins normally switch genes off. In KMT2A leukemia they behave strangely: one of them, CBX8, is actually needed to switch the fusion's target genes ON, and the enzyme everyone tries to drug (EZH2) turns out to be the dispensable part.

The same thing, for a scientist

Four results here contradict the textbook. (1) Core PRC2 is required but EZH2 is not: Ezh2-null MLL-AF9 AML maintains self-renewal through three rounds of transplantation with locus-specific H3K27me3 persisting via EZH1, whereas inactivating the essential PRC2 subunit Eed ablates PRC2 function and is incompatible with leukemia growth. (2) BMI1/PRC1 is dispensable specifically in MLL-AF9, because HOXA9 substitutes for its senescence-bypass function. (3) CBX8 is required for MLL-AF9-induced transcriptional ACTIVATION and leukemogenesis, and Cbx8-null mice are viable with no apparent haematopoietic defect — an unusually favourable therapeutic-index profile. (4) But the direct CBX8-AF9 interface is dispensable for proliferation, and MLL-ENL appears to antagonise CBX8/PRC1 rather than partner with it. The non-canonical CBX8 role is strong but mechanistically contested.

Go deeper

PRC2: the core is required, the drug target is not1 deeper

This distinction has direct consequences for whether EZH2 inhibitors belong in KMT2A-r.

Ezh2-null MLL-AF9 AML failed to accelerate on secondary transplantation but maintained self-renewal up to the third round of transplantation, indicating Ezh2 is not strictly required for MLL-AF9 AML and plays a role in progression rather than maintenance; H3K27me3 persisted locus-specifically despite Ezh2 inactivation, attributable to EZH1 compensation. By contrast, inactivating the essential PRC2 gene Eed completely ablated PRC2 function and was incompatible with leukemia growth, with downregulation of Myc targets and upregulation of PRC2 targets 1.

An independent shRNA screen in an MLL-AF9;Nras-G12D model reached a compatible conclusion from the other side: suppression of Eed, Suz12 or Ezh1/Ezh2 together caused proliferation arrest and differentiation with minimal impact on non-transformed haematopoietic cell lines, partly through direct PRC2 repression of self-renewal-limiting genes including Cdkn2a 2.

So: PRC2 is a genuine dependency; EZH2 alone is not the right node. An EZH2-selective inhibitor is predicted to be buffered by EZH1. This is a concrete, cheap prediction — dual EZH1/EZH2 inhibitors (valemetostat class) should outperform EZH2-selective agents (tazemetostat class) in KMT2A-r, and nobody appears to have run that head-to-head in this disease.

Note also the awkward direction of the Cdkn2a finding: PRC2 is required because it represses a tumour suppressor. That means PRC2 inhibition works by de-repression, which is the opposite logic to the rest of this graph and predicts that CDKN2A-deleted KMT2A-r cases — which exist — should be resistant.

BMI1/PRC1 is dispensable, and for an instructive reason

AML1-ETO and PLZF-RARalpha, which do not activate Hox, triggered senescence in Bmi1-null cells. MLL-AF9, which drives Hoxa7 and Hoxa9, readily transformed Bmi1-null cells. MLL-AF9 could not initiate leukemia in Bmi1-/-;Hoxa9-/- mice, but either gene alone restored leukemia-stem-cell establishment, and HOXA9 regulates the p16Ink4a/p19ARF locus 3.

The mechanism is clean: BMI1's essential contribution to leukemogenesis is senescence bypass via Cdkn2a repression, and HOXA9 does that job independently. So the KMT2A fusion arrives with its own senescence-bypass module built in, and BMI1 becomes redundant. This is a nice worked example of the graph's recurring theme — the fusion's output program creates redundancy that removes candidate targets from the list.

CBX8: the non-canonical case, and the contradiction at its heart

The positive result. CBX8, a Polycomb-group protein, interacts with MLL-AF9 and TIP60 and is required for MLL-AF9-induced transcriptional activation and leukemogenesis. Both CBX8 ablation and point mutations in MLL-AF9 that specifically disrupt the CBX8 interaction abrogate HOX upregulation and abolish leukemic transformation. And — the therapeutically important part — Cbx8-deficient mice are viable and display no apparent haematopoietic defects 4.

