Phase-2 dossier: the fusion junction - a negative feasibility finding, with the one surviving modality named
also called fusion junction readiness · KMT2A::AFF1 neo-junction dossier · why we cannot drug the seam
The join between the two halves of the fusion protein is the only thing that is unique to the cancer, so it is the most attractive target imaginable. The problem is that there is no picture of it, both halves are floppy and shapeless, and a small molecule needs a solid pocket to hold on to. The one approach that could still work is showing a fragment of the join to the immune system.
The same thing, for a scientist
The KMT2A::AFF1 junction is the only truly tumour-specific epitope in this disease, and a phase-2 readiness review has to state plainly that it fails feasibility as a small-molecule target. The structural record is empty in a way that is verifiable rather than assumed: the AlphaFold database contains no model of full-length KMT2A (UniProt Q03164 returns HTTP 404), the AFF1 model that does exist has a global pLDDT of 52.34 over 1,210 residues - i.e. predominantly disordered - and no structure of any intact fusion oncoprotein exists in the PDB. A small-molecule campaign requires a pocket; two disordered partners joined at a breakpoint do not supply one. The surviving modality is immunological: a junction-spanning peptide presented on MHC, addressed by a TCR-mimic antibody, a TCR-engineered T cell or a peptide vaccine. That modality has its own hard gate, which is whether the junction peptide is presented at all.
What structure exists: verified emptiness
Checked directly this cycle, not recalled 1:
| Protein | UniProt | AlphaFold model | Global pLDDT |
|---|---|---|---|
| KMT2A | Q03164 | none - HTTP 404 | n/a |
| AFF1 | P51825 | AF-P51825-F1, 1,210 aa | 52.34 |
| menin | O00255 | AF-O00255-F1 | 84.44 |
A global pLDDT of 52 means the model is, in effect, a disorder prediction with coordinates attached. The contrast with menin at 84 is the whole argument: menin is a folded TPR solenoid with a cavity, and that is why menin has drugs.
No PDB entry exists for an intact KMT2A fusion protein or for any fusion segment spanning a breakpoint. Any structural plan for this target must therefore work on fragments, and there is no fragment at the junction that is known to fold.
Why the disorder is not an accident
The retained KMT2A N-terminus contributes AT-hooks, the CXXC domain and the menin-binding motif - short functional modules embedded in long low-complexity linkers. The retained AFF1 portion contributes the ALF, pSER and CHD regions, and AFF-family proteins are intrinsically disordered scaffolds whose job is to be a flexible platform for the super elongation complex. The fusion is a joining of two tethering surfaces, not a joining of two enzymes, and the mechanism of the disease is precisely that the join creates a new combination of tethers rather than a new folded entity. There is no reason to expect a pocket at the seam, and no published evidence of one.
The verdict for small molecules, and the surviving modality
Small molecule: STOP at G1. This is a feasibility failure, not a biology failure. It should not be revisited unless a structure of a folded junction-proximal module appears.
Immunological: conditional GO, one gate. The junction peptide is a genuine neoantigen and the same argument that makes it undruggable by chemistry makes it attractive to the immune system: it is present in every blast and absent from every normal cell. The gate is presentation. Infant ALL blasts are known to express MHC class I at variable and often low levels, and the junction sequence must contain a peptide with adequate affinity for a common HLA allele, be produced by the proteasome, and survive to the cell surface in sufficient copy number. None of that can be assumed; all of it can be measured by immunopeptidomics on a KMT2A::AFF1 line in a single experiment.
The breakpoint problem that makes even the immunological route hard
There is no single KMT2A::AFF1 junction. Breakpoints are distributed across the KMT2A breakpoint cluster region and across AFF1 introns, so the junction peptide differs between patients. A TCR-mimic or vaccine against one junction is a personalised product, not a platform product - unless a common junction class covers a large enough fraction of infants. That fraction is computable from breakpoint distributions the field already has, and computing it is a near-free desk analysis that should precede any immunological investment.
| cheapest next step | A desk calculation of junction-class coverage from published KMT2A and AFF1 breakpoint distributions. No new data, no cost, and it determines whether the immunological route is a platform or a bespoke product before a single experiment is funded. |
|---|---|
| phase2 verdict | NEGATIVE FEASIBILITY for small molecules, stated as a finding rather than a gap. Conditional continuation for the immunological route, gated on presentation. |
| compute backend | Laptop for the coverage calculation; Boltz co-folding is explicitly NOT recommended here, because co-folding two disordered chains produces confident-looking nonsense. |
target nodes in graph
[ "target-fusion-junction-epitope", "fusion-kmt2a-aff1" ]
structural record
{
"kmt2a_alphafold": {
"uniprot": "Q03164",
"model": null,
"api_status": 404
},
"aff1_alphafold": {
"uniprot": "P51825",
"model": "AF-P51825-F1",
"length": 1210,
"global_plddt": 52.34
},
"menin_alphafold_comparator": {
"uniprot": "O00255",
"model": "AF-O00255-F1",
"global_plddt": 84.44
},
"intact_fusion_structures": "none in the PDB",
"junction_spanning_structures": "none"
}gnomad aff1
{
"pLI": 0.765,
"loeuf": 0.502
}go no go
{
"small_molecule": {
"verdict": "STOP at G1 (feasibility)",
"criterion": "No folded, ligandable module has been demonstrated within 200 residues of any recurrent KMT2A::AFF1 junction. Reopen only if such a module is solved experimentally."
