start / graph / ana-reanalysis-of-ml2-mll1n-chip-for-gained-peaks-under-vtp-50469

AnalysismoderateQ83-fusion-fate

The gain question is answerable today for free: GSE127508 contains MLL1n ChIP-seq in a fusion-only line, plus and minus VTP-50469, and nobody has called peaks on the treated sample

also called GSE127508 reanalysis · ML2 gained-peak call · free relocation test

The data needed to ask whether the fusion moves instead of leaving has been sitting in a public archive since 2019. It includes a cell line whose only copy of the protein is the fusion, treated and untreated. Nobody has run the analysis that looks for new sites.

The same thing, for a scientist

Krivtsov et al. deposited their chromatin data as **GEO GSE127508**, public since 2019-11-06 and last updated 2025-07-30 [E1]. Reading the 78 sample titles, the series contains exactly the arms this question needs [E1]: - `ML2_ChIP_MLL1n_DMSO_d3` and `ML2_ChIP_MLL1n_VTP_d3` - **the fusion-only line**, day 3, paired - `ML2_ChIP_MENIN_DMSO` / `_VTP`, `ML2_ChIP_DOT1L_DMSO` / `_VTP`, `ML2_ChIP_input_DMSO` - `RS411_ChIP_MLL1n_DMSO_d3` / `_VTP_d3` - the index fusion KMT2A::AFF1, shared-epitope antibody - `MOLM13_ChIP_MLL1n_DMSO_d3` / `_VTP_d3`, plus MOLM13 inputs for both DMSO and VTP - `MOLM13-Cas9_CUTRUN_MLL1_sgMEN1 / sgNull, d6, two replicates each` **The published analysis quantified TSS-anchored signal and tabulated 2-fold losses. The proposed analysis calls peaks de novo in both arms and tests the treated-only set.** Concretely: align to hg38, call MACS2 narrow peaks in DMSO and VTP separately against the matched input, build a union peak set, quantify both arms over it, and report (a) the number and genomic distribution of VTP-only peaks, (b) their overlap with published unmethylated-CpG maps in the same lineage, (c) whether the fusion's share of all uCpG sites rises from the baseline 20% reported in SEM, and (d) the residual fraction at DMSO peaks, which is the number the whole field is missing. **This analysis is honest about being underpowered, and that is the point of writing it down.** The ML2 arm has **one sample per condition**, no biological replicate, and **an input for DMSO only**. So it cannot support a statistical claim about differential binding. What it can do is discriminate between two very different pictures - a handful of spurious treated-only peaks versus thousands of them concentrated at unoccupied uCpG islands - and that is the difference between the eviction reading and the relocation reading. It is a screen, it costs a day of compute, and it should be run before anyone spends money on new sequencing.

Go deeper

What each arm can and cannot answer

arm	attribution	replicates	input	answers
ML2 MLL1n	fusion only (wild-type KMT2A allele lost)	1 per condition	DMSO only	is there a treated-only peak set at all, and where
RS4;11 MLL1n	fusion + wild-type MLL1	1 per condition	DMSO only	the same question in the index fusion, with an attribution ceiling
MOLM13 MLL1n	fusion + wild-type MLL1	1 per condition	both DMSO and VTP	the only arm with a matched treated input, so the cleanest peak calling
MOLM13-Cas9 CUT&RUN MLL1	fusion + wild-type	2 per condition	none listed	the only replicated arm; genetic rather than pharmacological menin removal

The sensible primary analysis is MOLM13 ChIP-seq (matched inputs) with ML2 as the attribution arm and the replicated MOLM13 CUT&RUN as the orthogonal confirmation. If all three show the same treated-only peak population, the relocation reading is live despite n=1 in two of them. If none does, the eviction-plus-retention reading stands and the question is closed cheaply.

Why this was not done in 2019, and why that is reasonable

The paper's question was whether a menin inhibitor is worth taking into humans. For that, losses at target genes and PDX survival are the right readouts, and they were done well. Gained peaks were not part of the question, the fusion-relocation hypothesis did not exist in the literature, and no reviewer would have asked for it. This is an analytic gap, not an experimental failure, which makes it the cheapest open question in this beat by a wide margin.

The same reasoning applies to Perner's dataset, where MACS2 peaks were called with input controls in every condition and the differential comparison was simply never made 2.

