Oncofusion

start / graph / ana-q117-the-residual-fusion-occupancy-fraction-under-a-menin-inhibitor-is-about-18-percent

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q117-gained-fusion-occupancy

The residual fraction this graph says is absent from the entire literature, measured: FLAG-MLLAF9 enrichment at its own top 1,228 sites falls from 15.9x the genome median to 3.0x under VTP50469 - about 18% residual - and 12.8% of those sites stay in the top 0.1%

also called Q117 residual occupancy fraction · 82% of fusion enrichment lost under VTP50469

When the menin drug is applied, the leukaemia fusion protein keeps only about a fifth of the grip it had on its favourite stretches of DNA. That fifth is the number every occupancy-based drug-response test would have to be built on, and until now nobody had published it.

The same thing, for a scientist

`mechanism-vtp-50469-does-not-globally-displace-the-fusion-from-chromatin` states in its own open questions: *"What fraction of fusion peaks survives saturating menin inhibition? The pivotal paper states 'not globally displaced' and names the losses, but never reports the residual fraction, and that number is what any occupancy-based pharmacodynamic endpoint would have to be built on."* **This entry supplies a number for it, in murine MLLAF9 rather than in a human line.** In GSE260742's FLAG ChIP of FLAG-tagged MLLAF9 (KMT2A::MLLT3) in a murine myeloid line, 2 kb bins, mm10, ENCODE-blacklisted, universe 1,227,708 bins, held set = top 0.1% by FLAG DMSO (n = 1,228) [E1]: | quantity, all WITHIN-sample and therefore scale-invariant | DMSO | VTP50469 500 nM 48 h | SGC0946 5 uM 48 h | |---|---:|---:|---:| | median enrichment at held sites over the track's own genome median | **15.9x** | **3.0x** | **6.1x** | | share of the track's total coverage inside the held set | **1.975%** | **0.354%** | **0.589%** | | share ratio vs DMSO | 1.000 | **0.179** | **0.298** | | held sites still in the track's own top 0.1% | 1,228/1,228 | **157 (12.8%)** | **431 (35.1%)** | | held bins losing within-sample percentile rank | - | **92.3%** | **78.9%** | **So the residual is ~18% by signal and ~13% by rank membership**, and the per-bin median enrichment ratio is 0.174 with 95.5% of held bins below 0.5 [E1]. The genome-wide median bin is essentially unchanged (372 -> 375 coverage-bases) and the two libraries have near-identical totals (537.6 M vs 541.5 M coverage-bases, ratio 1.007), so **this result does not depend on the choice between share, rank and raw depth normalisation** - which matters, because no spike-in scale factor is recoverable for this series [E2]. **This is a large displacement, and the graph currently holds that VTP-50469 does not globally displace the fusion.** Both can be true and the reconciliation is in the sections below: Krivtsov's statement is about peaks persisting genome-wide with a shared-epitope MLL1n antibody in human lines, and this is about how much enrichment remains at the strongest sites of an epitope-tagged fusion in mouse, where wild-type MLL1 cannot contribute a single read.

Go deeper
Why this is not simply a contradiction of the pivotal paper

Four differences, each of which pushes the two numbers apart in a predictable direction:

  1. Antibody. Krivtsov et al. used an MLL1n antibody in MOLM13 and RS4;11, which detects wild-type MLL1 as well as the fusion. Wild-type MLL1 occupancy that does not move under the drug dilutes the measured loss. The control for this in their paper is ML2, which has lost the wild-type allele. Here the antibody is anti-FLAG against a FLAG-tagged fusion, so every read is fusion by construction and there is no dilution.
  2. What is being counted. Not globally displaced is a statement about peaks continuing to exist. 18% residual enrichment is a statement about magnitude. A site at 15.9x that falls to 3.0x is still 3x enriched and would very plausibly still be called a peak.
  3. Species, partner and lineage. Murine MLLAF9 in a myeloid line, 48 h at 500 nM; their human lines carry KMT2A::MLLT3, KMT2A::AFF1 and KMT2A::AFDN and were treated for 3 days.
  4. Summit bins, not gene bodies. The held set is the top 0.1% of 2 kb bins. It is deliberately the part of the genome where the fusion is most concentrated, which is where the largest absolute loss is available.

The honest synthesis is that the drug does not remove the fusion from the genome, and it does remove most of the fusion from the places the fusion is most concentrated. Those are compatible and they have different clinical readings.

