Oncofusion

start / graph / fusion-cbfa2t3-glis2

Fusionstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q11-comparative

CBFA2T3::GLIS2, inv(16)(p13.3q24.3)

also called CBFA2T3-GLIS2 · ETO2-GLIS2 · CBFA2T3/GLIS2 · inv(16)(p13.3;q24.3) · C/G fusion · RAM-phenotype AML

In some infants and toddlers, a single chromosome folds back on itself and joins a gene that normally silences blood genes to a gene that normally relays a developmental signal, and the hybrid switches on a growth signal that makes immature platelet-precursor cells copy themselves endlessly. It is one of the most aggressive childhood leukaemias, it has a distinctive fingerprint on the flow cytometer, and it has produced the most promising immune target in this whole family of leukaemias.

The same thing, for a scientist

A cryptic **inv(16)(p13.3q24.3)** — not to be confused with the classical inv(16)(p13;q22) that makes CBFB::MYH11 — fuses **CBFA2T3/ETO2** (O75081), a member of the ETO corepressor family, to **GLIS2** (Q9BZE0), a GLI-similar zinc-finger transcription factor in the Hedgehog/Gli family. It was found by transcriptome sequencing of non-Down-syndrome acute megakaryoblastic leukaemia (non-DS-AMKL) in **27% of paediatric cases**, and expression in Drosophila and murine haematopoietic cells **induces BMP signalling and markedly increases progenitor self-renewal** [E1]. It is **overwhelmingly paediatric** — a single adult case has now been reported [E2] — and carries a grim prognosis with reported overall survival of **15-30%** across essentially all case series [E2]. Its flow-cytometric correlate is the **RAM immunophenotype**, defined in part by bright **CD56**: in the COG AAML0531 cohort the RAM cluster (n=16) was 63% CBFA2T3-GLIS2-positive and had a **5-year EFS of 19%**, versus 69% and 39% for the two other CD56-positive clusters [E5]. Modelling the fusion in human cord-blood HSPCs on an endothelial-cell niche recapitulates the transcriptome, morphology and immunophenotype of the human disease and identified **FOLR1** as a CAR-T target with preclinical efficacy [E3].

What it looks like
loading 9DE2…
PDB 9DE2 ↗
shape

This is real experimental data — atoms placed by crystallography or cryo-EM, deposited by the people who did the experiment.

I have never read one of these pictures before

A protein is one long chain of amino acids that folds up. The ribbon traces that chain from one end to the other; it is not a wire, it is a simplification of where the backbone runs. Drag to rotate, scroll to zoom.

Domains are the stretches of chain that do a particular job, and they are what the coloured blocks above the picture mark out. Click one and the view flies to it. In this project the important question about any domain is usually whether the fusion still has it — the labels say lost where it does not.

Anything in yellow sticks is a bound small molecule, which is usually the drug.

Go deeper
What is kept and what is lost1 deeper

Kept from CBFA2T3/ETO2 (5-prime partner): the nervy homology regions NHR1, NHR2 and NHR3. Two of these matter enormously:

  • NHR2 is an oligomerisation domain — in the paralogue RUNX1T1/ETO it forms a tetramer whose structure is solved and whose integrity is critical for AML1-ETO activity (PDB 1WQ6) 6. CBFA2T3 is the closest ETO-family paralogue, so the fusion is expected to be an oligomer.
  • NHR4 / the MYND domain binds the corepressors NCOR/SMRT — the ETO-family MYND-SMRT complex structure exists for AML1-ETO (PDB 2ODD) and an ETO2 MYND structure bound to a GATAD2A peptide has been deposited (PDB 9DE2) 6.

Whether NHR4/MYND is retained depends on the exact junction; the published architecture retains the ETO2 N-terminal and central NHRs.

Lost from CBFA2T3: depending on breakpoint, part or all of the C-terminal MYND region.

Kept from GLIS2 (3-prime partner): the five C2H2 zinc fingers that constitute GLIS2's DNA-binding domain, plus the C-terminus.

Lost from GLIS2: the N-terminal region.

The architecture is therefore: an oligomerising corepressor scaffold fused to a Gli-family DNA-binding domain. That makes the fusion, on paper, a tf-repression-corruption device of the same broad class as ETV6::RUNX1 and RUNX1::RUNX1T1 — a DNA-binding module bolted onto a corepressor-recruiting, self-oligomerising platform. But the measured output is not simple repression: the fusion induces BMP signalling and drives self-renewal 1, and it produces a distinctive activated transcriptional programme with aberrant surface-antigen expression (CD56-bright, FOLR1) 35. The graph records the class as tf-repression-corruption with an explicit note that the functional readout is a gain of a BMP/self-renewal programme.

