What fills the space: RING1B and BCOR arrive at the fusion's sites under BOTH drugs, but the DOT1L inhibitor puts them everywhere on the fusion's territory while the menin inhibitor puts them only in the quartile where the fusion is most completely evicted - and at Meis1 RING1B rises 16-fold
also called Q117 PRC1 in-fill · displacement-coupled versus displacement-independent PRC1 recruitment
When the fusion protein is knocked off the DNA, a repressive complex moves in behind it. Which drug is used changes the pattern: the DOT1L drug brings the complex in across all the fusion's sites equally, while the menin drug brings it in only at the sites the fusion actually vacates.
The same thing, for a scientist
The depositing paper states that *"following treatment, MLLAF9 target genes acquired strong KDM2B and BCOR occupancy"* and that PRC1.1 deposition is *"independent of MLL-FP displacement"* [E3]. This beat re-derives the in-fill from the deposited tracks with a scale-invariant within-sample share and finds the two drugs behave differently, which the pooled statement conceals [E1]. Splitting the fusion's 1,228 held bins into quartiles by **their own** FLAG VTP/DMSO ratio (Q1 = the fusion most completely lost, median ratio 0.057; Q4 = least lost, 0.345): | share ratio vs DMSO at fusion-held bins | Q1 | Q2 | Q3 | Q4 | |---|---:|---:|---:|---:| | **RING1B, VTP50469** | **1.38** | 0.87 | 0.81 | 0.88 | | **RING1B, SGC0946** | 1.96 | 1.69 | 1.59 | 1.62 | | **BCOR, VTP50469** | **1.44** | 1.31 | 1.33 | 1.27 | | **BCOR, SGC0946** | 1.91 | 1.79 | 1.79 | 1.68 | | **KDM2B, VTP50469** | 1.09 | 1.01 | 0.99 | 0.99 | | **KDM2B, SGC0946** | 1.12 | 1.06 | 1.04 | 1.01 | **Under the DOT1L inhibitor the PRC1 gain is flat across quartiles; under the menin inhibitor RING1B rises only in Q1 and falls slightly everywhere else.** So *independent of displacement* is right for SGC0946 and wrong for VTP50469, where the in-fill is displacement-coupled. **At the locus level the effect is much larger than the summit-bin averages suggest**, because PRC1 spreads over gene bodies while the fusion sits at a summit [E2]. Over gene body minus 5 kb / plus 2 kb, VTP50469 / DMSO share ratios: **RING1B Meis1 16.4x, Jmjd1c 7.5x, Six1 3.6x**; BCOR **Jmjd1c 2.4x, Six1 1.8x, Erbin 1.7x, Tsc22d2 1.5x**. Those are the three loci the paper itself names as its example PRC1.1-rescued genes (*"n = 11; for example, Meis1, Jmjd1c and Six1"*), recovered here independently and without knowing the eleven-gene list [E2][E3]. They are also three of the four loci at which the fusion is most completely lost.
Why the pooled number is small and the locus numbers are large
Pooled over all 1,228 fusion-held 2 kb bins, the VTP in-fill is KDM2B 1.01, BCOR 1.33, RING1B 0.93 - close to nothing. Over gene bodies it reaches 16x. Three reasons, and they are not in conflict:
- The held set is dominated by sites the fusion keeps. The Hoxa cluster contributes five of the top twenty held bins and is the best-retained locus in the table. Averaging retained and vacated sites together dilutes the in-fill to nothing.
- The unit is wrong for the biology. A fusion summit is 2 kb; H2AK119ub and the PRC1 complexes that write it spread across a gene body, which is what the paper's own phrase "selective induction or spreading of H2AK119ub across regions previously bound by MLLAF9" describes. A summit bin is the one place spreading is least visible.
- Directional stratification recovers it. Q1 alone - the 307 bins where the fusion falls to 5.7% of its DMSO share - is the only quartile where RING1B rises under VTP.
The transferable point is that a pooled test over a set defined by baseline occupancy will miss an effect that is conditional on change in that occupancy. The pre-registered secondary test in this beat did exactly that and returned a null; the stratified version, added as a recorded deviation, did not.
The two drugs are not interchangeable and the difference is mechanistic
Both drugs strip the fusion from its own sites - VTP50469 to a share ratio of 0.179 and SGC0946 to 0.298 - so displacement alone cannot explain why SGC0946's PRC1 in-fill is uniform and VTP50469's is confined to the most-displaced quartile. The paper's model supplies the explanation: H3K79me2 is the direct antagonist of PRC1 ubiquitin-ligase activity. SGC0946 removes H3K79me2 everywhere the mark exists, which licenses PRC1 across the whole fusion territory whether or not the fusion left; VTP50469 removes the fusion, and H3K79me2 then decays only where transcription actually stops, so PRC1 arrives only behind a complete eviction.
