Oncofusion

start / graph / exp-q117-spike-in-normalised-tagged-fusion-chip-under-a-menin-inhibitor-in-a-human-kmt2a-aff1-line

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

The experiment this beat could not do: an epitope-tagged KMT2A::AFF1 ChIP with a Drosophila spike-in, in a human infant-derived B-lineage line, at three replicates and two menin-inhibitor exposures - the design that would turn a within-sample ranking into an absolute occupancy measurement

also called Q117 proposed spike-in fusion ChIP · the human index-fusion version of the gain call

Do the same experiment in the human infant leukaemia the lab actually studies, with a measuring stick added to every sample so that losses and gains can be compared properly, and with three repeats so the numbers have error bars.

The same thing, for a scientist

Every quantitative statement this beat produced is a **within-sample** statement, because GSE260742 deposits no spike-in and no scale factors, and every one is in **murine MLLAF9** with **n = 1** per condition. Three deficits, one design. **Design.** A human KMT2A::AFF1 line - ideally an infant-derived one, otherwise SEM or RS4;11 - carrying an endogenously **knock-in-tagged** fusion allele (3xFLAG or dTAG at the KMT2A N-terminus, so the tag marks both wild-type KMT2A and the fusion, plus a second line tagged at the AFF1 C-terminus so the fusion alone is marked). Treat with **revumenib or VTP-50469 at two exposures** (a sub-saturating and a saturating dose) and DMSO, **three biological replicates each**, 48 h. **Add Drosophila S2 chromatin at a fixed ratio to every ChIP before immunoprecipitation** and a fixed amount of Drosophila-specific antibody, i.e. a standard ChIP-Rx. Sequence ChIP and matched input for every sample. **Deposit the scale factors and the spike-in read counts, not only the tracks.** **Primary readout.** Absolute occupancy at the fusion's DMSO peak set, expressed in spike-in-normalised units, plus a de novo peak call in each arm against its own input and a count of treated-only peaks with a replicate-based FDR. **Secondary.** KDM2B, BCOR, RING1B and H2AK119ub in the same cells and arms, so the in-fill can be measured on the same scale. **What it settles that this beat could not.** (1) Whether the ~18% residual fraction is an absolute residual or an artefact of a share statistic. (2) Whether the 52 low-amplitude gained bins are reproducible across replicates - with n = 3 this becomes a testable question rather than an impression. (3) Whether the index fusion behaves like MLLAF9 at all: the whole of this beat is one partner, one species, one lineage. (4) Whether fusion protein abundance changes, via the tag, which decides whether 'eviction' is displacement or depletion. **Cost tier: medium.** Two knock-in lines, 18 ChIP-seq libraries for the fusion arm plus 54 for the secondary readouts, and no animal work. The tagging is the long pole at roughly three months.

Go deeper
The controls that are not optional
  • A spike-in that is added before the IP, not after. A spike-in added to the library rather than to the chromatin controls for sequencing depth and nothing else, which is what the deposit already gives for free.
  • An untreated time-matched arm at each timepoint, because DMSO for 48 h in a proliferating line is not a no-op.
  • An immunoblot of the tagged fusion at every timepoint, to separate displacement from depletion. This is the cheapest control in the design and its absence is the largest single caveat on this beat's conclusions.
  • A second tag position. An N-terminal KMT2A tag marks wild-type KMT2A and the fusion together, which is the situation every published MLL1n experiment is in; a C-terminal AFF1 tag marks the fusion alone. Running both in the same experiment measures the size of the attribution problem directly, which nobody has done.
  • Report the spike-in fraction per sample. If the fraction shifts with treatment-induced cell death, apparent gains and losses both scale and the result will look like a mechanism.
The cheaper things to do first

Three of them, all compute:

