Tested in an independent deposit with an epitope-tagged fusion: of the pivotal paper's three named losers under a menin inhibitor, MEIS1 and JMJD1C reproduce and MEF2C does not; of its three named retained loci, the HOXA cluster reproduces and MYB does not - and GAPDH, the negative control, loses half its share too
also called Q117 locus table · Meis1 0.117, Jmjd1c 0.167, Hoxa9 0.660, Gapdh 0.508
The paper that launched menin inhibitors listed which genes lose the fusion protein under the drug and which keep it. Re-measuring that list in a different laboratory's public data, two of the three named losers and the famous retained cluster come out the same, two do not, and even the housekeeping control loses half its signal - which is the scale everything else has to be read against.
The same thing, for a scientist
`mechanism-vtp-50469-does-not-globally-displace-the-fusion-from-chromatin` records that Krivtsov et al. report **>2-fold MLL1n loss at MEIS1, MEF2C and JMJD1C** under VTP-50469 and **no loss at the HOXA cluster, MYB and GAPDH**, in human MOLM13, RS4;11 and ML2 with an MLL1n antibody that also sees wild-type MLL1 [E3]. This beat re-measures that list in **GSE260742**, a different laboratory, a **murine** MLLAF9 line, and an **anti-FLAG** ChIP against a FLAG-tagged fusion, so that no read can come from wild-type MLL1 [E1]. The statistic is the fusion's **within-sample share** of its own track over the gene body minus 5 kb / plus 2 kb, expressed as a ratio VTP50469 / DMSO. Because shares sum to one across the genome, **1.0 means 'moved with the genome average', not 'unchanged in absolute terms'** [E2]: | locus | Krivtsov's call | FLAG VTP/DMSO | FLAG SGC/DMSO | verdict here | |---|---|---:|---:|---| | **Meis1** | >2-fold LOSS | **0.117** | 0.232 | **reproduces, strongly** | | **Jmjd1c** | >2-fold LOSS | **0.167** | 0.224 | **reproduces, strongly** | | **Mef2c** | >2-fold LOSS | **0.870** | 1.201 | **does not reproduce** | | **Hoxa9** | no loss | **0.660** | 0.614 | reproduces, at the housekeeping floor | | **Hoxa10** | no loss | **0.561** | 0.554 | reproduces, at the housekeeping floor | | **Myb** | no loss | **0.362** | 0.568 | **does not reproduce** - loses more than the floor | | **Gapdh** | no loss (control) | **0.508** | 0.659 | **the control itself falls by half** | | *Actb* (added here) | - | 0.571 | 0.836 | second housekeeping reference | **The Gapdh and Actb rows are the most useful numbers in the table.** They set the floor: under VTP50469 a gene-proximal window loses roughly half its share of the FLAG track whatever it is, because the evicted signal redistributes into genome-wide background. Read against that floor, **Meis1 at 0.117 and Jmjd1c at 0.167 are three- to four-fold below the floor and are real locus-specific losses, while Hoxa9 at 0.660 is ABOVE the floor and is a genuine relative retention.** The largest locus-specific losses this beat found are not all on the published list: **Tsc22d2 0.062, Erbin 0.076, Epha7 0.178, Arid2 0.186, Pds5a 0.188, Six1 0.199, Nipbl 0.200, Mbnl1 0.248, Cdk6 0.281** [E1].
The three disagreements, and which way each one could be wrong
MEF2C, 0.870. The published call is a >2-fold loss in human lines. Here it barely moves. MEF2C's FLAG DMSO share is one of the lowest in the table (0.77 x 1e-4 against Meis1's 5.57), so this is a weak-signal locus in the mouse line and the ratio is correspondingly unstable; the murine and human MLLAF9 programmes are also not identical. This is the disagreement most likely to be a property of the system rather than of either measurement.
MYB, 0.362. Published as retained; here it loses more than the housekeeping floor. Myb is a documented MLL-fusion target in both species and its DMSO share here (1.36 x 1e-4) is real but modest. If this is right, one of the pivotal paper's three named 'spared' loci is not spared for the fusion itself once wild-type MLL1 is removed from the readout - which is exactly the direction the epitope-tag design predicts, because wild-type MLL1 at a shared promoter would mask fusion loss.
GAPDH, 0.508. Published as a negative control with no loss. Here it falls as much as the average gene-proximal window. This is not a disagreement about GAPDH; it is a statement that a share-based readout has no invariant reference locus, because the redistribution has to land somewhere. Anyone using GAPDH as an internal ChIP control under a menin inhibitor should know that.
