Oncofusion

start / graph / hyp-no-surface-antigen-is-both-fusion-driven-and-normal-tissue-clean-in-kmt2a-aff1

Hypothesismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q138-kmt2a-aff1-surfaceome

REFUTED IN THE CYCLE IT WAS WRITTEN - the proposal that fusion-driven and normal-tissue-clean surface antigens are disjoint in KMT2A::AFF1 because of what the fusion's output programme is; an expression-matched permutation says it is the base rate

also called the disjointness hypothesis

This entry proposed that the surface proteins this leukaemia makes because of its fusion gene are unusually likely to also appear in healthy organs. A fair comparison against other proteins made at the same level in the same cells found no difference at all, so the proposal is recorded here as tested and wrong.

The same thing, for a scientist

🟠 **FIRST QUANTITATIVE DATA 2026-09-24 (cycle 136), and it cuts both ways.** This hypothesis asserted an empty intersection between fusion-enriched and parenchymally clean surface antigens among 169 candidates, and named NG2/CSPG4 and LAMP5 as its cases - but this project's normal-tissue filter had scored **22 chromatin targets and not one surface antigen**, so the central claim had never been tested against the class it is about. It has now been, on the same panel and metric (controls reproduced: ACTB 1.21 against 1.2). | antigen | bone marrow nTPM | highest normal tissue | ratio | |---|---:|---|---:| | **PROM1 / CD133** | 3.7 | kidney 19.5 | **5.27** | | **LAMP5** | 2.1 | cerebral cortex 41.1 | **19.57** | | **CSPG4 / NG2** | **0.0** | lung 18.1 | *undefined* | | TAPT1 | 24.2 | liver 28.5 | **1.18** | 🔵 **Supporting the empirical claim:** **PROM1/CD133 is fusion-driven (direct MLL-AF4 binding, published) AND not normal-tissue clean** - 5.27x kidney over marrow, worse than MEN1 at 3.3. **LAMP5 at 19.57 exceeds the worst of the previous 22 (MEIS1, 17.2).** Both are the shape this hypothesis predicts, and PROM1 is the antigen it had omitted. 🟠 **Arguing against the MECHANISTIC version:** the hypothesis reasons that fusion-driven surface genes are off-compartment *because* they are developmental. **TAPT1 sits at 1.18, marrow-selective, despite sharing the intragenic H3K79me2/3 enhancer that drives PROM1.** Co-regulation by the same element therefore does not force an off-compartment profile, so the mechanism is not doing the work the reasoning gives it - even as the empirical pattern holds. **Confidence raised weak -> moderate on the empirical claim only.** Three antigens is not 169, and the remaining 14 fusion-enriched candidates have still never been scored. --- A surfaceome-wide screen found 15 fusion-enriched surface proteins and 3 parenchymally clean ones among 169 candidates expressed in the index lines, with an empty intersection [E1]. The hypothesis is that this disjointness is **mechanistic rather than incidental**. The reasoning: KMT2A::AFF1 is a transcription-elongation complex that de-represses a **developmental** programme - HOXA, MEIS1, and the surface genes that travel with it. Developmental programmes are by definition used elsewhere in the body. **NG2/CSPG4 is the clearest case**: it is a direct MLL-AF4 target gene [E2] and it is a pericyte and oligodendrocyte-precursor antigen, i.e. a normal developmental surface protein the fusion has switched on out of context. **LAMP5** is the second: a direct MLL-fusion target and a brain protein [E3]. By contrast the antigens that are clean in normal parenchyma - CD19, SIT1, S1PR4 - are **leukocyte-restricted** proteins whose cleanliness comes precisely from being lineage markers, which is why they cannot be fusion-specific [E1]. **The prediction that makes it a hypothesis rather than a restatement**: the same disjointness should hold for other fusion oncoproteins that act by de-repressing developmental programmes, and should NOT hold for fusions that create a genuinely novel gene product or that drive a tissue-restricted programme. CBFA2T3::GLIS2 is the test case on the other side - and FOLR1, its antigen, **also** has high normal-tissue expression (choroid plexus 2,860.9 nTPM), which is evidence FOR the hypothesis and against the analogy this project drew from it [E4].

Go deeper
Why the alternative explanations are weaker but not excluded

Alternative 1: the screen is underpowered. Two index lines in the B-arm, mRNA only, and a 5 nTPM cleanliness threshold that killed CD19 on its first pass. A larger or gentler screen might find the intersection. The rule-of-three bound of 1.78% limits how large the missed class can be within this pool, but says nothing about proteins the pool excluded.

