Oncofusion

start / graph / ana-q139-the-degron-clears-the-fusion-from-chromatin-and-two-independent-batches-agree

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q139-fusion-degron

The positive control this corpus has never had: degrading MLL::AF9 clears its own tag from 93-97% of its strongest chromatin sites within 3 hours, and two INDEPENDENT sequencing batches agree to within 4.6 points - the first technical noise floor ever measured for a ChIP eviction comparison in this project

also called the HA degron control · 2.5% and 7.1% · the technical floor is 4.6 points · does the degron work

Every claim in this project about a drug knocking a cancer protein off DNA has been measured against another drug, never against simply deleting the protein - so we checked a public dataset that did the deletion, and three hours after the protein is destroyed its own molecular label is gone from about 95% of the places it used to sit. The same experiment was run twice independently and the two runs agree closely, which is the first time anyone here has been able to say how much of such a number is real and how much is noise.

The same thing, for a scientist

**The experiment.** HCB1 cells (human CD34+ cord blood transformed by **MLL-AF9-HA-FKBP12**) treated with **dTAG-VHL 500 nM**, with paired DMSO/dVHL **HA ChIP** on either side [E1, E2]. HA tags the fusion itself, so this is the one arm in the deposit whose expected answer is known in advance. **Result, scored as summit containment against the arm's own DMSO peak set (top decile by -log10 p):** | arm | batch | DMSO peaks | dVHL peaks | top-decile summits retained | |---|---|---:|---:|---:| | HA 3 h | 1909x2 | 12,017 | 997 | **2.5%** | | HA 3 h | 190814 | 5,049 | 1,290 | **7.1%** | | HA 24 h | 190814 | 2,240 | 1,193 | 15.6% | | HA 4 d | 200908 | 11,044 | 320 | **1.0%** | | HA 4 d | 1912x2 | 3,204 | 929 | 10.3% | **Five of five arms clear the pre-registered 33% threshold. The degron works, and the beat that rests on it is licensed.** **THE PART THAT IS WORTH MORE THAN THE CONTROL ITSELF.** The 3 h experiment exists **twice, in two independent sequencing batches with different barcodes and different library depths** (14.79 M vs 16.20 M fragments). They give **2.5% and 7.1% — a 4.6-point spread.** That spread is a **technical noise floor**, and it is the only one measurable anywhere in this deposit. This project has stated for many cycles that its floors span roughly 20x across arms and that none may be imported; here is one that was actually measured, on the strongest-possible-effect channel, and it is small. **Read it as a floor and not a licence.** 4.6 points is the reproducibility of a near-total signal collapse. It does not bound the reproducibility of a 20% change, which is where most of this corpus's contested numbers sit, and it is technical (two libraries from the same biology) rather than biological (two independent transductions).

Go deeper
Why this is the reference point the whole eviction thread lacked

Every eviction claim in this corpus is stated relative to a drug — a menin inhibitor moves X% of sites, a DOT1L inhibitor moves Y%. None was stated relative to the fusion simply being absent.

That matters most for the specificity question the corpus cannot otherwise reach: a menin inhibitor could displace menin by occupying its pocket rather than by dismantling the complex, and no menin-inhibitor experiment can distinguish those. Removing the fusion tests whether a component's residence depends on the fusion being present at all. This entry establishes that the removal worked; the sibling entries score what followed.

The 4 d arms, and a wrong reading of mine that the metric produced

Under the 1 kb bin-overlap measure this project normally uses, the two independent 4 d experiments read 0.9% and 90.2% — which I first recorded as a qualitative between-batch contradiction and as the strongest argument against reading any n=1 ChIP arm as a measurement.

That reading was wrong and it was the metric talking. Under summit containment the same two experiments read 1.0% and 10.3% — both cleared, in agreement. The contradiction was manufactured by the measure. It is recorded here rather than deleted because the wrong intermediate reading is itself the evidence that the measure is broken (see ana-q139-the-one-kb-bin-overlap-retention-measure-fails-its-own-positive-control).

What HA can and cannot tell you

HA is an epitope on the transgene. It reads the tagged fusion and nothing else — not endogenous wild-type MLL1, not an untagged allele. And a ChIP signal disappearing is consistent with the protein being degraded or with the epitope being masked or the chromatin being inaccessible. The dTAG system is designed to degrade and the authors show protein loss by immunoblot 1, so degradation is the parsimonious reading, but the ChIP alone does not establish it.

