The fusion junction itself as a small-molecule target: a candid verdict
also called KMT2A::AFF1 junction · chimeric breakpoint surface · fusion-specific epitope
The seam where KMT2A is glued to its partner is the one piece of the leukemia protein that exists nowhere else in the body, so it looks like the perfect drug target. Unfortunately the seam sits inside a floppy, shapeless stretch of protein with no pocket for a drug to grip, and its exact sequence differs from patient to patient.
The same thing, for a scientist
A fusion-specific small-molecule site requires (i) a surface present only in the chimera, (ii) enough order and hydrophobic burial to form a pocket, and (iii) sequence invariance across patients. The KMT2A::partner junction fails (ii) and (iii). KMT2A residues immediately 5' of the breakpoint lie in a low-complexity linker between the CXXC domain and the (lost) PHD fingers, and the AFF-family N-terminus contributed 3' is the intrinsically disordered ALF region. Breakpoints scatter across KMT2A introns 9-11 and across >100 partner genes, so any junction-specific chemotype would be an N-of-few drug. The junction is a far better immunological target (a linear peptide needs no pocket) than a chemical one - see target-fusion-neoantigen-tcr.
Why there is no pocket1 deeper
Druggable small-molecule sites need buried, shape-complementary volume. The KMT2A breakpoint cluster region maps to introns 9-11, so the retained KMT2A protein ends in the linker downstream of the CXXC domain 1. The AFF1 N-terminal ALF homology region contributed by the partner is predicted-disordered and works through multivalent low-affinity contacts rather than by folding into a globular domain. Drugging disordered regions is not impossible (see target-aff1-pser-elongation-surface) but there is no reported ligandable pocket at any KMT2A fusion junction, and no co-crystal or NMR structure of a junction region exists in the PDB 3.
What would change the verdict
A cryo-EM, AlphaFold3 or Boltz model showing that the junction region folds back onto a stable partner surface (the retained CXXC domain, or menin) to create a composite groove absent from wild-type KMT2A. That is a testable prediction, written up as hyp-fusion-composite-interface and ana-boltz-cofold-kmt2a-partner.
The heterogeneity problem
Even a perfect junction binder would be partner- and breakpoint-specific. Within the commonest partner (AFF1) the KMT2A breakpoint distributes across at least two introns, and >100 partner genes are documented across the KMT2A recombinome 1. Pediatric oncology cannot support a separate chemical series per junction; this is the economic argument that pushes the lab toward shared partner interfaces such as menin.
The one genuine advantage the junction has
Infant KMT2A-r ALL has among the lowest somatic mutation burdens of any sequenced cancer (mean 1.3 non-silent mutations in the dominant clone) 2. There are therefore almost no conventional neoantigens, so the fusion junction is essentially the only tumour-specific protein sequence available to the immune system. A modality that reads a linear peptide - TCR-T, TCR-mimic antibody, peptide vaccine - has an unusually clean target space here, whereas small molecules do not.
| target class | fusion-intrinsic |
|---|---|
| modality assessed | small molecule |
| feasibility verdict | low |
| verdict rationale | No pocket (disordered junction), no invariant sequence (breakpoint and partner heterogeneity), no structural precedent. |
| chemical matter | none reported |
| better modality | TCR-T / peptide immunotherapy (target-fusion-neoantigen-tcr) |
structures
[]
- A junction-specific agent would be useless at relapse if the clone changes breakpoint or fusion isoform usage.
- Chasing the junction consumes medicinal-chemistry resources with far higher expected value at the menin or ENL interfaces.
- Does the junction region fold against the retained CXXC domain or against menin to create a composite pocket absent from wild-type KMT2A? A Boltz co-fold of KMT2A(1-1400)::AFF1(1-300) versus wild-type KMT2A(1-1400) would answer this cheaply.
- Is there any conserved sub-sequence shared by the commonest five partners just 3' of the junction that a single chemotype could read?
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Leukemogenic KMT2A fusions retain the CXXC domain and lose the PHD fingers, with breakpoints in a defined intronic cluster region; the retained C-terminal boundary of KMT2A is the linker after CXXC.
Multiple interactions recruit MLL1 and MLL1 fusion proteins to the HOXA9 locus in leukemogenesis · Milne TA, Kim J, Wang GG, et al. · Molecular Cell 2010What this evidence cannot show
Domain-mapping study in cell lines and mouse models; defines which domains are retained but says nothing directly about ligandability of the junction.
