Cell state in KMT2A-rearranged leukaemia is organised by lineage, not by the fusion — five independent axes split the same way, and every one of them splits AML from ALL rather than partner from partner
also called lineage-not-fusion hypothesis · metabolic split by lineage · the AML/ALL divide in cell state
Across five completely different kinds of measurement — energy use, fat requirements, oxygen response, how cells respond to the newest drug, and which survival protein they depend on — the myeloid and lymphoid forms of this leukaemia behave differently, while different fusion partners within one form behave alike. That pattern suggests the cell's identity, not the fusion, is what sets these states.
The same thing, for a scientist
This beat wrote nine state entries without setting out to test anything, and the same split appeared in all of them. **Metabolism:** MLL/AF9 AML lines are OXPHOS-dependent and metformin-sensitive, while in MLL-rearranged B-ALL, OXPHOS inhibition promotes leukaemia-initiating-cell emergence and hypoxia/glycolysis inhibition impairs it [E1][E2]. **Lipid:** the KMT2Ar AML dependency is ACSL4 and polyunsaturated fatty acid utilisation; the MLL-AF4 infant ALL requirement is ELOVL1, saturated and monounsaturated elongation [E3]. **Hypoxia:** every genetic HIF experiment in this lesion is myeloid and negative; every supportive functional result is lymphoid. **Drug response:** menin inhibition kills KMT2A-rearranged ALL rapidly and differentiates KMT2A-rearranged AML slowly [E4]. **Apoptosis:** KMT2A-rearranged infant ALL is BCL-2-dependent where most childhood ALL is BCL-XL-co-dependent, and the whole apoptotic-priming literature in this disease is lymphoid. The hypothesis is that **the fusion sets which genes are addressable and the cell's lineage sets what it then does with them** — so a state-level therapeutic claim should be stratified by lineage first and by partner second, which is the reverse of how this field and this graph have organised their thinking. The hypothesis is falsifiable and the falsifiers are cheap, because most of the required comparisons are re-analyses of existing public data.
The pattern, laid out
| Axis | Myeloid finding | Lymphoid finding | Same experiment in both? |
|---|---|---|---|
| Energy metabolism | OXPHOS-dependent; metformin-sensitive 1 | OXPHOS inhibition promotes LIC emergence; glycolysis/hypoxia inhibition impairs LICs 2 | No |
| Lipid | ACSL4 / PUFA utilisation 3 | ELOVL1 / saturated VLCFA elongation [E3 counter] | No |
| Hypoxia | Hif-1a and Hif-2a deletion accelerates leukaemogenesis; dispensable for maintenance | hypoxia pathway active in residual cells; inhibition sensitises to chemotherapy | No |
| Menin inhibition | slow differentiation 4 | rapid death 4 | Yes |
| Apoptotic dependence | MCL-1 emphasis in TP53-mutant resistance | BCL-2-dependent, BCL-XL-independent | No |
The fourth row is the only one where a single study measured both arms side by side, and it is the row where the split is cleanest. That is suggestive in two directions at once: it is the best evidence for the hypothesis, and it is also a warning that the other four rows may be splitting because nobody ran the comparison rather than because the biology differs.
The confound that must be named. Fusion partner distribution differs systematically between the two diseases: KMT2A::AFF1 is overwhelmingly lymphoid, KMT2A::MLLT3 predominantly myeloid. So every AML-versus-ALL comparison in this table is also a partner comparison, and the hypothesis as stated cannot be separated from its rival — 'state splits by partner' — using any of the cited studies.
The rival hypothesis, and the discriminating case
Rival: state splits by fusion partner, and the apparent lineage split is an artefact of partner distribution.
The discriminating cases are the fusions that occur in both lineages in appreciable numbers. KMT2A::MLLT1 (t(11;19)) presents as both ALL and AML; KMT2A::MLLT10 and KMT2A::AFDN likewise occur across lineages, and mixed-phenotype acute leukaemia carries the same lesions in a cell that is both at once.
So the test is: take one partner, compare its lymphoid and myeloid cases on one of these state axes. If the state follows lineage, the hypothesis holds. If it follows partner, the rival holds. If MPAL cases are intermediate, the state is being set continuously by lineage programme rather than switched by the lesion.