What this says about the attribution problem
The recurring objection to every published measurement of fusion occupancy under a menin inhibitor is that the antibody sees wild-type KMT2A too. Krivtsov et al. handled it with ML2, a line that has lost the wild-type allele. This deposit handles it differently and independently: the fusion carries a FLAG tag and the antibody is anti-FLAG, so the attribution is exact by construction rather than by cell-line genotype.
That the two designs agree on MEIS1, JMJD1C and the HOXA cluster is the strongest available evidence that those three calls are about the fusion and not about the antibody. That they disagree on MYB is the most informative row in the table, because MYB is precisely where a shared-epitope antibody would be expected to mask a fusion-specific loss.
| script | analyses/q117-gained-fusion-occupancy/locus117.py |
|---|---|
| status | run |
| result summary | Of Krivtsov's three named losers, Meis1 (0.117) and Jmjd1c (0.167) reproduce and Mef2c (0.870) does not. Of the three named retained, Hoxa9/Hoxa10 (0.660/0.561) reproduce at the housekeeping floor and Myb (0.362) does not. Gapdh, the control, falls to 0.508. |
| statistic | within-sample share of the track's mappable non-blacklisted coverage over gene body -5 kb/+2 kb, ratio VTP50469 / DMSO |
| attribution | anti-FLAG against a FLAG-tagged fusion in mouse - no wild-type MLL1 can contribute a read, which is the design difference from the MLL1n antibody used in the published human experiments |
| shown for fusion | KMT2A::MLLT3 (MLLAF9) in a murine myeloid line, FLAG-tagged; the published comparison is human KMT2A::MLLT3 (MOLM13), KMT2A::AFF1 (RS4;11) and KMT2A::AFDN (ML2) |
inputs
[ "GSM8123359", "GSM8123360", "GSM8123361", "GSM8549823", "UCSC mm10 refGene via api.genome.ucsc.edu" ]
flag vtp over dmso — the numbers behind the chart
{
"Meis1": 0.117,
"Jmjd1c": 0.167,
"Epha7": 0.178,
"Arid2": 0.186,
"Pds5a": 0.188,
"Six1": 0.199,
"Nipbl": 0.2,
"Mbnl1": 0.248,
"Cdk6": 0.281,
"Myb": 0.362,
"Tsc22d2": 0.062,
"Erbin": 0.076,
"Gapdh": 0.508,
"Hoxa10": 0.561,
"Actb": 0.571,
"Ncor1": 0.592,
"Hoxa9": 0.66,
"Mef2c": 0.87,
"Bcl2": 0.905,
"Pbx3": 0.913,
"Flt3": 0.919,
"Eya1": 1.159,
"Zfp521": 1.34,
"Sox4": 0.526
}housekeeping floor
{
"Gapdh": 0.508,
"Actb": 0.571,
"interpretation": "roughly half the share is lost at any gene-proximal window; loci far below this are locus-specific losses and loci at or above it are relative retentions"
}- The HOXA cluster being the best-retained locus is the uncomfortable part of this table for the drug. If HOXA occupancy is what maintains the differentiation block and HOXA is where the fusion stays, then clinical benefit has to be coming from the loci that do lose it, and a patient whose disease depends more on HOXA than on MEIS1 would be predicted to respond less.
- A correlative assay that picks GAPDH as its internal control will report a smaller drug effect than one that picks a genome-wide reference, because GAPDH itself moves. The choice of reference locus is a free parameter that can change the reported potency by a factor of two, and nothing currently requires it to be declared.
- Is the MYB disagreement an attribution effect? The prediction is specific and testable: if wild-type MLL1 masks fusion loss at MYB, then an MLL1n ChIP and a FLAG ChIP in the SAME cells should differ at MYB and agree at MEIS1. GSE260742 contains an MLL1 ChIP in the same MLLAF9 line under the same three conditions (GSM8123395/6/7) and this beat did not download it.
- What is the distribution of the VTP/DMSO share ratio over several thousand expression-matched, fusion-unbound promoters? That, and not two housekeeping genes, is the floor every row in this table should be read against.
- Does MEF2C behave differently in mouse MLLAF9 than in human KMT2A-rearranged lines generally, or only here? Its DMSO share in this line is low enough that the locus may simply not be a strong fusion target in this model.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
In murine MLLAF9 cells, the FLAG-tagged fusion's within-sample share of its own ChIP track over gene-body windows falls to 0.117 at Meis1, 0.167 at Jmjd1c, 0.660 at Hoxa9, 0.561 at Hoxa10, 0.362 at Myb, 0.870 at Mef2c and 0.508 at Gapdh under VTP50469 relative to DMSO.
