Oncofusion

start / graph / hyp-q118-the-drug-removes-menin-and-leaves-an-mll1-n-complex-at-the-fusions-own-sites

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q118-menin-at-the-human-index-fusion

In the human index fusion the menin inhibitor subtracts MENIN and leaves MLL1-N-reactive material in place at the same sites - so the drug's chromatin action is decoupling rather than eviction, and what this literature calls fusion eviction is a menin measurement

also called decoupling, not eviction · Q118 hypothesis · menin out, MLL1-N in place

The drug takes menin off the DNA almost completely but leaves the MLL1 protein where it was, at the very same spots. If that surviving material includes the cancer fusion protein, the drug is separating two partners rather than clearing the fusion away, and a different kind of drug would be needed to remove it.

The same thing, for a scientist

**Statement.** In human KMT2A::AFF1 cells (RS4;11) at 330 nM VTP-50469 for 3 days, menin's within-sample share at its own strongest 2 kb bins falls to **0.1999** while MLL1-N-reactive material at the *same* bins falls only to **0.7996** - a **4.00-fold** discrepancy - and at the 692 bins where the two antibodies' held sets intersect the figures are **0.2005** and **0.8305**. **The proposed reading is that the drug dissociates menin from a chromatin-bound MLL1-N-containing complex rather than removing that complex from chromatin** [E1]. **Origin: data-pattern.** Prior probability **0.45**, and the number is low deliberately: the competing reading - that the fusion leaves with menin and the flat MLL1-N track is wild-type MLL1 remaining behind - is equally consistent with every measurement in this beat, and the reagents in GSE127507 cannot separate them. **What makes this more than a restatement of the attribution problem** is three quantitative facts that sit awkwardly with pure dilution [E1][E2]: 1. **At MLL1-N-only sites the signal does not move at all (1.0422)** while menin at those same sites falls to 0.4014. If the MLL1-N track were fusion plus wild-type, sites with no menin at baseline should be the wild-type-dominated ones and should be flat - which they are - but sites with menin should show a fusion-shaped loss, and the intersection set shows only 0.8305. 2. **MLL1-N enrichment at its own held set rises slightly in MOLM13 (13.00x to 14.10x)** and falls only 14.20x to 12.18x in RS4;11, with the input-corrected MOLM13 figure at **0.9891**. A track that is a mixture losing one of its two components should fall by the component's share. 3. **53 bins GAIN MLL1-N-reactive material**, reproducibly in a second line (1.7716 against a class-matched null of 1.0134, one-sided p = 5.0e-4) with the input flat at 1.0115. **Wild-type MLL1 has no reason to relocate when a menin inhibitor is added; a released fusion does.** **None of the three is decisive** and the entry says so. Fact 1 has a dilution-compatible reading if wild-type MLL1 is the dominant species at the intersection set too; fact 2 is a share statistic that cannot see a uniform loss; fact 3 is 0.05% of the track. **The competing hypothesis this must beat** is Q117's, generalised: *the fusion is evicted with menin, and the MLL1-N flatness is wild-type MLL1.* Under that model the numbers in this beat are all correct and the word for what happens is still eviction. **This graph also already holds `hyp-the-fusion-stays-on-chromatin-and-stops-working`, which is the same idea reached from a different direction, and this entry is the quantitative version of it in the index fusion.**