  1. Re-align GSE260742's FLAG arm from SRA (SRX23818722/23/24) with the two deposited input BAMs, and call peaks properly. This tests the 52-bin gain set against a real peak caller with a matched input and costs one node-day.
  2. Run the identical within-sample analysis on GSE127508, which contains MLL1n ChIP plus and minus VTP-50469 in human RS4;11 (index fusion), MOLM13 and ML2 (fusion-only). That gives the human comparison without any new sequencing, and this graph already carries a written proposal for it.
  3. Download the H3K79me2 arm of GSE260742 and test whether its loss under VTP50469 is quartile-graded in the same way the RING1B gain is. One afternoon.
Facts and figures
designHuman KMT2A::AFF1 line with an endogenously knock-in-tagged fusion (3xFLAG at the KMT2A N-terminus in one line, at the AFF1 C-terminus in a second), treated with a menin inhibitor at a sub-saturating and a saturating exposure plus DMSO, 48 h, three biological replicates, Drosophila S2 chromatin spike-in added to the chromatin before IP, ChIP and matched input sequenced for every sample.
readoutSpike-in-normalised absolute occupancy at the DMSO peak set; de novo peak calls per arm against matched input; count of treated-only peaks at a replicate-based FDR; parallel KDM2B, BCOR, RING1B and H2AK119ub ChIP in the same arms; immunoblot of the tagged fusion at every timepoint.
controlstime-matched DMSO at each timepoint; per-sample spike-in fraction reported; two tag positions to measure the wild-type-MLL1 attribution problem directly; matched input for every ChIP; immunoblot to separate displacement from depletion.
cost tiermed
timelineapproximately 3 months for the knock-in lines, 6 weeks for the ChIP series and sequencing, 2 weeks of analysis
cheaper first steps
[
  "re-align the GSE260742 FLAG arm from SRX23818722/23/24 with the two deposited input BAMs and call peaks properly - one node-day",
  "run this beat's within-sample analysis on GSE127508's human arms - no new sequencing",
  "download the GSE260742 H3K79me2 arm and test whether its loss is quartile-graded like the RING1B gain - one afternoon"
]
tests hypotheses
[
  "hyp-q117-menin-inhibition-evicts-the-fusion-without-redistributing-it-to-a-recognisable-programme"
]
what it settles
[
  "whether the ~18% residual is absolute or an artefact of a share statistic",
  "whether the 52 low-amplitude gained bins reproduce across replicates",
  "whether the index human fusion behaves like murine MLLAF9",
  "whether eviction is displacement or depletion"
]
How this could bite us back
  • A spike-in fixes the scale and not the biology. If treatment-induced cell death changes the ratio of spike-in to sample chromatin, the normalisation introduces its own artefact, and the fix - reporting the spike-in fraction per sample and treating a shifting fraction as a failed experiment - has to be written into the protocol in advance rather than checked afterwards.
  • Two tag positions in one experiment will quantify the attribution problem, and if the answer is that a shared-epitope antibody understates fusion loss by a large factor, a decade of published MLL1n occupancy measurements become harder to read rather than easier.
Still open
  1. Is there an existing endogenously tagged human KMT2A-fusion line in any laboratory? Building one is the long pole of this design and a collaboration would remove it.
  2. Would a degron system be better than a drug for the mechanistic half? Acute fusion degradation separates 'the fusion is gone' from 'the interface is blocked', and the two predict different PRC1 in-fill patterns under the model this beat proposes.
Every source behind this entry (2)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    The deficits this design addresses are present and measured: GSE260742 deposits no spike-in scale factors for any of 158 samples, the FLAG arm has n = 1 per condition, and every sample in the arm is murine MLLAF9 rather than human KMT2A::AFF1.

    What this evidence cannot show

    The proposal assumes an endogenously taggable KMT2A::AFF1 line can be made; knock-in tagging at the KMT2A locus in a leukaemia line is not routine and the fusion allele specifically may be refractory. If it is, the fallback is a dTAG or degron line, which changes the question from occupancy under a drug to occupancy under acute degradation and is a different experiment.

    Evidence that argues against this (1)
    • A tagged fusion is an engineered allele and its expression level, stability and interactions may differ from the endogenous product. This beat's own data are from a FLAG-tagged, exogenously expressed murine fusion, and the same objection applies to them - which is an argument for doing the tagging endogenously, and an argument that no tagged system fully escapes the criticism. Nature Cell Biology 2026
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    A public dataset already exists that would answer the human half of the question by compute alone: GSE127508 contains MLL1n ChIP-seq plus and minus VTP-50469 in RS4;11 (KMT2A::AFF1), MOLM13 and ML2, the last of which has lost its wild-type KMT2A allele.

    PMID 31821784DOI
    What this evidence cannot show

    PMID and DOI fetched from a live PubMed esummary on 2026-09-23. The GSE127508 sample inventory is taken from this graph's existing entry ana-reanalysis-of-ml2-mll1n-chip-for-gained-peaks-under-vtp-50469 and was not re-verified against GEO in this beat. That entry records n = 1 per condition in the ML2 arm and an input for DMSO only, so the public route is underpowered in the same way this beat was.

    Evidence that argues against this (1)
    • If GSE127508's processed files are also browser tracks without scale factors, the human re-analysis inherits exactly the constraint that limited this beat, and the only thing it adds is species and partner. That would still be worth having, but it would not settle the absolute-occupancy question, which needs the new experiment. NCBI Gene Expression Omnibus 2019
Connected to