The active chromatin landscape does not lose a single peak call within 24 hours of the fusion being destroyed: H3K79me2 holds 99.7% at both 3 h and 24 h, H3K9ac 100.0% at both, H3K27ac 83.3%/67.2%, H3K4me3 99.6%/78.9% - while the authors, measuring signal at fusion target genes, report a marked H3K79me2 reduction by 24 h
also called the marks do not move · H3K79me2 at 99.7% twice · K3 fires on three of four marks · the sharpest instrument contrast in the beat
We looked at four chemical tags on DNA packaging one day after the cancer protein was destroyed and found that essentially none of their detectable locations had disappeared, while the original authors - looking only at the specific genes the protein controls, and measuring signal strength rather than presence - did see a clear drop. Both can be true, and the gap between them is the most instructive number in this analysis.
The same thing, for a scientist
**Pre-registered branches K1/K2/K3, top-decile summit retention against each mark's own DMSO:** | mark | 3 h | 24 h | 4 d | pre-registered branch | |---|---:|---:|---:|---| | H3K79me2 | **99.7%** | **99.7%** | — | **K3 — does not move** | | H3K9ac | **100.0%** | **100.0%** | — | **K3 — does not move** | | H3K27ac | 83.3% | 67.2% | 99.9% (different batch) | **K3** | | H3K4me3 | 99.6% | 78.9% | — | **K1 — falls 20.7 points** | **Three of four marks fire K3.** Only H3K4me3 crosses the pre-registered 20-point threshold for a downstream kinetic effect, and it does so by 0.7 points. **THE CONTRAST WITH THE AUTHORS IS THE RESULT.** Olsen *et al.* report that *"while we saw DOT1L loss at MLL::AF9 target genes after only 3 hours of dTAG-VHL treatment, we did not see major changes in H3K79me2 at this early timepoint. However, after 24 hours of dTAG-VHL treatment there was a marked reduction of H3K79me2 at the MLL::AF9 target genes, again with more prominent loss at the 'highly sensitive genes'"* [E1]. So at 24 h: **their target-restricted signal measure reports marked reduction; a genome-wide peak-presence measure on the same libraries reports 99.7% retention — a difference of 0.0 percentage points from the 3 h timepoint.** This is the widest gap between the two instruments anywhere in this beat, and it is a clean illustration of the rule in `concept-q139-a-target-restricted-signal-measure-and-a-genome-wide-peak-measure-are-different-instruments`: **a heavily deposited, slow-turnover mark covering large domains genome-wide is the single worst case for a presence measure**, because the marked loci are a vanishing fraction of the calls. **The ordering the two instruments DO agree on** is the paper's central claim, and this analysis independently supports it: the **protein layer moves first** (HA to 2.5-7.1% at 3 h) and the **mark layer does not follow within 24 h** at the peak-call level. That is the title claim — *"subsequent loss of an active chromatin landscape"* — and *subsequent* is doing real work.
Why H3K79me2's immobility is expected and still worth recording
H3K79 methylation has no known demethylase and is diluted mainly by replication, so its turnover is slow. A mark that cannot be actively removed should be the last thing to change after its writer's tether is destroyed, and the DOT1L result in this beat is consistent: DOT1L is the most-displaced protein in the panel (68.7% retained, loss/HA 0.33) while its mark is the least-moved (99.7%).
Writer displacement and mark persistence are separable, and this deposit separates them cleanly within one experiment. This graph already holds mechanism-q120-dot1l-eviction-and-h3k79me2-persistence-are-separable-at-three-days from a drug experiment; the degron gives the same dissociation under the cleaner perturbation of removing the fusion entirely, at 3 h and 24 h rather than 3 days.
What not to conclude
Do not conclude that the marks are fusion-independent. The authors' own measurement at the loci where the fusion sits shows H3K79me2 falling markedly by 24 h. The genome-wide null is a statement about the genome, not about the target genes, and treating it as the latter would invert the biology.
Do not conclude that H3K4me3's 20.7-point drop is a finding. It clears the pre-registered threshold by 0.7 points, on n = 1 per arm, with a measured technical floor of 4.6 points from the only replicated comparison in the deposit. It is at the edge of what this design can resolve and is reported as the branch firing, not as an effect.
Do not compare H3K27ac's 4 d figure to its 3 h and 24 h figures. The 4 d arm is from a different sequencing batch (201007, hg38-mapped) than the 3 h/24 h arms (190814, not hg38-mapped). Constraint U5 forbids it.
| script | analyses/what-leaves-when-the-fusion-is-degraded/run.py |
|---|---|
| status | run |
| result summary | Top-decile summit retention: H3K79me2 99.7%/99.7% (3 h/24 h), H3K9ac 100.0%/100.0%, H3K27ac 83.3%/67.2%, H3K4me3 99.6%/78.9%. Pre-registered K3 fires on three of four marks; only H3K4me3 reaches K1 and only by 0.7 points over the threshold. The authors' target-restricted measure reports marked H3K79me2 loss by 24 h where this measure reports none. |
| preregistration | analyses/what-leaves-when-the-fusion-is-degraded/PREREGISTRATION.md |
inputs
[ "GSE173599 nine histone-mark DMSO/dVHL pairs across 3 h, 24 h and 4 d" ]
- If the active chromatin landscape genuinely does not change genome-wide within a day of removing the oncoprotein, then any pharmacodynamic assay reading a bulk histone mark will show nothing while the drug is working. This project has repeatedly been drawn to mark-based readouts, and this is a direct argument that they are the wrong biomarker for acute target engagement.