This graph already holds mechanism-kmt2a-mpal-mixed-phenotype and a KMT2A::MLLT1 control-arm folder with 26 entries, so the material for this comparison is partly assembled.
A third possibility should be kept open: neither, with the split tracking the model system. Almost all myeloid work is mouse MLL-AF9 and almost all lymphoid work is human primary or PDX, so 'myeloid versus lymphoid' is partly 'mouse versus human'.
Why it matters practically
If true, the operational consequences are specific:
- Trials that enrol 'KMT2A-rearranged acute leukaemia' as one stratum are pooling two different state biologies. This graph already records partner-blind eligibility as a risk; lineage-blind combination selection is a second and possibly larger one.
- Metabolic and apoptotic combination partners should be chosen per lineage. The clearest case: an OXPHOS-directed agent is supported in myeloid disease and is predicted to expand the initiating compartment in lymphoid disease.
- Read-across from MLL-AF9 mouse work to infant ALL is unsafe on state questions specifically, even where it is safe on chromatin questions. That is a narrower and more defensible claim than a blanket warning.
- The menin-interface hypothesis is not directly threatened — the interface is a protein-protein contact, not a state — but the therapeutic argument built on it is, because 'release the fusion and the cell differentiates' is a myeloid sentence.
The five axes are five different laboratories, and that was checked rather than assumed
This hypothesis rests on five metabolic and pharmacological axes all splitting AML from ALL. "Five independent axes" is only worth anything if they are independent groups, and this project has just built the instrument to check that - the author-clustering analysis in ana-how-many-laboratories-is-this-graph-standing-on.
Last authors, fetched 2026-09-20:
| axis | last author |
|---|---|
| OXPHOS dependence, myeloid | Khandanpour C |
| hypoxic/glycolytic vulnerability, lymphoid | Rowe RG |
| ACSL4 / PUFA lipid state | Wurm AA |
| fetal origin and its metabolic programme | Ottersbach K |
| menin-inhibitor response splits by lineage | Stam RW |
Five distinct groups. The convergence is not one laboratory's framework reappearing in five papers, which is the failure mode the ledger warned about and which this graph has found before.
One qualification worth keeping visible. The persister evidence cited elsewhere in this beat also comes from Stam RW, so that group contributes two of the beat's load-bearing results - and Stam (14 papers) and Pieters (13) are the fifth and seventh largest groups in the whole graph. The lineage-split axis they supply is the one about the drug, which is also the most consequential. It is independent of the other four; it is not independent of this graph's general reliance on the Rotterdam/Utrecht programme.
| statement | In KMT2A-rearranged leukaemia, cell-state properties (metabolic, lipid, hypoxic, apoptotic, and drug-response) are determined primarily by the lineage the blast occupies rather than by the fusion partner it carries; the fusion determines which genes are addressable, the lineage determines what is done with them |
|---|---|
| discriminating test | take a fusion that occurs in both lineages in appreciable numbers — KMT2A::MLLT1, KMT2A::MLLT10, KMT2A::AFDN — and compare its lymphoid and myeloid cases on one state axis. State following lineage supports this hypothesis; state following partner supports the rival; MPAL cases intermediate suggests continuous setting by lineage programme |
| cheapest version | re-analyse the ACSL4 dependency call in the same public CRISPR screen data restricted to KMT2A-rearranged LYMPHOID lines; the data exist and the comparison has not been reported |
| what it does not threaten | the menin-KMT2A interface itself, which is a protein-protein contact and not a state; what it threatens is the therapeutic narrative built on it, because 'release the fusion and the cell differentiates' is a myeloid sentence |
rival hypotheses
[ "state splits by FUSION PARTNER, and the lineage split is an artefact of partner distribution (supported by fusion-partner-specific binding patterns correlating with ALL/AML prevalence, PMID 39472576)", "state splits by MODEL SYSTEM: myeloid work is mostly mouse MLL-AF9 and lymphoid work is mostly human primary or PDX", "state splits by ASSAY ENDPOINT: bulk proliferation versus limiting-dilution engraftment moved in opposite directions in the one study measuring both" ]
