Oncofusion

start / graph / ana-q120-three-of-the-four-channels-replicate-across-two-fusion-partners-and-h3k79me2-does-not

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q120-h3k79me2-second-line-and-the-mll1-epitope

The four-way ordering given a second cell line: menin, DOT1L and the MLL1-N channel reproduce between KMT2A::AFF1 and KMT2A::MLLT3 to within 0.04 of a share ratio, and H3K79me2 - the one channel the ordering was built around - diverges by 0.21 to 0.30 and changes places with MLL1-N

also called the second line · the ordering does not fully replicate · H3K79me2 is the odd channel out

Repeating last cycle's four-antibody measurement in a second leukaemia cell line shows that three of the four answers come out almost identically, which is about as good as this kind of experiment gets. The fourth - the chemical tag on the histone - behaves completely differently, and it is the one the whole ordering was built around.

The same thing, for a scientist

**Q118 measured a four-way ordering of retained occupancy under VTP-50469 in ONE cell line** and the king log called a second line the single most valuable missing measurement in the queue. This is that measurement: the same four antibodies, the same drug and dose, the same series, the same 2 kb genome-wide instrument, in **MOLM13 (KMT2A::MLLT3)** instead of **RS4;11 (KMT2A::AFF1)** [E1]. | aggregate within-sample share ratio VTP/DMSO | RS4;11 | MOLM13 | \|delta\| | |---|---:|---:|---:| | **at the MLL1-N held set** (1,390 / 1,389 bins) | | | | | MLL1-N | 0.9069 | 0.9438 | **0.037** | | MENIN | 0.2560 | 0.2578 | **0.002** | | DOT1L | 0.2455 | 0.2773 | **0.032** | | H3K79me2 | 0.7008 | **0.9980** | **0.297** | | **at the MENIN held set** (1,390 / 1,389) | | | | | MLL1-N | 0.7996 | 0.7788 | 0.021 | | MENIN | 0.1999 | 0.1602 | 0.040 | | DOT1L | 0.1959 | 0.1974 | **0.002** | | H3K79me2 | 0.6391 | 0.7107 | 0.072 | | **at the co-occupied set** (692 / 467) | | | | | MLL1-N | 0.8305 | 0.8020 | 0.029 | | MENIN | 0.2005 | 0.1657 | 0.035 | | DOT1L | 0.2003 | 0.2219 | 0.022 | | H3K79me2 | 0.6578 | **0.8667** | **0.209** | **All nine non-H3K79me2 comparisons agree within 0.04 of a ratio (the largest is 0.0397); all three H3K79me2 comparisons disagree by 0.07 to 0.30.** Two cell lines, two different C-terminal fusion partners, two different lineages (B-ALL and AML), n = 1 everywhere, and three of four channels land on top of each other. That is the strongest cross-line agreement this project has measured, and it makes the fourth channel's disagreement hard to write off as noise. **The ordering itself does not replicate, and the pre-registered refutation fired.** Criterion **F2** - *H3K79me2 is as inert as the shared-epitope channel*, fires if `R_K79 >= R_MLL1n - 0.10` at the co-occupied set - **FIRES**: 0.8667 against 0.8020, i.e. H3K79me2 is not merely as inert as MLL1-N, it is *above* it. The observed order low to high in MOLM13 is `MENIN < DOT1L < MLL1-N < H3K79me2` on both the mixed-exposure and the pure 3-day set, against RS4;11's `MENIN ~ DOT1L < H3K79me2 < MLL1-N`. **Criterion O6 (the ordinal pattern) DOES NOT FIRE and criterion O2 (H3K79me2 strictly intermediate) DOES NOT FIRE, on both sets.** **So the transferable statement is narrower than Q118's and better supported.** What replicates across two fusion partners is: *menin and DOT1L leave together at about a fifth, and the MLL1-N-reactive channel does not move.* What does **not** replicate is where the histone mark sits, and the mark is the functional readout.

Go deeper
What the pre-registered criteria did

Scored exactly as written in PREREGISTRATION.md §5, before the bins existed.

idcriterionverdict
O0positive control - MENIN's own share at its own top 0.1% falls (0.1602 against an abort threshold of 0.80)PASS
O1menin and DOT1L leave together, |difference| <= 0.10FIRES on S_AB (0.056); AT THRESHOLD on S_d3 (0.080, the band's own edge)
O2H3K79me2 strictly intermediate, both gaps >= 0.10DOES NOT FIRE on both sets
O3the MLL1-N channel does not move, floor-normalised >= 0.90FIRES (1.0359)
O4H3K79me2 falls by about a third, 0.45-0.85AT THRESHOLD (0.8667, inside the declared upper band 0.80-0.90)
O5eviction magnitude replicates, MENIN <= 0.35FIRES (0.1657)
O6the full ordinal pattern reproducesDOES NOT FIRE on both sets
O7both sets give the same verdictDOES NOT FIRE, and only because O1 crosses a band edge between 0.056 and 0.080; O2 and O6 agree on both sets
F1the mark tracks menin one-for-onedoes not fire
F2the mark is as inert as the shared-epitope channelFIRES
F3DOT1L does not leave with menindoes not fire
F4the MLL1-N channel does move in this linedoes not fire
F5instrument failuredoes not fire

O7's failure is reported as a band-edge artefact and not as a set dependence, which is what the indifference band exists to make visible: 0.056 and 0.080 are the same answer and a fixed cutoff at 0.10 would have called both FIRES, while a cutoff at 0.06 would have called them opposite things.

