Oncofusion

start / graph / hyp-aff1-chd-dimerisation-in-kmt2a-fusion

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q2b-atlas2

KMT2A::AFF1 oligomerises through the retained AFF1 CHD, putting the commonest fusion in the oligomerisation class

also called CHD dimerisation hypothesis · AFF1 C-terminal homology domain oligomerisation

The bit of AFF1 kept by the fusion includes a module that, in its closest relative, makes two copies of the protein stick together. If the fusion does the same, then the most common childhood leukemia fusion works partly by the same trick as the rare ones, and it has a well-shaped surface that a drug could grip.

The same thing, for a scientist

The AFF4 C-terminal homology domain is a novel eight-helix fold that mediates AFF4 homodimerisation and AFF1-AFF4 heterodimerisation (PDB 6R80, 2.2 Å). The CHD is conserved across AFF1-4 and, in AFF1, spans residues 949-1209 — downstream of every observed KMT2A::AFF1 breakpoint and therefore fully retained in the fusion protein. We hypothesise that KMT2A::AFF1 self-associates and/or heterodimerises with endogenous AFF4 through this domain, and that this contributes materially to its activity. If true, oligomerisation is not a minority mechanism confined to cytoplasmic partners like AFDN and EPS15 but is available to the partner in roughly half of infant KMT2A-rearranged ALL — with the difference that the CHD is a folded domain, not a coiled coil, and therefore a better small-molecule prospect.

Go deeper
Where the hypothesis comes from1 deeper

Three separately established facts that do not appear to have been put together:

  1. The CHD is a dimerisation domain. Chen and Cramer solved the AFF4 CHD at 2.2 Å and showed it mediates AFF4 homodimers and AFF1-AFF4 heterodimers, binds RNA and DNA, and carries a CDK9-substrate loop 1.
  2. The CHD is conserved and is in AFF1 at 949-1209 (Pfam PF18876 / InterPro IPR043640) 2.
  3. Every clinically observed KMT2A::AFF1 break is far upstream of it. Breaks cluster in KMT2A BCR1 and the AFF1 contribution begins in the 5' part of AFF1, so the fusion retains the entire C-terminal third of AFF1 including the CHD 3.

The combination is a straightforward prediction that nobody seems to have tested: the fusion carries an intact, structurally characterised dimerisation module.

The reason it may have been missed is the labelling error this beat found. The CHD has been widely described as the AF9/ENL-binding domain (the actual AF9/ENL element is the short motif at AFF1 761-775, PDB 2LM0). If you already believe the CHD's job is recruiting AF9, you have no reason to ask what else it does.

Why it would matter

It reclassifies the dominant fusion. This graph treats oligomerisation as the mechanism of the rare cytoplasmic partners: AFDN's RA1 (minimal region sufficient for transformation, acting by self-association) and EPS15's coiled coil (necessary and sufficient, substitutable by synthetic dimerisers). Those together are about 2.5% of infant ALL. AFF1 is 48.3%. If AFF1 also oligomerises, the mechanism is not a curiosity of the tail of the distribution.

It is a better-shaped target than the other oligomerisation interfaces. A coiled-coil homodimer has no pocket and, per the synthetic-dimeriser result, no unique pharmacophore. An eight-helix folded domain with a defined dimerisation surface and a TPR-like architecture is the kind of thing fragment screening handles.

It changes what 'the fusion recruits the SEC' means. If KMT2A::AFF1 heterodimerises with endogenous AFF4, the fusion is a subunit of the super elongation complex rather than a client that recruits it, and its activity should be stoichiometrically sensitive to AFF4 dose.

It is orthogonal to menin. A CHD-interface agent would act on the partner side of the fusion, downstream of and independent of the menin-MBM1 interaction — which makes it a combination candidate and a route that MEN1 resistance mutations would not touch.

