Oncofusion

start / graph / hyp-the-writer-reaches-the-kmt2a-anchor-only-through-ctcf-and-only-after-the-break

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q110-h3k9me3-writer-layer

The H3K9me3 writer reaches the KMT2A anchor only through CTCF and only after the break - which would make this graph's CTCF arm and its H3K9 arm one arm, with CTCF upstream, and would make every resting-state H3K9me3 measurement at that element the wrong assay

also called CTCF-first H3K9 hypothesis · the writer arrives with the break

Two of this graph's arguments about the KMT2A break may be the same argument. A 2026 paper shows that CTCF is what brings the H3K9-silencing enzymes to a DNA break - and CTCF sitting on the KMT2A gene is already the centre of our story. If that is what happens here, the silencing mark appears because the DNA broke, not before it.

The same thing, for a scientist

**The observation that generates this.** CTCF recruits HP1-gamma, KAP1, SUV39H1, SETDB1 and KDM5A to double-strand breaks; CTCF depletion abrogates their enrichment and the H3K9 methylation that goes with it, and the consequence is defective BRCA1/BARD1 loading, defective end resection and dysfunctional homologous recombination [E1]. The older mechanism agrees and adds the kinetics: a KAP1/HP1/SUV39H1 complex loads within minutes, spreads H3K9me3 for tens of kilobases by chromodomain re-reading, activates Tip60 and ATM, and is then stripped off when ATM phosphorylates KAP1 [E2]. **Why that lands exactly on this graph's locus.** `mechanism-h3k9me1-2-ctcf-balance-gates-kmt2a-rearrangement` makes CTCF occupancy at the conserved *KMT2A* exon-11 site the load-bearing node - CTCF depletion alone is sufficient for the rearrangement phenotype and KDM3B co-depletion adds nothing. Q108 then found that KDM3B depletion lowers CTCF **only** at that element while raising H3K9me1 broadly, and concluded the locus address sits *on the CTCF arrow, not the mark arrow*. And this graph holds the element as a TOP2-sensitive, cohesin-bound loop anchor with 2- to 10-fold elevated breakage in the normal remission cells of KMT2A-rearranged patients. **The hypothesis.** At the *KMT2A* exon-11 element, the H3K9me3 writers have **no constitutive address**. They arrive when the element breaks, brought by CTCF, deposit a transient H3K9me3 domain that licenses accurate HR, and are removed by the ATM/KAP1 feedback. Therefore: 1. **Lowering CTCF at the anchor lowers writer recruitment there, and so lowers the repressive-chromatin step that licenses HR at that break.** That is a route from CTCF loss to *rearrangement* that does not require H3K9 methylation to be a gate at all - it requires it to be a repair licence. 2. **The two arms are one arm.** `mark-h3k9me3` and `mechanism-h3k9me1-2-ctcf-balance-gates-kmt2a-rearrangement` need not predict opposite signs for the same intervention if the H3K9me3 at this element is downstream of CTCF and conditional on damage. 3. **A resting-state H3K9me3 measurement at this element is the wrong assay.** Q108's finding that H3K9me3 does not move at the CTCF element in unperturbed, undamaged RPE cells would be the *expected* result, not an informative one - and `risk-the-h3k9me3-exoneration-of-g9a-inhibition-is-epithelial-only` would gain a fifth independent reason. 4. **And it inverts the safety sign of the proposed prophylaxis.** `mechanism-h3k9me1-2-ctcf-balance-gates-kmt2a-rearrangement` proposes UNC0642 alongside doxorubicin or etoposide. If H3K9 methylation at a break licenses HR, then lowering it during anthracycline or epipodophyllotoxin exposure removes a repair licence at the moment TOP2 breaks are being made. **Prior probability: 0.2.** The mechanism is real and well-published; what is entirely unevidenced is that it operates at *this* element, in a haematopoietic cell, at a TOP2 break. Both source papers are non-haematopoietic - U2OS and HEK-293T - with enzyme-induced breaks at reporters, and in a G0/G1 lymphoid progenitor an HR-licensing cascade may not run at all.

