Against a correctly measured same-condition floor, a menin inhibitor's effect on H3K4me3, H3K79me2 and H3K27me3 is a genome-wide redistribution - up in the decile below the peaks, down monotonically to the sparsest decile - while its effect on the FUSION stays confined to the top 0.1%. The drug displaces the complex at the peaks and moves the marks everywhere
also called the marks move everywhere, the fusion does not · two effects with two different footprints · is the bulk shift real or compositional
A menin inhibitor knocks the cancer-driving protein complex off a small number of very strong sites. This proposes that its effect on the chemical tags on DNA packaging is much broader - shifting across nearly the whole genome - which would mean the drug's consequences are far less targeted than the protein-displacement picture suggests.
The same thing, for a scientist
**Statement.** VTP-50469's effect on fusion occupancy is confined to the top 0.1% of fusion sites (0.7554 there, 0.9937 at D1, both verdicts surviving a measured floor). Its effect on the histone marks is not. Against a measured same-condition floor of **2.42% at D1 and 2.28% at D10**, all three scoreable marks move at both ends of the bulk and monotonically in between - **up ~5-10% at D1, down ~6-16% at D10**. The two footprints are different, and the mark footprint is genome-wide. **Origin: data-pattern, arising from a floor correction rather than a new measurement.** Every number was on disk; what changed is the ruler. **The hypothesis has a serious and fully-stated alternative, and separating them is the whole point.** Because these are WITHIN-SAMPLE shares and no spike-in exists, a uniform global LOSS of a mark is invisible, and a purely peak-localised gain would appear as exactly this shape: shares rising near the peaks and falling everywhere else by arithmetic. **A compositional shadow and a real redistribution are not distinguishable in any data this project currently holds.** The hypothesis is that the redistribution is real; the falsifiers below are written to kill it if it is not.
The three marks behave alike, which is the part arithmetic does not force
Compositionality forces a rebalance. It does not force three chemically unrelated marks - an activating H3K4 trimethyl, an elongation-associated H3K79 dimethyl and a repressive H3K27 trimethyl, deposited by three different enzyme systems - to rebalance in the same direction with the same monotone shape.
H3K4me3 runs 1.0982 at D1 down to 0.8744 at D10. H3K79me2 runs 1.0777 down to 0.8383. H3K27me3 runs 1.0478 down to 0.9375. Three antibodies, three enzymes, one shape. That is either one shared compositional cause - which is the alternative hypothesis - or one shared biological cause.
What would settle it, in order of cost
- Cheapest, no new data. Correlate the per-bin changes of the three marks across the bulk. A shared compositional shift predicts near-identical per-bin patterns scaled by each mark's peak concentration; three independent biological redistributions do not. Every bin is on disk.
- One download. Test whether the shape survives a different stratification reference - cut the strata on H3K4me3 DMSO rather than on FLAG DMSO. A compositional shadow of a peak-localised gain should follow whichever track defines the peaks; a real genome-wide redistribution should not.
- Needs new data. A spike-in-normalised histone ChIP under a menin inhibitor. This is the only design that sees a global abundance change, and no such deposit is known in this thread.
| statement | A menin inhibitor's chromatin-mark consequence is a genuine genome-wide redistribution - marks gained in the high-signal bulk immediately below the peaks and lost monotonically toward the sparsest decile - and is not confined to the top 0.1% of fusion sites where the fusion itself is evicted. |
|---|---|
| origin | data-pattern |
| prior probability | 0.35 |
predictions
[ "In a spike-in-normalised H3K4me3 ChIP in a KMT2A-rearranged line at 72 h of a menin inhibitor, absolute H3K4me3 in the sparsest decile of the parental fusion universe falls by more than 10% while total H3K4me3 changes by less than 10%.", "The per-bin changes of H3K4me3, H3K79me2 and H3K27me3 across deciles D2-D10 correlate with each other at Spearman rho below 0.5 - i.e. three partly independent redistributions rather than one shared compositional rescaling.", "Re-cutting the strata on the H3K4me3 DMSO track instead of the FLAG DMSO track leaves the D1-up / D10-down shape intact for all three marks." ]
experiments
[ "exp-spike-in-normalised-histone-chip-under-a-menin-inhibitor" ]
analyses
[ "ana-cycle-104s-m1-fired-on-a-floor-applied-at-the-wrong-stratum-and-the-marks-do-move-in-the-bulk", "ana-a-third-mark-breaks-the-floor-in-the-sparsest-decile-and-the-d1-versus-d10-puzzle-dissolves", "ana-the-same-condition-noise-floor-measured-at-last-nine-percent-at-the-peaks-and-two-percent-everywhere-else" ]
falsifiers
