Is KMT2A::AFF1 leukemia epithelial or endothelial? No — it is haematopoietic. But the question is a good one

An hESC-based in vitro differentiation model with enforced (lentiviral/transgenic) fusion expression, which is a very imperfect proxy for prenatal human haematopoiesis: expression level, timing and cellular context are all artificial. The readout is precursor emergence and lineage output, not leukemia; no leukemic transformation is reported. The reciprocal AF4-MLL fusion is expressed in only about half of t(4;11) patients, so the cooperativity described applies to a patient subset at best.

• Enhanced hemato-endothelial specification in a hESC system is not evidence that the disease originates there. The same laboratory found MLL-AF4 in human cord-blood CD34+ HSPCs insufficient to initiate leukaemia in vivo, and adding activated KRAS did not change that. Specification phenotypes in directed differentiation and transformation in a patient are different endpoints. Read: abstract only. Montes R et al. 2011

Ectopic constitutive expression in hESCs, a system with no immune microenvironment, no niche and a differentiation trajectory that only approximates human embryonic haematopoiesis. The endothelial skewing is a differentiation-output phenotype in vitro and has never been recapitulated in a human embryo or in a patient sample. The negative result (no transformation) is the most robust part of this paper and is the part most often left out when it is cited.

• Explaining lineage or cell-type specificity by the fusion's DNA-binding module is not supported. Replacing the MLL CXXC domain in MLL-AF9 with CXXC or MBD domains that cannot bind the same DNA still targeted the fusion to the Hoxa9 locus in primary bone-marrow progenitors; what differed was whether specific CpGs were protected from methylation, and hence whether leukaemia developed. Locus addressing and leukaemogenic function separate, so an endothelial-versus-haematopoietic argument built on where the fusion can bind is arguing from the wrong property. Read: abstract only. Risner LE et al. 2013
• No KMT2A-fusion model reproduces the human disease from this route: KMT2A fusions 'fail to phenocopy short latency leukaemogenesis in mammalian models, suggesting additional factors are necessary for transformation', and loss of a different family member, KMT2C, blocks hemogenic endothelium specification in hPSCs altogether. Hemato-endothelial output in these systems is sensitive to genetic background as well as to the fusion. Read: abstract only. Maurya S et al. 2022

Mouse, explanted/ex vivo imaging of a dissected embryo preparation rather than a live intact animal; the emerging cells are defined by surface phenotype in the imaging field rather than by transplantation of the individual imaged cell, so 'HSC' is phenotypic. Human EHT has not been imaged directly and is inferred from single-cell transcriptomics and marker studies.

• Haemogenic endothelium is plural, not singular. Fate mapping in mouse identified a separate early HE wave in the vitelline and umbilical arteries generating fetal-restricted lympho-myeloid progenitors distinct from the definitive HSCs that emerge from the aortic floor. 'Blood is born from a vessel wall' is true of more than one vessel wall and more than one output. Read: abstract only. Barone C et al. 2026

Mouse conditional genetics; Cre-driver specificity (VE-cadherin-Cre, Vav1-Cre) sets the resolution of the conclusion and both drivers have documented off-lineage activity. Does not prove that every definitive HSC passes through an endothelial state, only that Runx1 is required in an endothelial-cadherin-expressing compartment.

• Mouse conditional deletion establishing Runx1's requirement in VE-cadherin-positive endothelium is developmental biology being used to argue about tissue restriction of a leukaemogenic fusion. The two are connected only if the fusion's permissive cell is the haemogenic endothelial one, which is the hypothesis rather than the evidence. Oncofusion knowledge graph (internal analysis) 2026

A commercial clinical-sequencing cohort, so ascertainment is biased toward advanced/refractory disease and the denominators are not population-based. Detection by hybrid-capture DNA and targeted RNA panels; functional oncogenicity of KMT2A::MAML2 was not tested. n = 11 is small, and this is the only verified recurrent KMT2A fusion in a non-haematopoietic tumour located during this search — it should not be generalised into a claim that KMT2A fusions are common in solid tumours.

• A KMT2A fusion in an epithelial tumour does not show that the leukaemic fusion could act in epithelium. KMT2A::MAML2 carries a different 3' partner with a different effector mechanism, and domain-swap work shows that what makes an MLL fusion leukaemogenic is not simply where the N-terminal portion goes — substituted CXXC domains still targeted MLL-AF9 to Hoxa9 yet abolished leukaemogenicity. Partner identity, not the shared KMT2A N-terminus, is doing the tissue-specific work. Read: abstract only. Risner LE et al. 2013

A 1994 cloning paper: it establishes the fusion and its structure, not oncogenic sufficiency, which came later. 'Most' ALCL is an overstatement by current classification — ALK-positive ALCL is one of several ALCL entities, and NPM1::ALK accounts for the majority but not all ALK rearrangements in it.

• These are KINASE fusions and the subject of this graph is a chromatin fusion. A constitutively active tyrosine kinase transforms by catalysis in the cytoplasm and works in almost any cell that expresses its substrates, which is why ALK and NTRK fusions recur across lymphoma, lung, thyroid, breast and soft tissue. A KMT2A fusion works by occupying chromatin and sustaining elongation, which requires the target loci to be present and the partner's complexes to be available. The two classes are not evidence about each other, and using the kinase-fusion literature to argue about tissue restriction of KMT2A fusions crosses that boundary. Nature 2012

The discovery frequency of 6.7% comes from a small Japanese cohort (n = 75) and overstates the ~3-5% seen in unselected NSCLC populations. Transformation was demonstrated in 3T3 fibroblasts, not in lung epithelium, so this paper proves the fusion is transforming, not that it is transforming specifically in an epithelial context.