That combination (required by the leukemia, dispensable in normal blood) is the exact dependency profile this project is hunting for, and it is arguably better than menin's, since menin-null mice have a long-term HSC defect under transplantation stress.

Chemical validation exists. SW2_110A, a DNA-encoded-library-derived CBX8 chromodomain inhibitor (Kd 800 nM, at least 5-fold selective over other CBX paralogues, cell-permeable), inhibits CBX8-chromatin association in cells and inhibits proliferation of MLL-AF9-driven THP-1 cells with significant decrease in fusion target genes including HOXA9 — validating the activating role and defining the H3K27me3-reading chromodomain as necessary for it 5. A repressive-mark reader domain being required for gene activation is a genuinely odd and interesting result.

The contradiction. Structures of AF9 complexes with CBX8 and BCOR show that binding of AF4, DOT1L, BCOR and CBX8 to AF9 is mutually exclusive. Using selective point mutants, disrupting the direct CBX8/MLL-AF9 interaction did NOT impact cell proliferation, whereas loss of direct BCOR/MLL-AF9 binding caused partial differentiation and increased proliferation and abrogated leukemogenic potential in vivo 6.

And a third position exists: MLL-ENL antagonises CBX8/PRC1. ENL's contacts with elongation machinery and with CBX8 are mutually exclusive, CBX8 inhibited elongation in a reporter assay and this was neutralised by direct association with ENL, and a CBX8-binding-defective MLL-ENL could not fully activate the loci needed for transformation 7. That is a titration model — the fusion sequesters PRC1 away from its normal sites — which is the mechanistic inverse of 'CBX8 is a positive cofactor'.

How to hold these together. CBX8 protein and its chromodomain are required 4 5, but the direct CBX8-AF9 interface is not 6. The only reconciliation is that CBX8 acts substantially off the AF9 interface — through its chromodomain at chromatin, and/or by being titrated as in 7. For drug discovery that is good news, because it points at the chromodomain (druggable, validated by a probe) rather than at a protein-protein interface that the genetics says is dispensable.

Finally, the fusion also uses Polycomb for targeted repression rather than merely evading it: it represses BIM through CBX8/PRC1 recruitment 8, and the PROM1 locus is PRC2-occupied when silent 9. Polycomb here is a tool the fusion wields, not only an obstacle it overcomes.

Facts and figures

chain	mechanism-kmt2a-fusion-causal-chain
other dependency surfaced	BCOR-MLL-AF9 interface (required in vivo; largely absent from standard target maps)

prc2

{
  "core_required": true,
  "required_subunit": "EED (and SUZ12)",
  "ezh2_required": false,
  "ezh1_compensation": true,
  "mechanism": "represses self-renewal-limiting genes including Cdkn2a",
  "drug_prediction": "dual EZH1/EZH2 inhibitors should outperform EZH2-selective agents; untested in KMT2A-r"
}

prc1

{
  "bmi1_required": false,
  "bmi1_bypass_explanation": "HOXA9 independently represses the p16Ink4a/p19ARF locus",
  "cbx8_protein_required": true,
  "cbx8_chromodomain_required": true,
  "cbx8_af9_interface_required_for_proliferation": false
}

cbx8

{
  "role": "non-canonical — required for fusion-driven ACTIVATION of HOXA9",
  "normal_tissue": "Cbx8-null mice viable, no apparent haematopoietic defect",
  "chemical_probe": "SW2_110A, Kd 800 nM, >=5-fold paralogue-selective, cell-permeable",
  "contested_by": [
    "Schmidt 2020 interface mutants",
    "Maethner 2013 antagonism model"
  ],
  "tip60_mechanism_confidence": "weak"
}

fusion uses polycomb

[
  "represses BIM via CBX8/PRC1",
  "PROM1 locus PRC2-repressed when silent"
]