Go deeper
What this predicts, in the order the tests become cheap
  1. A tagged KMT2A::AFF1 ChIP in RS4;11 under VTP-50469 gives a share ratio above 0.6 at the fusion's own held set, not menin's 0.20. That is the whole hypothesis in one experiment and it is exp-q118-a-tagged-kmt2a-aff1-chip-in-rs4-11-under-vtp-50469.
  2. An MLL1 C-TERMINAL ChIP in RS4;11, which sees wild-type MLL1 only, gives a share ratio near 1.0 and a DIFFERENT held set from MLL1-N. If instead the C-terminal held set is nearly the same as the MLL1-N held set, the MLL1-N track is wild-type-dominated and this hypothesis is in trouble. No such sample exists anywhere known to this graph.
  3. H3K79me2 falls at the held set while MLL1-N occupancy does not. A decoupled complex should stop working. GSM3635697/98 are the RS4;11 H3K79me2 pair in this same series and they were not downloaded; this is one afternoon.
  4. AFF1, ENL and DOT1L co-occupancy at the held set behaves like MLL1-N and not like menin. If the SEC stays when menin goes, the retained complex is functional and this hypothesis becomes more interesting; if the SEC leaves, the retained material is an inert MLL1-N remnant. GSE127507 has DOT1L for RS4;11 (GSM3635691/92, not downloaded) and no AFF1 or ENL.
  5. The 53 gained bins are fusion-positive in a tagged experiment. If the gain is the fusion, redistribution is real and small; if it is wild-type MLL1, it is a different phenomenon.
Facts and figures
statementA menin inhibitor dissociates menin from a chromatin-bound MLL1-N-containing complex in human KMT2A::AFF1 cells rather than removing that complex from chromatin, so the drug's chromatin action is decoupling rather than eviction.
origindata-pattern
prior probability0.55
prior probability noteRaised from 0.45 to 0.55 on 2026-09-23 after the FUNCTION ARM falsifier was run and did not fire. The original 0.45 is recorded here rather than overwritten. The move is deliberately small and it is a HALF-move: the four-way ordering (MENIN 0.2005, DOT1L 0.2003, H3K79me2 0.6578, MLL1-N 0.8305 at 692 co-occupied bins) shows that something MLL1-N-reactive is left behind after its partners and most of its catalytic output have gone, which is the shape this hypothesis predicts. It does NOT show that the something is the fusion - the competing dilution reading (the fusion left with menin, and what remains is wild-type MLL1 which never had menin or DOT1L in the first place) survives every measurement here and would predict the same ordering. Only the tagged-fusion experiment moves this prior decisively.
competing hypothesisThe fusion is evicted together with menin - Q117's reading - and the flat MLL1-N track is wild-type MLL1 remaining behind, in which case the masking factor of 4.00 is a dilution factor and the fusion's true share ratio is near 0.20.
shown for fusionKMT2A::AFF1 in human RS4;11, and KMT2A::MLLT3 in human MOLM13 - via a shared-epitope MLL1 N-terminus antibody in both, so the fusion is not distinguished from wild-type MLL1 in any measurement supporting this hypothesis
predictions
[
  "a tagged KMT2A::AFF1 ChIP in RS4;11 under VTP-50469 gives a within-sample share ratio above 0.6 at the fusion's own top 0.1% of 2 kb bins, not menin's 0.20",
  "an MLL1 C-terminal ChIP in RS4;11 gives a share ratio near 1.0 AND a held set substantially different from the MLL1-N held set",
  "H3K79me2 falls at the MLL1-N held set under VTP-50469 while MLL1-N occupancy there does not",
  "DOT1L and SEC co-occupancy at the MLL1-N held set behaves like MLL1-N rather than like menin. REFUTED IN CYCLE 68: DOT1L behaves like MENIN, to three decimal places at the co-occupied set (DOT1L 0.2003 against MENIN 0.2005, with MLL1-N at 0.8305). The retained MLL1-N-reactive material has lost menin, lost DOT1L and lost a third of its H3K79me2, so it is a remnant and not a functional complex minus menin. The SEC itself (AFF1, ENL) is not measured in GSE127507 at all and that half of the prediction is still untested.",
  "the 53 gained bins are fusion-positive in a tagged experiment"
]
experiments
[
  "exp-q118-a-tagged-kmt2a-aff1-chip-in-rs4-11-under-vtp-50469"
]
analyses
[
  "ana-q118-the-mll1-n-antibody-masks-the-occupancy-collapse-four-fold-in-rs4-11",