NCBI GEO GSE260742, GSM8123359 / GSM8123361 (MLLAF9 FLAG ChIP-seq, DMSO and VTP50469 500 nM 48 h), re-binned to 2 kb from the deposited igvtools .tdf · NCBI Gene Expression Omnibus 2026What this evidence cannot show
n = 1 per condition. Windows are gene body minus 5 kb / plus 2 kb around the longest refGene transcript, with coordinates fetched live from the UCSC mm10 API on 2026-09-23 and matched on the refGene track's own posName field - an earlier version of the lookup took the longest match across all tracks and returned chr1 intervals for Hoxa9 and Meis1, so the coordinates in this table are the second, checked set. Bin resolution is 2 kb, so a window's value is the sum over 3 to 70 bins depending on gene length and short genes are noisier. Gapdh's and Actb's windows are small and their DMSO shares are among the lowest in the table, so the 'floor' they define is itself imprecise.
Evidence that argues against this (1)
- A two-gene housekeeping floor is a weak reference. Gapdh and Actb are both highly transcribed promoters and could share whatever makes gene-proximal FLAG signal fall; a proper floor would be the distribution of the ratio over several thousand expression-matched, fusion-unbound promoters, which this beat did not compute. Until that exists, calling Hoxa9 'retained' because 0.660 exceeds 0.508 rests on two numbers. NCBI Gene Expression Omnibus 2026
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
The comparison is a within-sample share, so a ratio of 1.0 means the locus tracked the genome average; there is no invariant reference locus in such a readout because an evicted protein's signal redistributes into background.
NCBI GEO GSE260742 - per-sample SOFT records showing no spike-in and no deposited scale factors for any of 158 samples · NCBI Gene Expression Omnibus 2026What this evidence cannot show
This is a property of the statistic, not a measurement. It is stated as evidence because it is the reason the Gapdh row reads as it does, and because a reader who expects a control locus to sit at 1.0 will otherwise conclude the analysis is broken.
Evidence that argues against this (1)
- If a spike-in were available, the absolute question - did GAPDH-proximal fusion signal actually fall, or did everything else just fall further - would be answerable and the floor would not be needed. That the floor is needed is a consequence of the deposit, not of the biology, and a spike-in-normalised repeat could overturn the direction of every row in this table by a common factor. Nature Cell Biology 2026
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The pivotal menin-inhibitor paper reports greater than 2-fold MLL1n loss at MEIS1, MEF2C and JMJD1C and no loss at the HOXA cluster, MYB and GAPDH under VTP-50469.
A Menin-MLL Inhibitor Induces Specific Chromatin Changes and Eradicates Disease in Models of MLL-Rearranged Leukemia. · Krivtsov AV, Evans K, Gadrey JY, et al. · Cancer Cell 2019What this evidence cannot show
PMID, title, journal, year and DOI fetched from a live PubMed esummary on 2026-09-23. The locus list is taken from this graph's entry mechanism-vtp-50469-does-not-globally-displace-the-fusion-from-chromatin, whose author read the paper; the paper was not re-opened in this beat, so the comparison is against what this graph records rather than against the figures.
Evidence that argues against this (1)
- If the graph's rendering of the locus list is incomplete - for instance if the paper qualifies the HOXA call by cell line, or reports the losses at a different timepoint than the one compared here - then two of the four agreements and both disagreements could be artefacts of the comparison rather than of the data. The paper should be re-opened before this table is quoted as a replication. Cancer Cell 2019
re-measures that entry's named locus list in an independent deposit with an epitope-tagged fusion: MEIS1 and JMJD1C reproduce, MEF2C and MYB do not
Meis1 loses the most fusion share of any named locus, 0.117 of DMSO under VTP50469
- related to →HOXA9 — required for survival of established MLL-rearranged leukemia, and dispensable for starting it
Hoxa9 at 0.660 sits above the housekeeping floor of 0.508-0.571 and is a relative retention, reproducing the published call
the locus ratios are only interpretable against the genome-wide collapse that entry quantifies
the trap fired inside that entry's locus table and the coordinates there are the second, checked set
Q117's named-locus pattern is re-tested in human orthologues: MEIS1 agrees to 6%, MYB diverges 1.69-fold in RS4;11 and agrees to 12% in MOLM13
Q117 recorded murine Mef2c as retained (1.713 over its floor) and a break with the published account; MEF2C is lost in all three human lines (0.181 / 0.248 / 0.040), so the murine retention does not generalise
closes a gap Q117's own memo named - a floor from several thousand promoters rather than two housekeeping genes
Q117's own stated next step, run: MLL1 and FLAG differ at Myb (1.83x) and agree at Meis1 (0.96x) in the same murine nuclei, exactly the direction it predicted
q117 independently puts Epha7 in the top twenty murine MLL-AF9 fusion-occupied bins at a 0.178 VTP50469/DMSO ratio