How this could bite us back

PRC2 inhibition works partly by de-repressing tumour suppressors such as CDKN2A, so CDKN2A-deleted cases — which occur in KMT2A-r — are predicted to be intrinsically resistant.
EZH2-selective inhibition is predicted to be buffered by EZH1, so a negative EZH2-inhibitor result in this disease should not be read as a negative PRC2 result.
The fusion uses CBX8/PRC1 to repress BIM; disabling CBX8 could de-repress BIM (helpful) but also de-repress other Polycomb targets unpredictably.
Because the fusion titrates PRC1 away from its normal sites, removing the fusion redistributes Polycomb rather than simply restoring the pre-leukemic state — a poorly characterised transition that could be its own source of instability.
Cbx8-null mice being viable has not been stress-tested by competitive transplantation, which is exactly the test that exposed the menin HSC defect.

Still open

Do dual EZH1/EZH2 inhibitors outperform EZH2-selective inhibitors in KMT2A-r? The EZH1-compensation result predicts yes, both compound classes are clinically available, and the head-to-head has apparently never been run in this disease.
Is CBX8 the best therapeutic-index target in the whole KMT2A-r landscape? Required by the leukemia, viable knockout in normal blood, and a validated chromodomain probe exists. It deserves a full dossier (DepMap, GTEx/HPA, gnomAD, ChEMBL) — arguably ahead of several better-known partners.
How can CBX8 protein and its chromodomain be required while the direct CBX8-AF9 interface is dispensable? The reconciliation determines whether the drug should be a chromodomain ligand or an interface disruptor — currently the data favour the chromodomain.
Is BCOR a missed target? Its interface with MLL-AF9 is required in vivo and it is nearly absent from published KMT2A-r target maps.
Do CDKN2A-deleted KMT2A-r cases lose the PRC2 dependency, as the Cdkn2a mechanism predicts? This is directly testable in existing PDX panels with known CDKN2A status.

Every source behind this entry (9)

