The GMP-like state is being used as a therapeutic anchor in two independent arguments, and its chromatin definition rests on one post-lineage-switch sample that also defined it
Two separate lines of reasoning treat a particular myeloid-like state as the place where a new drug class works best, and as the place leukaemias go when they change identity. The chromatin version of that state is defined from one patient, and outside that patient this deposit cannot see it.
The same thing, for a scientist
The GMP-like state carries therapeutic weight in this corpus from two directions: the deposit's paper makes it the destination of therapy-associated lineage switch and the state in which ENL persists, motivating a **menin-plus-ENL combination proposal**; and an independent proteomic drug screen finds **GMP-like AML the most bleximenib-sensitive class** ex vivo. `hyp-gmp-destination-paradox-menin` holds that these two should imply each other, and records their overlap as **untested**. **The risk this entry records is narrower and now measured.** The chromatin-side definition of the state - the authors' GMP-like gene export - is enriched over a coverability-matched null in **1 of 27** GSE252378 samples, and that one is **p318**, the sample the class was derived from. The deposit's only other post-switch AML sample, **152985**, is revumenib-exposed with every channel collapsed. **So the number of independent, drug-naive, post-lineage-switch samples in which the chromatin GMP-like state can be observed in this deposit is zero**, and the third lineage-switch pair (R3595/R3652) is labelled B-lineage on both halves and contributes no post-switch sample at all. A therapeutic anchor with n = 1 and that one circular is not an anchor. It may still be correct - the paper's peak-level analysis is richer than a 40-promoter proxy - but a patient-selection strategy built on it would be built on one patient.
What makes this a risk rather than a criticism
Nothing here says the GMP-like state does not exist or that bleximenib's GMP-like sensitivity is wrong. The proteomic class and the chromatin class are different operational definitions of the same name, measured on different material with different instruments, and cell-state-gmp-like already records their overlap as untested. The risk is that the two are cited together as if they were one state, and the chromatin half - which is the half that carries the mechanism - is supported by a single sample.
Mitigation, in order of cost. (1) Score the authors' GMP-like intervals, not their promoters, in all 27 samples against a matched null - compute only. (2) Define a GMP-like gene set from normal haematopoiesis rather than from one patient pair and re-score; a reference signature is a less circular reagent. (3) Test the overlap the paradox hypothesis names: score the chromatin GMP-like class and the proteomic GMP-like class on the same cells.
| kind | resistance |
|---|---|
| severity | moderate - it does not invalidate either published result, but it means a patient-selection strategy built on chromatin GMP-like occupancy currently rests on one circular sample |
| mitigation | score the authors' GMP-like INTERVALS rather than promoters in all 27 samples against a matched null (compute only); re-score with a GMP signature defined from normal haematopoiesis instead of from one patient pair; and test the chromatin-versus-proteomic overlap the paradox hypothesis names |
| n independent drug naive post switch samples in gse252378 | 0 |
| status | run |
affects
[ "drug-revumenib", "drug-bleximenib" ]
measurements
[
{
"quantity": "independent_drug_naive_post_lineage_switch_samples",
"subject": "GSE252378 samples in which the chromatin GMP-like state could be observed non-circularly",
"value": 0,
"units": "samples",
"n": 27,
"universe": "all GSE252378 KMT2A-fusion patient samples; post-switch AML aliquots are p318 (circular) and 152985 (revumenib-exposed)",
"null": null,
"note": "the third lineage-switch pair R3595/R3652 is labelled B-lineage on both halves"
}
]- Selecting patients for a menin inhibitor on a GMP-like signature would select on a signature whose chromatin definition comes from one patient and whose proteomic definition has never been shown to be the same thing.
- The combination proposal - menin inhibitor plus ENL inhibitor - is aimed at a state observed in one post-switch sample. If the state is not general, the combination is being designed for a population that has not been shown to exist.
- Do the chromatin GMP-like class and the bleximenib-sensitive proteomic GMP-like class overlap on the same cells? This is the principal falsifier of the paradox hypothesis and it remains unrun.
- Is there any post-lineage-switch, drug-naive KMT2A-rearranged sample with deposited occupancy outside GSE252378? If not, the chromatin GMP-like state has n = 1 in the world.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
The authors' GMP-like promoter class is enriched over a coverability-matched null in 1 of GSE252378's 27 KMT2A-fusion samples, and that sample is the one the class was derived from.
analyses/q148-what-the-hoxa-negative-blast-occupies - results148_matched_null.json; results148.json · Q148 (Oncofusion Knowledge Graph) · this repository 2026What this evidence cannot show
A 40-promoter proxy for a 45-interval class, at 2 kb rather than the ~20 kb the class was built at. The authors' peak-level analysis could give a different answer and has not been re-run in all 27 samples by anyone.
Evidence that argues against this (1)
- The paper's GMP claim is specifically about lineage-switching samples, and this deposit deposits only two post-switch AML aliquots, one circular and one drug-collapsed. So a negative across the other 25 samples is a negative about generalisation, not about the claim as the authors stated it. Janssens DH et al. 2024
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
An independent quantitative-proteomic drug screen finds GMP-like AML the most sensitive class to the menin-KMT2A inhibitor bleximenib ex vivo.
Bleximenib, the novel menin-KMT2A inhibitor JNJ-75276617, impairs long-term proliferation and immune evasion in acute myeloid leukemia. · Hogeling SM, Le DM, La Rose N, Kwon MC, Wierenga ATJ, Van den Heuvel FAJ, Van den Boom V, Kuchnio A, Philippar U, Huls G, Schuringa JJ · Haematologica 110(6):1278-1291 (issue Jun 2025; epub 2024-12-19) 2025What this evidence cannot show
The GMP-like class there is a proteomic clustering of primary AML, not a chromatin-occupancy class, and the two definitions have never been scored on the same cells. The screen is preclinical and ex vivo.
Evidence that argues against this (1)
- If the proteomic GMP-like class and the chromatin GMP-like class are not the same population, then the inference 'lineage switch goes to GMP-like, GMP-like is menin-sensitive, therefore switching predicts response' has a broken middle term, and the paradox hypothesis's own principal falsifier is exactly this overlap. Hogeling SM et al. 2025
the chromatin half of the paradox has n = 1 and that one is circular
records the measured support for that state's chromatin operational definition
the destination the switch is said to reach is observable in one sample here
bears on patient selection for menin inhibitors on a GMP-like signature