Hypothesis: menin inhibition pushes cells toward the GMP-like state, and the GMP-like state is where menin inhibition works best — so a lineage switch on therapy should predict response, not failure
also called GMP destination paradox · does menin-inhibitor-driven lineage switch help or hurt
Two separate studies say things that do not fit together: leukaemias that change lineage under treatment end up looking like a particular kind of myeloid cell, and that same kind of cell is the one a menin inhibitor kills best. If both are true, a lineage switch on the drug should be good news, yet the one patient profiled switched and did badly.
The same thing, for a scientist
Two published observations point in opposite directions and have never been placed side by side. (1) In KMT2A-rearranged leukaemias that lineage-switch in response to therapy, oncoprotein levels are reduced and the oncoprotein preferentially activates **granulocyte-monocyte progenitor (GMP) genes**; the one sample that switched during revumenib treatment did so into a GMP-like program [E1]. (2) A large-scale drug screen across primary AML patient cells, comparing genetic alterations and quantitative proteomics with drug sensitivity, found that **GMP-like AML display the greatest sensitivity** to the menin-KMT2A inhibitor bleximenib [E2]. Taken together these predict that a cell moving toward a GMP-like state under menin inhibition is moving into the state where menin inhibition is most effective — that is, lineage switch on a menin inhibitor should be a **transitional and self-limiting** event, not an escape. We hypothesise this is true, that menin-inhibitor-associated lineage switch is therefore prognostically different from blinatumomab-associated lineage switch, and that the discordant patient sample is explained by a separate maintenance mechanism (ENL-marked epigenetic memory) rather than by the lineage change itself.
Why the two results have never been reconciled
They sit in different literatures. 1 is a chromatin-occupancy paper about lineage identity; its audience is developmental epigenetics. 2 is a drug-sensitivity and proteomics paper about a specific compound; its audience is AML pharmacology. Neither cites the other's question. But the object they both describe is the same — a GMP-like leukaemic state — and they assign it opposite roles: destination of escape, and locus of maximal drug effect.
Only one of three things can be true. Either (a) the GMP-like state reached by lineage switching is not the same GMP-like state that is bleximenib-sensitive, (b) it is the same state and lineage switch on a menin inhibitor is a favourable event that the field is misclassifying as progression, or (c) it is the same state but something else in the switched cell — the ENL-marked lesion 3 — makes it drug-insensitive despite its lineage. Each of these has a different clinical consequence and they are distinguishable by experiment.
Why this matters clinically and not just conceptually
Lineage switch on therapy is currently scored as progression. In CD19-directed therapy that is correct, because the switched cell has lost the antigen the drug depends on and the outcome is dismal — remission rates under 40% and median survival of 4.8 months are recorded in this graph. A menin inhibitor is different in kind: the switched cell retains the fusion, and on 2 may actually be more drug-sensitive than the cell it came from.
If hypothesis (b) is right, then stopping a menin inhibitor because the marrow has turned myeloid is the wrong response, and the right response is to continue and reassess. That is a decision a clinician will face and there is currently no evidence base for it at all. A 2026 paediatric letter reporting successful management of lineage-switch infant leukaemia with revumenib plus transplant is consistent with (b) but is a single title-level observation 4.
The prediction that separates this from the pessimistic reading
This graph already holds hyp-menin-release-lineage-destination, which predicts that releasing the differentiation block with a menin inhibitor is not lineage-neutral and will drive a fraction of cases toward a myeloid program — and treats that as a hazard, arguing for closing the myeloid exit before releasing the block. The present hypothesis does not contradict that prediction; it contradicts its sign. It says the myeloid exit may lead somewhere the drug can follow.
That is a rare situation in this beat, where two hypotheses make the same mechanistic prediction and opposite clinical ones, and a single experiment — serial single-cell phenotyping with matched drug sensitivity on the switched population — separates them.