experiments
[ "exp-dynamic-bh3-profiling-under-menin-inhibition", "exp-elutriation-phase-and-size-resolved-kmt2a" ]
analyses
[ "ana-cell-state-enumeration" ]
falsifiers
[ "ACSL4 is an equally selective dependency in KMT2A-rearranged ALL lines as in KMT2A-rearranged AML lines in the same screen dataset", "within KMT2A::MLLT1, lymphoid and myeloid cases show the SAME metabolic or apoptotic state, while KMT2A::AFF1 and KMT2A::MLLT3 lymphoid cases differ from each other", "limiting-dilution engraftment after OXPHOS inhibition in a KMT2A-rearranged MYELOID model shows the same LIC-promoting inversion reported in B-ALL, making the split an endpoint artefact", "menin inhibition produces rapid death rather than differentiation in a KMT2A-rearranged AML sample, or slow differentiation in a KMT2A-rearranged ALL sample, when partner is held constant", "an isogenic system expressing one fusion in a lymphoid and a myeloid background shows no state difference on any of the five axes" ]
atlas refs — the numbers behind the chart
[
{
"label": "OXPHOS dependence in MLL/AF9 AML",
"pmid": "35158754",
"title": "High Metabolic Dependence on Oxidative Phosphorylation Drives Sensitivity to Metformin Treatment in MLL/AF9 Acute Myeloid Leukemia.",
"authors": "Liu L, Patnana PK, Xie X, et al.",
"venue": "Cancers (Basel)",
"year": 2022,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"label": "OXPHOS inhibition promotes LIC emergence in MLL-r B-ALL",
"pmid": "35476984",
"title": "Hypoxic, glycolytic metabolism is a vulnerability of B-acute lymphoblastic leukemia-initiating cells.",
"authors": "Morris V, Wang D, Li Z, et al.",
"venue": "Cell Rep",
"year": 2022,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"label": "ACSL4/PUFA dependency in KMT2Ar AML",
"pmid": "41712378",
"title": "ACSL4-associated lipid metabolism is a distinct therapeutic vulnerability in KMT2A-rearranged acute myeloid leukemia.",
"authors": "Schäfer S, Rahimian E, Schmitz-Hübsch L, et al.",
"venue": "Cell Rep",
"year": 2026,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"label": "ELOVL1 fatty acid requirement in MLL-AF4 infant ALL",
"pmid": "34706236",
"title": "Defining the fetal origin of MLL-AF4 infant leukemia highlights specific fatty acid requirements.",
"authors": "Symeonidou V, Jakobczyk H, Bashanfer S, et al.",
"venue": "Cell Rep",
"year": 2021,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"label": "ALL dies, AML differentiates, under revumenib",
"pmid": "38892207",
"title": "Distinct Responses to Menin Inhibition and Synergy with DOT1L Inhibition in KMT2A-Rearranged Acute Lymphoblastic and Myeloid Leukemia.",
"authors": "Adriaanse FRS, Schneider P, Arentsen-Peters STCJM, et al.",
"venue": "Int J Mol Sci",
"year": 2024,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
},
{
"label": "Fusion-partner-specific binding correlates with ALL versus AML prevalence",
"pmid": "39472576",
"title": "MLL oncoprotein levels influence leukemia lineage identities.",
"authors": "Janssens DH, Duran M, Otto DJ, et al.",
"venue": "Nat Commun",
"year": 2024,
"title_provenance": "Title/authors/venue backfilled from PubMed esummary on 2026-09-20 because this citation stored a bare PMID with no title, which made it unscoreable by verify-citations.py. BACKFILLING IS NOT VERIFICATION: it confirms the PMID resolves to a real paper, not that this is the right paper for the claim. That still requires reading."
}
]- If true, combination partners for a menin inhibitor must be chosen per lineage, and the current practice of enrolling 'KMT2A-rearranged acute leukaemia' as one stratum pools two state biologies into one efficacy estimate.
- An OXPHOS-directed agent is supported in the myeloid arm and predicted to expand the initiating compartment in the lymphoid arm, so a single trial of such an agent across the class could show net benefit while harming the infants in it.
- The hypothesis licenses scepticism about read-across from mouse MLL-AF9 to infant ALL on state questions, which if applied too broadly would discard a large and otherwise sound literature — the claim should stay narrow.
- It also cuts against this graph's own habit of organising by fusion partner, which means accepting it would require re-reading a substantial number of existing entries for lineage confounding.