E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Ezh2-null MLL-AF9 AML failed to accelerate on secondary transplantation but maintained self-renewal to the third round, indicating Ezh2 is not strictly required and acts in progression, with H3K27me3 persisting locus-specifically via EZH1; by contrast inactivation of the essential PRC2 gene Eed completely ablated PRC2 function and was incompatible with leukemia growth, with downregulated Myc targets and upregulated PRC2 targets.
Polycomb repressive complex 2 is required for MLL-AF9 leukemia · Neff T, Sinha AU, Kluk MJ, et al. (Orkin SH, Armstrong SA, senior authors) · Proceedings of the National Academy of Sciences USA 109(13):5028-5033 2012
PMID 22396593 DOI
What this evidence cannot show
Mouse conditional deletion in MLL-AF9 myeloid leukemia; Eed loss ablates all PRC2 function, which is a more severe perturbation than any drug achieves, so 'PRC2 required' does not translate directly into a dosing window. EZH1 compensation was inferred from persisting H3K27me3 rather than measured by EZH1 deletion in the same experiment.
Evidence that argues against this (1)
- This is murine MLL-AF9 in the myeloid compartment and the index disease of this graph is human infant KMT2A::AFF1 B-lymphoid leukaemia. The two differ in the partner, in the lineage, in the cell of origin and — on this beat's own evidence — in HOXA output, since infant t(4;11) is HOXA-LOW while the mouse models are built on HOXA activation. Retroviral MLL-AF9 also expresses the fusion far above the single rearranged allele a patient carries. The result is also internally mixed: Ezh2 is dispensable with H3K27me3 persisting through EZH1, while Eed loss is disabling. That makes PRC2 as a COMPLEX required and its catalytic subunits interchangeable, which is close to the opposite of a drug target. Oncofusion knowledge graph (internal analysis) 2026
E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In an MLL-AF9;Nras-G12D AML shRNA screen, suppression of PRC2 components Eed, Suz12 or Ezh1/Ezh2 caused proliferation arrest and differentiation with minimal impact on several non-transformed haematopoietic cell lines, mediated partly by direct PRC2 repression of self-renewal-limiting genes including Cdkn2a.
The Polycomb complex PRC2 supports aberrant self-renewal in a mouse model of MLL-AF9;Nras(G12D) acute myeloid leukemia · Shi J, Wang E, Zuber J, et al. (Lowe SW, Vakoc CR, senior authors) · Oncogene 32(7):930-938 2013
PMID 22469984 DOI
What this evidence cannot show
Compound MLL-AF9;NrasG12D model, so the RAS lesion may modify the PRC2 requirement; shRNA knockdown with variable depletion. The Cdkn2a mechanism predicts that CDKN2A-deleted cases would be resistant, which was not tested.
Evidence that argues against this (1)
- This is murine MLL-AF9 in the myeloid compartment and the index disease of this graph is human infant KMT2A::AFF1 B-lymphoid leukaemia. The two differ in the partner, in the lineage, in the cell of origin and — on this beat's own evidence — in HOXA output, since infant t(4;11) is HOXA-LOW while the mouse models are built on HOXA activation. Retroviral MLL-AF9 also expresses the fusion far above the single rearranged allele a patient carries. Comparing against "several non-transformed haematopoietic cell lines" is the weak control: immortalised non-transformed lines are not the normal progenitors whose sparing would establish a window, and shRNA screens in this era carry off-target effects that a small set of hairpins cannot exclude. Oncofusion knowledge graph (internal analysis) 2026
E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
AML1-ETO and PLZF-RARalpha, which do not activate Hox, triggered senescence in Bmi1-null cells, whereas MLL-AF9, which drives Hoxa7 and Hoxa9, readily transformed Bmi1-null cells; MLL-AF9 could not initiate leukemia in Bmi1-/-;Hoxa9-/- mice but either gene restored leukemia-stem-cell establishment, and HOXA9 regulates the p16Ink4a/p19ARF locus.
Functional crosstalk between Bmi1 and MLL/Hoxa9 axis in establishment of normal hematopoietic and leukemic stem cells · Smith LL, Yeung J, Zeisig BB, et al. (So CWE, senior author) · Cell Stem Cell 8(6):649-662 2011
PMID 21624810 DOI
What this evidence cannot show
Germline Bmi1-null and Hoxa9-null mice, so developmental compensation is possible; myeloid MLL-AF9 context. The functional substitution argument rests on the shared Cdkn2a axis and is inferred rather than directly measured at that locus in every genotype.
Evidence that argues against this (1)
- This is murine MLL-AF9 in the myeloid compartment and the index disease of this graph is human infant KMT2A::AFF1 B-lymphoid leukaemia. The two differ in the partner, in the lineage, in the cell of origin and — on this beat's own evidence — in HOXA output, since infant t(4;11) is HOXA-LOW while the mouse models are built on HOXA activation. Retroviral MLL-AF9 also expresses the fusion far above the single rearranged allele a patient carries. The logic also runs through Hox: MLL-AF9 transformed Bmi1-null cells while AML1-ETO and PLZF-RARalpha did not, and the difference is attributed to Hoxa7/Hoxa9. In infant t(4;11), which is HOXA-low, the premise of that explanation is absent. Oncofusion knowledge graph (internal analysis) 2026
E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
CBX8, a Polycomb-group protein, interacts with MLL-AF9 and TIP60 and is required for MLL-AF9-induced transcriptional activation and leukemogenesis; both CBX8 ablation and point mutations in MLL-AF9 that disrupt the CBX8 interaction abrogate HOX upregulation and abolish leukemic transformation, and Cbx8-deficient mice are viable with no apparent haematopoietic defects.
CBX8, a polycomb group protein, is essential for MLL-AF9-induced leukemogenesis · Tan J, Jones M, Koseki H, et al. (Hess JL, senior author) · Cancer Cell 20(5):563-575 2011
PMID 22094252 DOI
What this evidence cannot show
The TIP60/KAT5 arm is an interaction only — the abstract does not demonstrate TIP60 catalytic dependence for the activating function, so any 'CBX8 acts via TIP60' claim should be graded weak. Mouse MLL-AF9; the point-mutant result is directly contradicted for the proliferation endpoint by [E6]. 'No apparent haematopoietic defects' in a viable knockout has not been stress-tested by competitive transplantation.
Evidence that argues against this (1)
- Structures of AF9 complexes with CBX8 and BCOR show AF4, DOT1L, BCOR and CBX8 binding to AF9 is mutually exclusive, and selective point mutants show that disrupting the DIRECT CBX8/MLL-AF9 interaction does not impact cell proliferation, whereas loss of BCOR/MLL-AF9 binding causes partial differentiation, increased proliferation and abrogates leukemogenic potential in vivo. Schmidt CR et al. 2020