Facts and figures
statementAt the KMT2A exon-11 CTCF/cohesin element, SUV39H1 and SETDB1 have no constitutive occupancy; they are recruited by CTCF when the element breaks, deposit a transient H3K9me3 domain that licenses homologous-recombination repair, and are removed by ATM-mediated KAP1 phosphorylation. Consequently CTCF occupancy is upstream of H3K9 methylation at this element, the graph's CTCF arm and H3K9 arm are one arm, and a resting-state H3K9me3 measurement there cannot test either.
originliterature-gap
prior probability0.2
what makes it more than a restatementThis graph already holds 'CTCF is upstream' (from Q108's finding that KDM3B depletion lowers CTCF only at the element) and 'H3K9 methylation matters at this locus' (from the founding paper). What is new is a PHYSICAL mechanism connecting them in that direction - CTCF as the recruiter of the writers - published in 2026 and cited nowhere in this graph before this beat. Without it, 'CTCF is upstream of the mark' was an inference from a correlation of magnitudes.
why the prior is only 0 2Both source papers are in non-haematopoietic cells with enzyme-induced breaks at engineered reporters (U2OS AsiSI and FokI systems, HEK-293T), neither touches the KMT2A locus, and the HP1 paralogue in the 2026 cascade is HP1-gamma rather than HP1-alpha. A cascade whose readout is BRCA1/BARD1-dependent resection is cell-cycle-restricted, and the index disease's putative cell of origin is a fetal progenitor rather than a cycling osteosarcoma line. The mechanism is strong; its applicability here is unevidenced.
localizationchromatin at chr11q23.3 - the KMT2A exon-11 CTCF/RAD21/SMC3 element (GRCh38 chr11:118,488,159-118,489,206) and the 7,466 bp breakpoint cluster region around it
subnuclear localizationnot addressed by this hypothesis and not measured at this element in any cell type. The published cascade is a local chromatin event at a break, not a compartment change, and no imaging, Hi-C or nuclear-compartment assay has ever placed the KMT2A anchor relative to the lamina or a speckle.
cell typethe hypothesis is ABOUT a haematopoietic cell - a CD34+ progenitor or a B-lineage progenitor - and the evidence it is built on is in NONE: U2OS osteosarcoma and HEK-293T for the CTCF cascade, unnamed lines for the KAP1/HP1/SUV39H1 loading, hTERT-RPE for everything this graph knows about the locus. That gap is the hypothesis's whole risk.
cell stateDNA DAMAGE is the state this hypothesis turns on, and it is the state nobody has measured at this locus. Secondary state dependence: an HR-licensing cascade is cell-cycle-restricted, so the hypothesis predicts different behaviour in a quiescent progenitor than in a cycling blast.
epigenetic readoutH3K9me3 as a DAMAGE-INDUCED, TRANSIENT, ATM-REVERSIBLE domain rather than as a steady-state silencing mark - which is a different object from the H3K9me3 that mark-h3k9me3 describes at fusion-target promoters, and the distinction is the hypothesis's main content.
shown for fusionNO - nobody has looked, in any fusion. Neither source paper involves a fusion protein, a leukaemia, or the KMT2A locus, and no damage-state H3K9 measurement at the KMT2A breakpoint exists in any cell type.
predictions
[
  "In a cell with an induced double-strand break at or near the KMT2A exon-11 element, SUV39H1 and SETDB1 ChIP signal at the element RISES within minutes and returns to baseline; in the same cell without a break, neither is detectable above input.",
  "CTCF depletion abolishes that damage-induced writer recruitment at the element, while KDM3B depletion does not abolish it directly but reduces it in proportion to the CTCF it removes.",
  "H3K9me3 at the element is damage-inducible and ATM-reversible: an ATM inhibitor should PROLONG the H3K9me3 domain rather than prevent it, because ATM's role in the published mechanism is to strip the complex off.",
  "The H3K9me3 domain induced at a break spreads tens of kilobases, so it will be visible as a whole-gene or whole-domain change and NOT as a BCR-versus-flank contrast - which predicts that the flank-normalised designs used in this graph so far are structurally unable to see it.",
  "In an unperturbed cell, the writers' occupancy at the element is at or below the local background - the prediction this beat's own ENCODE analysis tests in K562 and GM12878.",
  "Prophylactic G9a/GLP inhibition given with etoposide reduces HR efficiency at TOP2 breaks rather than protecting the locus, measurable as a shift toward end-joining junction signatures."
]
experiments
[
  "exp-h3k9me3-and-ctcf-at-the-kmt2a-anchor-in-a-cd34-progenitor"
]
analyses
[
  "ana-q110-h3k9-writer-and-reader-occupancy-at-the-kmt2a-locus",
  "ana-q108-h3k9me3-at-the-kmt2a-bcr-measured"
]
falsifiers
[
  "CONSTITUTIVE OCCUPANCY. If SUV39H1 or SETDB1 is reproducibly enriched at the KMT2A exon-11 element in an undamaged cell - above the local background and in independent replicates - then the writers have a resting address there and the 'only after the break' half of this hypothesis is dead. This beat's ENCODE analysis is a partial test of it for SETDB1 and CBX5 in two blood cells.",