That makes a prediction this deposit can test and this beat did not: under VTP50469, the H3K79me2 loss should itself be quartile-graded and should track the RING1B gain, while under SGC0946 it should be flat. The H3K79me2 samples are in the series (GSM8123382-8123384 and the rep2 set) and were not downloaded.
| script | analyses/q117-gained-fusion-occupancy/run117.py + locus117.py |
|---|---|
| status | run |
| result summary | PRC1 in-fill at fusion-held sites is displacement-coupled under VTP50469 (RING1B 1.38 in the most-depleted quartile, 0.81-0.88 elsewhere) and displacement-independent under SGC0946 (1.59-1.96 in all quartiles). Locus-level RING1B gains: Meis1 16.4x, Jmjd1c 7.5x, Six1 3.6x. |
| statistic | within-sample share of each track's mappable non-blacklisted coverage; ratio treated/DMSO |
| stratification | the 1,228 fusion-held bins split into quartiles by their own FLAG VTP/DMSO share ratio; Q1 median 0.057, Q4 median 0.345, 307 bins each |
| epigenetic readout | H2AK119ub is the PRC1.1 product; this entry measures the ligase machinery (RING1B/RNF2, BCOR, KDM2B) rather than the mark, because the mark's arm in this series is the one panel with a spike-in and the machinery's arms are not |
| replicates | n = 1 per antibody per condition |
| shown for fusion | KMT2A::MLLT3 (MLLAF9) in a murine myeloid line, FLAG-tagged; nothing in human KMT2A::AFF1 |
inputs
[ "GSM8123392", "GSM8123393", "GSM8123394", "GSM8549824", "GSM8549825", "GSM8549826", "GSM8549827", "GSM8549828", "GSM8549829", "GSM8123359", "GSM8123361", "GSM8549823" ]
quartile table
{
"RING1B_VTP": [
1.381,
0.871,
0.807,
0.879
],
"RING1B_SGC": [
1.956,
1.687,
1.585,
1.624
],
"BCOR_VTP": [
1.443,
1.314,
1.33,
1.27
],
"BCOR_SGC": [
1.911,
1.793,
1.79,
1.68
],
"KDM2B_VTP": [
1.087,
1.007,
0.986,
0.987
],
"KDM2B_SGC": [
1.117,
1.063,
1.037,
1.009
]
}pooled over all held bins VTP — the numbers behind the chart
{
"KDM2B": 1.009,
"BCOR": 1.327,
"RING1B": 0.935
}flank plus minus 10kb VTP — the numbers behind the chart
{
"KDM2B": 0.962,
"BCOR": 1.134,
"RING1B": 1.21
}locus level VTP over DMSO — the numbers behind the chart
{
"RING1B": {
"Meis1": 16.42,
"Jmjd1c": 7.544,
"Six1": 3.637,
"Erbin": 1.689,
"Hoxa10": 1.571,
"Epha7": 1.545
},
"BCOR": {
"Jmjd1c": 2.407,
"Six1": 1.803,
"Erbin": 1.664,
"Tsc22d2": 1.526,
"Mbnl1": 1.436,
"Meis1": 1.413,
"Cdk6": 1.403
},
"KDM2B": {
"Six1": 1.869,
"Hoxa10": 1.31,
"Hoxa9": 1.247,
"Jmjd1c": 1.134
}
}- If PRC1.1 in-fill is what converts menin-inhibitor-driven eviction into durable silencing, then a patient whose blasts carry a BCOR or PCGF1 loss-of-function lesion - both recurrent in myeloid disease - would be predicted to evict the fusion normally and re-express the programme on withdrawal. That is a resistance mechanism invisible to any assay that measures fusion occupancy alone.
- The same logic makes combined DOT1L plus menin inhibition attractive and also identifies its ceiling: if DOT1L inhibition already licenses PRC1 everywhere on the fusion's territory, adding a menin inhibitor may add eviction without adding silencing, and the combination's benefit would then be smaller than the sum of the parts.
- A displacement-coupled in-fill means the silencing endpoint and the target-engagement endpoint are not independent readouts. Reporting both as if they were would double-count the same event in a pharmacodynamic package.
- Is the H3K79me2 loss under VTP50469 itself quartile-graded, and does it track the RING1B gain? The samples exist in this series (GSM8123382-8123384 plus a rep2 set) and were not downloaded. This is the single cheapest test of the mechanism proposed here.
- Does the displacement-coupled versus displacement-independent distinction hold in a human KMT2A::AFF1 line? Nothing in this beat is human and nothing is B-lineage.
- RING1B falls slightly (0.81-0.88) at the three quartiles where the fusion is only partly lost under VTP50469. Is that a real redistribution of PRC1 away from partly-occupied sites, or the reciprocal of the Q1 gain in a share statistic that must sum to one?
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
At the fusion's held bins stratified by how much fusion was lost, RING1B's within-sample share ratio under VTP50469 is 1.38 in the most-depleted quartile and 0.81-0.88 in the other three, while under SGC0946 it is 1.59-1.96 in all four; BCOR rises modestly in all quartiles under both drugs and KDM2B is essentially flat.
