What is actually conserved across KMT2A partner classes — forced self-association, with two documented exceptions that break the rule
also called oligomerisation as the conserved core · dimerisation model of MLL fusions · partner-class unification
Partners as different as an elongation scaffold and a cytoskeletal GTPase both cause leukaemia when joined to KMT2A, and the usual explanation is that all of them simply glue copies of the KMT2A fragment together. That explanation is right often enough to be useful and wrong in at least two places that have been directly tested.
The same thing, for a scientist
The field's answer to 'what do AFF1 and SEPT9 have in common?' is forced self-association. The evidence is genuinely strong for several partners: the coiled-coil oligomerisation domains of AF1p/EPS15 and GAS7 are necessary AND sufficient, and synthetic dimerisation modules substituted for them reproduce Hox activation and enhanced progenitor self-renewal [E1]; a 15-amino-acid tubulin-binding region of gephyrin is necessary and sufficient to convert MLL into an oncogene and confers oligomerisation capacity despite gephyrin having no transcriptional activity [E2]; the AFDN/AF6 RA1 domain is the minimal sufficient region for immortalisation and short-latency leukemogenesis, acting through self-association, with Ras-binding-dead mutants still transforming and heterologous RA domains not sufficient [E3]. Dimerisation was already being described as a general oncogenic switch across fusion families in 2004 [E4]. But the model has two documented failures. First, substituting a synthetic 4-OHT-inducible dimerisation module for the SEPT6 coiled coil did NOT restore myeloid immortalisation, and the GTP-binding domain, the P-loop and the coiled coil were each independently required — the authors' conclusion was that 'homo-oligomerization with proper whole structure is essential' [E5]. Second, heterologous RA domains did not substitute for AFDN RA1 either [E3]. So the conserved element is not 'any self-association module'. It is a partner-specific assembly with the right geometry and stoichiometry, of which a synthetic dimeriser is sometimes an adequate mimic and sometimes not.
The case for oligomerisation, stated fairly
Four independent lines support it.
- Necessity and sufficiency in cis. For AF1p/EPS15 and GAS7, the coiled-coil oligomerisation domain alone is necessary and sufficient for transformation by the respective MLL fusion 1.
- Sufficiency of a synthetic substitute. Homodimerisation of MLL by synthetic dimerisation modules mimics bona fide MLL fusion proteins, activating Hox genes and enhancing progenitor self-renewal 1.
- A partner with no transcriptional activity at all. Gephyrin is a neuronal receptor-assembly protein with no transactivation function, and a 15-aa tubulin-binding domain is necessary and sufficient for MLL-GPHN to transform progenitors in vitro and in vivo; that domain also confers oligomerisation capacity 2. The fusion produced both myeloid and lymphoid leukaemias, implying a multipotent target cell.
- A cytoplasmic junctional protein reduced to its self-association module. AFDN RA1 is the minimal region sufficient for immortalisation and short-latency leukemogenesis, Ras-binding-dead point mutants still transform, and direct fusion of MLL to activated RAS or RAP1 is not sufficient 3.
Taken together these make forced self-association the best single generalisation available, and they justify treating the elongation hijack as one of several ways to satisfy it rather than as the mechanism.
The two results that break it
Both are negative controls inside papers that otherwise support the model, which is why they are easy to miss.
MLL-SEPT6. The coiled coil was replaced with the 4-OHT-inducible mutant oestrogen-receptor ligand-binding domain, the same class of synthetic dimeriser that works for EPS15 and GAS7. It failed to immortalise with or without 4-OHT. Separately, deleting the GTP-binding domain, deleting the P-loop and the single S56N substitution each abolished immortalisation, and only intact MLL-SEPT6 self-associated by co-immunoprecipitation 5. Septins are hetero-oligomers that assemble into filaments, not dimers, so the natural reading is that the required assembly is of higher order or different geometry than a forced dimer.
MLL-AF6. Heterologous RA domains did not substitute for RA1 3. The transforming requirement is not 'an RA domain' and not 'Ras binding' but that specific self-association surface.
A model that predicts a synthetic dimeriser will always work is falsified by the first result. A model that predicts any homologous self-association domain will work is falsified by the second.