Facts and figures

script	analyses/q83-ml2-gained-peaks/run.py (not yet written)
status	planned
result summary	Planned. Primary output: number, genomic distribution and uCpG overlap of VTP-only MLL1n peaks, plus the residual occupancy fraction at DMSO peaks.
cost tier	low
compute	one node-day; roughly 20 FASTQ files, no GPU
primary arm	MOLM13 (matched treated input)
attribution arm	ML2 (fusion-only)
orthogonal arm	MOLM13-Cas9 MLL1 CUT&RUN sgMEN1 vs sgNull, 2 replicates
shown for fusion	KMT2A::AFDN in ML2 is the only fusion-attributable arm; KMT2A::AFF1 in RS4;11 is index-fusion but shared-epitope

inputs

[
  "GEO GSE127508 FASTQ - ML2_ChIP_MLL1n_DMSO_d3, ML2_ChIP_MLL1n_VTP_d3, ML2_ChIP_input_DMSO, MOLM13_ChIP_MLL1n_DMSO_d3, MOLM13_ChIP_MLL1n_VTP_d3, MOLM13_ChIP_input_DMSO_d3, MOLM13_ChIP_input_VTP, RS411_ChIP_MLL1n_DMSO_d3, RS411_ChIP_MLL1n_VTP_d3, RS411_ChIP_input_DMSO_d3",
  "GEO GSE83671 for the SEM uCpG reference"
]

limits

[
  "n=1 per condition in the ChIP arms",
  "no treated input for ML2 or RS4;11",
  "no matched uCpG map in the reanalysed lines",
  "hg19 in the original processing; realign rather than lift over"
]

How this could bite us back

If the reanalysis finds thousands of treated-only peaks, the class's mechanism of action needs rewriting while the drug is in paediatric trials - and the finding would come from the sponsor-adjacent dataset that supported the IND, which is a communication problem as much as a scientific one.
If it finds none, the relocation hypothesis is not dead: three days is late, and the analysis would then need the early timepoints that exp-calibrated-early-timecourse-fusion-profiling-under-menin-inhibition supplies. A negative here must be reported as 'no persistent gain at day 3', not as 'no relocation'.
n=1 per condition means a positive result is a lead, not a result. Publishing or acting on it without the replicated confirmation arm would repeat the class of error this project has documented in its own analyses.

Still open

Do treated-only MLL1n peaks exist at all, and if so are they enriched at unmethylated CpG islands that were unoccupied at baseline?
What is the residual fraction - what percentage of DMSO MLL1n peaks retain a called peak under drug? This single number, which the deposited data can supply today, is absent from the entire literature.
Does the replicated genetic arm (MEN1 CRISPR CUT&RUN, two replicates) show the same treated-only population as the pharmacological arms? Agreement would make n=1 in the drug arms much less troubling.

Every source behind this entry (3)

E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
GEO GSE127508, the SuperSeries deposited with the pivotal VTP-50469 paper, is public and contains 78 samples including paired DMSO and VTP-50469 MLL1n ChIP-seq at day 3 in ML2 - a line that has lost its wild-type KMT2A allele so that all MLL1n signal is fusion - as well as in RS4;11 and MOLM13, with a DMSO input for ML2 and both DMSO and VTP inputs for MOLM13, and two replicates per arm of MLL1 CUT&RUN after MEN1 CRISPR deletion.
GEO GSE127508 - A novel Menin-MLL inhibitor induces specific chromatin changes and eradicates disease in models of MLL-rearranged leukemia (SuperSeries, 78 samples; public 2019-11-06, last updated 2025-07-30) · Krivtsov AV, et al. (depositors) · NCBI Gene Expression Omnibus 2019
What this evidence cannot show
Sample composition read from the GEO accession display on 2026-09-21; FASTQ files were not downloaded and read depth, fragment size and library quality were not assessed. The ML2 and RS4;11 ChIP arms have no biological replicate and no treated input, which caps any differential analysis at the level of a screen. Alignment in the original work was to hg19; a reanalysis should realign rather than lift over. Antibody attribution for RS4;11 and MOLM13 remains shared-epitope.
Evidence that argues against this (1)
- A treated-only peak set called without a treated input can be produced entirely by changes in background and chromatin accessibility under drug, which is a real risk here because menin inhibition induces differentiation over three days. The MOLM13 arm, which does have a VTP input, is the control for that and should gate the interpretation of the other two. Krivtsov AV et al. 2019
E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
A second independent dataset suitable for the same reanalysis exists: Perner et al. called MACS2 peaks with matched input controls for menin and MLL1 ChIP-seq across a revumenib dose ladder in MV4;11 and OCI-AML3, and reported only TSS-window losses.
MEN1 mutations mediate clinical resistance to menin inhibition. · Perner F, Stein EM, Wenge DV, et al. · Nature 2023
PMID 36922589 DOI
What this evidence cannot show
The accession is not recorded in this entry and should be resolved before the analysis is planned around it; the methods state MACS2 with appropriate input samples but do not state whether raw data were deposited. Dose ladder without a stated treatment duration.
Evidence that argues against this (1)
- If Perner's raw data are not deposited, this second dataset is unavailable and the reanalysis rests on GSE127508 alone, which would leave the conclusion resting on one laboratory's reagents and one antibody. Krivtsov AV et al. 2019
E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The comparison the reanalysis would make is meaningful because the fusion occupies only about 20% of available unmethylated CpG sites at baseline, against 50% for CFP1 and 89% for KDM2B in the same cells, so there is a large unused substrate for a redistribution to move onto.
MLL-AF4 Spreading Identifies Binding Sites that Are Distinct from Super-Enhancers and that Govern Sensitivity to DOT1L Inhibition in Leukemia. · Kerry J, Godfrey L, Repapi E, et al. · Cell Rep 2017
PMID 28076791 DOI
What this evidence cannot show
The 20/50/89 figures are from SEM, an infant-derived KMT2A::AFF1 B-ALL line, and the reanalysis arms are ML2, RS4;11 and MOLM13 - different lineages, partners and methylomes. The uCpG landscape would have to be measured or imported per line, and no Bio-CAP or equivalent uCpG map exists in GSE127508.
Evidence that argues against this (1)
- Without a matched uCpG map in the reanalysed lines, the 'did the fusion move onto unused uCpG' question degrades to 'did it gain peaks anywhere', which is a weaker test. A published CpG-island annotation is a poor substitute because the relevant quantity is unmethylated in these cells, not annotated as an island. Kerry J et al. 2017