NCBI GEO GSE260742, GSM8549824/8549825/8549826 (RING1B DMSO/VTP/SGC), GSM8549827/8549828/8549829 (BCOR DMSO/VTP/SGC), GSM8123392/8123393/8123394 (KDM2B DMSO/SGC/VTP), re-binned to 2 kb from the deposited .tdf · NCBI Gene Expression Omnibus 2026What this evidence cannot show
n = 1 per antibody per condition; no spread is reportable. The quartiles are defined by the FLAG ratio measured in the same single sample pair, so a bin's quartile assignment carries that sample pair's noise, and regression to the mean will place genuinely noisy bins in Q1 preferentially. The RING1B SGC library is 29% larger in total coverage than the RING1B DMSO library (2.06e9 vs 1.60e9 coverage-bases), which the share statistic handles but which also means the two libraries differ enough that a depth-normalised fold change would have given a materially different answer. The RING1B, BCOR and input tracks come from a second processing batch that is NOT duplicate-filtered, unlike the FLAG and KDM2B tracks; the ENCODE mm10 blacklist removes the resulting pileups but cannot be assumed to remove all of them.
Evidence that argues against this (1)
- The stratified analysis is a deviation from the pre-registration, added after the pooled test returned a null. That is the textbook shape of a post-hoc rescue and it is recorded as such in PREREGISTRATION.md DEVIATIONS. The justification offered - that the pooled set mixes retained and vacated sites - is sound in principle and was not stated in advance, so the quartile result should be treated as hypothesis-generating until it is repeated on a second dataset with the stratification declared beforehand. Iqbal Lab oncofusion knowledge graph 2026
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
Over gene-body windows, RING1B's within-sample share rises 16.4-fold at Meis1, 7.5-fold at Jmjd1c and 3.6-fold at Six1 under VTP50469, and BCOR rises 2.4-fold at Jmjd1c and 1.8-fold at Six1 - the same three loci the depositing paper names as example PRC1.1-rescued genes.
NCBI GEO GSE260742 locus table over UCSC mm10 refGene windows, analyses/q117-gained-fusion-occupancy/locus-table.json · NCBI Gene Expression Omnibus 2026What this evidence cannot show
A 16-fold ratio computed from n = 1 with a DMSO denominator that is small is the least stable kind of number in this beat. Meis1's RING1B DMSO share is low, which is exactly why the ratio is large; the absolute post-treatment share is a more defensible quantity and is in locus-table.json. Coordinates from the live UCSC mm10 refGene track, matched on posName. No peak calling and no input subtraction were performed.
Evidence that argues against this (1)
- Recovering the paper's three named example genes is a weaker confirmation than it looks, because those three are also among the most-studied MLL-fusion targets and would be near the top of almost any ranking of leukaemia-relevant loci. The confirmation that carries information is that they are the loci where the FUSION falls furthest in the same data, which is a relationship the paper does not report. Nature Cell Biology 2026
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The depositing paper reports that MLLAF9 target genes acquired strong KDM2B and BCOR occupancy following treatment, with selective induction or spreading of H2AK119ub across regions previously bound by MLLAF9, and concludes that PRC1.1 deposition is independent of MLL-FP displacement.
DOT1L provides transcriptional memory through PRC1.1 antagonism. · Neville D, Ferguson DT, Heikamp EB, et al. · Nature Cell Biology 2026What this evidence cannot show
Quoted from the retrieved fullTextXML (PMC12904788) on 2026-09-23. The sentence says 'following treatment' without separating SGC0946 from VTP50469; that is the ambiguity this entry resolves, and it is possible the authors' figures do separate them in a panel whose legend text does not say so. Figures were not viewed as images.
Evidence that argues against this (1)
- If the authors' Fig. 4e-g already shows the two drugs separately and shows the same quartile gradient, this entry is a re-derivation rather than a refinement. The prose does not say, and the honest position is that the PROSE pools the two drugs and this analysis does not. Nature Cell Biology 2026
measures RING1B, BCOR and KDM2B occupancy at fusion-held sites under both drugs
RING1B/RNF2 share rises 1.38x in the most fusion-depleted quartile under VTP50469 and 1.59-1.96x in every quartile under SGC0946
- related to →BCOR — the other side of the CBX8 conflict, and a target you would be inhibiting in the direction of disease
BCOR share rises 1.27-1.44x under VTP50469 and 1.68-1.91x under SGC0946 at fusion-held sites
the DOT1L inhibitor SGC0946 produces a displacement-independent PRC1 in-fill, consistent with H3K79me2 being the direct antagonist
the ligase machinery measured here writes H2AK119ub; the mark itself was not downloaded in this beat
the quartiles are defined by the per-bin fusion loss that entry measures
KDM2B flat while BCOR and RING1B rise is the pattern that distinguishes assembly-on-a-resident-address from recruitment
the BCOR and RING1B tracks are the unfiltered batch; the in-fill ratios in that entry are computed after removing the blacklist
the local-substitution half rests on the stratified PRC1 in-fill
the control-genotype half of clause 2 is already measured here (RING1B 1.38x in the most-evicted quartile against 0.81-0.88x elsewhere); the knockout half is the twelve-download H2AK119ub arm this entry prices
supplies the knockout half of a measurement whose control half is already here: Q117's RING1B 1.38x in the most-evicted quartile against 0.81-0.88x elsewhere is the comparator, and criteria H1 and H3 are written against it
Q117 measured RING1B occupancy across these quartiles; this measures the mark itself