  "ana-q118-menin-occupancy-in-rs4-11-falls-to-one-fifth-and-reproduces-q117s-murine-numbers",
  "ana-q118-fifty-three-non-island-bins-gain-mll1-n-signal-in-two-lines-with-a-flat-input",
  "ana-q118-molm13-is-the-only-line-with-both-inputs-and-its-mll1-n-occupancy-does-not-move"
]
falsifiers
[
  "TAGGED-FUSION ARM, the decisive one. If a FLAG-, HA- or degron-tagged KMT2A::AFF1 ChIP in RS4;11 at 330 nM VTP-50469 for 3 days gives a within-sample share ratio BELOW 0.35 at the tagged fusion's own top 0.1% of blacklist-filtered 2 kb hg19 bins - that is, if the fusion behaves like menin (0.1999) rather than like the MLL1-N track (0.7996-0.9069) - then this hypothesis is dead and Q117's eviction reading generalises to the index fusion unchanged. NOT YET RUN; no tagged human KMT2A::AFF1 occupancy dataset exists in any archive known to this graph.",
  "WILD-TYPE-DOMINANCE ARM. If an MLL1 C-TERMINUS ChIP in RS4;11 (wild-type-only, because the C-terminus is lost in the fusion) returns a held set whose Jaccard with the MLL1-N held set exceeds 0.60 at top 0.1% of 2 kb bins, the MLL1-N track is wild-type-dominated, the 4.00-fold discrepancy is dilution, and this hypothesis loses its main support. NOT YET RUN; no such sample exists in GSE127507 or in any dataset known to this graph.",
  "FUNCTION ARM. If H3K79me2 at the 1,390 RS4;11 MLL1-N held bins has a within-sample share ratio above 0.75 under VTP-50469 - i.e. if the mark does NOT fall where occupancy is retained - then the retained complex is still functional and 'decoupling' understates what survives. The samples are GSM3635697 (DMSO) and GSM3635698 (VTP 330 nM) in this same series and they were NOT downloaded in this beat. RUN IN CYCLE 68 - DOES NOT FIRE. GSM3635697/98 were downloaded, binned to 2 kb and deleted. H3K79me2's within-sample share ratio at the 1,390 RS4;11 MLL1-N held bins is 0.7008, floor-normalised 0.7114 against a genome-wide mid-occupancy floor of 0.9851 over ~555,800 bins - below the 0.75 threshold. At the 692 bins where MLL1-N and MENIN co-occupy the four-way ordering is MENIN 0.2005, DOT1L 0.2003, H3K79me2 0.6578, MLL1-N 0.8305. So the mark DOES fall where occupancy is retained, and the retained material is a remnant rather than a working complex. See ana-q118-menin-and-dot1l-leave-together-h3k79me2-falls-and-mll1-n-stays. **RUN IN CYCLE 79 BY Q120 - SPLIT VERDICT ACROSS TWO CELL LINES, WHICH IS ITSELF THE FINDING.** H3K79me2 at the MLL1-N held set: **RS4;11 0.7008 - DOES NOT FIRE** against the 0.75 threshold; **MOLM13 0.9980 - FIRES**, and fires by a wide margin. Same series, same dose, same instrument, same construction; the only difference is the fusion partner (KMT2A::AFF1 versus KMT2A::MLLT3). **THIS FALSIFIER WAS PERFECTLY SCOREABLE AND STILL FLIPPED, BECAUSE IT NAMED AN INSTRUMENT AND A THRESHOLD BUT NOT A POPULATION.** Cycle 76's lesson was that a falsifier is only as good as the reagent it names; this is the adjacent failure - a falsifier must name the population it is scored over, not just the instance it was written from. **NO PRIOR MOVE IS RECORDED HERE**, because a split verdict across the only two lines measured does not support one: the honest reading is that the quantity this arm tests is partner-dependent, which is what `hyp-q120-whether-h3k79me2-survives-a-menin-inhibitor-is-set-by-the-fusion-partner-not-by-menin` now proposes. Source: q120 `results-arm1.json` / `results-addendum.json`, GSM3635687/88 and GSM3635685/86, figures re-grepped by the king.",
  "GAIN ARM. If the 53 gained bins fail to reproduce in a second independent RS4;11 VTP-50469 MLL1-N experiment at a share ratio above 1.3 against a class-matched null, the redistribution half is noise. NOT YET RUN; there is no replicate in GSE127507.",
  "DEPLETION ARM. If an immunoblot of total MLL1 N-terminus-reactive protein in RS4;11 at 330 nM VTP-50469 for 3 days shows a reduction above 40%, then the flat MLL1-N share is a flat SHAPE over a smaller amount and the word 'leaves in place' is wrong. NOT YET RUN; a within-sample share is blind to a uniform loss by construction, which is this hypothesis's largest single vulnerability.",
