Oncofusion

start / graph / ana-q140-vtp-50469-depletes-h3k4me3-at-menin-bound-sites-by-23-percent-in-sem-and-epz-5676-does-not

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q140-sem-menin-arm

In SEM, the infant-derived KMT2A::AFF1 line, VTP-50469 lowers H3K4me3 at menin-bound sites by 23% relative to H3K4me3-matched menin-free sites (95% CI 13-33%), while the DOT1L inhibitor EPZ-5676 changes it by 1% (CI -4 to +6%)

A menin inhibitor is a drug that pries a leukaemia-driving protein off DNA, and we asked what it does to a chemical mark called H3K4me3 in cells from an infant leukaemia. Comparing the places menin sits against places with the same amount of the mark but no menin - inside the same sample, so the measurement itself cannot be blamed - the mark falls by about a quarter where menin was, while a drug aimed at a different enzyme does nothing to it.

The same thing, for a scientist

Measured from the deposited GSE159608 SEM H3K4me3 AutoCUT&Tag libraries (DMSO n=2, 30 uM EPZ-5676 n=2, 30 uM VTP-50469 n=2, 72 h). The quantity reported is the **ratio of a library's read share at menin-bound sites to its read share at menin-free sites**, both drawn from the same top-0.1% DMSO H3K4me3 peak set, matched decile-by-decile on DMSO H3K4me3 level, and both read **inside one library** - so library depth, global scale and fraction-of-reads-in-peaks all cancel. Menin-bound is defined on **menin enrichment over IgG**, not raw menin occupancy. **VTP-50469 / DMSO = 0.766 (bootstrap 95% CI 0.667-0.870, 2,000 draws over 5 kb bins, seed 20260923). EPZ-5676 / DMSO = 1.010 (CI 0.956-1.064).** The separation is complete at the library level: all six non-VTP H3K4me3 libraries give 0.988-1.327, both VTP libraries give 0.777 and 0.800. The same contrast built on **KMT2A N-terminal** occupancy over a same-batch IgG control gives VTP/DMSO = 0.871 (CI 0.762-0.985), EPZ/DMSO = 0.990 (CI 0.950-1.031) - same direction, smaller, and the interval only just excludes 1.

Go deeper
What the effect size means, stated carefully

A 23% relative depletion is not a collapse. The pre-registered prediction P1 was that VTP-50469 would reduce H3K4me3 at HOXA9/MEIS1 by more than 1.5-fold; on the only measure that survives the confounds, it does not. The honest headline is that H3K4me3 at menin-bound sites is partially and selectively reduced, and that the effect is small enough that it was invisible to every cruder measure we tried first. Both of those facts matter for the menin-interface hypothesis and they point in opposite directions: the selectivity supports it, the magnitude limits how much of the drug's lethality H3K4me3 can be carrying. 1

Why the contrast is built this way, and what it is immune to

Three confounds had to be removed, and each removal was forced by a control that fired.

  1. Compositional normalisation. The deposited tracks are depth-scaled with no spike-in, so a global change is invisible by construction and a raw ratio between arms is uninterpretable. A ratio of two site classes within one library is not affected by the global scale at all.
  2. Fraction of reads in peaks. The two VTP libraries hold 3.3% and 3.8% of their reads in the DMSO-defined H3K4me3 peak set against 8.9% for DMSO - a 2.5-fold difference that would, on a raw comparison, look like a drug effect at every locus in the genome. It does exactly that: see ana-q140-the-pre-registered-housekeeping-control-fired. A within-library class ratio cancels a uniform peak-share change.
  3. Accessibility masquerading as occupancy. Defining menin-bound on raw menin signal produces a class that the IgG libraries also pile into (see ana-q140-a-menin-high-bin-class-is-an-accessibility-class-until-you-divide-by-igg). Defining it on menin-over-IgG produces a class the IgG libraries are depleted from - the IgG bound/control ratio is 0.076 and 0.078 - which is the signature of a clean class. 1
How the matching works and why it is not optional

Menin-bound sites carry more H3K4me3 than menin-free sites do. A class with more to lose loses more, and regression to the mean alone would produce the predicted result. So the bound and control classes are drawn within deciles of DMSO H3K4me3 signal: the realised means are 235.7 and 229.0 CPM (menin contrast) and 231.9 and 230.3 CPM (KMT2A-N contrast), while the enrichment separation is 44.6-fold vs 5.2-fold and 39.4-fold vs 6.4-fold. The classes differ in menin, not in H3K4me3. 1

The comparator arm is the part that makes this interpretable

Without EPZ-5676 this would be one number with no scale. EPZ-5676 is a potent, selective DOT1L inhibitor at the same 30 uM, same 72 h, same cells, same submitter, processed identically 4. It moves the menin-bound contrast by 1.0% with a CI that comfortably spans zero, and the KMT2A-N contrast by -1.0%. The DOT1L arm behaves exactly as a negative control should, which is what licenses reading the VTP number as an effect of menin displacement rather than of being treated with something. This is the clean comparator pair that this project has repeatedly found missing elsewhere. 14

