Oncofusion

start / graph / hyp-elongation-hijack-is-a-special-case-of-forced-oligomerisation

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q-nonelongation-partners

Hypothesis — the elongation hijack is how one partner class satisfies a more general requirement, and the general requirement is what should be drugged

also called elongation hijack as special case · oligomerisation-first model of KMT2A leukaemia

This graph explains KMT2A leukaemia by saying the partner brings in the machinery that reads genes too fast. That cannot be the whole story, because some partners are cytoskeleton proteins with no such machinery. The proposal here is that what all partners really supply is a way of sticking copies of the KMT2A fragment together, and that the gene-reading machinery is just one way of cashing that in.

The same thing, for a scientist

**Statement.** The conserved oncogenic requirement across all KMT2A fusion partner classes is forced, partner-specific, higher-order self-association of the retained KMT2A N-terminus. The super-elongation-complex/DOT1L hijack is the mechanism by which the elongation partner class (AFF1, MLLT1, MLLT3, MLLT10, ELL) satisfies that requirement, not the requirement itself. Non-elongation partners — the five to six septins, ABI1, GPHN, GAS7, EPS15, AFDN — satisfy it by supplying a self-association module directly. **Why it is worth stating as a hypothesis rather than a fact.** The oligomerisation evidence is strong but has two documented failures of substitution (SEPT6 synthetic dimeriser; AFDN heterologous RA domains), and five of six septin partners plus ABI1 have never been tested at all [see mech-forced-oligomerisation-as-the-conserved-core-across-kmt2a-partner-classes]. The prediction that discriminates it from the standard model is a therapeutic one, and it is checkable. **Predictions.** (1) A non-elongation fusion should be menin-dependent, because menin binds the retained N-terminus, which is the part the general model says matters. (2) A non-elongation fusion should NOT depend on DOT1L, because H3K79 methylation is the elongation class's implementation detail. (3) Menin-inhibitor sensitivity should be roughly partner-class-independent while DOT1L-inhibitor sensitivity should be partner-class-dependent. (4) Oligomerisation order, not merely dimerisation, should track transforming strength. **What it changes.** If correct, the drug target this graph should privilege is the retained KMT2A N-terminus and its obligate interactions (menin, LEDGF, CXXC), which are partner-agnostic, over partner-side elongation machinery. It also predicts that the attraction of oligomerisation interfaces as a target class is weaker than it looks, because the two substitution failures say the interfaces are not interchangeable and would each need their own drug.

Go deeper
How this differs from the standard model in practice

The standard model and this one make the same prediction for AFF1 and MLLT3. They diverge on three testable points:

elongation-hijack modeloligomerisation-first model
septin fusion menin dependenceexpected but unexplainedrequired — menin binds the retained N-terminus, which is the whole mechanism
septin fusion DOT1L dependenceexpected, because H3K79 is taken as generalnot expected — no elongation module is present
synthetic dimeriser sufficiencynot addressedexpected, and it FAILS for SEPT6, which is the hypothesis's own strongest problem

The third row is why this is a hypothesis and not a conclusion. The MLL-SEPT6 Δcoil-ER result is a direct negative against the simplest version of the model and it is recorded here as such rather than buried.

What would kill it

Listed as falsifiers in data. The cleanest single result: if a KMT2A-septin fusion turned out to recruit the super elongation complex and depend on DOT1L to the same degree as KMT2A::AFF1, the partner classes would not be mechanistically distinct and the reframing would add nothing. The second cleanest: if a KMT2A-septin fusion proved menin-INDEPENDENT, the general requirement would not be about the retained N-terminus and the hypothesis would be wrong in the direction that matters most clinically.