Facts and figures
scriptanalyses/what-leaves-when-the-fusion-is-degraded/run.py
statusrun
result summaryHA top-decile summit retention 2.5% / 7.1% at 3 h in two independent batches, 15.6% at 24 h, 1.0% / 10.3% at 4 d. Pre-registered H-PASS in 5 of 5 arms. Between-batch spread at 3 h = 4.6 points, the only measured technical floor in the deposit.
preregistrationanalyses/what-leaves-when-the-fusion-is-degraded/PREREGISTRATION.md, written before any peak table was opened
inputs
[
  "GSE173599 GSM5271662/GSM5271665 (HA 3 h batch 1909x2)",
  "GSE173599 GSM5271707/GSM5271709 (HA 3 h batch 190814)",
  "GSE173599 GSM5271706/GSM5271708 (HA 24 h)",
  "GSE173599 GSM5271664/GSM5271667 (HA 4 d batch 200908)",
  "GSE173599 GSM5271663/GSM5271666 (HA 4 d batch 1912x2)"
]
How this could bite us back
  • A degron is not a drug. Nothing about how completely dTAG-VHL removes the fusion predicts how completely any small molecule will, and this entry should not be read as setting an achievable ceiling for a therapeutic.
  • If the residual 2.5-7.1% of sites are the therapeutically important ones - the highest-occupancy, most transcriptionally consequential loci - then even total protein degradation leaves the drivers engaged, and a 95% figure is the wrong statistic. Which sites survive has not been characterised here.
Still open
  1. The measured 4.6-point technical floor is for a near-total signal collapse. What is the floor for a 20-40% change, which is where this corpus's contested eviction numbers actually sit? Nothing in any public deposit answers this, because it needs replicate ChIP at an intermediate effect size.
  2. Is the 4 d arm's difference between batches (1.0% vs 10.3%) biological drift in the transduced population over four days, or library preparation? A third batch would separate them.
  3. Does an HA ChIP signal loss track the immunoblot loss quantitatively, or does chromatin-bound protein have a different half-life from the bulk pool?
Every source behind this entry (2)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    HCB1 cells carry MLL-AF9-HA-FKBP12 and dTAG-VHL 500 nM degrades the fusion, with ChIP-seq before and after at 3 h, 24 h and 4 d.

    MLL::AF9 degradation induces rapid changes in transcriptional elongation and subsequent loss of an active chromatin landscape. · Olsen SN, Godfrey L, Healy JP, Choi YA, Kai Y, Hatton C, Perner F, Haarer EL, Nabet B, Yuan GC, Armstrong SA · Molecular Cell 2022
    PMID 35245435DOI
    What this evidence cannot show

    n = 1 library per arm except HA, which has two independent batches at 3 h and two at 4 d. No spike-in, so nothing is on an absolute scale. HCB1 is KMT2A::MLLT3, not the index fusion KMT2A::AFF1; this project has established that between MLLT3 and AFF1 lines per-gene ordering transfers but magnitude does not, so every figure here is partner-imported. The published title differs from the GEO series title, which is what the dispatch brief quoted.

    Evidence that argues against this (1)
    • The authors' own orthogonal measurement is ChIP-qPCR at selected loci and normalised read-count fold changes at a defined MLL::AF9 target-gene set, not genome-wide peak-call presence. Their Figure 4A/4B numbers are therefore not directly comparable to the retention figures here, and where they differ the instrument difference is the first explanation to test. Olsen SN et al. 2022
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset

    The MACS2 peak tables for every HA, MEN1, DOT1L, AFF4, CDK9, RNA Pol II and histone-mark arm are deposited in GEO under GSE173599 and were re-verified sample by sample against the live series file list.

    What this evidence cannot show

    Peak tables only; the 23 GB of TDF coverage tracks was not downloaded. A MACS peak call is a threshold crossing, not an occupancy measurement, so a site can lose most of its signal and keep its call. Arms come from seven sequencing batches and some are hg38-mapped while others are not; within any DMSO/dVHL pair the batch and genome build are identical, but cross-mark ranking crosses batches.

    Evidence that argues against this (1)
    • The GEO series overall_design describes 4-day DOT1L and MENIN ChIP in HCB1, but no such peak table is deposited - only HA and H3K27ac have 4-day drug arms. The deposit's description is broader than its contents. NCBI Gene Expression Omnibus 2022
Connected to