Facts and figures
statementKMT2A::AFF1 self-associates and/or heterodimerises with endogenous AFF4 through the retained AFF1 C-terminal homology domain (residues 949-1209), and this oligomerisation contributes materially to the fusion's transforming activity — placing the commonest KMT2A fusion in the same mechanistic class as AFDN and EPS15 fusions, and exposing a folded, druggable interface.
originliterature-gap
prior probability0.4
why nowThe enabling facts have been available since 2019 but sit in an HIV-transcription structural paper, while the leukemia literature has been describing the CHD as the AF9/ENL-binding domain — a labelling error this beat documents against PDB 2LM0 and Pfam PF18875. The hypothesis is cheap to test and the reagents (fusion constructs, PDB 6R80 coordinates, AFF4 antibodies) all exist.
predictions
[
  "Purified or immunoprecipitated KMT2A::AFF1 will show CHD-dependent self-association and co-precipitate endogenous AFF4.",
  "A CHD-deleted KMT2A::AFF1 (stopping at residue ~948) will retain the AF9/ENL-binding motif and the pSER/TAD but will be attenuated in colony-forming and self-renewal assays relative to full-length fusion.",
  "Point mutations at the AFF4 CHD dimer interface, transplanted onto the equivalent AFF1 residues by structural alignment to PDB 6R80, will phenocopy CHD deletion without destabilising the domain.",
  "AFF4 knockdown will reduce KMT2A::AFF1 activity in a dose-dependent manner beyond what is explained by loss of SEC scaffolding, if heterodimerisation is load-bearing.",
  "A CHD-interface agent will be additive or synergistic with a menin inhibitor, and will retain activity in cells carrying MEN1 interface resistance mutations."
]
experiments
[
  "exp-aff1-chd-dimerisation-separation-of-function",
  "exp-aff1-idr-condensate-separation-of-function"
]
analyses
[
  "ana-phase2-boltz-cofolding-plan"
]
falsifiers
[
  "A CHD-deleted KMT2A::AFF1 transforms as efficiently as full-length fusion in serial replating and in vivo — oligomerisation through this domain is then not load-bearing.",
  "Interface point mutants that abolish AFF4 CHD dimerisation in vitro have no effect on fusion activity when transplanted to AFF1.",
  "The fusion protein does not self-associate or co-precipitate AFF4 in a CHD-dependent manner in leukemia cells — the paralogue result then simply does not transfer.",
  "An AFF1 CHD structure, when solved, turns out not to dimerise despite the AFF4 CHD doing so, i.e. the conserved fold has diverged in function."
]
How this could bite us back
  • If CHD dimerisation is load-bearing, wild-type AFF1 and AFF4 use the same interface for normal super-elongation-complex function. An interface inhibitor would hit normal pause release, with a toxicity profile closer to a CDK9 inhibitor than to a menin inhibitor.
  • Heterodimerisation with endogenous AFF4 means a CHD blocker must out-compete a constitutive cellular interaction present in every cell, not just leukemic ones. Selectivity would have to come from something other than the interface itself.
  • If the hypothesis is right, it also predicts that fusions retaining the CHD are harder to kill by menin inhibition alone than the simple 'the target is in the retained KMT2A half' model implies — because part of the fusion's function is being supplied by a surface menin inhibition does not touch.
Still open
  1. Does the AFF1 CHD dimerise at all? No AFF1 CHD structure exists; the entire hypothesis rests on transfer from the AFF4 paralogue, and the two could have diverged.
  2. If the fusion heterodimerises with AFF4, does that make AFF4 itself a target — and is AFF4 tolerable to inhibit, given its role in normal pause release?
  3. The AFF4 CHD binds RNA and DNA. If the AFF1 CHD does too, the retained fusion carries a second nucleic-acid-binding surface in addition to KMT2A's AT-hooks and CXXC, which would change the model of how the fusion finds its targets.
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The AFF4 C-terminal homology domain (PDB 6R80, chain A 899-1163, 2.2 Å) is a novel eight-helix fold distantly related to tetratricopeptide repeats that mediates AFF4 homodimerisation or AFF1-AFF4 heterodimerisation, binds RNA and DNA by fluorescence anisotropy, and contains a CDK9-substrate surface loop; the CHD is conserved across AFF1, AFF2, AFF3 and AFF4.

    PMID 31147444DOI
    What this evidence cannot show

    All crystallography is on AFF4; AFF1 enters only through the heterodimerisation biochemistry. Purified-domain experiments do not establish that the dimer forms at physiological concentrations in cells, and nucleic-acid binding by anisotropy with naked RNA/DNA says nothing about chromatin. Nothing in this paper concerns leukemia or fusion proteins.