Why the delta table is the result and the criteria are the scaffolding

A pre-registered criterion gives a verdict; it does not give a size. The nine agreements at 0.002-0.040 and the three disagreements at 0.072-0.297 are the measurement, and they are what makes the H3K79me2 result something other than an n = 1 fluctuation.

The rank statistics say the same thing from a direction where a ceiling cannot help. At the MOLM13 MLL1-N held set, the fraction of bins losing rank under drug is MENIN 0.9150, DOT1L 0.5652, MLL1-N 0.6184 and H3K79me2 0.2498 - i.e. three quarters of those bins gain H3K79me2 rank under a menin inhibitor - and the per-bin median share ratio there is 1.0845 for the mark against 0.3405 for menin. In RS4;11 the mark fell at the same construction.

And the input says it is not a coverage artefact. MOLM13 is the only line in GSE127507 with a matched input in both conditions. The input share ratio is 0.9272 at the co-occupied set, 0.9542 at the MLL1-N held set and 1.0173 at the H3K79me2 held set - so input-correcting moves every figure by under 8% and changes no verdict.

Facts and figures
scriptanalyses/q120-h3k79me2-second-line-and-the-mll1-epitope/arm1-molm13-ordering.py
statusrun
result summaryMENIN 0.1657, DOT1L 0.2219, MLL1-N 0.8020, H3K79me2 0.8667 at the 467 co-occupied MOLM13 bins. Three channels replicate RS4;11 within 0.04; H3K79me2 diverges by 0.209 and overtakes MLL1-N, so the ordering has three levels in MOLM13 and not four.
cell lineMOLM13
fusionKMT2A::MLLT3 (MLL-AF9)
lineageAML
drugVTP-50469 330 nM; 3 days for MLL1n, DOT1L, H3K79me2 and both inputs, 4 days for MENIN
genome buildhg19, verified from each of the ten files' own chromosome extents
bin bp2000
n bins universe1389438
n held1389
antibodyMLL1n Bethyl A300-086A; MENIN Bethyl A300-105A; DOT1L CST 77087S clone 1W4Z; H3K79me2 CST 5427 lot 4
epigenetic readoutH3K79me2 (CST 5427) at the MOLM13 MLL1-N held set has a share ratio of 0.9980 under VTP-50469, floor-normalised 1.0082 - no change - against 0.7008 (0.7114) at the identical construction in RS4;11
mark writtenH3K79me2, by DOT1L, whose own occupancy at the same bins falls to 0.2773
writer of that markgene-dot1l
shown for fusionKMT2A::MLLT3 in human MOLM13, compared against KMT2A::AFF1 in human RS4;11; neither the MLL1-N nor the MENIN arm distinguishes the fusion from wild-type MLL1, so the retained species is unidentified in both lines
absence classmeasured-not-analysed
absence class noteThe gap this entry closes is an ANALYSIS gap, not a data gap. GSE127507's ten MOLM13 .deduped.tdf files including the H3K79me2 pair GSM3635687/88 have been public since 2019; the depositors' own processed table excludes the H3K79me2 samples by their own statement, so the arm was reachable only by binning the raw coverage files, which nobody had done. Cost: twelve downloads and about four minutes of compute.
absence class assigned2026-09-23 (Q120)
inputs
[
  "GSM3635685/86 MOLM13 MLL1n d3",
  "GSM3635683/84 MOLM13 MENIN d4",
  "GSM3635681/82 MOLM13 DOT1L d3",
  "GSM3635687/88 MOLM13 H3K79me2 d3",
  "GSM3635689/90 MOLM13 input DMSO and VTP",
  "ENCODE hg19 blacklist v2",
  "GSE127507 series-level ROSE TSS windows, used only as gene annotation"
]
How this could bite us back
  • If H3K79me2 does not move under a menin inhibitor in an MLL-AF9 line, then the mark is useless as a pharmacodynamic biomarker in exactly the disease where menin inhibitors are furthest advanced - and Q118 proposed it as a better biomarker than the MLL1 ChIP on the strength of the RS4;11 number alone.
  • It also weakens the combination rationale from the other side. Q118 argued a DOT1L inhibitor adds little to a menin inhibitor in RS4;11 because DOT1L is already at 0.20. In MOLM13 DOT1L is also at 0.22 but the MARK is untouched, so there is a large H3K79me2 pool left for a DOT1L inhibitor to remove and the combination rationale is stronger in AML than the occupancy numbers alone suggest.
  • A patient responding to a menin inhibitor in an MLL-AF9 disease may carry chromatin that still looks fully H3K79-methylated. Any trial using H3K79me2 as a target-engagement readout for a MENIN inhibitor would read that as non-engagement, when menin occupancy at the same sites has fallen to 0.16.
Still open
  1. Is the H3K79me2 difference a fusion-partner property or a lineage property? RS4;11 is B-ALL with KMT2A::AFF1 and MOLM13 is AML with KMT2A::MLLT3, and the two are confounded in this series. ML-2 (KMT2A::AFDN, AML) would break the confound in one direction and GSE127507 deposits NO H3K79me2 sample for ML-2 - the only one of the three lines missing it.
  2. If DOT1L occupancy falls to 0.22 in MOLM13 while H3K79me2 does not move at all, what is the mark's decay half-life in these cells? A single 3-day timepoint cannot distinguish 'the mark is stable for longer than three days' from 'the mark is being actively maintained'. This graph already records the H3K79me2 decay half-life as never measured.
  3. MOLM13's partner is AF9, which contacts DOT1L directly, and RS4;11's is AFF1, which reaches DOT1L through the SEC. Naively the direct-contact fusion should lose the mark faster when menin evicts the complex, and the opposite is observed. Does an AF9 fusion leave behind a DOT1L-independent maintenance route, or is H3K79me2 simply more abundant and so better buffered in AML chromatin?
Every source behind this entry (1)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    In MOLM13 (KMT2A::MLLT3), 330 nM VTP-50469, 3 days for every track except the 4-day MENIN pair, 2 kb genome-wide bins, universe 1,389,438 bins (MOLM13 input DMSO > 0, ENCODE hg19 blacklist v2 removed), top 0.1% = 1,389 bins: at the 467 bins in BOTH the MLL1-N and MENIN top 0.1% the within-sample share ratios are MENIN 0.1657, DOT1L 0.2219, MLL1-N 0.8020 and H3K79me2 0.8667 (floor-normalised 0.179, 0.243, 0.880, 0.876 against genome-wide mid-occupancy floors of MENIN 0.9251, DOT1L 0.9132, MLL1-N 0.9111, H3K79me2 0.9899); at the MLL1-N held set alone MLL1-N is 0.9438 (1.036 over floor) and H3K79me2 is 0.9980 (1.008 over floor); the matched input share ratio at the co-occupied set is 0.9272.