  "UNIFORM-LOSS ARM, and it is now this hypothesis's largest single vulnerability. A within-sample share is blind to a uniform global loss by construction, so 'MLL1-N stays at 1.036 over floor' is a statement about DISTRIBUTION and not amount. If a spike-in-normalised or immunoblot-calibrated measurement of total MLL1 N-terminus-reactive chromatin-bound protein in RS4;11 at 330 nM VTP-50469 for 3 days shows a reduction above 40%, then the flat share is a flat shape over a smaller amount and the word 'leaves in place' is wrong. NOT YET RUN; no calibrant exists in GSE127507 because the depositors ran a Drosophila spike-in and deliberately did not use it."
]
function arm outcome cycle 79
{
  "verdict": "SPLIT ACROSS LINES - does not fire in RS4;11 (0.7008), fires in MOLM13 (0.9980)",
  "threshold": 0.75,
  "what_differs": "the fusion partner only - AFF1 vs MLLT3; series, dose, instrument and construction are identical",
  "why_no_prior_move": "a split verdict over the only two lines measured supports neither direction; the partner-dependence reading is now its own hypothesis",
  "the_general_lesson": "A FALSIFIER MUST NAME ITS POPULATION, NOT ONLY ITS INSTRUMENT AND THRESHOLD. This one was scoreable, pre-registered and cheap, and it still flipped between cell lines because the population it applied to was never stated.",
  "source": "q120 results-arm1.json and results-addendum.json; re-grepped by the king"
}
How this could bite us back
  • If the drug decouples rather than evicts, an interface inhibitor is the wrong modality and a degrader is indicated - and a degrader of the fusion would also degrade wild-type MLL1, which is essential, so the therapeutic window argument has to be made again from scratch.
  • A decoupled complex sitting on chromatin is a substrate for re-binding the moment drug exposure dips. That predicts rapid transcriptional recovery on missed doses and a steep dependence on trough concentration, neither of which a pharmacodynamic assay based on menin ChIP would detect.
  • This hypothesis's prior is 0.45 and its competitor's is therefore 0.55, which means the honest state of the field is that nobody knows whether a menin inhibitor removes the fusion oncoprotein from chromatin in the human index fusion. That is a strong claim about a drug in the clinic and it rests on the absence of one tagged-fusion experiment.
Still open
  1. The four-way ordering is equally consistent with the dilution reading: if the retained MLL1-N signal is wild-type MLL1, it would carry no menin and no fusion-recruited DOT1L to begin with, so menin and DOT1L falling while MLL1-N holds is exactly what dilution predicts too. The test that separates them is whether the retained MLL1-N-positive bins were MENIN-NEGATIVE at baseline - and the 692-bin intersection is by construction menin-positive, so the measurement that matters is the 698 MLL1n-only bins, where MLL1-N is 1.0422 and menin was already low. That comparison exists in this beat's data and was not framed as this test.
  2. What is the fusion:wild-type MLL1 ratio in an MLL1-N ChIP signal in RS4;11? Every number in this hypothesis and in its competitor depends on it and nothing measures it.
  3. If the complex stays and menin goes, what holds it there? The candidates are the retained AT-hooks, the CXXC domain and LEDGF/PSIP1, and this graph holds entries on all three - but no experiment in this beat touches any of them.
  4. Does the decoupled state revert on drug withdrawal faster than an evicted state would? A washout time course would distinguish a complex that is still there from one that has to be rebuilt, and it is a straightforward experiment nobody has published.
Every source behind this entry (2)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    In RS4;11 2 kb bins, menin's within-sample share ratio under VTP-50469 is 0.1999 at its own held set and 0.2560 at MLL1-N's, while MLL1-N's is 0.7996 and 0.9069 respectively; at the 692 bins in both held sets the figures are 0.2005 and 0.8305; at the 698 MLL1-N-only bins MLL1-N is 1.0422 and menin 0.4014; and MLL1-N's enrichment over its own genome median falls only 14.20x to 12.18x in RS4;11 and rises 13.00x to 14.10x in MOLM13 (input-corrected share ratio 0.9891).

