start / graph / concept-q139-a-target-restricted-signal-measure-and-a-genome-wide-peak-measure-are-different-instruments

ExplainerstrongQ139-fusion-degron

"Menin was lost" and "menin did not move" are both true of the same experiment: a target-gene-restricted signal measure and a genome-wide peak-presence measure answer different questions, and the majority of the genome that does not move is exactly what the second one is made of

also called the two instruments · 50% lost and 80% retained are both right · restricted signal versus global presence · which denominator

Two honest analyses of one experiment can say opposite-sounding things: one asks how much signal was lost at the handful of genes the cancer protein controls, the other asks whether detectable signal is still present anywhere in the genome. Because most of the genome is unaffected the second answer is dominated by places nothing happened, so neither is wrong - they are different instruments pointed at different questions.

The same thing, for a scientist

**The worked case.** In GSE173599, three hours after MLL::AF9 degradation: - Olsen *et al.*, using **normalised read-count fold change restricted to a pre-defined MLL::AF9 target-gene set** (TSS −1kb/+3kb or gene bodies), report *"more than 50% of MENIN and almost 70% of DOT1L was lost at MLL::AF9 target genes, **while there were no changes in occupancy at non-MLL::AF9-bound genes**"* [E1]. - This project, using **genome-wide peak-call presence against each channel's own DMSO peak set**, finds menin retaining **80.0%** and DOT1L **68.7%** of their strongest calls. **Both are correct, and the authors' own clause explains why.** If the non-target genome does not move, and the non-target genome is where nearly all of a factor's peak calls live, then a genome-wide presence measure will be dominated by unchanged sites **by construction**. The target-restricted measure deliberately excludes them. **Three axes separate the instruments, and they are independent:** | axis | authors | this analysis | |---|---|---| | **denominator** | a curated target-gene set | every peak in the control | | **quantity** | continuous normalised signal | binary threshold crossing | | **normalisation** | housekeeping-TSS or total-read ratio | none; each channel vs its own control | A disagreement can arise from **any one** of the three, so reconciling two ChIP analyses requires naming all three — not just "they used different software".

Go deeper

The rule

Before comparing two occupancy results, write down the denominator, the quantity and the normalisation for each. If any differs, the two numbers are not the same measurement and disagreement between them is not evidence that either is wrong.

And state your denominator in the claim itself. "Menin does not leave chromatin" is unreadable; "menin retains 80% of its genome-wide peak calls, while losing more than half its signal at fusion target genes" is a single sentence that cannot be misread. This graph has published occupancy claims without their denominator and they are the ones that later needed correcting.

The specific trap here is the one this project is most exposed to. A genome-wide presence measure is the cheapest thing to compute from a deposited peak table, and it is the instrument least able to see a target-restricted effect. That is the exact combination that produces a confident, well-controlled, published null about biology that is real.

Which instrument for which question

"Does this factor's chromatin residence depend on the fusion existing?" — genome-wide presence is appropriate, and its answer for menin, DOT1L, AFF4, CDK9 and Pol II is no, mostly not.
"Does this factor leave the loci the fusion occupies?" — target-restricted signal is required, and the genome-wide measure is close to blind to it.
"How much of a factor is at a site?" — neither. Presence is a threshold; restricted fold change is relative to a normalisation that no spike-in anchors here.

The practical bridge, available from the data already in hand, is to partition the genome-wide measure by the control HA peak set and report retention inside and outside fusion-bound regions separately. That gives the target-restricted question a presence-based answer on the same footing as the global one.

How this could bite us back

This concept can be used to dissolve any inconvenient disagreement by asserting the instruments differ. It should only be invoked when the denominator, quantity and normalisation are actually named and actually differ - otherwise it is a way of never having to resolve a contradiction.

Still open

How many occupancy claims in this corpus state their denominator? A field-level audit would show whether 'X% retained' entries name the peak set they are a percentage of, and the suspicion is that most do not.
Does the genome-wide presence measure agree with the authors' restricted signal measure once restricted to the same gene set? This is the direct test of the reconciliation and it has not been run.

Every source behind this entry (1)

E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Olsen et al. report >50% MENIN and ~70% DOT1L signal loss at MLL::AF9 target genes at 3 h with explicitly no change at non-MLL::AF9-bound genes, using normalised read-count fold change over a defined target-gene set.
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 35245435 DOI
What this evidence cannot show
Read in full text at PMC9044330. The 'no changes at non-MLL::AF9-bound genes' clause is load-bearing for this entire reconciliation and is the authors' summary of a boxplot comparison, not a formal test of equivalence - absence of a significant difference in a Mann-Whitney comparison is not proof that nothing moved. If the non-target genome moved slightly, the reconciliation would be weaker than stated here.
Evidence that argues against this (1)
- The reconciliation would fail if the two measures disagreed even WITHIN the target-gene set. That has not been tested: this analysis never partitioned its genome-wide measure by the target set, so the claim that the instruments agree once the denominator is matched is argued, not demonstrated. NCBI Gene Expression Omnibus 2022

Connected to

← depends onMenin does NOT leave chromatin when the fusion is removed: three hours after MLL::AF9 is degraded to 2.5-7.1% of its own chromatin sites, menin still holds 80.0% of its strongest peak calls genome-wide - so menin occupancy is not a readout of fusion occupancy, and a menin inhibitor that displaces menin is doing something fusion loss does not do
THE CLAIM IS ONLY DEFENSIBLE WITH ITS DENOMINATOR: 80% genome-wide retention and the authors' >50% loss at MLL::AF9 target genes are both true, and separating them is what makes the menin result a finding rather than a contradiction of the paper.
← depends onThree hours after the fusion is degraded to 2.5-7.1% of its own sites, NOT ONE of the eight complex components profiled beside it falls below 68% retention genome-wide - menin 80.0%, DOT1L 68.7%, CDK9 83.6%, AFF4 92.7%, RNA Pol II 99.6%, pS2 98.3%, pS5 86.8% - so nothing in the complex is held on chromatin by the fusion at genome scale
Required to reconcile this genome-wide null with the authors' report that AFF4, CDK9 and Pol II ARE specifically lost at MLL::AF9 target genes. Without the denominator distinction this entry would read as contradicting the paper.
related to →Plain-English: chromatin and epigenetics — the sticky notes on the DNA, and which ones matter here
A methodological companion to the chromatin explainer: two honest ChIP analyses of one experiment can report opposite-sounding answers when their denominator, their quantity and their normalisation differ.
supports →A chromatin-occupancy readout has one dimension and the question has three: eviction, degradation and relocation all lower the signal at a target gene
Same family of limitation, one layer up: that entry is about what an occupancy measurement cannot distinguish; this is about two occupancy measurements of the same data not being the same quantity at all.
← depends onThe 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
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.
← supportsMenin loses half its promoter signal where the fusion sat, and nothing where it did not
quantifies the claimed instrument difference: 47.1 points between strata on the signal measure
← supportsA binary peak call cannot see a two-fold change, so 'the peaks are still there' is not 'the protein is still there'
supplies the mechanism: the peak instrument saturates at strong sites and cannot resolve a 2-fold change