| statement | The GMP-like state that KMT2A-rearranged leukaemias enter when they lineage-switch under therapy is the same state that shows the greatest ex vivo sensitivity to menin-KMT2A inhibition; therefore lineage switch occurring on a menin inhibitor is a transitional event with a better prognosis than lineage switch on CD19-directed therapy, and cases that nonetheless progress do so because of a separate maintenance mechanism (ENL-marked epigenetic memory) rather than because of the lineage change. |
|---|---|
| origin | literature-gap |
| prior probability | 0.3 |
| decision value | Determines whether a clinician who sees a marrow turn myeloid on a menin inhibitor should stop the drug or continue it. There is currently no evidence base for that decision and it will be faced in the paediatric trials now enrolling. |
| relationship to existing hypotheses | hyp-menin-release-lineage-destination predicts the same mechanistic event (menin release drives a myeloid destination) and assigns it the opposite clinical sign. The two are distinguishable by the same experiment, so testing one tests both. |
predictions
[ "Blasts sorted from a KMT2A-rearranged leukaemia after a menin-inhibitor-associated lineage switch are MORE sensitive to menin inhibition ex vivo than matched pre-switch blasts from the same patient.", "The GMP-like signature defined chromatin-wise in lineage-switching samples and the GMP-like proteomic class that is bleximenib-sensitive overlap substantially when scored on the same cells.", "Menin-inhibitor-associated lineage switch carries a materially better outcome than blinatumomab- or CAR-associated lineage switch in KMT2A-rearranged disease, once enough cases are collected to compare.", "Switched samples that do progress on a menin inhibitor carry ENL retained at oncoprotein-target loci; switched samples that respond do not.", "In a patient-derived xenograft driven to switch by menin inhibition, continuing the drug through the switch produces deeper eventual clearance than stopping it at the point of switch." ]
experiments
[ "exp-menin-inhibitor-lineage-destination", "exp-residual-occupancy-cutrun-saturating-menin" ]
analyses
[ "ana-lineage-switch-signature-scoring" ]
falsifiers
[ "Post-switch blasts are LESS sensitive to menin inhibition ex vivo than matched pre-switch blasts — which would establish lineage switch as true escape under this drug class and retire the hypothesis outright.", "The chromatin-defined GMP-like program of lineage-switching samples and the proteomic GMP-like class that is bleximenib-sensitive do not overlap when scored on the same cells, meaning the two literatures are describing different states and there was never a paradox.", "A series of menin-inhibitor-associated lineage switches shows outcomes as poor as the immunotherapy-associated ones.", "ENL retention is found equally in switched samples that respond and switched samples that progress, removing the proposed explanation for the discordant case." ]
- If the hypothesis is wrong in the pessimistic direction, the clinical advice it suggests — continue the menin inhibitor through a lineage switch — would keep a child on an ineffective drug while an aggressive myeloid leukaemia expands. That is a real harm and the reason the prior is set below 0.5.
- It rests on comparing two studies that used different compounds, different diseases (AML versus a mixed KMT2A-rearranged cohort), different assays and possibly different definitions of the same cell state. Cross-study state matching has failed before in this field.
- Even if true, it applies only to switches that occur on a menin inhibitor. A child who switched on blinatumomab and is then given a menin inhibitor is a different situation and this hypothesis says nothing about it.
- The favourable reading could be an artefact of survivorship: the reported successful case is published because it succeeded, and the profiled failing case is published because it was profiled.
- Are the two GMP-like definitions the same object? Everything rests on this and it is a bioinformatics question that could be answered this week if both studies' signatures were scored on one dataset.
- How many menin-inhibitor-associated lineage switches exist in total? One is profiled and one is reported in a letter. Even a case series of five would move this from speculation to signal.
- Does the direction of switch matter? Both documented KMT2A-rearranged switches in this graph are lymphoid-to-myeloid; a myeloid-to-lymphoid switch under a menin inhibitor would test the hypothesis from the other side and has not been reported.
- If GMP-like cells are the most menin-sensitive, why is GMP-like the destination of escape under other pressures? That may be the deepest question here: the same state may be simultaneously the safest place to hide from an antibody and the worst place to hide from a menin inhibitor.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In KMT2A-rearranged leukaemia samples that lineage-switched in response to therapy, oncoprotein levels are reduced and the oncoproteins preferentially activate granulocyte-monocyte progenitor genes; in a sample that lineage switched during revumenib treatment, the oncoprotein and menin were reduced to undetectable levels while ENL persisted on numerous oncoprotein-target loci including genes in the GMP-like lineage-switching program.
MLL oncoprotein levels influence leukemia lineage identities. · Janssens DH, Duran M, Otto DJ, et al. · Nat Commun 2024What this evidence cannot show
Three lineage-switching samples, one of them on revumenib. Abstract read in full via NCBI efetch; full text not obtained. 'GMP-like program' is defined by gene-set enrichment in that study and may not be the same operational definition used in [E2].
