Analysis (runnable): is there a shared cell-state signature behind non-genetic menin-inhibitor and BET-inhibitor resistance?
also called WNT signature menin resistance · shared plasticity signature · cross-modality resistance signature
Two different drug classes fail in this leukemia for reasons that are not mutations. We can check, using published data, whether the surviving cells look the same in both cases — and if they do, that shared state becomes the real target.
The same thing, for a scientist
A cross-dataset signature analysis testing whether non-genetic menin-inhibitor resistance in KMT2A::AFF1 and BET-inhibitor resistance in MLL-AF9 converge on a common transcriptional and chromatin state. Inputs: published transcriptomes from upfront-menin-inhibitor-resistant KMT2A::AFF1 samples and PDX models, BET-resistant MLL-AF9 clone expression data, the WNT/beta-catenin and MYC-restoration signatures from the two 2015 BET-resistance studies, single-cell CUT&Tag chromatin heterogeneity data at KMT2A oncofusion binding sites, and KMT2A-r patient RNA-seq for clinical anchoring. Outputs: a shared-state signature score, its enrichment in each resistant dataset, and a ranked list of pharmacologically addressable pathways. Directly tests hyp-cell-state-plasticity-limits-interface-disruption.
The three datasets that have never been compared, and what the comparison would show1 deeper
Dataset 1 — non-genetic menin resistance. Serial KMT2A::AFF1 samples from the same patients, menin-inhibitor-class-naive, markedly less sensitive later in the treatment course, with transcriptomics showing sustained on-target efficacy in the resistant cells and no single sufficient comutation 1. This is the phenotype to explain.
Dataset 2 — BET resistance via WNT. Resistant leukemias uniformly characterised by rapid restoration of MYC transcription via activation and recruitment of WNT signalling components and a focal MYC enhancer, with PRC2 suppression promoting resistance by remodelling pathways that restore MYC 2. Independently, I-BET-resistant MLL-AF9 clones emerged from leukaemia stem cells, were cross-resistant to genetic BET depletion, were not efflux-mediated, showed globally reduced chromatin-bound BRD4 with unaltered Myc, and were resensitised by negative regulation of WNT 3.
Dataset 3 — chromatin heterogeneity. Single-cell CUT&Tag at KMT2A oncofusion binding sites shows a subset carry bivalent signatures with cell-to-cell heterogeneity suggestive of lineage plasticity 4, and gene-body H3K4me3 at those sites is menin-inhibitor-sensitive — giving both a candidate reservoir and a measurable chromatin readout.
The analysis. Build a WNT/beta-catenin plus MYC-restoration plus stemness composite score from datasets 2 and 3, then test its enrichment in the menin-resistant samples of dataset 1 by GSEA and by single-sample scoring. Then invert: derive a signature de novo from dataset 1 and test for reciprocal enrichment in dataset 2. Concordance in both directions is the result that matters; one-directional enrichment is suggestive only.
Anchoring. Score the composite in bulk KMT2A-r patient RNA-seq (TARGET, St. Jude Cloud) and test association with outcome and with age/HOXA status. A state that predicts relapse independently of known risk factors is a clinically usable biomarker whether or not the mechanistic hypothesis holds.
Pre-registered interpretation, because this analysis is easy to over-read
Signature-overlap analyses across different drugs, species and platforms generate spurious concordance readily. Three guards, set before running:
- Bidirectional requirement. Enrichment must hold in both directions (composite-in-menin-resistant and menin-derived-in-BET-resistant) to count as support.
- Negative-control signatures. Test unrelated signatures of comparable size (e.g. interferon response, cell cycle, hypoxia) to establish the false-positive rate for this dataset pairing. Proliferation signatures in particular will correlate with almost anything in drug-resistance data.
- Species and platform. Dataset 2/3 are largely murine MLL-AF9; dataset 1 is human KMT2A::AFF1 ALL. Ortholog mapping and batch structure must be handled explicitly, and a failure to concord may reflect species or lineage rather than the hypothesis.
If all guards pass and concordance holds, the deliverable is a ranked list of pharmacologically addressable nodes in the shared state — WNT/beta-catenin (porcupine and tankyrase inhibitors exist), plus whatever else emerges — to be tested in combination with a menin inhibitor in resistant PDX models.