    What this evidence cannot show

    THE MENIN PAIR IS A 4-DAY EXPOSURE (GSM3635683/84, `timepoint: 4 days`) AND EVERY OTHER MOLM13 SAMPLE IS 3 DAYS. A longer exposure can only push MENIN further down, so a low MENIN ratio is uninterpretable as biology while a high one would be conservative; the ordering is therefore reported on two sets, S_AB (Q118's exact definition, which mixes the exposures in its own set definition) and S_d3 (MLL1-N and DOT1L tops, pure 3-day), and both are given. NO SPIKE-IN NORMALISATION. All 26 GSE127507 records state a Drosophila spike-in was run, measured, and deliberately not used, and no scale factor is deposited; no GSE260742 record contains spike-in vocabulary at all. Every cross-condition number here is therefore a WITHIN-SAMPLE share or rank, which cancels the unknown per-sample scale factor exactly. No raw fold change is reported as a biological quantity anywhere in this beat. n = 1 per antibody per condition per line in both series. No number in this beat carries a spread and none is precise. Reagents and what each cannot distinguish: Bethyl A300-086A, deposited as 'anti-MLL1n', an MLL1 N-TERMINUS epitope. The N-terminus is retained in every KMT2A fusion, so this track is the sum of the fusion and wild-type MLL1 in unknown proportion and CANNOT distinguish them. The immunogen residue range was not verified: three vendor routes were tried this session (fortislife.com two URL forms, bethyl.com/product/A300-086A) and all returned HTTP 404, so 'N-terminus' rests on the depositors' own target label and on the standard fact that MLL-FPs lack the C-terminal SET domain. Bethyl A300-105A anti-Menin cannot distinguish menin bound to the fusion from menin bound to wild-type KMT2A/KMT2B, nor from menin's non-KMT2A functions. CST #77087S clone 1W4Z anti-DOT1L cannot distinguish DOT1L held by the fusion's partner moiety from DOT1L in wild-type complexes - and the two lines compared here differ in exactly that moiety, MLLT3/AF9 (direct DOT1L contact) versus AFF1 (indirect, through the AF4-family/SEC axis). CST #5427 anti-H3K79me2 reads a histone MARK, i.e. a consequence. It cannot distinguish halted deposition from active demethylation from replicative dilution, and it cannot attribute any of them to the fusion.

    Evidence that argues against this (1)
Connected to