},
"immunological": {
"G1_feasibility": {
"question": "Is the junction peptide presented?",
"GO": "Immunopeptidomics on a KMT2A::AFF1 line (SEM or MV4-11) detects a junction-spanning peptide on MHC class I",
"STOP": "No junction peptide detected at a detection limit of roughly 10 copies per cell, with a positive control peptide recovered in the same run"
},
"G2_biology": {
"question": "Does a TCR-mimic or TCR-T recognise it?",
"GO": "Specific killing of a junction-positive line at effector:target 5:1 with no killing of an HLA-matched junction-negative line",
"STOP": "Killing of the junction-negative control at more than 20% of the junction-positive level - off-target reactivity in a child is unacceptable"
},
"G3_window": {
"question": "Coverage",
"GO": "One junction class covers at least 30% of infant KMT2A::AFF1 cases in published breakpoint distributions",
"STOP": "Below 10% coverage - the product is per-patient and cannot be a platform"
}
}
}- A junction-directed immunotherapy selects for blasts that lose the epitope through MHC down-regulation or antigen processing defects - the same escape route that defeats CD19 CAR-T in this disease, arriving faster because the antigen is a single peptide.
- Lineage switch, already documented in KMT2A-rearranged disease, does not remove the fusion junction, which is an argument *for* this modality; but a switched myeloid blast may present a different immunopeptidome entirely.
- A per-patient junction product cannot be manufactured on the timescale of infant ALL relapse, so a bespoke result is equivalent to a negative result for most children.
- Confident-looking co-folding predictions on two disordered chains are a real trap: a plausible model of a junction pocket would justify spending that the underlying physics does not.
- Does any locally folded module exist within a few hundred residues of a recurrent junction, and has anyone looked with a method other than AlphaFold?
- What fraction of infant KMT2A::AFF1 cases share a junction class at peptide resolution, and does that fraction change with HLA restriction?
- Do KMT2A-rearranged infant blasts down-regulate MHC class I, and if so does that pre-empt every junction-directed immunological approach before it starts?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
The AlphaFold database contains no model of full-length KMT2A (UniProt Q03164 returns HTTP 404) and the AFF1 model that exists (AF-P51825-F1, 1,210 residues) has a global pLDDT of 52.34, indicating predominant disorder - both verified by direct query of the AlphaFold prediction API this cycle.
AlphaFold Protein Structure Database prediction API (Q03164 returns 404; P51825 returns AF-P51825-F1 with globalMetricValue 52.34) · EMBL-EBI 2026What this evidence cannot show
The absence of a public model does not mean a model cannot be computed locally; KMT2A at 3,969 residues exceeds the length cut-off for the public bulk predictions. A low global pLDDT for AFF1 does not exclude small locally folded modules.
Evidence that argues against this (1)
- If a junction-proximal module were found to fold - for example a locally structured segment of the AFF1 ALF domain immediately 3' of the breakpoint - the small-molecule verdict would have to be revisited. Nothing in the current record suggests one. source 2026
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
Menin, by contrast, is a well-folded protein (AF-O00255-F1, global pLDDT 84.44) with a deep peptide-binding cavity resolved to 1.55 A in complex with an MLL peptide - which is why the partner interface, not the fusion, carries this project's drugs.
RCSB PDB data API entry record for 4GQ6 (Human menin in complex with MLL peptide, X-ray 1.55 A), with AlphaFold AF-O00255-F1 · RCSB PDB / EMBL-EBI 2026What this evidence cannot show
This is an argument by contrast, not a proof that no fusion-specific pocket exists anywhere.
Evidence that argues against this (1)
- Contrasting a well-folded menin with a disordered fusion partner frames the junction as the hard case, which it is, and understates how hard: the junction's position MOVES with the breakpoint across introns 9 to 11, and the minor-BCR partner class retains domains the major-BCR class loses. There is no single junction to target. RCSB PDB 2026
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database
AFF1 is the least loss-of-function-constrained gene among all candidates in this beat (pLI 0.765, LOEUF 0.502), consistent with a flexible scaffold rather than a tightly constrained folded enzyme.
gnomAD GraphQL API, gnomad_constraint for AFF1 · Broad Institute 2026What this evidence cannot show
Constraint is an organismal measure and is only weakly informative about foldedness; it is cited here as consistent with, not evidence for, the disorder argument.
Evidence that argues against this (1)
- A gnomAD constraint metric is not a safety signal for a drug, and this beat uses it as one across every dossier. pLI and LOEUF measure intolerance to HETEROZYGOUS loss-of-function across a whole lifetime from conception, in a population sample. A drug gives partial, reversible occupancy in an adult for weeks. High constraint says the gene matters in development; it does not predict on-target toxicity of transient pharmacological inhibition, and this project has already had to make the same correction for germline MEN1 penetrance in the endocrine beat. AFF1's comparatively low constraint is being read as reassurance, when low constraint equally means the gene tolerates loss — which for a fusion partner says little about the fusion. gnomAD, Broad Institute 2026
registers the negative feasibility finding and the immunological gate
readiness layer on top of the junction-epitope entry
- related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
the junction whose structural record was checked and found empty
AFF1 AlphaFold global pLDDT 52.34 over 1210 residues - predominantly disordered
- ← supportsDataset: PDB and AlphaFold - the structural census, including the two absences that shape every plan
supplied the verified structural absences that make the junction a negative feasibility finding
co-folding the junction is explicitly forbidden: two disordered chains give a confident-looking pose that is not a pocket