Status is planned. Nothing has been run; no result should be cited from this node.
| script | analyses/menin-resistance-state-signature/run.py |
|---|---|
| method | Composite WNT/MYC-restoration/stemness score built from the BET-resistance and single-cell chromatin datasets, tested for enrichment in menin-resistant samples by GSEA and single-sample scoring; reciprocal test of a de-novo menin-resistance signature in BET-resistant data; bidirectional concordance required |
| status | planned |
| result summary | Not yet run. No result may be cited from this node. |
| deliverable | ranked list of pharmacologically addressable nodes in the shared resistance state, for combination testing with a menin inhibitor in resistant PDX models |
| cost tier | low |
| compute needs | modest — single Explorer CPU node; no GPU required |
inputs
[ "Transcriptomes from upfront-menin-inhibitor-resistant KMT2A::AFF1 patient samples and PDX models (public availability to be confirmed)", "BET-resistant MLL-AF9 clone expression data from the two 2015 Nature studies", "WNT/beta-catenin and MYC-restoration signature gene sets from those studies", "Single-cell CUT&Tag chromatin heterogeneity data at KMT2A oncofusion binding sites", "KMT2A-r patient RNA-seq with outcome (TARGET, St. Jude Cloud) for clinical anchoring" ]
guards
[ "Bidirectional enrichment required for a positive call", "Negative-control signatures of matched size (interferon, cell cycle, hypoxia) to estimate the false-positive rate; proliferation signatures correlate with nearly everything in resistance data", "Explicit ortholog mapping and batch handling for the mouse-to-human comparison" ]
tests hypotheses
[ "hyp-cell-state-plasticity-limits-interface-disruption", "hyp-aff1-condensates-dispensable-for-kmt2a-fusion-function" ]
- Cross-dataset signature concordance is easy to manufacture and hard to trust; without the pre-registered negative controls this analysis is more likely to produce a confident wrong answer than no answer.
- If the shared state is WNT-driven, the obvious combination partner (WNT inhibition) targets a pathway essential to normal intestinal and haematopoietic stem cells, so the fix for efficacy may create the toxicity.
- A positive result identifies a state, not a target; states are harder to drug than proteins, and the analysis should not be presented as if it yields a candidate molecule.
- Are the upfront-resistance transcriptomes publicly deposited? With only two patients and five PDX models, if the data are not available the analysis cannot proceed and the priority shifts to generating equivalent data in-house.
- How should the discrepancy between the two BET-resistance studies (MYC restored versus Myc unaltered) be handled? It may reflect different resistance routes converging on WNT, which would actually strengthen the state hypothesis, or it may mean the composite signature is not coherent.
- Does WNT/beta-catenin activation appear in menin-inhibitor-resistant samples at all? This is the single most direct question the analysis answers and the one whose answer most changes what the lab does next.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
Heavily pretreated KMT2A::AFF1 ALL samples were markedly less menin-inhibitor-sensitive than earlier samples from the same patients with no prior class exposure, with transcriptomics showing sustained on-target efficacy in resistant cells; RAS-pathway and TP53 comutations were present but neither was sufficient in vitro, and KMT2C/D downregulation was implicated in one patient.
Upfront menin-inhibitor resistance in multiply pretreated leukemias · Mahdavi L, Alikarami F, Goodrow H, et al. (Bernt KM, senior author) · Experimental Hematology 153:105268 2025What this evidence cannot show
Two serial patients plus five PDX models — the small n is the main limit on this analysis's power, and public availability of the underlying transcriptomes must be confirmed before committing.
Evidence that argues against this (1)
- The best-documented clinical resistance mechanism to menin inhibition is genetic, not a cell state. Perner et al. identified somatic MEN1 mutations at the revumenib-menin interface in patients with acquired resistance, reproduced them in xenografts and in an unbiased base-editor screen, and showed they attenuate drug binding while preserving the MLL1 interaction and preventing inhibitor-induced eviction of menin and MLL1 from chromatin. A transcriptional cell-state signature will not detect that mechanism and risks mislabelling MEN1-mutant samples. Perner F et al. 2023
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Resistant leukemias under BET inhibition are uniformly characterised by rapid restoration of MYC transcription through activation and recruitment of WNT signalling components and a focal MYC enhancer, and PRC2 suppression promotes BET-inhibitor resistance by remodelling pathways that restore MYC rather than by altering BRD4-dependent transcripts.