What a pharmacodynamic assay would have to do with this number

If the intended readout is fusion occupancy at a target locus, an 82% reduction at the strongest sites is a usable dynamic range, but only if the assay is scale-controlled. Depth normalisation alone converts a global loss into an apparent redistribution, which is the failure mode this beat was dispatched to avoid; and the deposit this measurement rests on carries no spike-in 2. Any clinical correlative built on ChIP signal under a menin inhibitor needs a spike-in or a cell-number-referenced normalisation written into the protocol, not added at analysis time.

Facts and figures
scriptanalyses/q117-gained-fusion-occupancy/run117.py
statusrun
result summaryResidual fusion occupancy under VTP50469 at the fusion's own top 0.1% of 2 kb bins: ~18% by signal share, 12.8% by top-0.1% membership, per-bin median enrichment ratio 0.174.
systemmurine MLLAF9 (KMT2A::MLLT3) myeloid line, FLAG-tagged fusion, anti-FLAG ChIP-seq
treatmentVTP50469 500 nM 48 h and SGC0946 5 uM 48 h versus DMSO
genomemm10, verified from each file's own chromosome extents against live UCSC mm10/mm39 sizes
bin bp2000
universe bins1227708
held set bins1228
replicatesn = 1 per condition; no spread can be reported
answers open question inmechanism-vtp-50469-does-not-globally-displace-the-fusion-from-chromatin
shown for fusionKMT2A::MLLT3 (MLLAF9) in a murine myeloid line with a FLAG-tagged fusion; nothing in human KMT2A::AFF1 and nothing in an infant-derived line
enrichment over own genome median
VTP504693.000VTP50469: 3SGC09466.100SGC0946: 6.1DMSO15.9DMSO: 15.9
share of track total in held set
VTP504690.004VTP50469: 0.003541SGC09460.006SGC0946: 0.005888DMSO0.020DMSO: 0.019751
held bins retained in own top 0.1pct
VTP50469157.0VTP50469: 157SGC0946431.0SGC0946: 431denominator1,228denominator: 1228
per bin enrichment ratio VTP over DMSO
q250.099q25: 0.099median0.174median: 0.174q750.271q75: 0.271fraction_below_0.50.955fraction_below_0.5: 0.955
inputs
[
  "GSM8123359",
  "GSM8123360",
  "GSM8123361",
  "GSM8549823",
  "ENCODE mm10 blacklist v2"
]
enrichment over own genome median — the numbers behind the chart
{
  "DMSO": 15.9,
  "VTP50469": 3,
  "SGC0946": 6.1
}
share of track total in held set — the numbers behind the chart
{
  "DMSO": 0.019751,
  "VTP50469": 0.003541,
  "SGC0946": 0.005888
}
share ratio vs DMSO
{
  "VTP50469": 0.179,
  "SGC0946": 0.298
}
held bins retained in own top 0.1pct — the numbers behind the chart
{
  "VTP50469": 157,
  "SGC0946": 431,
  "denominator": 1228
}
per bin enrichment ratio VTP over DMSO — the numbers behind the chart
{
  "median": 0.174,
  "q25": 0.099,
  "q75": 0.271,
  "fraction_below_0.5": 0.955
}
library totals coverage bases
{
  "FLAG_DMSO": 537618204,
  "FLAG_SGC": 595298580,
  "FLAG_VTP": 541500899,
  "note": "printed so the reader can see that the DMSO/VTP comparison is not resting on a large depth difference: ratio 1.007"
}
How this could bite us back
  • An 18% residual is exactly the size of thing that supports resistance. If the surviving sites are the ones that matter - and the locus table for this beat shows the Hoxa cluster is among the best-retained - then a drug that removes 82% of fusion occupancy genome-wide may leave the transforming subset almost untouched, and the clinical response would come from somewhere other than the sites the assay is watching.
  • If a correlative assay is calibrated on the 82% loss and a patient's blasts show only 40%, the natural reading is under-dosing. The alternative reading - that the assay's normalisation drifted - is indistinguishable without a spike-in, and this deposit shows how easily a real dataset ships without one.
  • A residual measured at the strongest sites flatters the drug's apparent potency, because the strongest sites have the most to lose. The same drug measured at mid-ranking sites would show a smaller proportional loss, and which number reaches a protocol is a choice nobody currently has to declare.
Still open
  1. Does the same residual fraction hold for human KMT2A::AFF1? GSE127508 contains RS4;11 MLL1n ChIP plus and minus VTP-50469 and the equivalent within-sample enrichment ratio has never been computed on it - that is a desk job of the same size as this one.
  2. Is the residual 18% the same sites in every replicate, or a different 18% each time? With n = 1 this cannot be asked, and the answer decides whether the residual is a stable pharmacological floor or sampling noise at the top of a collapsed distribution.
  3. At what dose and time does the residual fraction plateau? The series contains a VTP50469 time course (4, 8, 16, 24 h) but only for H3K79me2, H2AK119ub and H3K27me3 - there is no FLAG sample in the time course, so the kinetics of fusion eviction are unmeasured in this deposit.
Every source behind this entry (3)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    In GSE260742 murine MLLAF9 cells, FLAG-MLLAF9 enrichment at its own top-0.1% 2 kb bins falls from 15.9-fold over the track's genome median in DMSO to 3.0-fold under VTP50469 and 6.1-fold under SGC0946; the held set's share of total coverage falls from 1.975% to 0.354% (VTP) and 0.589% (SGC); 157 of 1,228 held bins remain in the top 0.1% under VTP against 431 under SGC.