E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
SW2_110A, a DNA-encoded-library-optimised CBX8 chromodomain inhibitor (Kd 800 nM, at least 5-fold selective over other CBX paralogues, cell-permeable), inhibits CBX8-chromatin association in cells and inhibits proliferation of MLL-AF9-driven THP-1 leukemia cells with significant decrease in MLL-AF9 target genes including HOXA9, defining the chromodomain as necessary for the activating function of CBX8.
Optimization of Ligands Using Focused DNA-Encoded Libraries To Develop a Selective, Cell-Permeable CBX8 Chromodomain Inhibitor · Wang S, Denton KE, Hobbs KF, et al. (Dykhuizen EC, Krusemark CJ, senior authors) · ACS Chemical Biology 15(1):112-131 2020
PMID 31755685 DOI
What this evidence cannot show
Modest potency (800 nM) and modest paralogue selectivity (at least 5-fold) in a single cell line; at those parameters off-target CBX engagement cannot be excluded, so the chromodomain-requirement conclusion is suggestive rather than definitive. No in-vivo data.
Evidence that argues against this (1)
- An 800 nM chromodomain binder with 5-fold paralogue selectivity is a tool compound, and 5-fold is not selectivity in a family whose members share a fold and a ligand. Antiproliferative activity in THP-1 at concentrations needed to engage an 800 nM target is where off-target effects live, and this graph's own DepMap analysis finds the CBX8 co-dependency in KMT2A-rearranged lines indistinguishable from zero. Oncofusion knowledge graph (internal analysis) 2026
E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Binding of AF4, DOT1L, BCOR and CBX8 to AF9 is mutually exclusive; selective point mutants show disruption of the direct CBX8/MLL-AF9 interaction does not impact cell proliferation, whereas loss of direct BCOR/MLL-AF9 binding causes partial differentiation and increased proliferation and abrogates leukemogenic potential in vivo, acting through altered EYA1, SIX and MYC activity.
BCOR Binding to MLL-AF9 Is Essential for Leukemia via Altered EYA1, SIX, and MYC Activity · Schmidt CR, Achille NJ, Kuntimaddi A, et al. (Zeleznik-Le NJ, Bushweller JH, senior authors) · Blood Cancer Discovery 1(2):162-177 2020
PMID 32954361 DOI
What this evidence cannot show
Structure-guided point mutants in MLL-AF9 mouse models; the 'no effect on proliferation' readout is in-vitro proliferation, which is a less sensitive endpoint than leukemogenesis, so a modest CBX8-interface contribution could be missed. Also identifies BCOR as a KMT2A-fusion dependency that is largely absent from the field's target maps.
Evidence that argues against this (1)
- This is the entry's clearest negative and it is stated without consequence: disrupting the direct CBX8/MLL-AF9 interaction does NOT affect proliferation, while losing BCOR binding does. An entry built around CBX8 has to reckon with the point mutant that removes the interaction and changes nothing. Oncofusion knowledge graph (internal analysis) 2026
E7strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
ENL is a scaffold contacting both elongation machinery and the PRC1 component CBX8, these contacts are mutually exclusive in vitro, MLL-ENL and CBX8 behave antagonistically in vivo, CBX8 inhibited elongation in a reporter assay and this was neutralised by direct association with ENL, and a CBX8-binding-defective MLL-ENL could not fully activate the gene loci needed for transformation.
MLL-ENL inhibits polycomb repressive complex 1 to achieve efficient transformation of hematopoietic cells · Maethner E, Garcia-Cuellar MP, Breitinger C, et al. (Hess JL, Slany RK, senior authors) · Cell Reports 3(5):1553-1566 2013
PMID 23623499 DOI
What this evidence cannot show
KMT2A::MLLT1 rather than MLL-AF9 or KMT2A::AFF1, which may explain the apparent conflict with the CBX8-as-positive-cofactor model; reporter assays and in-vitro exclusivity carry the mechanistic weight.
Evidence that argues against this (1)
- Mutual exclusivity was established in vitro with isolated domains, where competition reflects the concentrations the experimenter chose. In a nucleus the same surfaces can be occupied by different partners on different molecules at the same locus, so "mutually exclusive" constrains a single AF9 molecule and not the behaviour of the complex at a gene. Nature 2012
E8strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
KMT2A::AFF1 directly activates BCL2 and MCL1 while repressing BIM via binding of the PRC1 component CBX8, showing the fusion uses Polycomb for targeted repression rather than only evading it.
MLL-AF4 binds directly to a BCL-2 specific enhancer and modulates H3K27 acetylation · Godfrey L, Kerry J, Thorne R, et al. (Milne TA, senior author) · Experimental Hematology 47:64-75 2017
PMID 27856324 DOI
What this evidence cannot show
Single-locus observation in t(4;11) cell lines; the genome-wide extent of fusion-directed CBX8/PRC1 repression is not established.
Evidence that argues against this (1)
- Direct activation of BCL2 and MCL1 with repression of BIM via CBX8 is the entry's most specific mechanistic claim and rests on binding data at three loci. Apoptotic rheostat genes are among the most frequently reported fusion targets in any leukaemia, and the CBX8 requirement is contradicted within this same entry by the point mutant that abolishes CBX8/MLL-AF9 binding without a proliferation phenotype. Oncofusion knowledge graph (internal analysis) 2026
E9strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In PROM1/CD133 non-expressing cells the PROM1 locus is repressed by PRC2 binding with reduced TAPT1 expression, whereas in expressing cells KMT2A::AFF1 activates it via an intragenic H3K79me2/3 enhancer element — showing PRC2 as the default state at a key fusion target locus.
H3K79me2/3 controls enhancer-promoter interactions and activation of the pan-cancer stem cell marker PROM1/CD133 in MLL-AF4 leukemia cells · Godfrey L, Crump NT, O'Byrne S, et al. (Roy A, Milne TA, senior authors) · Leukemia 35(1):90-106 2021
PMID 32242051 DOI
What this evidence cannot show
Cell-line based; PRC2 occupancy in non-expressors is correlative with silence and was not tested by PRC2 removal at that locus.
Evidence that argues against this (1)
- One locus. PRC2 as the default state at PROM1 in non-expressing cells is the expected configuration of any silent developmental gene, so it does not distinguish a fusion-specific mechanism from ordinary Polycomb silencing of an unexpressed locus. Oncofusion knowledge graph (internal analysis) 2026