Facts and figures
scriptanalyses/sem-arm-identity-from-signal/run3.py
statusrun
cell lineSEM
fusionKMT2A::AFF1
genomehg19
bin size bp5000
n bound bins130
n control bins130
prereganalyses/sem-arm-identity-from-signal/PREREGISTRATION.md; P1 (>1.5-fold at HOXA9/MEIS1) did NOT fire; P2 (EPZ smaller than VTP) fired; P3 (housekeeping unmoved) FAILED on the raw measure and forced this one.
arms
{
  "DMSO": [
    "GSM5139709",
    "GSM5139710"
  ],
  "EPZ5676": [
    "GSM5139711",
    "GSM5139712"
  ],
  "VTP50469": [
    "GSM5139713",
    "GSM5139714"
  ]
}
menin contrast
{
  "VTP_over_DMSO": 0.7657,
  "VTP_95CI": [
    0.6668,
    0.8704
  ],
  "EPZ_over_DMSO": 1.0095,
  "EPZ_95CI": [
    0.9562,
    1.0635
  ],
  "per_library": {
    "GSM5139709": 1.0059,
    "GSM5139710": 1.0532,
    "GSM5139711": 1.0136,
    "GSM5139712": 1.0651,
    "GSM5139713": 0.7767,
    "GSM5139714": 0.8
  },
  "IgG_negative_control_ratio": [
    0.0763,
    0.0781
  ]
}
kmt2a nterm contrast
{
  "VTP_over_DMSO": 0.8707,
  "VTP_95CI": [
    0.7624,
    0.9851
  ],
  "EPZ_over_DMSO": 0.9896,
  "EPZ_95CI": [
    0.9498,
    1.0312
  ],
  "IgG_negative_control_ratio": [
    0.1669,
    0.1352
  ]
}
bootstrap
{
  "draws": 2000,
  "seed": 20260923,
  "unit": "5 kb genomic bin"
}
How this could bite us back
  • If H3K4me3 at menin-bound sites falls only 23% while the drug is curative in these models, H3K4me3 is largely a BYSTANDER and targeting the mark - or using it as a pharmacodynamic biomarker - would mislead. A 23% change is close to the noise of most clinical chromatin assays.
  • Resistance: a MEN1 binding-pocket mutation restores menin occupancy without touching H3K4me3 machinery, so a biomarker built on this 23% would read as 'drug working' in a resistant patient whose menin has re-engaged.
  • Compensation: the non-fusion KMT2A/SETD1 complexes deposit most of the genome's H3K4me3 and are untouched by a menin inhibitor. That is exactly why the effect is small, and it means any attempt to deepen it pharmacologically runs into essential-function toxicity in normal haematopoiesis.
  • Delivery is not the constraint here; interpretation is. This entry's number comes from a 5 kb reanalysis of two libraries per arm and should not be carried into a trial design without the spike-in experiment named in the experiment nodes.
Still open
  1. Is 23% the ceiling, or the 72-hour value? A menin inhibitor displaces menin within hours; H3K4me3 turns over slowly. The same contrast across a time course would say whether H3K4me3 is lagging or genuinely partial.
  2. There is no H3K79me2 readout in the SEM drug arms of this deposit, so EPZ-5676 target engagement at 30 uM / 72 h is assumed, not shown. A DOT1L-inhibitor arm whose engagement is unverified is a weaker negative control than it looks.
  3. Does the 23% concentrate at a subset of menin-bound sites - i.e. is it 100% loss at a quarter of them, or a quarter loss everywhere? At 130 bins and n=2 we cannot tell, and the two possibilities imply different mechanisms.
  4. Menin-bound sites here are defined by a menin antibody in DMSO cells. If VTP-50469 changes where menin is rather than only how much, the class is a pre-treatment definition being scored post-treatment, which is the right design but bounds what 'menin-bound' means.
Every source behind this entry (4)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    In SEM H3K4me3 AutoCUT&Tag, the within-library read-share ratio between menin-over-IgG-enriched sites and H3K4me3-matched menin-free sites falls to 0.766 of DMSO under VTP-50469 (95% CI 0.667-0.870) and is 1.010 under EPZ-5676 (CI 0.956-1.064).

    What this evidence cannot show

    n=2 libraries per arm, one experiment, one cell line, one time point (72 h), one dose (30 uM), 5 kb bins, 130 bound and 130 control bins. The bootstrap resamples bins, not biological replicates, so the interval describes sampling of the genome and NOT biological reproducibility - with n=2 per arm there is no interval on the latter. Confidence is capped at moderate for that reason. Our own reanalysis; not peer-reviewed and not reported in this form by the depositors.