    What this evidence cannot show

    NEITHER ANTIBODY IN THIS SERIES IS FUSION-SPECIFIC AND THE ANTIBODY NAME SAYS SO. MLL1n is anti-MLL1 N-terminus (Bethyl A300-086A, lots 5 or 6); the N-terminus is retained in KMT2A::AFF1, so that track is the sum of the fusion and wild-type MLL1 in unknown proportion. MENIN is Bethyl A300-105A Lot 8; menin binds the fusion, wild-type MLL1 and non-KMT2A partners. Q117 could attribute its numbers because its fusion was FLAG-tagged; nothing here can. Every occupancy statement in this beat is about MLL1-N-reactive material or menin-occupied sites, never about the fusion. No scale factor exists for any sample. The depositors state that a Drosophila spike-in was run on one batch, that fly/human ratios were 'found to be small', and that NO spike-in normalisation was applied; no number for that measurement is reported and none is deposited. Every cross-condition figure here is therefore a WITHIN-SAMPLE share or a within-sample rank, which cancel an unknown per-sample scale factor exactly, and never a fold change. n = 1 per antibody per condition per line - there is no replicate anywhere in GSE127507 - so no number here carries a spread and none is precise. RS4;11 is an ADULT-derived t(4;11) line. A share ratio near 1.0 is consistent with total retention AND with a uniform global loss, and this dataset has no statistic that distinguishes them, so 'leaves in place' is a claim about DISTRIBUTION and not about amount. There is no VTP input for RS4;11; the one input-corrected figure is MOLM13's. The MLL1n-AND-MENIN intersection is a proxy for fusion-occupied chromatin and not an attribution.

    Evidence that argues against this (1)
    • The dilution model explains every number here without any decoupling: if wild-type MLL1 dominates the MLL1-N signal at all three partitions, then a complete loss of the fusion component would produce exactly this pattern, and the fusion's true share ratio would be near menin's 0.20. Nothing in GSE127507 measures the fusion:wild-type ratio in the ChIP signal, so the dilution model cannot be excluded and this hypothesis's prior is set at 0.45 for that reason. Krivtsov AV et al. 2019
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    53 2 kb bins rise into the RS4;11 MLL1-N top 0.1% from below the 90th DMSO percentile, and the same 53 coordinates give a MOLM13 MLL1-N share ratio of 1.7716 against a class-matched null median of 1.0134 (2,000 draws, 95% interval 0.893-1.147, one-sided p = 5.0e-4) with the MOLM13 input share ratio at those coordinates 1.0115; they hold 0.0503% of the RS4;11 VTP track against the held set's 1.4070%.

    What this evidence cannot show

    The cross-line test was not pre-registered (DEVIATION 7). The empirical p of 5.0e-4 is the floor for 2,000 draws. Fifty-three bins holding 0.05% of a track is a small quantity. The input control excludes library depth, mappability and copy number but not a shared immunoprecipitation background, and GSE127507 deposits no IgG sample in any line. One of the 53 bins is pericentromeric chr1, 226 kb from the worst blacklisted bin in the series. n = 1 per antibody per condition per line - there is no replicate anywhere in GSE127507 - so no number here carries a spread and none is precise.

    Evidence that argues against this (1)
    • Q117's equivalent 52 murine bins were LOW-amplitude and it used that to override its own pre-registered pass; these 53 are median 15.28x the genome median, HIGHER than the held set's 12.18x, so the amplitude argument that carried Q117's negative is unavailable here. A reader who weights Q117's framing would say the two beats simply measured different things; a reader who weights the controls would say the human gain is real and the murine one was not. Iqbal Lab oncofusion knowledge graph 2026
Connected to