Connected to

← part ofKDM2B — 513 potent ChEMBL rows that are mostly patent bins, and no structure of the domain they claim to inhibit
regulates →The primed chromatin state — what is actually established, and why the word 'bivalent' should be used carefully here
← depends onMYC reactivation through PRC1.1 loss — a named, druggable bypass out of menin dependence
← related toPolycomb bodies — indirect evidence via CBX8, undercut by the finding that CBX8-PRC1 condensates are dissolved rather than built by the rest of PRC1
adds the compartment question and the CBX2-versus-CBX8 condensate asymmetry
← related toATM phosphorylates ENL at conserved SQ sites and flips it from elongation to Polycomb repression - on sequence every KMT2A::MLLT1 fusion keeps
supplies the missing kinase input to the PRC1 axis
← supportsEZH2 - the Polycomb enzyme that collaborates with the fusion instead of opposing it, and that no single-agent inhibitor can switch off
supplies the PRC2 subunit entry that entry's id already named
← related toPRC2 - the complex, not the subunit, is the unit every experiment in this disease actually resolves
supplies the PRC2 half of an entry whose id named PRC2 while no PRC2 node existed
part of →The KMT2A-fusion causal chain: six links, one measured ordering, and two arms that are parallel rather than sequential
modifier, not a step - contested repressive arm of the chain