Evidence that argues against this (2)
- 'Lose the PHD fingers' is true of MAJOR-BCR fusions and false as a general statement, which matters for an epitope-design entry because it changes what the junction looks like. The minor breakpoint cluster region, documented in this graph at 47 of 3,401 cases and dominated by USP2, AFDN and USP8, produces fusions that RETAIN PHD1-3, the bromodomain and ePHD4. The retained C-terminal boundary also moves with the breakpoint across introns 9 to 11, so 'the linker after CXXC' describes a common case rather than a fixed junction. SOURCING CORRECTION 2026-09-20 (Q42 citation-integrity repair). Two things were wrong with this note as written. (1) The third minor-BCR partner was given as SEPTIN6; the source says USP8. Read from PMC10169636 full text on 2026-09-20: 'A total of 47 breakpoints were found in the minor BCR of KMT2A. The most frequent partner genes are USP2 (n = 29), AFDN (n = 9), and USP8 (n = 3).' SEPTIN6 is a MAJOR-BCR partner (n = 19) and does not appear in the minor BCR at all. (2) None of these facts is in Milne et al. 2010, the paper this note cites; they come from the 2023 KMT2A recombinome, added as the note below. The domain-retention sentence ('retain PHD1-3, the bromodomain and ePHD4') is a structural inference from the intron 20 - exon 24 breakpoint window and is not asserted by either paper. Molecular Cell 2010
- Primary source for the minor-BCR numbers used in the note above, fetched 2026-09-20: of 3,401 genomically characterised KMT2A-rearranged patients, 3,336 (98%) break in the major BCR between intron 7 and exon 13 and 47 (1.4%) in the minor BCR between intron 20 and exon 24; the minor-BCR partners are USP2 (n=29), AFDN (n=9) and USP8 (n=3), with AFF1, ARHGAP32, CREBBP, ELL, MLLT3 and MLLT10 once each. The minor BCR is ALL-specific: only one AML patient. So a junction-epitope design keyed to 'CXXC linker to partner' covers 98% of patients and misses a defined 1.4% whose junction sits several hundred residues further into KMT2A. Meyer C et al. 2023
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Infant KMT2A-rearranged ALL carries a mean of 1.3 non-silent somatic mutations in the predominant clone, one of the lowest mutation burdens of any sequenced cancer.
The landscape of somatic mutations in infant MLL-rearranged acute lymphoblastic leukemias · Andersson AK, Ma J, Wang J, et al. · Nature Genetics 2015What this evidence cannot show
n=65 infants; WGS/WES depth limits detection of subclonal events, and structural or epigenetic lesions are not counted.
Evidence that argues against this (1)
- Older children with KMT2A-r leukemia carry more mutations (mean 6.5/case, 45% in epigenetic regulators), so the 'no other neoantigens' argument is age-specific and weaker outside infancy. source
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database
Structural coverage of KMT2A in the PDB is restricted to isolated domains (CXXC-DNA, WIN peptide, menin-binding motif); no entry contains a fusion junction region.
RCSB PDB - full-text search for KMT2A/MLL entries (verified 2026-09-17) · RCSB Protein Data Bank 2026What this evidence cannot show
Absence of evidence from a database search on one date; new depositions could change this.
Evidence that argues against this (1)
- A counted absence over a structural database needs its query recorded, and this one has a mechanistic explanation that bounds the whole proposal: the junction sits in a region that is predicted disordered and is precisely the kind of sequence that does not crystallise. The absence is therefore weak evidence that nobody has tried and strong evidence that a junction-spanning experimental structure may not be obtainable — which an epitope programme should price in rather than treat as an open opportunity. RCSB Protein Data Bank 2009
the minor-BCR partner named SEPTIN6 in that entry's counter-note is USP8 (PMID 37019990, Results, 'The minor BCR of KMT2A'); SEPTIN6 is a major-BCR partner and appears nowhere in the minor BCR
same junction-specificity logic applied at the RNA rather than the peptide level
the defensible n-of-few application in this disease
correcting the junction rather than drugging it
the only surface unique to the fusion, and a hard one
the same epitope argued for therapeutically, here argued for as a research reagent
- part of →KMT2A / MLL1 — the gene that breaks
the neo-surface created where the two genes meet
the shared-pocket alternative to a fusion-selective surface
- ← related toThomas Milne (MRC WIMM, Oxford) — MLL-AF4 target biology, enhancer architecture and PROM1/CD133
PROM1/CD133 is the worked example of turning a fusion-driven target into an immunotherapy antigen
- ← related toPhase-2 dossier: the fusion junction - a negative feasibility finding, with the one surviving modality named
readiness layer on top of the junction-epitope entry