Alternative 2: a base-rate effect. If most surface proteins are expressed somewhere in 51 normal tissues, then any subset of 15 would mostly fail the tissue filter, and the disjointness is arithmetic rather than biology. This is the serious objection and it needs its null: the measured base rate is that only 3 of 169 index-expressed surface genes are parenchymally clean - 1.8%. Under that null, the expected number of clean genes among 15 is 0.27, so observing 0 is entirely unsurprising. The screen as run cannot distinguish the hypothesis from the base rate, and saying so is more useful than the hypothesis.

That is what the falsifiers below have to fix: the discriminating test is not whether the intersection is empty, but whether fusion-driven antigens are more normal-tissue-exposed than lineage antigens matched for expression level.

Facts and figures
statementIn KMT2A::AFF1 leukaemia, the set of fusion-driven surface antigens and the set of surface antigens absent from normal parenchyma are disjoint, because the fusion's output is a developmental programme whose surface genes are re-used elsewhere in the body.
origindata-pattern
prior probability0.3
predictions
[
  "A surface-proteomic screen of an isogenic human KMT2A::AFF1 model will return antigens that are individually normal-developmental proteins (pericyte, neural, endothelial) rather than leukaemia-restricted ones.",
  "Fusion-driven antigens across paediatric fusion leukaemias will show HIGHER maximum normal-tissue expression than lineage antigens matched for expression level in the tumour.",
  "Fusions that create a chimeric extracellular product, or that drive a tissue-restricted rather than developmental programme, will violate the disjointness."
]
experiments
[
  "exp-surface-proteomics-of-an-isogenic-human-kmt2a-aff1-model"
]
analyses
[
  "ana-the-kmt2a-aff1-surfaceome-screen-and-its-zero",
  "ana-lamp5-the-published-mll-r-immunotherapy-target-sits-at-298-ntpm-in-basal-ganglia"
]
falsifiers
[
  "MATCHED-COMPARISON FALSIFIER, and the one that actually discriminates: take the 15 fusion-enriched surface genes and 15 lineage-restricted surface genes matched on index-arm nTPM to within 20%, and compare maximum HPA parenchymal nTPM. If the fusion-enriched group's median parenchymal maximum is NOT at least 2-fold higher than the lineage-matched group's, the hypothesis is refuted - the disjointness would then be the 1.8% base rate and nothing more. RUN IN CYCLE 115, THE SAME HOUR THIS HYPOTHESIS WAS WRITTEN, AND IT REFUTED IT. analyses/kmt2a-aff1-surface-target-screen/falsifier_matched.py, 20,000 permutations, seed 20260923: median parenchymal maximum 97.4 nTPM for the 15 fusion-enriched genes against 106.9 nTPM for expression-matched non-fusion-enriched controls from the same 169-gene pool, ratio 0.911, permutation p = 0.789. 14 of 15 matched; IFITM2 at 1,603 nTPM had no expression match and is reported dropped. REFUTED: fusion-enriched surface antigens are NOT more normal-tissue-exposed than expression-matched controls, and the screen's empty intersection is the 1.8% base rate (3 of 169 parenchymally clean) and nothing more. The null at n=14 excludes a 2-fold effect, not a 1.3-fold one, so the correct reading is that this design cannot support the hypothesis rather than that no such effect exists. Scored in ana-the-disjointness-is-the-base-rate-a-matched-permutation-refuted-this-beats-own-hypothesis-the-hour-it-was-written.",
  "A single surface protein that is >= 4x fusion-enriched in SEM and RS4;11 against a B-precursor comparator AND below 5 nTPM in all 45 non-lymphohaematopoietic HPA tissues refutes the disjointness outright.",
  "If a cell-surface-capture proteomic screen of an isogenic KMT2A::AFF1 model returns a leukaemia-restricted antigen with no normal-tissue counterpart, the hypothesis fails at the level it was proposed.",
  "If FOLR1-directed CAR-T for CBFA2T3::GLIS2 proves tolerable in children at therapeutic doses, the normal-tissue threshold used here is wrong and the hypothesis's evidence base collapses."
]
quantified cycle 136
{
  "PROM1": 5.27,
  "LAMP5": 19.57,
  "CSPG4": "undefined (marrow 0.0)",
  "TAPT1": 1.18,
  "note": "first normal-tissue numbers for surface antigens; the earlier filter scored 22 chromatin targets and no surface antigen. PROM1 and LAMP5 support the empirical claim; TAPT1 at 1.18 argues against the mechanistic one.",
  "confidence_change": "weak -> moderate, on the empirical claim only"
}
How this could bite us back
  • This hypothesis is retained with its refutation attached rather than deleted, because the mechanistic story is tempting and a later beat would otherwise propose it again. Anyone quoting it must quote the falsifier outcome with it.
  • If true, this says the fusion-specific-antigen strategy that worked for CBFA2T3::GLIS2 will not transfer to KMT2A::AFF1, and a group that spends two years building an isogenic model to find one will find developmental proteins. That is an expensive prediction to get wrong in either direction.
  • The hypothesis as stated is currently indistinguishable from the 1.8% base rate, and a reader who takes it as established would be taking a base rate for a mechanism - which is a failure this project has recorded repeatedly.
  • A hypothesis that predicts 'no good target exists' is unfalsifiable in practice if nobody looks, because absence of a discovery confirms it. The matched-comparison falsifier is written to avoid that, and it can be run today.
Still open
  1. REFUTED, so the live question is the successor one: if fusion-driven antigens are no dirtier than anything else, and 98.2% of leukaemia-expressed surface proteins fail the normal-tissue filter, is the filter measuring biology or measuring its own 5 nTPM threshold? A threshold sweep from 1 to 50 nTPM would answer it and has not been run.
  2. Is the matched-comparison falsifier above already answerable from results.json without any new data? It appears to be, and if so this hypothesis should be scored rather than carried.
  3. Does the disjointness hold for KMT2A::MLLT3, whose output programme overlaps but is not identical? Three MLLT3 lines are already in the arms table.
  4. What would a fusion that violates the hypothesis look like? Naming the counter-example class in advance is the difference between a hypothesis and a description.
Every source behind this entry (4)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    Among 169 surface proteins expressed in the KMT2A::AFF1 B-lineage arm, 15 are fusion-enriched and 3 (CD19, SIT1, S1PR4) are parenchymally clean, and the intersection is empty; the three clean ones are all leukocyte-restricted, with PTPRC profile correlations of 0.818, 0.882 and 0.811.