- Is there a single published dataset in which one fusion partner is compared across lineages on any state axis? If not, the field has never run the discriminating experiment, and the hypothesis and its rival are both currently unfalsified.
- Does the ACSL4 dependency call hold in KMT2A-rearranged lymphoid lines within the same screen data? This is a re-analysis, not an experiment, and it is the cheapest available falsifier.
- How much of the apparent split is mouse-versus-human? A census of model systems by lineage across the cited studies would quantify the confound before any biology is inferred.
- If the fusion sets addressability and the lineage sets use, what is the molecular mediator? The obvious candidates are the lineage transcription factor repertoire at the same fusion-bound loci, which is measurable by footprinting in existing data.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
AML cells with an MLL/AF9 genotype showed upregulated mitochondrial activity and depended mainly on oxidative phosphorylation for energy production within a metabolically heterogeneous AML cell-line panel, and metformin repressed their proliferation by inhibiting mitochondrial respiration.
High Metabolic Dependence on Oxidative Phosphorylation Drives Sensitivity to Metformin Treatment in MLL/AF9 Acute Myeloid Leukemia · Liu L, Patnana PK, Xie X, Frank D, Nimmagadda SC, Rosemann A, et al., Khandanpour C · Cancers (Basel) 2022What this evidence cannot show
Cell lines only, small genotype arm, lineage not controlled for; monocytic identity is an unexcluded alternative explanation. Metformin's in vitro concentrations exceed achievable plasma levels. Read: abstract only.
Evidence that argues against this (1)
- The lymphoid counterpart is not merely different in degree but opposite in direction for the compartment that matters: in MLL-rearranged B-ALL, inhibiting oxidative phosphorylation blunted blast proliferation while PROMOTING leukaemia-initiating cell emergence, and inhibiting hypoxia and glycolysis impaired LICs and gave therapeutic benefit in xenotransplantation. A single metabolic instruction for 'KMT2A-rearranged leukaemia' would therefore be right in one lineage and actively harmful in the other. Read: abstract only. Morris V et al. 2022
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In MLL-rearranged B-ALL studied by single-cell transcriptomics and quantitative xenotransplantation, inhibiting oxidative phosphorylation blunted blast proliferation but promoted leukaemia-initiating cell emergence, whereas inhibiting hypoxia and glycolysis impaired LICs and provided therapeutic benefit; engraftable LICs were abundant relative to reported AML frequencies and could emerge from more mature fractions as well as from phenotypically undifferentiated cells.
Hypoxic, glycolytic metabolism is a vulnerability of B-acute lymphoblastic leukemia-initiating cells · Morris V, Wang D, Li Z, Marion W, Hughes T, Sousa P, et al., Rowe RG · Cell Rep 2022What this evidence cannot show
Murine xenograft niche; agents not named or separated in the abstract; fusion partner not reported, which is exactly the confound this hypothesis has to contend with, since MLL-rearranged B-ALL is dominated by KMT2A::AFF1. Read: abstract only.
Evidence that argues against this (1)
- The difference may be one of assay compartment rather than of lineage. This is the only cited study that measured both bulk proliferation and limiting-dilution engraftment under the same drug, and those two endpoints moved in opposite directions within a single lineage. If the myeloid studies had measured initiating-cell frequency rather than proliferation, they might have found the same inversion, in which case the split in the table is a split between endpoints and not between diseases. No myeloid study in this lesion reports limiting-dilution engraftment after OXPHOS inhibition. Read: abstract only (same paper). Morris V et al. 2022
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
ACSL4 is a selective vulnerability of KMT2A-rearranged AML: its loss impairs growth of KMT2Ar but not non-KMT2Ar AML cells, reduces colony formation in patient-derived KMT2Ar AML and murine MLL-AF9 cells, delays leukaemia onset in vivo, depletes polyunsaturated lipid species, and is rescued by exogenous polyunsaturated fatty acid supplementation.
ACSL4-associated lipid metabolism is a distinct therapeutic vulnerability in KMT2A-rearranged acute myeloid leukemia · Schafer S, Rahimian E, Schmitz-Hubsch L, Shaikh M, Brilloff S, Kufrin V, et al., Wurm AA · Cell Rep 2026What this evidence cannot show
Myeloid only, screen-derived with the monocytic lineage confound unexcluded, and the accompanying prognostic signature is derived from the genotype it correlates with. Read: abstract only.