[ "SCOREABLE: if the three marks' per-bin VTP/DMSO log ratios across deciles D2-D10 of the parental FLAG universe in the GSE260742 NT genotype correlate at Spearman rho above 0.8 with each other AND each is proportional to that mark's own peak share, the bulk pattern is one compositional rescaling and this hypothesis is REFUTED. Every bin is on disk; this costs no download.", "SCOREABLE: if re-cutting the strata on the H3K4me3 DMSO track (rather than the FLAG DMSO track) in the same cells abolishes the D1-up / D10-down shape - specifically, if H3K4me3's D10 ratio moves inside 1 +/- 0.023 - then the shape was a shadow of the FLAG stratification and this hypothesis is REFUTED.", "SCOREABLE: if a spike-in-normalised H3K4me3 ChIP in a KMT2A-rearranged human line at 72 h of a clinical menin inhibitor shows total genome-wide H3K4me3 falling by more than 25% while the decile-wise ABSOLUTE profile is flat, the redistribution is a compositional artefact of a global loss and this hypothesis is REFUTED.", "SCOREABLE: if a same-condition histone-mark replicate pair in any menin-inhibitor deposit gives a bulk-decile floor wider than 10%, then cycle 104's original M1 verdict was right after all, the mark movements at D1 are inside the floor, and this hypothesis is REFUTED." ]
- If the redistribution is real, a menin inhibitor is a genome-wide chromatin perturbagen and its therapeutic index rests on the leukaemic cell's dependence at the peaks, not on the drug's spatial selectivity. That is a weaker safety argument than the one this project has been making, and it predicts on-target toxicity in any normal cell carrying the same complexes.
- If instead it is compositional, then the peak-stratum results the whole thread rests on are also partly compositional, and an apparent GAIN of a mark at the peaks under drug could be a LOSS everywhere else - which would invert several published readings of the same tables, including this project's own.
- Either way, within-sample share cannot tell the difference, and every chromatin conclusion in this thread uses within-sample share. The missing reagent is a spike-in, not a better statistic.
- Is the bulk mark movement a property of menin inhibition or of any perturbation that reduces transcription at a large gene programme? No non-menin drug arm with matched histone ChIP exists in this thread, so the specificity of the effect is untested.
- The fusion and the marks now have different footprints. If the marks move where the fusion does not, what carries the signal to the bulk - diffusion of a freed enzyme pool, a change in nuclear concentration of SAM or acetyl-CoA, or a transcription-coupled feedback? None of these has been examined here.
- Every falsifier above except the third is computable from data already on disk and none has been run.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
Against a same-condition floor measured at 2.42% (D1) and 2.28% (D10) in vehicle-versus-vehicle pairs, cycle 104's control-genotype VTP/DMSO mark ratios exceed the floor at both strata for H3K4me3 (1.0982, 0.8744), H3K79me2 (1.0777, 0.8383) and H3K27me3 (1.0478, 0.9375), while the fusion's own VTP/DMSO ratio is 0.7554 at the top 0.1% and 0.9937 at D1, the latter inside the floor.
What this evidence cannot show
The floor is measured in human MV4;11 (KMT2A::AFF1 but acute MONOCYTIC with FLT3-ITD, not infant B-ALL) with a menin-complex factor ChIP; the marks are mouse mm10 histone ChIPs in an MLL-AF9 line. n = 1 per mark per treatment. Two of the three D1 verdicts also hold against GSE260742's own stratum-matched input floor of 5.68% and so do not depend on the transfer.
Evidence that argues against this (1)
- The decisive counter is compositional and is stated in the hypothesis itself: within-sample shares sum to 1, so a purely peak-localised gain produces a fall everywhere else with no real bulk change. Cycle 103 recorded this as its own U2 - 'SHARES ARE COMPOSITIONAL - a deficit FORCES a surplus, so its existence is arithmetic and only its SHAPE can be informative'. Nothing in this project can currently distinguish the two, and the hypothesis is filed as speculative for that reason. source 2026
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
GSE260742 contains no spike-in that would permit absolute normalisation, and the within-sample share convention was adopted across cycles 101-104 for that reason.
What this evidence cannot show
An internal prior finding, re-read rather than re-run. GSE299100 was not searched for a spike-in by this beat; its tracks are submitter-processed .tdf with no normalisation method documented, so the same limitation is assumed rather than established there.
Evidence that argues against this (1)
- If GSE299100 or another menin-inhibitor deposit does carry a spike-in, this hypothesis becomes directly testable without new wet work. That search has not been run and is the cheapest step listed in the experiments below. source