ETO2::GLIS2 rewires chromatin in stages and hijacks the osteogenic homeobox factor DLX3 as its first required step — a genome-edited human iPSC model that resolves leukaemia initiation into an ordered sequence
also called CBFA2T3::GLIS2 iPSC initiation model · DLX3 hijack in paediatric AMKL · progressive chromatin rewiring by a paediatric fusion
Researchers built the exact chromosome inversion that causes a rare, aggressive infant leukaemia into human stem cells, then watched what happened step by step as those cells tried to become blood. The fusion protein's very first move is to switch on DLX3, a gene that normally builds bone, and if DLX3 is deleted the leukaemia never starts.
The same thing, for a scientist
The closest published analogue to what this project wants for KMT2A::AFF1, in a different fusion family. Boudia et al. **precisely reproduced the chromosome 16 inversion generating ETO2::GLIS2 in human iPSCs**; the derived haematopoietic cells showed differentiation alterations in vitro and efficiently induced leukaemia in vivo that phenocopied human acute megakaryoblastic leukaemia **by flow cytometry and by single-cell transcriptomes** [E1]. Comparing iPSC-derived with patient-derived cells revealed **altered chromatin accessibility at early and later bona fide leukaemia stages** — i.e. the rewiring is *progressive*, with an order — and identified **aberrantly higher accessibility and expression of the osteogenic homeobox factor DLX3 that PRECEDED increased accessibility to ETS factors**. DLX3 overexpression in normal CD34+ cells reproduced the motif shift (up at ETS, down at GATA), a DLX3 module was enriched in ETO2::GLIS2 AMKL and in some aggressive paediatric osteosarcoma, and **DLX3 knockout abrogated leukaemia initiation in the model** [E1]. **Three things transfer to the KMT2A::AFF1 problem and one does not.** Transferable: (1) a genome-edited human iPSC carrying the exact rearrangement is a tractable way to watch initiation in a human cell, and this corpus already holds `model-crispr-t411-human-fetal-liver` and an IMSUT iPSC line carrying KMT2A::AFF1, so the reagents exist; (2) **chromatin change is ordered, and the earliest step is not the canonical output** — for ETO2::GLIS2 the canonical story is a megakaryocytic programme and the first required event is a homeobox factor from a non-haematopoietic lineage; (3) single-cell transcriptomes were the instrument that established phenocopy, not bulk. Not transferable: **DLX3 itself.** This is a fusion-specific hijack and nothing here predicts a homeobox hijack in KMT2A::AFF1, which already carries HOXA9/MEIS1 as its canonical output.
The structural lesson: the first required step need not be the famous one
ETO2::GLIS2 (also written CBFA2T3::GLIS2) is defined clinically by paediatric acute megakaryoblastic leukaemia and mechanistically by a megakaryocytic transcriptional programme. The step that was required for initiation in this model is DLX3 — an osteogenic homeobox factor — and it acts before the ETS accessibility shift that carries the recognisable programme 1.
Ask the same question of KMT2A::AFF1. Its canonical output is HOXA9/MEIS1, and this project has spent many cycles on that axis. Is HOXA9 the first required event in a human cell acquiring the fusion, or is it, like the ETS shift here, a later and more visible consequence of something else? Nobody has looked in an ordered human model, and this corpus's own reanalysis found the menin-inhibitor response in two KMT2A::AFF1 lines did not touch HOXA or MEIS1 — an observation that is much less surprising if HOXA is downstream scenery rather than the load-bearing step.
This is the analogy that earns its place: not the same molecule, but the same shape of error, and a demonstrated method for detecting it.
Why an iPSC model and not a cell line, and what it costs
What it buys: a human cell, the exact rearrangement rather than a cDNA overexpression, a defined normal starting point, a time axis, and an isogenic control. Those five together are what make an ordered claim possible. A cell line is a snapshot of an endpoint after decades of selection and can never answer 'what happened first'.
What it costs: iPSC-derived haematopoiesis is embryonic/yolk-sac-like rather than definitive adult, which for a paediatric fusion is arguably the right ontogeny and is certainly not the adult one. That is a confound for some questions and the entire point for others — and for infant KMT2A::AFF1, where every signal this corpus has says the permissive compartment is fetal, it is closer to the point than to the confound.