    What this evidence cannot show

    Cell-line mRNA, pre-registered thresholds, two index lines in the B-arm. The base rate of parenchymal cleanliness in this pool is 1.8%, which by itself predicts 0 clean genes among 15.

    Evidence that argues against this (1)
    • The base-rate calculation above shows the observation is consistent with chance. This hypothesis is therefore currently unsupported by its own motivating data and is recorded as a hypothesis for that reason. Oncofusion Knowledge Graph 2026
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    NG2/CSPG4 is an epigenetically regulated direct target gene of MLL-AF4, and is a canonical pericyte and oligodendrocyte-precursor antigen present at 124.5 nTPM in normal blood vessel.

    PMID 39093982DOI
    What this evidence cannot show

    The fusion-target mechanism is from the Blood 2024 abstract; the pericyte identity is from a large separate literature not read on this beat.

    Evidence that argues against this (1)
    • NG2 could be a developmental surface gene the fusion happens to reach through an open chromatin state rather than a gene the fusion's programme requires, in which case the mechanistic story does not hold even though the observation does. Blood 2024
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    LAMP5 is a direct target of the oncogenic MLL fusion protein and reaches 297.8 nTPM in normal basal ganglia and 323.89 TPM in GTEx nucleus accumbens.

    PMID 33910331DOI
    What this evidence cannot show

    Partner-agnostic MLL-rearranged signature; not shown for KMT2A::AFF1 specifically, and this beat measures only 1.18x over other KMT2A partners.

    Evidence that argues against this (1)
    • A brain protein and a pericyte protein are two data points. Two is not a pattern, and both were selected because they are the published fusion-driven antigens rather than by any unbiased procedure. Haematologica 2022
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    FOLR1, the fusion-specific antigen identified in a CBFA2T3::GLIS2 isogenic human model, is present at 2,860.9 nTPM in choroid plexus, 206.7 in salivary gland, 182.4 in lung and 142.6 in kidney.

    What this evidence cannot show

    Adult bulk tissue mRNA. The CBFA2T3::GLIS2 programme has now advanced to a CAR-T despite this, so the field's revealed judgement is that such expression is tolerable.

    Evidence that argues against this (1)
    • If FOLR1-directed CAR-T proves tolerable in children, the whole normal-tissue filter used here is calibrated too strictly and the hypothesis's central observation dissolves into a threshold artefact. J Clin Invest 2022
Connected to