  "CTCF-INDEPENDENT RECRUITMENT. If a break is induced at the element in CTCF-depleted cells and H3K9me3 still rises there, CTCF is not the route and the 'only through CTCF' half is dead.",
  "H3K9 METHYLATION AT A BREAK IS NOT A REPAIR LICENCE IN A LYMPHOID CELL. If SUV39H1 or SETDB1 loss does not impair HR at a defined break in a haematopoietic cell, the repair-licence reading fails and with it the inverted safety argument - which would be a relief, since that argument currently opposes a proposed prophylaxis.",
  "THE MARK ARM SURVIVES INDEPENDENTLY. If restoring H3K9me1/2 at the element in CTCF-depleted cells prevents the rearrangement phenotype, then the mark is downstream of CTCF and still causal in its own right, and 'one arm with CTCF upstream' is too simple.",
  "TIMING. If H3K9me3 at the element in a KDM3B-depleted cell is stable over 72 hours rather than transient, the ATM/KAP1 feedback is not operating there and the transient-domain model does not apply to this locus."
]
what it would change if true
{
  "for_the_dispute": "mark-h3k9me3 and mechanism-h3k9me1-2-ctcf-balance-gates-kmt2a-rearrangement stop predicting opposite signs for the same intervention, because they would be describing a mark at fusion-target promoters and a damage-conditional mark at one loop anchor - two different objects.",
  "for_the_experiments": "every resting-state H3K9 measurement at this element, including Q108's null, becomes uninformative rather than informative, and the right experiment acquires a damage arm it has never had.",
  "for_the_drug_proposal": "the sign of prophylactic G9a/GLP inhibition alongside anthracycline or etoposide inverts from protective to potentially harmful, because it would be removing a repair licence at the moment the breaks are made.",
  "for_the_menin_interface_hypothesis": "nothing directly. This is about the genesis of the fusion, not about the fusion's dependence on menin."
}
How this could bite us back
  • IF THIS HYPOTHESIS IS TRUE IT INVERTS THE SIGN OF A PROPHYLAXIS THIS GRAPH ALREADY HOLDS. mechanism-h3k9me1-2-ctcf-balance-gates-kmt2a-rearrangement proposes UNC0642, a dual G9a/GLP inhibitor, given before doxorubicin or etoposide in order to prevent KMT2A rearrangement. If H3K9 methylation at a break is what licenses accurate homologous-recombination repair, then lowering it during anthracycline or epipodophyllotoxin exposure is removing a repair licence exactly when TOP2 breaks are being made. The hypothesis is speculative and the drug proposal is for children receiving curative therapy, so the asymmetry of cost matters: this must be tested before the prophylaxis is advanced, not after.
  • It is not a licence for the opposite intervention either. 'Raise H3K9 methylation to license repair' would be acting on a mark that spreads tens of kilobases from a break, silences indiscriminately when raised globally, and is set by at least three enzyme-independent mechanisms in this graph's own entries. Nothing here is addressable at one element.
  • The hypothesis is hard to falsify with the assay this beat could run. A population ChIP in undamaged cells cannot see a transient, asynchronously induced, ATM-stripped domain, so a null from this beat's ENCODE analysis supports the 'no constitutive occupancy' half and cannot test the 'arrives with the break' half at all. A hypothesis whose interesting half needs an assay nobody has run should be held at low prior and said to be so.
Still open
  1. Does an induced break at the KMT2A exon-11 element recruit SUV39H1 or SETDB1 there? No damage-state occupancy measurement exists at this locus in any cell type, and a CRISPR or AsiSI-based break at the element with a short ChIP time course is the direct test.
  2. Is the H3K9me3 that this graph argues about at the locus damage-induced or steady-state? Every measurement in the corpus is a single timepoint in unirradiated cells, and the two possibilities predict opposite experiments.
  3. Is the flank-normalised design used by every H3K9 measurement here structurally blind to a break-induced domain? A domain spreading tens of kilobases would raise the BCR and its 20 kb flanks together and cancel in the ratio. That is checkable against the existing deposited data by widening the comparator.
  4. Does H3K9 methyltransferase loss impair HR at a TOP2 break in a haematopoietic cell? If it does, prophylactic G9a/GLP inhibition alongside etoposide is acting against repair at the moment of maximum break load, and the sign of the proposed prophylaxis inverts.
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In U2OS and HEK-293T cells, CTCF recruits the repressive chromatin proteins HP1-gamma, KAP1, SUV39H1, SETDB1 and KDM5A to double-strand breaks, thereby enabling rapid accumulation of BRCA1/BARD1; CTCF depletion abrogates the enrichment of these proteins with loss of H3K9 methylation and H3K4 demethylation at DSBs and defective rapid BRCA1/BARD1 recruitment; depriving any one component, or deregulating H3K9 methylation, compromises the immediate accumulation of the others and of BRCA1/BARD1; and the consequence is abrogated EXO1/DNA2 recruitment, defective DNA end resection and dysfunctional homologous recombination, with the authors describing CTCF as establishing a hierarchical recruitment cascade in HR.