- Conversely, a mark that does not move is a mark that does not need to move for the cell to stop transcribing. The transcriptional effect precedes the chromatin effect by the authors' own account, so a therapeutic strategy premised on resetting the epigenome may be aiming at the consequence rather than the cause.
- Restricted to the fusion's own sites - definable from the deposited HA peak table in the same cells - does H3K79me2 retention fall between 3 h and 24 h? This is the one analysis that would convert a genome-wide null into a test of the authors' claim, and it needs no new data.
- Why does H3K27ac show the largest movement of the four marks (83.3% to 67.2%) when it is also the mark the authors report moving fastest by ChIP-qPCR? If the presence measure has any target-gene sensitivity at all, H3K27ac is where it would show, and that is worth checking rather than assuming.
- Is there any mark that DOES lose peak calls genome-wide after fusion degradation? Only four were profiled. A repressive mark - H3K27me3 or H2AK119ub - was not, and this project has repeatedly found that the polycomb side carries the signal that the active side does not.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
H3K79me2 shows no major change at 3 h after MLL::AF9 degradation but a marked reduction at MLL::AF9 target genes by 24 h, and H3K27ac shows roughly 50% reduction by ChIP-qPCR after 3 h, establishing that chromatin changes follow rather than precede the transcriptional ones.
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 2022What this evidence cannot show
Read in full text at PMC9044330. The authors' H3K79me2 result is a target-gene-restricted, normalised read-count comparison; 'marked reduction' is their wording for a boxplot shift, and no effect size is given in the text. Their H3K27ac kinetics come from ChIP-qPCR at selected loci. Neither is genome-wide peak presence, so the contrast drawn in this entry is between instruments and is not a disagreement about the underlying data.
Evidence that argues against this (1)
- This is the direct counter-reading of the genome-wide null: at the loci the fusion occupies, H3K79me2 does fall by 24 h. Anyone taking '99.7% at both timepoints' as evidence that the mark is unaffected by fusion loss will be wrong about the biology that matters. Olsen SN et al. 2022
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
Paired DMSO/dTAG-VHL peak tables at both 3 h and 24 h are deposited for H3K4me3, H3K27ac, H3K79me2 and H3K9ac, and at 4 d for H3K27ac.
GSE173599, samples GSM5271659/61 and GSM5271658/60 (H3K4me3 3 h/24 h), GSM5271695/97 and GSM5271694/96 (H3K27ac 3 h/24 h), GSM5271699/701 and GSM5271698/700 (H3K79me2 3 h/24 h), GSM5271703/05 and GSM5271702/04 (H3K9ac 3 h/24 h), GSM5271653/54 (H3K27ac 4 d) · Olsen SN, Armstrong SA, et al. · NCBI Gene Expression Omnibus 2022What this evidence cannot show
n = 1 per arm, no replicates, no spike-in. A near-100% retention figure is the least informative reading a presence measure can return, because it is also what a measure with no sensitivity would return - the HA channel in the same deposit is the only thing establishing that the measure CAN detect a change. H3K4me3's 24 h arm triggers the pre-registered X-LIVE depth-confound rule (shallower dVHL library calling fewer peaks); for a retention finding that biases retention downward, so 78.9% is a lower bound.
Evidence that argues against this (1)
- A retention figure of 100.0% for H3K9ac at both timepoints is suspicious on its face and should be read as an instrument statement as much as a biological one: with ~19,000-30,000 broad calls per arm, a presence measure has almost no room to register loss. Oncofusion knowledge graph 2026
SAME DISSOCIATION UNDER A CLEANER PERTURBATION: that entry separates DOT1L eviction from H3K79me2 persistence under a drug at three days; here removing the fusion entirely gives DOT1L as the most-displaced protein (68.7%) and H3K79me2 as the least-moved mark (99.7% at both 3 h and 24 h).
A near-100% retention figure is only informative because the HA channel in the same deposit shows the measure CAN detect a collapse. Without the positive control these marks' immobility would be indistinguishable from an insensitive instrument.
The widest instrument gap in the beat: the authors report a marked H3K79me2 reduction at target genes by 24 h while genome-wide peak presence moves 0.0 points between 3 h and 24 h. A broad slow-turnover mark is the worst case for a presence measure.
The chromatin half of the two clocks: no active mark loses a genome-wide peak call by 24 h, against nascent transcription moving at 15 minutes - a latency gap of roughly two orders of magnitude.