Facts and figures
statementThe conserved oncogenic requirement across KMT2A fusion partner classes is forced, partner-specific, higher-order self-association of the retained KMT2A N-terminus; the SEC/DOT1L elongation hijack is how the elongation partner class satisfies it, not the requirement itself. Therefore the partner-agnostic drug target is the retained N-terminus and its obligate interactions (menin, LEDGF, CXXC), and partner-side targets such as DOT1L should be expected to work only within the elongation class.
originliterature-gap
prior probability0.55
predictions
[
  "KMT2A::SEPTIN6 is menin-dependent: a menin inhibitor reduces colony formation and Hoxa9/Meis1 expression in MLL-SEPT6-immortalised progenitors to a comparable degree as in MLL-AF9 controls.",
  "KMT2A::SEPTIN6 is NOT DOT1L-dependent to the same degree: pinometostat has a substantially smaller effect on MLL-SEPT6 colonies than on MLL-AF9 colonies.",
  "H3K79me2 does not spread over Hoxa in MLL-SEPT6 cells the way it does in elongation-partner fusions.",
  "A graded series of synthetic oligomerisation modules on one KMT2A backbone shows transforming activity increasing with oligomer order, with a dimer insufficient where a higher-order module suffices.",
  "Menin occupancy at HOXA loci scales with fusion oligomer order, making menin-inhibitor IC50 partner-dependent even though menin binding is not."
]
experiments
[
  "exp-menin-inhibitor-sensitivity-of-kmt2a-septin-fusions"
]
analyses
[
  "ana-menin-dependence-untested-in-septin-and-cytoskeletal-kmt2a-fusions",
  "ana-kmt2a-nonelongation-partner-systematic-enumeration"
]
falsifiers
[
  "A KMT2A-septin fusion is shown to recruit the super elongation complex and to depend on DOT1L to the same degree as KMT2A::AFF1 — then partner classes are not mechanistically distinct and the reframing adds nothing.",
  "A KMT2A-septin fusion is shown to be menin-INDEPENDENT — then the conserved requirement is not about the retained N-terminus and the hypothesis is wrong in the direction that matters clinically.",
  "A graded synthetic-oligomerisation series shows a simple dimer is fully sufficient on a septin-free KMT2A backbone AND that the SEPT6 Δcoil-ER failure was a construct artefact — then the 'higher-order, partner-specific' qualifier is unnecessary and the plain dimerisation model stands.",
  "A partner is found that transforms with no demonstrable self-association module of any kind and no elongation module either — MLLT11/AF1q at 90 aa is the obvious candidate — then neither model is general."
]
How this could bite us back
  • This hypothesis strengthens the case for menin and weakens the case for DOT1L, which is convenient for a graph already built around menin. That is a reason to test it rather than to adopt it, and the specific test — a septin fusion plus a menin inhibitor — has never been run.
  • If the hypothesis is right, the oligomerisation-interface target class loses its main attraction: one drug per interface, not one drug for many fusions, because two substitution experiments say the interfaces are not interchangeable.
  • If it is wrong in the second falsifier's direction — a menin-independent septin fusion — then a class of patients is currently eligible for an approved drug on a mechanism that does not apply to them.
Still open
  1. MLLT11/AF1q is 90 residues, has no chromatin domain, no annotated oligomerisation module, and 28 patients. If it transforms, what does it supply?
  2. Is there any partner class for which menin dependence is expected to fail under this model? The model says no, which makes it unusually easy to falsify.
  3. Does oligomer order change menin occupancy, and is that measurable by ChIP in a graded construct series?
Every source behind this entry (5)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    The coiled-coil oligomerisation domains of the cytoplasmic MLL partners AF1p/EPS15 and GAS7 are necessary and sufficient for leukemogenic transformation, and homodimerisation of MLL by synthetic dimerisation modules mimics bona fide MLL fusion proteins with Hox activation and enhanced progenitor self-renewal.

    PMID 12957285DOI
    What this evidence cannot show

    Two partners; mouse progenitor assays.

    Evidence that argues against this (1)
    • The synthetic module does not substitute for the SEPT6 coiled coil, so 'any dimeriser will do' is false and the hypothesis must be stated as partner-specific higher-order self-association rather than as dimerisation. Ono R et al. 2005
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Myeloid immortalisation by MLL-SEPT6 required the GTP-binding domain, the P-loop and the coiled coil individually, only intact MLL-SEPT6 self-associated by co-immunoprecipitation, a 4-OHT-inducible ER dimerisation module substituted for the coiled coil failed to rescue, MLL-SEPT6 activated Hoxa7, Hoxa9 and Meis1, and MLL-SEPT6 alone produced long-latency myeloproliferative disease requiring FLT3-ITD for acute leukaemia.

    PMID 15761502DOI
    What this evidence cannot show

    One partner, mouse retroviral system, single negative construct for the substitution experiment.

    Evidence that argues against this (1)
    • That MLL-SEPT6 alone gives only myeloproliferative disease, not acute leukaemia, weakens the claim that partner-supplied oligomerisation is sufficient for the disease phenotype at all — for this partner it is sufficient only for immortalisation. So CW et al. 2003
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The AFDN/AF6 RA1 domain is the minimal region sufficient for MLL-AF6-mediated immortalisation and short-latency leukemogenesis in vivo, acting through self-association, with Ras-binding-dead point mutants still transforming and heterologous RA domains and MLL fused to activated RAS or RAP1 not sufficient.

    PMID 20395419DOI
    What this evidence cannot show

    Mouse models; the non-substitutability of heterologous RA domains is a single set of constructs.

    Evidence that argues against this (1)
    • A second mechanism operates in the same fusion — nuclear sequestration of endogenous AF6 with RAS activation, reversed by fusion silencing — so even the cleanest self-association partner is not a pure case. Manara E et al. 2014
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    A 15-amino-acid tubulin-binding domain of gephyrin, a protein with no transcriptional activity, is necessary and sufficient for MLL-GPHN transformation of haematopoietic progenitors in vitro and in vivo and confers oligomerisation capacity on MLL; MLL-GPHN caused both myeloid and lymphoid leukaemias in mice.

    The small oligomerization domain of gephyrin converts MLL to an oncogene. · Eguchi M, Eguchi-Ishimae M, Greaves M · Blood 2004
    PMID 14751928DOI
    What this evidence cannot show

    A single-patient fusion with essentially no clinical weight; retroviral transduction in mice.

    Evidence that argues against this (1)
    • Building a general model on GPHN and GAS7 — partners with 0 and 3 patients respectively in a 3401-patient series — generalises from the most tractable rather than the most common biology; the partners that actually carry the clinical weight are the elongation partners the model is trying to demote. Meyer C et al. 2023
  5. E5established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Menin binds a high-affinity conserved motif in the amino-terminus retained by every oncogenic MLL fusion, and that interaction is required for initiation and for maintenance of MLL-associated leukemogenesis.

    PMID 16239140DOI
    What this evidence cannot show

    The abstract does not name the fusion proteins used and the full text was not accessible for this entry, so the partner classes tested are not asserted here.

    Evidence that argues against this (1)
    • Menin dependence has never been demonstrated for any non-elongation partner: a PubMed intersection of menin with septin terms returns zero records against denominators of 1264 and 1936, and the pivotal revumenib KMT2A-r trial names no septin translocation among its 57-patient efficacy population. Issa GC et al. 2025
Connected to