    PMID 42475683DOI
    What this evidence cannot show

    Abstract plus Materials and methods read from PMC13384258. Systems are HEK-293T and U2OS-based reporter lines - AsiSI-ER-U2OS, ER-mCherry-LacI-FokI-DD-U2OS, DR-GFP and EJ5-GFP - so every break is enzyme-induced at a reporter or an AsiSI site and nothing is at an endogenous fragile element, let alone the KMT2A locus. No blood cell. The reagent list includes etoposide and BIX-01294, a G9a/GLP inhibitor whose selectivity is the same open question this graph has about UNC0642. The HP1 paralogue is HP1-GAMMA/CBX3, not HP1-alpha. 2026 paper, no replication known to this project.

    Evidence that argues against this (2)
    • The direction is opposite to the mechanism this hypothesis is trying to connect to. Here repressive chromatin at a break PROMOTES accurate repair, so H3K9 methylation is protective of genome integrity - whereas mechanism-h3k9me1-2-ctcf-balance-gates-kmt2a-rearrangement has H3K9me1/2 accumulation at the BCR as the state that PRECEDES rearrangement. Reconciling them requires the marks or the timing to differ, and this hypothesis asserts that they do without having measured it.
    • The cascade's readout is BRCA1/BARD1-dependent end resection, which is S/G2-restricted. In a quiescent progenitor - the cell this hypothesis is about - a TOP2 break is more likely to be handled by end-joining, in which case the cascade may not run and the hypothesis is simply inapplicable in the cell that matters.
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    A KAP1/HP1/SUV39H1 complex is rapidly loaded onto chromatin at double-strand breaks; SUV39H1 methylates H3K9 there, facilitating loading of additional KAP1/HP1/SUV39H1 through HP1's chromodomain binding nascent H3K9me3, which initiates cycles of loading and methylation that spread H3K9me3 and the complexes for tens of kilobases from the break; these H3K9me3 domains activate Tip60, which acetylates ATM and histone H4; cells lacking suv39h1 show defective Tip60 and ATM activation, decreased DSB repair and increased radiosensitivity; and activated ATM phosphorylates KAP1, releasing the repressive complex from chromatin as a negative feedback loop that may limit DSB repair.