The model is also slow, low-throughput, and its leukaemia readout is a xenograft. Nothing about it is a screening platform.
What the KMT2A field would need to run the same play
| component | ETO2::GLIS2 (done) | KMT2A::AFF1 (status) |
|---|---|---|
| exact rearrangement engineered in a human pluripotent or progenitor cell | yes, inv(16) in iPSC | yes — CRISPR t(4;11) in human HSPCs and in human fetal liver, and an iPSC line carrying KMT2A-AFF1 (IMSUTi002-A-2) |
| haematopoietic differentiation with single-cell readout at multiple stages | yes | not reported as an ordered series |
| matched patient-derived comparison | yes | possible; primary infant samples exist and are single-cell profiled |
| accessibility profiling early AND late | yes | the gap |
| candidate first-step factor, then knockout | DLX3, knockout abrogates initiation | nothing nominated |
Four of the five components already exist for the index fusion. The missing piece is the ordered accessibility series across differentiation in the engineered human model, and that is an experiment rather than a discovery — which is what makes this analogy worth a node.
| mechanism class | ordered chromatin rewiring during leukaemia initiation |
|---|---|
| system | genome-edited human iPSC carrying inv(16) ETO2::GLIS2, differentiated to haematopoietic cells, with xenograft leukaemia readout and patient-derived comparison |
| first required factor | DLX3 (osteogenic homeobox) |
| motif shift | DLX3 overexpression increases accessibility to ETS motifs and reduces accessibility to GATA motifs |
| shown for fusion | ETO2::GLIS2 (CBFA2T3::GLIS2) only. NOT KMT2A, NOT KMT2A::AFF1. Every transfer to the index fusion in this entry is stated as analogy. |
| cross disease link | a DLX3 transcriptional module is enriched in some aggressive paediatric osteosarcoma |
| source pmid | 39656971 |
| cited in this corpus | 0 |
assays
[ "CRISPR genome editing of the exact inversion", "scRNA-seq", "chromatin accessibility profiling (early and late)", "flow cytometry", "xenotransplantation", "DLX3 overexpression in normal CD34+ cells", "DLX3 knockout" ]
- If the first required step of a fusion's programme is not its canonical output, then a drug targeting the canonical output arrives after the decisive event has already happened. For KMT2A::AFF1 that would mean a menin inhibitor is a maintenance agent rather than a curative one — consistent with the clinical pattern of deep responses followed by relapse.
- A factor required for INITIATION is not necessarily required for MAINTENANCE, and the DLX3 knockout here was done before initiation. An initiation-only dependency is useless as a therapeutic target in a patient who already has leukaemia, and this distinction is routinely blurred in papers that report both a knockout and a clinical enrichment.
- Is there a 'DLX3 step' for KMT2A::AFF1 — a non-canonical factor whose accessibility rises before HOXA and whose loss abrogates initiation? The engineered human models exist; the ordered accessibility series does not.
- Does HOXA9 accessibility rise first or late in a human cell newly acquiring KMT2A::AFF1? The whole menin-interface argument assumes early and load-bearing, and nothing in this corpus's evidence base orders it.
- ETO2::GLIS2 hijacks a factor from a non-haematopoietic lineage (bone). Are there non-haematopoietic lineage factors in the KMT2A::AFF1 accessibility landscape that have been dismissed as noise?
- Can the ETO2::GLIS2 iPSC model be run alongside a KMT2A::AFF1 iPSC line in one differentiation experiment, giving a within-experiment comparison of two paediatric fusions' initiation programmes? Both lines are published.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The chromosome 16 inversion generating ETO2::GLIS2 was precisely reproduced in human iPSCs; derived haematopoietic cells showed in vitro differentiation alterations and induced in vivo leukaemia phenocopying human AMKL by flow cytometry and single-cell transcriptomes; chromatin accessibility was altered at early and later leukaemia stages with aberrantly higher DLX3 accessibility and expression preceding increased ETS-factor accessibility; DLX3 overexpression in normal CD34+ cells increased ETS and reduced GATA accessibility; a DLX3 module was enriched in ETO2::GLIS2 AMKL and some aggressive paediatric osteosarcoma; and DLX3 knockout abrogated leukaemia initiation.