Why the ETO-family NHR2 oligomer is the comparative link to RUNX1::RUNX1T1

CBFA2T3 (ETO2) and RUNX1T1 (ETO) are paralogues in the same corepressor family, and both appear as fusion partners in paediatric AML: RUNX1T1 as the 3-prime partner in t(8;21) RUNX1::RUNX1T1, and CBFA2T3 as the 5-prime partner in inv(16)(p13.3q24.3) CBFA2T3::GLIS2.

That symmetry is a gift to a comparative beat. In both cases the ETO-family protein contributes:

  • NHR2-mediated oligomerisation (tetramer in AML1-ETO; PDB 1WQ6, whose primary citation states the tetramer is critical for AML1-ETO activity) 6;
  • NHR4/MYND-mediated NCOR/SMRT corepressor recruitment (PDB 2ODD for AML1-ETO MYND with SMRT; PDB 9DE2 for ETO2 MYND) 6.

What differs is which module supplies the DNA address: in t(8;21) it is RUNX1's Runt domain (a resident haematopoietic address); in inv(16)(p13.3q24.3) it is GLIS2's zinc fingers (a Hedgehog/Gli developmental address that a megakaryoblast has no business reading). That is the same 'foreign-specificity import' pattern seen in TCF3::HLF, and it correlates with the same clinical consequence: the fusion that imports a foreign developmental DNA-binding specificity is the more aggressive one. RUNX1::RUNX1T1 is a favourable-risk AML; CBFA2T3::GLIS2 has 15-30% survival 2.

This is a genuine, testable comparative hypothesis for the Queen: among fusions sharing a corepressor/oligomerisation scaffold, aggressiveness tracks with how foreign the imported DNA-binding specificity is to the cell of origin.

Mechanism: BMP signalling and progenitor self-renewal, in a niche-dependent way

The primary functional finding. Expression of CBFA2T3::GLIS2 in Drosophila and in murine haematopoietic cells induced bone morphogenic protein (BMP) signalling and produced a marked increase in the self-renewal capacity of haematopoietic progenitors 1. BMP is a developmental morphogen pathway, consistent with importing a Gli-family DNA-binding specificity: the fusion is running a developmental signalling programme inside a blood progenitor.

The niche dependency is a distinctive and under-appreciated feature. Human transformation required context: expression of the fusion in human cord-blood haematopoietic stem and progenitor cells in an endothelial-cell co-culture system recapitulated the transcriptome, morphology and immunophenotype of CBFA2T3::GLIS2 AML and produced highly aggressive leukaemia in xenografts 3. The authors explicitly conclude that the endothelial niche promotes leukaemic transformation by this fusion 3.

That matters for two reasons:

  1. Modelling. Fusion-transduced human HSPCs alone had not reliably transformed; the niche was the missing ingredient. Any comparative fusion programme that fails to transform human cells should ask whether it is missing a niche, not whether the fusion is insufficient.
  2. Biology. A niche requirement suggests the fusion's programme is incomplete without an extrinsic signal, which is itself a therapeutic hypothesis (target the niche interaction) and a bite-back (the niche may protect residual disease).

No catalytic activity. Like most fusions in this beat, this is a protein-protein and protein-DNA assembly with no enzyme active site and no ligand pocket.

Clinical phenotype: infant AMKL, the RAM phenotype, CD56, and a dismal outcome

Disease: non-Down-syndrome acute megakaryoblastic leukaemia (AMKL, FAB M7), though 2 notes it is 'not exclusively restricted to the FAB M7 subgroup' and is also found in non-megakaryoblastic paediatric AML.

Age — overwhelmingly, but not exclusively, paediatric. This entry previously said the fusion had never been identified in adults 2; that is withdrawn (see the correction section dated 2026-09-19). One adult case is now published, and an inducible model shows adult bone-marrow HSCs are transformable but with a myeloid-shifted phenotype and strikingly longer latency. The fusion characterises a paediatric leukaemia subtype and is vanishingly rare in adults. It occurs 'primarily in infants and early childhood' 3. In a beat about comparative developmental timing, a fusion this strongly skewed to infancy is evidence that its cell of origin is most permissive in a developmental window — most plausibly a fetal/infant megakaryocyte-erythroid-biased progenitor — rather than evidence that the adult compartment is non-permissive.

Frequency: 27% of paediatric non-DS-AMKL in the discovery cohort (transcriptome sequencing of 14 diagnostic samples plus a validation cohort of 34 paediatric and 28 adult AMKL) 1; 13/105 (12.4%) in an independent European paediatric AMKL screen 4. It is described as the most frequent chimeric oncogene identified to date in non-DS-AMKL 2.