    Evidence that argues against this (1)
    • Whether AFF1-AFF4 heterodimers exist inside a working super elongation complex in cells is explicitly unsettled, and the cellular evidence points the other way. Asking that question directly, one group concluded that AFF1 and AFF4 reside in SEPARATE super elongation complexes with largely distinct and non-overlapping gene target sets: the AFF1-containing complex is the more potent for HIV-1 Tat transactivation, the AFF4-containing one for HSP70 induction on heat shock. Purified-domain heterodimerisation is therefore a demonstrated biochemical capability, not a demonstrated cellular configuration, and a hypothesis that the fusion heterodimerises with endogenous AFF4 has to contend with that. Lu H et al. 2015
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    The AFF-family C-terminal homology domain maps to AFF1 (UniProt P51825) residues 949-1209 as Pfam PF18876 / InterPro IPR043640.

    InterPro entry-to-protein matches for UniProt P51825 · InterPro / EMBL-EBI · InterPro 2026
    What this evidence cannot show

    An HMM hit boundary on a family alignment, not an experimental edge; ends float by a few residues between releases. Retrieved 2026-09-18.

    Evidence that argues against this (1)
    • An HMM window is the wrong unit for this protein family's functional architecture. The AFF4 scaffold has been mapped experimentally as a structurally disordered axis that recruits CycT1, ELL2 and ENL/AF9 through short hydrophobic regions dispersed along its length, permitting dynamic and flexible assembly rather than a fixed arrangement of folded modules. A Pfam/InterPro call at 949-1209 names a conserved stretch of sequence; on a scaffold of that kind it does not follow that the stretch behaves as one structural or functional unit, and the hypothesis's interface arguments inherit that uncertainty. Chou S et al. 2013
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    KMT2A breaks cluster in BCR1 (98% of 3,401 patients, between KMT2A intron 7 and exon 13), far upstream of any position that would truncate the AFF1 C-terminus, so the AFF1 CHD is retained in the fusion protein in essentially all cases.

    The KMT2A recombinome of acute leukemias in 2023. · Meyer C, Larghero P, Almeida Lopes B, Burmeister T, Gröger D, Sutton R, et al., Marschalek R · Leukemia 37:988-1005 2023
    PMID 37019990DOI
    What this evidence cannot show

    The paper characterises the KMT2A side of the junction in detail and the AFF1 side much less so. 'The CHD is retained' follows from the AFF1 breakpoint region lying in the 5' part of the gene rather than from a direct enumeration of AFF1 exon usage, which we could not obtain. A rare 3'-shifted AFF1 break that truncated the CHD would not necessarily appear in this analysis.

    Evidence that argues against this (1)
    • Inferring what the fusion retains from which side of the gene the break falls on ignores what happens at the join. Sequencing the derivative chromosomes of 40 t(4;11) patients showed that the fusion sites themselves carry deletions, inversions and duplications of parental DNA sequence - the junction is not a clean cut-and-paste. A breakpoint position 5' of the AFF1 C-terminal homology domain is therefore good evidence that the domain's coding sequence is on the correct derivative, and weak evidence about whether the reading frame and the C-terminus survive intact in any individual patient. Reichel M et al. 2001
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Oligomerisation is an established transforming mechanism for KMT2A fusions: the coiled-coil oligomerisation domains of AF1p/EPS15 and GAS7 are necessary and sufficient for transformation by their MLL fusions, and synthetic dimerisation modules substitute, activating Hox genes and enhancing progenitor self-renewal.

    Dimerization contributes to oncogenic activation of MLL chimeras in acute leukemias · So CW, Lin M, Ayton PM, Chen EH, Cleary ML · Cancer Cell 4:99-110 2003
    PMID 12957285DOI
    What this evidence cannot show

    Demonstrated for cytoplasmic partners (EPS15, GAS7), not for AFF1. The synthetic-dimeriser result shows dimerisation is sufficient in a mouse progenitor transformation assay; it does not show that it is what AFF1 fusions actually do, and AFF1 fusions have a well-supported alternative mechanism (super-elongation-complex recruitment) that the cytoplasmic partners lack.

    Evidence that argues against this (1)
    • The strongest argument against this hypothesis is that KMT2A::AFF1's activity is already well explained without invoking oligomerisation. If SEC recruitment through the 761-775 motif and the ELLBow element accounts for the phenotype, CHD dimerisation may be present but epiphenomenal. Qi S et al. 2017
Connected to