What this implies for the elongation-hijack framing
The elongation hijack — SEC recruitment, DOT1L, H3K79me2 spreading over HOXA — is demonstrated in detail for AFF1, MLLT1, MLLT3 and MLLT10, and it is the basis on which this graph nominates drug targets. The partner census says it cannot be the general mechanism: septins (n=46 across five genes in a 3401-patient series), ABI1, GPHN, GAS7 and EPS15 supply no elongation module, and for septins no SEC or DOT1L measurement of any kind has ever been published.
The honest formulation is that the elongation hijack is the mechanism by which one large partner class satisfies a more general requirement. That requirement is best described as forced, partner-specific, higher-order self-association of the retained KMT2A N-terminus, which then does whatever the retained N-terminus and its obligate cofactors — menin, LEDGF, the CXXC domain — can do when present at high local concentration on chromatin.
That reformulation matters for target selection, and this beat states it plainly: the drug target implied by the general mechanism is the retained KMT2A N-terminus and its obligate interactions, not the partner's elongation machinery. Menin is a retained-N-terminus target, which is why it should survive the reformulation. DOT1L is a partner-mechanism target, which is why it should not be expected to.
Where the generalisation has never been tested at all
Of the six septin genes that form KMT2A fusions, exactly one (SEPTIN6) has ever been put into a transformation assay 5. SEPTIN9, SEPTIN5, SEPTIN2, SEPTIN11 and SEPTIN3 have none. ABI1 has none. So the class that most sharply challenges the elongation model is also the class with the least functional data, and the single experiment that does exist is the one that contradicts the simplest version of the alternative.
| mechanism class | forced-oligomerisation |
|---|---|
| verdict | Forced self-association is the best available generalisation and is better supported than the elongation hijack as a CLASS-WIDE mechanism, but it is not 'any dimerisation module': two direct substitution experiments failed, both in partners whose native biology is higher-order assembly (septin filaments) or a specific interface (AFDN RA1). The elongation hijack is best described as the way one partner class satisfies the general requirement, not as the mechanism. |
| target implication | The drug target implied by the general mechanism is the retained KMT2A N-terminus and its obligate interactions (menin, LEDGF, CXXC), which are partner-independent. Partner-side targets such as DOT1L are expected to work only for the elongation class, and the DOT1L clinical record is consistent with that. |
partners with positive cis minimal region
[ "EPS15/AF1p coiled coil", "GAS7 coiled coil", "GPHN 15-aa tubulin-binding domain", "AFDN RA1" ]
partners where a synthetic dimeriser substitutes
[ "EPS15/AF1p", "GAS7" ]
partners where a synthetic or heterologous module FAILS to substitute
[ "SEPTIN6 (4-OHT-inducible ER LBD)", "AFDN (heterologous RA domains; MLL fused to activated RAS or RAP1)" ]
partners with no functional assay of any kind
[ "SEPTIN9", "SEPTIN5", "SEPTIN2", "SEPTIN11", "SEPTIN3", "ABI1" ]
- If oligomerisation is the conserved core, an oligomerisation-interface drug would have to be designed per partner, because the two substitution experiments say the interfaces are not interchangeable. That removes the main attraction of the class — one drug for many fusions.
- Reframing the elongation hijack as a special case does not make menin a safer bet automatically; it makes menin's partner-independence a claim that now has to be demonstrated in the partner classes that motivated the reframing, and it has not been.
- The strongest support for the conserved-core model comes from GPHN and GAS7, partners with essentially no clinical weight. Building target strategy on mechanism derived from three-patient fusions is how a field generalises from its most tractable rather than its most common biology.
- Does MLL-SEPT6 require a higher-order oligomer than a dimer? A graded series of synthetic oligomerisation modules (dimer, trimer, tetramer, filament-forming) on one KMT2A backbone would answer this and has never been run.
- Is the retained KMT2A N-terminus's requirement for menin independent of the oligomerisation order the partner supplies? If oligomerisation increases local menin occupancy, then menin dependence should scale with oligomer order and be testable.
- For the partners where a synthetic dimeriser works, does it reproduce the fusion's TARGET GENE PROGRAM or only self-renewal? Hox activation was shown; genome-wide equivalence was not.