    DNA double-strand breaks promote methylation of histone H3 on lysine 9 and transient formation of repressive chromatin. · Ayrapetov MK, Gursoy-Yuzugullu O, Xu C, Xu Y, et al. · Proceedings of the National Academy of Sciences 2014
    PMID 24927542DOI
    What this evidence cannot show

    Abstract level; cell systems not named in the abstract and the HP1 paralogue is unspecified, so nothing here can be attributed to CBX5, CBX1 or CBX3. The mechanism is explicitly transient with ATM-driven release, and 'tens of kilobases' of spreading exceeds the entire 7,466 bp breakpoint cluster region - which is the basis for this hypothesis's fourth prediction that a BCR-versus-flank design cannot see it.

    Evidence that argues against this (2)
    • This paper is the strongest argument that H3K9me3 at any break is a CONSEQUENCE of breakage, which if true at the KMT2A locus dissolves this graph's dispute in the least useful way - both entries would be describing a downstream mark. Against that, the founding rearrangement paper's H3K9me1 changes are produced by KDM3B depletion in undamaged cells and precede the cytogenetic event, so the me1 arm at least is not purely a damage response.
    • It also predicts that a population-level ChIP in asynchronously damaged cells sees a small, averaged signal, which makes the hypothesis hard to falsify by the resting-state measurement this beat actually ran - and an unfalsifiable-by-the-available-assay hypothesis is a weak one.
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The element this hypothesis is about is the same element in three independent measurements already in this graph: KDM3B depletion reduces CTCF occupancy at the conserved KMT2A exon-11 CTCF site without changing total CTCF protein, CTCF depletion alone is sufficient for the site-specific copy gain and FISH break-apart phenotype and is not enhanced by KDM3B co-depletion; and low-input DSBCapture mapping in NORMAL non-blast cells from AML patients in remission found breakage at the KMT2A intron 10 / exon 11 region 2- to 10-fold higher in each KMT2A-rearranged patient than in each non-fusion patient, with low-dose etoposide ex vivo raising breakage there in the fusion patients and not the others.

    Epigenetic balance ensures mechanistic control of MLL amplification and rearrangement. · Gray ZH, Chakraborty D, Duttweiler RR, Alekbaeva GD, et al. · Cell 2023
    PMID 37788669DOI
    What this evidence cannot show

    Citation fetched live from efetch on 2026-09-22 and NOT copied from a sibling entry - and that mattered: the first draft of this block carried a title, first-author initials, venue and DOI written from recall ("Aberrant H3K9 methylation and CTCF loss drive amplification and rearrangement of the MLL/KMT2A locus", Gray LT, Nucleic Acids Research, 10.1093/nar/gkad769) and ALL FOUR were wrong. The paper is in Cell, by Gray ZH and colleagues from the Whetstine laboratory, 26 authors. The patient DSBCapture half of the claim is PMID 34405474, a different paper, named in the claim rather than given its own block. All chromatin measurements in 37788669 are in hTERT-RPE epithelium.

    Evidence that argues against this (2)
    • The rival published model for this same element makes CTCF occupancy a MARKER OF FRAGILITY rather than a protector: TOP2-mediated breaks are enriched at CTCF- and cohesin-bound loop anchors with high transcriptional output. If that is right in direction, then a hypothesis in which CTCF recruits the repair-licensing machinery has to explain why the same protein both attracts the break and licenses its repair - which is not impossible, but is not what either literature says.
    • The patient measurement identified 59 regions break-prone across four datasets with 34 having a break ratio at least as large as the KMT2A region itself, so the KMT2A element is not uniquely break-prone even in the patients - and this graph already holds hyp-the-kmt2a-anchor-is-hypersensitive-not-uniquely-targeted for that reason.
Connected to