Evidence that argues against this (1)
- The oncoprotein-level observation comes from a single human sample profiled under revumenib, and an undetectable oncoprotein in a lineage-switching sample is equally consistent with the fusion-expressing population having been replaced rather than the protein being lost. The GMP-destination reading requires the first interpretation and the data cannot distinguish them. Blood 2025
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
A large-scale drug screen in which genetic alterations and quantitative proteomics were compared with drug sensitivity in primary AML patient cells revealed that GMP-like AML display the greatest sensitivity to the menin-KMT2A inhibitor bleximenib (JNJ-75276617); the study also found NPM1c/DNMT3A-mutant AML sensitive, identified patient-specific epigenetic alterations on treatment, and showed upregulation of MHC class I and II via CIITA activation independent of MEIS1 loss, enhancing sensitivity to T-cell-mediated cytotoxicity.
Bleximenib, the novel menin-KMT2A inhibitor JNJ-75276617, impairs long-term proliferation and immune evasion in acute myeloid leukemia. · Hogeling SM, Lê DM, La Rose N, et al. · Haematologica 2025What this evidence cannot show
Abstract read in full via NCBI efetch; full text not obtained, so the number of primary samples, the definition of GMP-like and the magnitude of the sensitivity difference are unknown to this entry. Primary AML, not KMT2A-rearranged ALL, and bleximenib rather than revumenib. Ex vivo sensitivity, not clinical response. The GMP-like classification here is proteomic and transcriptomic, which is not obviously the same operation as the chromatin-based GMP program in [E1] — reconciling the two definitions is part of the work this hypothesis calls for.
Evidence that argues against this (1)
- The one KMT2A-rearranged sample actually profiled after switching into a GMP-like program under revumenib did NOT behave as a drug-sensitive state: the patient progressed, and ENL was retained on the GMP-program loci. A single discordant case does not refute an ex vivo sensitivity screen, but it is the only in-patient datapoint that exists and it points the wrong way for this hypothesis. Janssens DH et al. 2024
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.preprint
The preprint version of [E1] frames the finding as resistance: KMT2A oncoproteins induce epigenetic lesions, marked by ENL, that support resistance to targeted therapies.
KMT2A oncoproteins induce epigenetic resistance to targeted therapies. · Janssens DH, Duran M, Otto DJ, et al. · bioRxiv 2024What this evidence cannot show
Preprint of [E1]; not independent. Cited for the third arm of the hypothesis (the ENL explanation for the discordant sample).
Evidence that argues against this (1)
- This is the preprint of E1, not a second source. The hypothesis's evidence base is one study. Blood 2025
- E4weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper
A 2026 paediatric letter reports successful management of lineage-switch infant leukaemia with revumenib and haematopoietic cell transplantation.
Successful Management of Lineage Switch Infant Leukemia With Revumenib and Hematopoietic Cell Transplantation. · Granberg RE, Magee JA, Shenoy S, et al. · Pediatr Blood Cancer 2026What this evidence cannot show
Letter with no abstract in PubMed; only the title-level claim is made. Whether the switch preceded or followed revumenib is not known to this entry, and that distinction matters for whether the case supports this hypothesis or merely coexists with it.
Evidence that argues against this (1)
- One successful case managed with revumenib plus transplant cannot bear on a destination hypothesis, and the outcome is confounded by the transplant. The base rate is poor: median survival of 2.9 months after switch across 33 clonally confirmed cases. Blood Cancer Journal 2024
supplies the cell type that hypothesis rested on without one
defines the state the hypothesis is about and names the definitional escape route
bears on the destination half of that paradox: no GMP-like enrichment outside the defining pair
scores the chromatin-wise GMP-like signature across six partners; enriched nowhere but p318
the chromatin half of the paradox has n = 1 and that one is circular
same event, opposite prognostic sign; this hypothesis argues the switch may be transitional rather than escape
agrees on the mechanism, disagrees on whether the myeloid destination is a hazard
GMP-like primary AML showed the greatest ex vivo sensitivity to bleximenib
serial single-cell phenotyping under menin inhibition, with matched drug sensitivity, separates the three possible readings
- depends on →ENL stays after the fusion leaves — an epigenetic lesion that outlives the protein that made it
the ENL lesion is the proposed explanation for the one switched sample that did not behave as the hypothesis predicts
hypothesis raised by the tension found during this enumeration