Evidence that argues against this (1)
- The lymphoid lipid requirement identified in this disease is a different enzyme acting on a different substrate class: ELOVL1, a very-long-chain elongase of saturated and monounsaturated fatty acids, emerged from a conserved fetal signature and was validated as critical for survival of MLL-AF4-positive ALL cells. Two lipid dependencies that cannot be satisfied by the same intervention is the cleanest single illustration of this hypothesis - and also its weakest link, because the two studies used entirely different discovery strategies (dependency screen versus fetal signature) and would likely have found different genes in the same cells. Read: abstract only. Symeonidou V et al. 2021
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Testing revumenib side by side in KMT2A-rearranged ALL and AML, ALL samples showed rapid dose-dependent induction of leukaemic cell death while AML responses were much slower and promoted myeloid differentiation.
Distinct Responses to Menin Inhibition and Synergy with DOT1L Inhibition in KMT2A-Rearranged Acute Lymphoblastic and Myeloid Leukemia · Adriaanse FRS, Schneider P, Arentsen-Peters STCJM, Fonseca AMND, Stutterheim J, Pieters R, et al., Stam RW · Int J Mol Sci 2024What this evidence cannot show
This is the ONLY axis in the hypothesis where one study measured both lineages under one protocol, which is why it carries most of the weight - and it is in vitro, with unstated n and unresolved fusion partner, so lineage and partner are confounded within it as everywhere else. Read: abstract only.
Evidence that argues against this (1)
- Fusion-partner-specific chromatin binding, not lineage, is offered as the organising variable by the most direct measurement available. Profiling oncoprotein target sites across 36 MLL-rearranged samples found that at high expression the oncoprotein preferentially activates either a pro-B-enriched ALL programme or an HSC-enriched AML programme, and that the fusion-partner-specific binding patterns over those gene sets are HIGHLY CORRELATED with the prevalence of each mutation in ALL versus AML. That is the rival hypothesis with evidence: the partner determines which programme is activated, and the lineage distribution follows from it rather than causing it. Read: abstract only. Janssens DH et al. 2024
- E-independencestrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.analysis
The five axes supporting the lineage-split hypothesis come from five distinct research groups - Khandanpour, Rowe, Wurm, Ottersbach and Stam - so the convergence is not one laboratory's framework restated.
Clustering the graph's cited papers into research groups by author overlap, and recounting independence per entry · Oncofusion knowledge graph (internal analysis) 2026What this evidence cannot show
Group identity is assigned by last author, a proxy that merges two people sharing a surname and initial and splits a rotating senior-author slot. For these five the names are distinctive and the topics differ, so the risk of a spurious split or merge is low - but **five groups is not five independent confirmations of one claim**: each supplies a different axis, and no axis has been replicated by a second group.
Evidence that argues against this (2)
- Independence of *sources* is not independence of *reasoning*. All five axes split AML from ALL, and the named confound - that partner distribution differs between the lineages, with MLLT3 myeloid-weighted and AFF1 lymphoid - applies to all five equally. Five groups making the same confounded comparison is still one confound. Adriaanse FRS et al. 2024
- The most consequential axis, the drug-response split, rests on a single in vitro study in a low-impact venue. The finding is stated plainly in its abstract and the group is a leader in infant KMT2A-r ALL, but no independent replication of the kill-versus-differentiate contrast exists. Adriaanse FRS et al. 2024
metabolic axis: OXPHOS dependence in myeloid, LIC-promoting OXPHOS inhibition in lymphoid
lipid axis: ACSL4/PUFA in myeloid against ELOVL1/saturated elongation in lymphoid
the only axis measured in both lineages under one protocol, and the cleanest split: ALL dies, AML differentiates
apoptotic axis: the BCL-2-dependent, BCL-XL-independent exception is lymphoid and the whole priming literature in this disease is lymphoid
MPAL carries the same lesions in a cell that is both lineages at once, so it is the discriminating case between this hypothesis and the fusion-partner rival
a lineage-stratified priming measurement is a direct test of the hypothesis on the apoptotic axis
the enumeration is where the lineage split was noticed: it recurred across five independently searched axes