• These are KINASE fusions and the subject of this graph is a chromatin fusion. A constitutively active tyrosine kinase transforms by catalysis in the cytoplasm and works in almost any cell that expresses its substrates, which is why ALK and NTRK fusions recur across lymphoma, lung, thyroid, breast and soft tissue. A KMT2A fusion works by occupying chromatin and sustaining elongation, which requires the target loci to be present and the partner's complexes to be available. The two classes are not evidence about each other, and using the kinase-fusion literature to argue about tissue restriction of KMT2A fusions crosses that boundary. The EML4-ALK transformation assay is also 3T3 fibroblasts and nude mice, a system in which many oncogenes transform. Nature 2012

Three cases with FISH and immunohistochemistry plus four previously reported cases with 2p22-24 involvement — a very small series, and the specific 5' fusion partners (later shown to include TPM3, TPM4, CLTC and others) were not identified here. 'Recurrent' rests on a handful of cases.

• These are KINASE fusions and the subject of this graph is a chromatin fusion. A constitutively active tyrosine kinase transforms by catalysis in the cytoplasm and works in almost any cell that expresses its substrates, which is why ALK and NTRK fusions recur across lymphoma, lung, thyroid, breast and soft tissue. A KMT2A fusion works by occupying chromatin and sustaining elongation, which requires the target loci to be present and the partner's complexes to be available. The two classes are not evidence about each other, and using the kinase-fusion literature to argue about tissue restriction of KMT2A fusions crosses that boundary. Nature 2012

Three tumours in the original series; frequency and specificity were established by later work. The paper proposes rather than demonstrates a transforming mechanism.

• These are KINASE fusions and the subject of this graph is a chromatin fusion. A constitutively active tyrosine kinase transforms by catalysis in the cytoplasm and works in almost any cell that expresses its substrates, which is why ALK and NTRK fusions recur across lymphoma, lung, thyroid, breast and soft tissue. A KMT2A fusion works by occupying chromatin and sustaining elongation, which requires the target loci to be present and the partner's complexes to be available. The two classes are not evidence about each other, and using the kinase-fusion literature to argue about tissue restriction of KMT2A fusions crosses that boundary. ETV6-NTRK3 is the closest of these to a chromatin fusion in having a transcription-factor dimerisation domain, and what that domain contributes is oligomerisation of a kinase — which is a mechanism KMT2A::AFF1 does not use. Nature 2012

Secretory breast carcinoma is rare (13 cases here), so this is the strongest available series but still small. The mouse mammary epithelial transformation used retroviral overexpression, which does not reproduce endogenous fusion stoichiometry. This is the single most useful citation in this node for the cell-of-origin argument, because the identical fusion is shown to produce both a mesenchymal and an epithelial tumour.

• These are KINASE fusions and the subject of this graph is a chromatin fusion. A constitutively active tyrosine kinase transforms by catalysis in the cytoplasm and works in almost any cell that expresses its substrates, which is why ALK and NTRK fusions recur across lymphoma, lung, thyroid, breast and soft tissue. A KMT2A fusion works by occupying chromatin and sustaining elongation, which requires the target loci to be present and the partner's complexes to be available. The two classes are not evidence about each other, and using the kinase-fusion literature to argue about tissue restriction of KMT2A fusions crosses that boundary. The breast result is the strongest form of the entry's argument and the weakest analogy for it: a kinase fusion crossing from mesenchymal to epithelial tissue is expected, because catalysis does not require a lineage-specific environment. Nature 2012

Molecular characterisation of a small number of tumours; NTRK1 rearrangement is a minority event in papillary thyroid carcinoma (RET/PTC and BRAF V600E dominate). No DOI is indexed for this 1997 article; the record was verified by PMID.

• These are KINASE fusions and the subject of this graph is a chromatin fusion. A constitutively active tyrosine kinase transforms by catalysis in the cytoplasm and works in almost any cell that expresses its substrates, which is why ALK and NTRK fusions recur across lymphoma, lung, thyroid, breast and soft tissue. A KMT2A fusion works by occupying chromatin and sustaining elongation, which requires the target loci to be present and the partner's complexes to be available. The two classes are not evidence about each other, and using the kinase-fusion literature to argue about tissue restriction of KMT2A fusions crosses that boundary. TRK-T1 and TRK-T2 are described as thyroid-SPECIFIC, which cuts against the entry: even among kinase fusions, tissue restriction exists and is not explained by the catalytic mechanism alone. Nature 2012

An in vitro hESC/OP9 co-culture system; marker boundaries defined there do not automatically transfer to the human embryo, and CD43 is not used this way in routine clinical flow cytometry. Useful for interpreting hemato-endothelial experiments, not for diagnosing a patient's leukemia.

• CD43 separating haematopoietic from endothelial and mesenchymal populations was established in human embryonic stem cell differentiation cultures, so it is a marker validated in a model of the transition. The separation it achieves is operational for that culture and does not establish a sharp lineage boundary in an embryo. Oncofusion knowledge graph (internal analysis) 2026