Oncofusion

start / graph / hyp-q144-the-n-over-c-score-reports-where-the-wild-type-n-terminal-fragment-sits-as-well-as-the-fusion

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q144-unbounded-magnitude

The N-over-C oncoprotein score is not fusion-specific by construction, because taspase1 cleaves wild-type KMT2A into an N320 and a C180 fragment that need not occupy the same chromatin - so a fusion-negative cell has a real N/C signal

The test for the cancer fusion protein compares antibodies against the front and back of a normal protein, assuming both halves sit in the same places. But the normal protein is cut in two inside the cell, and the two halves can end up in different places, which would make the test fire even where there is no fusion.

The same thing, for a scientist

**The stated logic of the published score** is that the N-terminal antibodies see the fusion **plus** the remaining wild-type KMT2A allele, while the C-terminal antibodies see wild-type only; therefore N/C > 1 marks the fusion. **That logic assumes the wild-type N and C halves are one object on chromatin. They are not.** KMT2A is cleaved by taspase1 into **MLL-N320** and **MLL-C180**, and UniProt Q03164 carries *separate* subcellular-location comments for the full-length protein, for the N320 cleavage product and for the C180 cleavage product [E1]. Two fragments with their own annotated behaviour are two species, and nothing requires their genome-wide distributions to be identical. **The prediction, and it is already partly measured in this beat.** If the wild-type N and C fragments differ in distribution, a **fusion-negative** cell scored with the same criterion will call real sites - not scattered noise but a structured, reproducible set. Run on the CD34+ HSPC control (GSE159608, same four antibody catalogue numbers), the published criterion at TSS +/- 1 kb promoters calls a substantial promoter set, while the **same-reagent** controls in that sample (N1 vs N2, C1 vs C2) call far fewer [E2]. **The N-vs-C asymmetry is therefore not antibody-pair noise**, because holding the protein end fixed and swapping the reagent does not produce it. **Alternative explanations that must be excluded before this is believed** - and none is excluded yet: (i) the C antibodies simply have lower efficiency, giving a uniform N/C offset that a promoter-level threshold converts into calls; (ii) the N antibodies cross-react; (iii) the CD34+ donor carries a cryptic rearrangement. The first is the most likely and the cheapest to test.

Go deeper
Why this matters to this corpus specifically

Several entries in this graph define fusion-target promoters by the N/C criterion and then reason about how that set moves - between a patient's B-ALL and its lineage-switched AML, between patients, and under a menin inhibitor. If a fixed fraction of every such set is wild-type N320 signal, then set-overlap statistics computed on them carry a shared contaminant, and a low overlap is exactly what two partly-random sets produce. The remedy is not to discard those analyses but to score every set-overlap statistic against the overlap two independent sets of the same sizes would give.

What would kill this hypothesis

A uniform, promoter-independent N/C offset in the fusion-negative control. If the CD34+ calls are simply the upper tail of a distribution whose whole mass is shifted - i.e. the log N/C distribution is displaced but unstructured, and the called promoters are the highest-coverage ones rather than a coherent gene set - then the effect is antibody efficiency, not fragment biology, and this hypothesis is wrong.

Facts and figures
statementThe wild-type KMT2A taspase1 fragments N320 and C180 do not occupy identical chromatin, so the N-over-C ratio has a fusion-independent component and calls sites in fusion-negative cells.
origindata-pattern
prior probability0.3
predictions
[
  "A fusion-negative CD34+ sample scored with the published criterion calls a structured, reproducible promoter set rather than a random one.",
  "The same-reagent controls (N1 vs N2, C1 vs C2) in that sample call far fewer promoters than N vs C does.",
  "The CD34+ calls are enriched for promoters, CpG islands and H3K4me3 - the wild-type MLL1 address - rather than for the HOXA/MEIS1 fusion program.",
  "A taspase1-uncleavable KMT2A knock-in would collapse the fusion-negative N/C signal."
]
experiments
[
  "exp-q144-recalibrate-the-n-over-c-threshold-on-the-fusion-negative-control-at-the-interval-scale-it-will-be-used-at"
]
analyses
[
  "analyses/q144-unbounded-magnitude/run.py"
]
falsifiers
[
  "In the CD34+ HSPC control (GSE159608, GSM5139616-GSM5139619), the promoters called by the published criterion at TSS +/- 1 kb are NOT a coherent set: their identity does not reproduce across a seeded 50/50 split of the fragments in each library, or they are simply the top decile by total N+C coverage. Either result means the fusion-negative calls are power, not biology, and this hypothesis is dead.",
  "The log2 N/C distribution in CD34+ is unimodal and merely shifted, with the shift equal to the global ratio of library efficiencies, so that a single per-sample offset removes every call. That would make the effect an antibody-efficiency constant rather than a per-promoter property.",
  "Published MLL-N320 and MLL-C180 ChIP or CUT&RUN in any primary human haematopoietic cell shows the two fragments at the same intervals within 1.25-fold across the top 1,000 sites."
]
How this could bite us back
  • If the N/C score has a wild-type component, a menin inhibitor that evicts the fusion will leave that component behind, and the residual signal will read as incomplete target engagement when it is in fact the wild-type protein. A pharmacodynamic assay built on the score would under-report response.
Still open
  1. Is there any published genome-wide comparison of MLL-N320 against MLL-C180 occupancy in the same cells, as opposed to N- and C-terminal antibodies used as a fusion discriminator?
Every source behind this entry (2)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    UniProt Q03164 (KMT2A) defines three chains - full-length 1-3969, MLL cleavage product N320 (1-2718) and MLL cleavage product C180 (2719-3969) - and carries a separate subcellular-location comment for each, so the database models the two taspase1 fragments as distinct species.

    UniProt Q03164 - KMT2A_HUMAN, Histone-lysine N-methyltransferase 2A · UniProt Consortium · UniProt Knowledgebase 2026
    What this evidence cannot show

    Verified against live UniProt REST on 2026-09-24: the full-length comment carries ECO:0000269 and BOTH fragment comments are uncoded, i.e. curator statements with no experimental evidence code attached. All three comments give the same location, 'Nucleus'.

    Evidence that argues against this (1)
    • UniProt gives all three chains the SAME location - Nucleus - so the database does not say the fragments are anywhere different, and this evidence item establishes only that they are modelled as separate species. Separate chains are fully compatible with identical chromatin occupancy: N320 and C180 re-associate into a stable complex after cleavage, and on that model the N and C antibodies report the same sites and the score is clean. This hypothesis needs the re-association to be incomplete or site-dependent, which is an additional and unestablished claim. UniProt Consortium 2026
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    Applying the published N/C criterion to TSS +/- 1 kb promoters in a fusion-negative CD34+ HSPC sample calls a substantial set of promoters, while N1-vs-N2 and C1-vs-C2 in the same sample call an order of magnitude fewer.

    What this evidence cannot show

    One donor, one CD34+ sample, one interval definition, hg19. The comparison to the patient samples is a comparison of COUNTS between two accessions and two genome builds, never a per-promoter join across builds. No spike-in exists in either deposit.

    Evidence that argues against this (1)
    • The paper calibrated this same criterion on this same CD34+ sample to yield only three false positives, on its own peak intervals. The most economical reading of the discrepancy is therefore that promoters are the wrong interval scale for a threshold calibrated on peaks - an instrument-scale explanation that needs no fragment biology at all, and that is strictly simpler than this hypothesis. Janssens DH et al. 2024
Connected to