    What this evidence cannot show

    n = 1 per condition. There is no replicate of the FLAG ChIP in any arm of this series, so none of these numbers carries a spread and none should be quoted to three significant figures as if it did; the project's own record of ChIP percentile spread across replicates of one antibody in one line is 51.8 to 96.3. The measurement is a MAGNITUDE of enrichment, not a peak call - no peak caller was run, and the FLAG arm's four matched input samples carry no deposited processed file, so a proper input-referenced peak call is not possible from the processed deposit. Genome build verified from the chromosome extents written inside each .tdf against live UCSC mm10 and mm39 sizes: every file is within 64 bp of mm10 on every chromosome and ~318 kb from mm39 on chr1. Confidence is moderate rather than strong because of n=1 and because the held set is defined by the DMSO sample itself, so regression to the mean contributes some unknown part of the fall.

    Evidence that argues against this (1)
    • Regression to the mean is a real and unquantified contributor here. The held set is the top 0.1% of the DMSO track, so even two technical replicates of DMSO would show some fall at that set. The SGC0946 arm bounds it: SGC is a different drug with a different mechanism and its held-set share ratio is 0.298, so at most about a third of the observed VTP fall could be shared regression-plus-nonspecific-drug-effect, leaving the VTP-specific part still large. A same-condition replicate would settle it and does not exist in this deposit. NCBI Gene Expression Omnibus 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    No spike-in scale factor is recoverable for the FLAG arm of GSE260742, so the comparison is made as a within-sample enrichment and a within-sample share, both invariant to an unknown per-sample scale factor.

    What this evidence cannot show

    See ana-q117-gse260742-deposits-no-spike-in-scale-factors-and-152-of-its-158-samples-are-tdf-only for the full check. In this particular comparison the missing scale factor happens not to matter much, because the DMSO and VTP libraries have total coverage within 0.7% of each other; that is luck, not method, and it would not hold for the SGC arm (+11%) or for the RING1B arm (-25%).

    Evidence that argues against this (1)
    • A within-sample share can fall without any eviction if the ChIP simply became less efficient, because a higher background dilutes every real site. There is no way to separate 'the fusion left the DNA' from 'the antibody pulled down proportionally more background' from processed coverage alone. The distinction matters clinically - one is target engagement, the other is assay drift - and needs a spike-in or an orthogonal readout to break. Krivtsov AV et al. 2019
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    The pivotal menin-inhibitor paper reports that VTP-50469 did not globally displace MLL1 from chromatin in MOLM13, RS4;11 and ML2, with locus-restricted losses, and does not report a residual occupancy fraction.

    PMID 31821784DOI
    What this evidence cannot show

    PMID, title, journal, year and DOI fetched from a live PubMed esummary on 2026-09-23. The characterisation of its contents is taken from this graph's existing entry mechanism-vtp-50469-does-not-globally-displace-the-fusion-from-chromatin, whose author read the paper; the paper itself was NOT re-opened in this beat. That is a deliberate and stated limitation: the comparison drawn here is against what this graph records the paper as saying.

    Evidence that argues against this (1)
    • If the pivotal paper does report a residual fraction somewhere in a supplementary table, this entry's framing as 'the number the literature is missing' is wrong even though the measurement stands. The graph's existing entry asserts the absence and this beat did not independently re-check it; rule 80 applies, because a residual fraction is exactly the kind of quantity that lives in a supporting panel rather than in an abstract. Cancer Cell 2019
Connected to