Transcriptional plasticity promotes primary and acquired resistance to BET inhibition · Rathert P, Roth M, Neumann T, et al. (Vakoc CR, Zuber J, senior authors) · Nature 525(7570):543-547 2015What this evidence cannot show
MLL-AF9;NrasG12D murine model plus human lines; a BET inhibitor is not a menin inhibitor, so cross-modality concordance is the hypothesis under test rather than an assumption.
Evidence that argues against this (1)
- Resistance to an epigenetic inhibitor in this disease has also arisen through plain pharmacology. Campbell et al. showed pinometostat treatment-emergent resistance in KOPN-8 was driven by upregulation of the ABCB1/P-glycoprotein efflux pump, with no target mutation; NOMO-1 resistance was efflux-independent and unresolved. An efflux mechanism produces no chromatin or cell-state signature at all, so a signature-based classifier would call those cells sensitive. Campbell CT et al. 2017
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
I-BET-resistant MLL-AF9-immortalised murine HSPC clones emerged from leukaemia stem cells, were cross-resistant to JQ1 and genetic BET knockdown, were not efflux- or metabolism-mediated, showed globally reduced chromatin-bound BRD4 with unaltered Myc expression, and were resensitised in vitro and in vivo by negative regulation of WNT/beta-catenin signalling.
BET inhibitor resistance emerges from leukaemia stem cells · Fong CY, Gilan O, Lam EYN, et al. (Dawson MA, senior author) · Nature 525(7570):538-542 2015What this evidence cannot show
Murine clones under in-vitro selection. Note the apparent tension with the companion study: Myc unaltered here versus MYC restoration there, which the analysis must address rather than paper over.
Evidence that argues against this (1)
- Same caution applied to the I-BET leukaemia-stem-cell model: the finding that resistance emerges from LSCs with globally reduced chromatin-bound BRD4 and unaltered Myc is a mouse MLL-AF9 result, and the clinical KMT2A-r resistance that has actually been observed in patients on this pathway's drugs is either a target-interface mutation (Perner 2023) or, in pretreated ALL, resistance with the drug still fully on target (Mahdavi 2025). Neither maps cleanly onto a BRD4-occupancy cell state. Mahdavi L et al. 2025
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Single-cell CUT&Tag at KMT2A oncofusion binding sites shows a subset carry bivalent H3K4me3/H3K27me3 signatures with cell-to-cell heterogeneity suggestive of lineage plasticity, and aberrant gene-body H3K4me3 enrichment at these sites is sensitive to menin inhibitors.
Automated CUT&Tag profiling of chromatin heterogeneity in mixed-lineage leukemia · Janssens DH, Meers MP, Wu SJ, et al. (Ahmad K, Henikoff S, senior authors) · Nature Genetics 53(11):1586-1596 2021What this evidence cannot show
The paper does not study drug resistance; the link from heterogeneous sites to a resistance reservoir is entirely this analysis's hypothesis.
Evidence that argues against this (1)
- Single-cell CUT&Tag bivalency is a snapshot of heterogeneity, not a demonstration that the heterogeneous state is selected by drug. Whether a bivalent subpopulation survives menin inhibition has not been shown; the resistance that has been demonstrated in patients on this drug is an MEN1 interface mutation (Perner 2023), which is invisible to a chromatin-state signature. Perner F et al. 2023
supplies the shared non-genetic escape signature needed to score route 4
- confers resistance to →MEN1 / menin — the adaptor the fusion cannot work without
Counter-evidence (Perner 2023, PMID 36922589): the documented clinical resistance mechanism is a somatic MEN1 interface mutation, which a transcriptional or chromatin cell-state signature cannot detect.
- confers resistance to →Pinometostat (EPZ-5676) — the DOT1L inhibitor, and the most instructive failure in this field
Counter-evidence (Campbell 2017, PMID 28428443): ABCB1/P-glycoprotein efflux drove pinometostat resistance in KOPN-8 with no target mutation - a resistance route with no cell-state signature.
- ← testsAdaptive, non-genetic resistance — the leukemia keeps the fusion and rewires how it is maintained
the signature analysis is the readout that would detect this resistance class
a menin-dependency signature scored on paired diagnosis and relapse infant samples is the direct test