Connected to

← related toAsked for the first time whether a KMT2A fusion GAINS chromatin sites under a menin inhibitor: all three pre-registered criteria fire, and the set they fire on is 52 low-amplitude bins reaching 6x the genome median, 16 of them with no gene within 20 kb and 1 of 52 on a CpG island where 66% of the fusion's real sites are
the same question asked of a different deposit; that proposal is for human ML2/RS4;11 MLL1n in GSE127508 and remains unrun
← testsThe falsifier that decides the beat, run: 65.6% of the fusion's own top sites overlap a CpG island and 1 of the 52 sites it formally 'gains' under a menin inhibitor does - 1.35-fold over a class-matched null, Fisher p = 0.53 - so the gained set does not carry the signature of CXXC-driven binding and the redistribution model is refuted on its own predicted substrate
tests that entry's uCpG redistribution premise in a different deposit; the gained set is not CpG-island enriched
← supportsFifty-three 2 kb bins gain MLL1-N-reactive signal under VTP-50469 in RS4;11, and the same 53 coordinates gain 1.77-fold in MOLM13 against a class-matched null of 1.01 with the MOLM13 input flat at 1.01 - so there IS a small, reproducible, input-controlled redistribution, and it carries no CpG-island signature
this is the analysis that entry proposed - the within-sample gained-peak question run on the GSE127508 human lines
← related toML-2 (KMT2A::AFDN, a non-elongation partner) is the line where menin eviction is weakest (retains 37.2% of its own top windows against 20.5% and 13.0%) and where the DOT1L track moves most of any track in the beat (retains 31.6%, 13.69 percentile points below a rank-matched null at p = 3e-10) - the opposite pattern to what an AF9/ENL-free fusion would predict
that entry proposes the ML-2 MLL1n re-analysis; this is the TSS-window version of it, and the genome-wide version is still not run
← testsBounded absence: no published study has tested whether a KMT2A fusion GAINS chromatin sites under a menin inhibitor - and the same pipelines found gains for two other proteins when they looked
the analysis that would close this absence from public data
tests →KMT2A::AFDN (MLL-AF6), t(6;11)(q27;q23)
the ML2 arm is fusion-attributable by genetics
tests →Revumenib (SNDX-5613, Revuforj) — first approved menin–KMT2A interface inhibitor
the deposited data use the preclinical analogue of this drug
tests →Hypothesis: menin inhibition does not clear the fusion from chromatin but redistributes it onto the four fifths of the unmethylated CpG address it normally ignores
first and cheapest test of the relocation hypothesis
tests →Hypothesis: the menin inhibitor's effect is inactivation in place, not eviction - the fusion keeps most of its sites and loses its output
supplies the residual occupancy fraction that the inactivation-in-place reading needs
tests →The pivotal menin-inhibitor paper reports that the drug does NOT globally displace MLL1 or the fusion from chromatin - menin goes, the fusion largely stays
would convert a qualitative 'not globally displaced' into a number