Progressive chromatin rewiring by ETO2::GLIS2 revealed in a genome-edited human iPSC model of pediatric leukemia initiation · Boudia F, Baille M, Babin L, et al. (senior author Mercher T) · Blood 2025What this evidence cannot show
A different fusion family (ETO2::GLIS2 / CBFA2T3::GLIS2, paediatric AMKL) and a different lineage (megakaryoblastic, not B-lymphoid). iPSC-derived haematopoiesis is embryonic-like and not definitive adult haematopoiesis, so 'leukaemia initiation' here means initiation in that ontogeny. The leukaemia readout is a xenograft. DLX3's role as the FIRST required step rests on the ordering of accessibility changes in one model system plus a knockout in that same system; no human patient time-course exists and none could. Read at abstract level; figures not opened. Nothing in this paper concerns KMT2A, and every transfer to KMT2A::AFF1 stated in this entry is an analogy, flagged as such.
Evidence that argues against this (1)
- The strongest argument against importing this design to KMT2A::AFF1 is that the index fusion has repeatedly failed to transform human pluripotent- and adult-derived cells efficiently, and this corpus's own entries record that only cord-blood and fetal-liver HSPCs are permissive. An iPSC model that reproduces the rearrangement faithfully and produces no leukaemia would be uninterpretable — and that is the likelier outcome for KMT2A::AFF1 than for ETO2::GLIS2, which is precisely why the KMT2A field uses fetal liver rather than iPSC. Secker KA et al. 2020
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The reagents for the KMT2A::AFF1 version of this experiment already exist: genome-editing-induced t(4;11) translocations model B-cell precursor ALL with the KMT2A-AFF1 fusion, and a human iPSC line harbouring KMT2A-AFF1 has been generated and deposited.
Genome editing-induced t(4;11) chromosomal translocations model B cell precursor acute lymphoblastic leukemias with KMT2A-AFF1 fusion · Pan F, Sarno J, Jeong J, et al. (senior author Cleary ML) · Journal of Clinical Investigation 2024What this evidence cannot show
PMID, title, journal, DOI and author list all fetched from PubMed at the time of writing. An earlier draft of this item carried a recalled author string ('Ta VT') and a recalled DOI (10.1172/JCI161273); BOTH were wrong - the true values are Pan F, Sarno J, Jeong J, et al. (senior author Cleary ML) and 10.1172/JCI171030. Recorded because it is the third recall failure in this beat and it is exactly the failure mode the project's standing rules name. The companion iPSC line is IMSUTi002-A-2 (PMID 37639801, Stem Cell Research 2023, Takahashi M, doi 10.1016/j.scr.2023.103185). Neither paper reports an ordered accessibility series across differentiation, which is the component this entry says is missing - so this item establishes that the reagents exist, not that the experiment is partly done.
Evidence that argues against this (1)
- Possessing the reagents is not the same as the model working. A genome-edited t(4;11) that models BCP-ALL and an iPSC line carrying the fusion are both published, but neither is in routine use anywhere in this corpus's cited experimental literature, which is weak evidence that they are harder to work with than the papers imply. Takahashi M et al. 2023
- related to →CBFA2T3::GLIS2, inv(16)(p13.3q24.3)
the fusion this ordered-initiation model was built for
- analogous to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
the transferable lesson: the first REQUIRED chromatin step need not be the canonical output, and four of the five reagents for the same experiment already exist for the index fusion
- related to →CRISPR-engineered t(4;11) in human fetal liver — the only model that reproduces infant-specific biology
the KMT2A::AFF1 engineered human model that would carry the same ordered-accessibility design
- related to →HOXA9 — required for survival of established MLL-rearranged leukemia, and dispensable for starting it
raises the unasked question of whether HOXA9 is the first required step for KMT2A::AFF1 or a later consequence