Cytogenetically cryptic — the inv(16)(p13.3q24.3) is not seen on routine karyotype, and it is easily confused in name with the classical, prognostically favourable inv(16)(p13;q22)/CBFB::MYH11. These are completely different lesions with opposite prognoses. Any diagnostic or database pipeline that collapses 'inv(16)' into one category will mis-stratify these children catastrophically.

The RAM immunophenotype and CD56. Unsupervised hierarchical clustering of immunophenotypic expression profiles in 769 children with de novo AML (COG AAML0531) identified three distinct CD56-positive clusters:

  • Cohort 1 (n=77): 72% t(8;21) — 5-year EFS 69%
  • Cohort 2 (n=52): 69% 11q23/KMT2A — 5-year EFS 39%
  • Cohort 3, the RAM phenotype (n=16): 63% CBFA2T3-GLIS2 — 5-year EFS 19% 5

The RAM phenotype is characterised in part by high-intensity CD56 expression 5. The practical value is enormous: a flow-cytometry pattern available on day one flags a cryptic fusion that karyotype will miss, months before RNA-seq returns in many centres. The comparative point is that CD56 alone is uninformative — the same antigen appears in three clusters with EFS of 69%, 39% and 19% — and only the multidimensional immunophenotype separates them 5.

Prognosis: grim, with overall survival ranging between 15% and 30% in almost all reported case series 2, and 5-year EFS 19% for the RAM cluster 5. Within the paediatric AMKL cohort in 4 (whole-cohort 5-year OS 42 +/- 6%), CBFA2T3::GLIS2 was one of the four defining fusion classes.

Druggability: emerging, and the route is immunotherapy not small molecules

Status: emerging.

The fusion itself is undruggable for the familiar reasons: an oligomerising corepressor scaffold plus a zinc-finger DNA-binding domain, no enzyme, no ligand pocket. GLIS2 has no experimental structure at all (an RCSB search on Q9BZE0 returns zero entries), and CBFA2T3 has exactly one (9DE2, the ETO2 MYND domain bound to a GATAD2A peptide) 6.

The productive route has been surface-antigen discovery. Interrogating the transcriptomes of fusion-transduced cord-blood cells and of primary CBFA2T3::GLIS2 AML produced a library of fusion-specific candidate targets; folate receptor alpha (FOLR1) was selected, CAR-T cells were built against it, and preclinical efficacy was demonstrated in vitro and in xenografts 3.

This is the single most encouraging therapeutic result for any of the nine non-enzyme fusions in this beat, and the strategic lesson is explicit: when the driver has no pocket, read out the programme it enforces and attack a surface antigen it induces. The authors themselves generalise the approach to 'a variety of oncogenic fusion-driven pediatric leukemias' 3.

Other mechanism-derived hypotheses: the induced BMP signalling 1 is in principle targetable (BMP receptor kinase inhibitors exist preclinically), and the endothelial niche dependency 3 suggests niche-interaction targets. Neither is clinical.

The honest caveat, from the authors: FOLR1 is also expressed in renal and pulmonary epithelium, raising toxicity concerns that must be addressed before clinical application 3. That is the generic failure mode of fusion-derived surface antigens — they are fusion-enriched, not fusion-exclusive.

Correction 2026-09-19: the fusion is not paediatric-exclusive

This entry previously asserted, in its expert summary, in its clinical section, in evidence item E2, in data.frequency_note, in data.stage_of_hematopoiesis, in data.age_peak and in its open questions, that CBFA2T3::GLIS2 is "paediatric-exclusive — never identified in adults" and that it "characterises a leukaemia subtype specific to paediatrics, having never been identified in adults". All of those statements are withdrawn.

What settles it. Huang J, Wu T, Zhu H, et al. Concurrent BCR::ABL1 and CBFA2T3::GLIS2 fusions in an adult patient with acute leukaemia. Lancet Haematol 2026;13(9):e690. PMID 42660135, doi 10.1016/S2352-3026(26)00105-5. It is a single short report with no abstract, so it establishes existence and nothing about frequency — but the withdrawn statement was an existence claim.

Why the absence claim was fragile from the start, and the node said so. E2's own caveats already flagged that 'never identified in adults' is an absence claim that depends on how systematically adult AMKL has been screened with a cryptic-fusion-capable assay. Adult AMKL is rare; the inv(16)(p13.3q24.3) is cytogenetically cryptic; and adult series were not routinely screened by RNA sequencing. The absence was of ascertainment.

The biology never demanded exclusivity either. Lopez et al. (Cancer Discov 2019, PMID 31662298) showed with an inducible transgenic model that CBFA2T3-GLIS2 in fetal HSCs gives rapid megakaryoblastic leukaemia while the same fusion in adult bone-marrow HSCs still transforms, with a shift toward myeloid phenotype and strikingly delayed latency. Adult cells are less permissive, not non-permissive — which predicts exactly the case profile now reported.