- 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 by the respective MLL fusion proteins, and homodimerisation of MLL by synthetic dimerisation modules mimics bona fide MLL fusion proteins, producing Hox gene activation and enhanced self-renewal of haematopoietic progenitors.
Dimerization contributes to oncogenic activation of MLL chimeras in acute leukemias. · So CW, Lin M, Ayton PM, et al. · Cancer Cell 2003What this evidence cannot show
Mouse haematopoietic progenitor assays with retroviral constructs; two partners tested. Sufficiency was demonstrated against the KMT2A N-terminus, not against every partner backbone.
Evidence that argues against this (1)
- The synthetic-module result does not generalise: an equivalent 4-OHT-inducible dimerisation module substituted for the SEPT6 coiled coil failed to immortalise myeloid progenitors with or without ligand, and the authors concluded that oligomerisation 'with proper whole structure' is required. Ono R et al. 2005
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Gephyrin, a neuronal receptor-assembly protein and rare MLL fusion partner that lacks transcriptional activity, transforms haematopoietic progenitors as an MLL-GPHN chimera; a small 15-amino-acid tubulin-binding domain of GPHN is necessary and sufficient for this activity in vitro and in vivo and also confers oligomerisation capacity on MLL, and MLL-GPHN caused myeloid and lymphoid leukaemias in mice, implying it can target multipotent progenitors.
The small oligomerization domain of gephyrin converts MLL to an oncogene. · Eguchi M, Eguchi-Ishimae M, Greaves M · Blood 2004What this evidence cannot show
Retroviral transduction in mouse progenitors. 'Confers oligomerisation capacity' is inferred from the transforming activity plus biochemical association rather than from a structure of the oligomer.
Evidence that argues against this (1)
- GPHN is an extremely rare partner — the original report describes a single child with therapy-related acute undifferentiated leukaemia after etoposide — so the cleanest demonstration that a non-transcriptional partner works through a minimal oligomerisation module comes from a fusion of essentially no clinical weight, and its generality across recurrent partners is assumed rather than shown. Eguchi M et al. 2001
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The RA1 domain of AF6/AFDN is the minimal region sufficient for MLL-AF6-mediated myeloid progenitor immortalisation and short-latency leukemogenesis in vivo, acting through self-association: point mutations abolishing AF6-Ras binding did not abrogate oncogenesis, and neither heterologous RA domains nor direct MLL fusion to activated RAS or RAP1 were sufficient.
Self-association mediated by the Ras association 1 domain of AF6 activates the oncogenic potential of MLL-AF6. · Liedtke M, Ayton PM, Somervaille TC, et al. · Blood 2010What this evidence cannot show
Mouse models with retroviral constructs. The negative result on heterologous RA domains is the load-bearing part for this entry and is a single set of constructs.
Evidence that argues against this (1)
- A second, genuinely different mechanism operates in the same fusion: MLL-AF6-rearranged cells relocate endogenous AF6 from cytoplasm to nucleus with consequent RAS activation, and silencing the fusion restores cytoplasmic AF6 and reduces RAS-GTP — so self-association is not the whole account even for the partner where it was most cleanly mapped. Manara E et al. 2014
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.review
Forced dimerisation or oligomerisation is a general mechanism for unleashing the oncogenic properties of chimeric transcription factors in acute leukaemias, and a key mechanistic feature with major clinical implications is the nature of the transcriptional cofactors recruited — corepressors for chimeric RARalpha and AML1 proteins, coactivators or basal machinery for MLL chimeras.
Dimerization: a versatile switch for oncogenesis. · So CW, Cleary ML · Blood 2004What this evidence cannot show
A review, cited as a map of the field's position in 2004 rather than as primary evidence.
Evidence that argues against this (1)
- The review's own framing already contains the caveat this entry develops: what matters is which cofactors the dimerised chimera recruits, which means oligomerisation is a permissive switch and not a complete mechanism, and partner identity still determines the output. So CW et al. 2003
- E5strong 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; the S56N P-loop substitution that reduces GTP binding also abolished it; only intact MLL-SEPT6 self-associated by co-immunoprecipitation; and substituting a 4-hydroxytamoxifen-inducible oestrogen-receptor dimerisation module for the coiled coil failed to restore immortalisation with or without ligand, leading the authors to conclude that homo-oligomerisation with proper whole structure is essential.