    Evidence that argues against this (2)
    • The depositors' own published claim from this dataset is that aberrant H3K4me3 enrichment in GENE BODIES is menin-inhibitor sensitive - a different statistic from the site-class contrast measured here. Our gene-body:promoter measure at 5 kb resolution was too coarse to test it (GAPDH is shorter than one bin) and is reported as uninformative rather than as agreement or disagreement. Janssens DH et al. 2021
    • VTP-50469 displaces menin and reduces MLL chromatin occupancy at select genes; if occupancy loss were complete at these sites, a 23% H3K4me3 effect is smaller than that mechanism alone would predict, which is itself evidence that H3K4me3 is not a faithful readout of fusion occupancy. Krivtsov AV et al. 2019
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    The libraries used are SEM (t(4;11) KMT2A::AFF1) AutoCUT&Tag for H3K4me3 under DMSO (GSM5139709/10), 30 uM EPZ-5676 (GSM5139711/12) and 30 uM VTP-50469 (GSM5139713/14) for 72 h, with menin (GSM5139719/20) and IgG (GSM5139717/18) in the same line.

    GSE159608: Automated CUT&Tag profiling of chromatin heterogeneity in mixed-lineage leukemia (419 samples, hg19 fragment BEDs) · Janssens DH, Meers MP, Wu SJ, Babaeva E, Meshinchi S, Sarthy JF, Ahmad K, Henikoff S · NCBI Gene Expression Omnibus 2021
    What this evidence cannot show

    Sample identities read from the GEO characteristics fields and independently confirmed against the signal (see the antibody self-sorting entry). The arm assignment is the depositors'; we did not re-derive dose or duration from anything but the record.

    Evidence that argues against this (2)
    • The arm identities are taken from sample titles. This project spent the same cycle retracting a published finding that rested on mis-attributed sample titles, and although the titles here were confirmed from the signal by nearest-neighbour clustering, that confirmation is about ANTIBODY identity and not about which of two same-antibody libraries received which drug. No independent evidence distinguishes the DMSO pair from the EPZ pair.
    • n is two per arm, from one batch, on one day, in one cell line. The library-level separation is complete but with two points per group a complete separation is the most likely outcome under chance for any difference of this size, and the bootstrap interval is over genes rather than over libraries, so it does not bound the between-library variance that actually matters.
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    VTP-50469 is a potent, selective, orally bioavailable menin-MLL inhibitor that displaces menin from protein complexes and reduces MLL chromatin occupancy at select genes in MLL-rearranged leukaemia.

    A Menin-MLL Inhibitor Induces Specific Chromatin Changes and Eradicates Disease in Models of MLL-Rearranged Leukemia · Krivtsov AV, Evans K, Gadrey JY, Eschle BK, Hatton C, Uckelmann HJ, et al., Armstrong SA · Cancer Cell 36:660-673.e11 2019
    PMID 31821784DOI
    What this evidence cannot show

    The chromatin experiments in this paper are ChIP-seq in MLL-r models and PDX rather than CUT&Tag in SEM, so the mechanism is imported into this deposit rather than replicated within it.

    Evidence that argues against this (2)
    • The published potency is cellular-nanomolar; this deposit exposes at 30 uM for 72 h, which is orders of magnitude above the concentration at which selectivity was established. At that exposure off-target activity cannot be excluded, and attributing the H3K4me3 change to menin displacement specifically is an inference from the compound's NAME rather than from any selectivity evidence at the dose used.
    • 'Reduces MLL chromatin occupancy at select genes' was shown in adult-derived MLL-rearranged models, mostly KMT2A::MLLT3. This entry's system is KMT2A::AFF1, and this project has already measured that between fusion backgrounds per-gene ordering transfers while magnitude does not.
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    EPZ-5676 (pinometostat) inhibits DOT1L with Ki 80 pM and 37,000-fold selectivity over other methyltransferases, reducing H3K79 methylation and MLL-fusion target gene expression.

    Potent inhibition of DOT1L as treatment of MLL-fusion leukemia · Daigle SR, Olhava EJ, Therkelsen CA, Basavapathruni A, Jin L, Boriack-Sjodin PA, et al., Pollock RM · Blood 122:1017-1025 2013
    PMID 23801631DOI
    What this evidence cannot show

    Establishes that the comparator arm is a real, selective intervention on a different enzyme - which is what makes its null result here informative rather than merely absent. It does not establish that 30 uM for 72 h in SEM achieved target engagement; no H3K79me2 readout exists in the SEM arm of this deposit to confirm it, and that is a genuine gap (see the open questions).

    Evidence that argues against this (2)
    • The 80 pM Ki and 37,000-fold selectivity are ENZYMATIC, measured on isolated methyltransferase, and do not transfer to a 30 uM cellular exposure for 72 h. Using them to argue that the EPZ arm is a clean DOT1L-specific control overstates what the numbers cover.
    • A null result from EPZ-5676 is weak evidence of specificity unless the compound is shown to have worked in these very libraries. No H3K79me2 readout exists in the SEM drug arms of this deposit, so there is no positive control that the DOT1L inhibitor engaged its target at all - the EPZ arm could be a null because the drug did nothing.
Connected to