Dimerization of MLL fusion proteins and FLT3 activation synergize to induce multiple-lineage leukemogenesis. · Ono R, Nakajima H, Ozaki K, et al. · J Clin Invest 2005What this evidence cannot show
A single negative construct carries the weight of the falsification, and failure-to-rescue experiments are vulnerable to expression, folding and geometry artefacts. Full text read at jci.org for this entry.
Evidence that argues against this (1)
- The positive control for the same technique exists and works: synthetic dimerisation modules DO convert MLL into an oncoprotein when substituted for AF1p/EPS15 and GAS7 coiled coils, so either the septin backbone imposes an additional requirement or the MLL-SEPT6 construct was defective for an unrelated reason. The two results have never been reconciled in one system. So CW et al. 2003
- E6established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Six septin genes (SEPTIN2, SEPTIN3, SEPTIN5, SEPTIN6, SEPTIN9, SEPTIN11) and ABI1 are documented KMT2A translocation partners, together accounting for 46 septin and 8 ABI1 patients among 3401 genomically characterised KMT2A-rearranged acute leukaemias, of which all but three are myeloid.
The KMT2A recombinome of acute leukemias in 2023. · Meyer C, Larghero P, Almeida Lopes B, et al. · Leukemia 2023What this evidence cannot show
Counts read from Table 1 of the 2023 recombinome (PMC10169636 full text) for this beat: SEPTIN6 19, SEPTIN9 14, SEPTIN5 8, SEPTIN11 3, SEPTIN2 1, SEPTIN3 1, ABI1 8. A referral cohort, not population incidence.
Evidence that argues against this (1)
- Several of these fusions are cytogenetically cryptic or impersonate commoner fusions (SEPTIN6 on the der(X); SEPTIN11 mimicking t(4;11); ABI1 indistinguishable from MLLT10 at 10p11-12), so the referral-based counts are floors and the non-elongation class is larger than 54 of 3401. Stevens SJ et al. 2010
- ← related toCBFA2T3::GLIS2, inv(16)(p13.3q24.3)
Mechanism substitution (Thirant 2017, PMID 28292442): the fusion acts by ERG-dependent enhancer activation with GLIS2 conferring megakaryocytic identity, and interfering specifically with ETO2-GLIS2 oligomerisation reverses it - not by the BMP axis reported from Drosophila.
the falsifying case: synthetic dimeriser fails
- supports →KMT2A::AFDN (MLL-AF6), t(6;11)(q27;q23)
AFDN RA1 is the minimal sufficient self-association region
supplies the KMT2A-side evidence base and two documented exceptions
the general mechanism implies a retained-N-terminus target, i.e. menin
- related to →KMT2A::ABI1 (MLL-ABI1), t(10;11)(p11.2;q23) — the partner that binds an elongation partner
ABI1 is the one non-elongation partner with a possible trans route to elongation
- related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
the elongation hijack as a special case of the general requirement
the reframing's central prediction is exactly this untested dependency
built on the partner-class evidence audit
a 15-aa tubulin-binding domain is necessary and sufficient and confers oligomerisation on MLL
- ← related toKMT2A::MLLT11 (MLL-AF1q), t(1;11)(q21;q23) — 28 patients, and not one functional experiment
the untested hard case: no annotated oligomerisation module and no functional experiment ever performed
the EPS15 coiled coil is necessary AND sufficient for transformation, and synthetic dimerisers substitute
the GAS7 coiled coil is necessary AND sufficient; synthetic dimerisers substitute
proposes templating on a polymer rather than pairwise dimerisation; the synthetic-dimeriser result is the strongest argument against it
TESTS THE CONSERVED CORE IN REVERSE: if forced oligomerisation is what KMT2A partner classes share, LDB1 is a strong instance of the partner side -- it promotes connectivity among regulatory elements independently of CTCF, cohesin and YY1, and an engineered LDB1-driven loop is cohesin independent. But here it drives KMT2A's CATALYTIC C-terminus rather than its targeting module, so it asks whether the core operates when the oligomeriser is fused to SET rather than to CXXC. Untested, and no LDB1::KMT2A protein has ever been observed.