What survives. The extreme paediatric skew, the infant/early-childhood age peak, the cryptic cytogenetics, the RAM/CD56 immunophenotype and the poor prognosis are all unchanged. data.adult_amkl_pct has been changed from 0 to null: the correct statement is that no adult series large enough to estimate a rate has been screened with a fusion-capable assay, so the value is unknown, not zero. The developmental-window argument survives in the weaker and more defensible form — a permissiveness gradient, not an on/off switch.

Analogy: one cytogenetic label, two diseases, opposite risk - and a second analogue that shares only the cell

Closest analogue, and the axis. fusion-cbfb-myh11, on an axis that is not biological at all: shared nomenclature. Both lesions are written inv(16). They are different breakpoints on the same chromosome - p13.1q22 for CBFB::MYH11, p13.3q24.3 for CBFA2T3::GLIS2 - different genes, different diseases, and opposite risk assignments: CBFB::MYH11 is favourable risk in the 2022 ELN classification 7, while CBFA2T3::GLIS2 reports 15-30% overall survival across essentially all series 2.

This is a real analogy, and it is the useful kind, because the resemblance is entirely in the label and the consequence is clinical. Two features make it dangerous rather than merely untidy:

  • CBFA2T3::GLIS2 is cryptic - it was found by transcriptome sequencing and is invisible to routine karyotyping 1. So the failure mode is not that a technician mis-reads one inversion as the other; it is that the true lesion is not seen at all, and any inv(16) string that does appear in the record belongs to the other one.
  • The direction of the error is the worst available. Mistaking a 15-30%-survival infant AMKL for favourable-risk CBF-AML de-escalates therapy.

What the analogy predicts, and it has been acted on rather than tested. If the label is the problem, then subtype assignment must be made by a fusion-detecting assay and never by cytogenetic nomenclature. That is precisely how CBFA2T3::GLIS2 was discovered 1 and how it is now diagnosed. The unmeasured quantity is how many children were mis-assigned before RNA sequencing became routine, and this graph has no source for it.

Where the analogy breaks: everywhere except the string. There is no shared 5-prime partner, no shared 3-prime partner, no shared mechanism class (ETO-family corepressor plus a GLI-similar zinc finger driving BMP signalling, against a disrupted core-binding-factor heterodimerisation subunit), no shared lineage and no shared age. A nomenclatural analogy is the one kind that predicts nothing about biology and everything about data handling - which is why it belongs in a knowledge graph rather than in a review.

A second analogue on a real axis: fusion-nup98-kdm5a, shared CELL OF ORIGIN. CBFA2T3::GLIS2 (12-27% of paediatric non-Down-syndrome AMKL) and NUP98::KDM5A (10.5% of paediatric AMKL) are the two commonest cryptic fusions of the same disease, in the same cell, both karyotypically invisible, and mechanistically unrelated. The disanalogy is the one that matters for risk modelling: CBFA2T3::GLIS2 is dismal wherever it is measured, whereas NUP98::JARID1A was not an independent prognostic factor within the same already-poor AMKL cohort 4. Two drivers, one cell of origin, one bad disease - and only one of them adds risk on top of the lineage. That is an argument that a large part of the AMKL outcome belongs to the cell and the age rather than to the driver, and it is the same shape as the lineage-versus-fusion question the coordinator's cycle-39 log raises for KMT2A.

Facts and figures
gene5CBFA2T3
gene3GLIS2
cytobandinv(16)(p13.3q24.3)
cytoband noteCRITICAL: this is NOT the classical inv(16)(p13;q22) that produces CBFB::MYH11. Same chromosome, different breakpoints, opposite prognosis (favourable CBF-AML versus 15-30% survival). Any pipeline that collapses 'inv(16)' into one category will mis-stratify these children.
frequency noteThe two paediatric figures disagree and both are reported honestly: 27% in the 14-case transcriptome discovery cohort [E1] versus 13/105 (12.4%) in an independent European AMKL screen [E4]. The true value in non-DS-AMKL is probably between them; the discovery cohort was very small. adult_amkl_pct is null rather than 0: the fusion DOES occur in adults - one case is published [E2] - but no adult series large enough to estimate a percentage has been screened with a cryptic-fusion-capable assay, so the value is unknown rather than zero.
prognosisGrim, and overwhelmingly a paediatric disease (one adult case reported, see the correction section dated 2026-09-19): overall survival 15-30% across almost all reported case series [E2], and 5-year EFS 19% for the RAM-immunophenotype cluster that is 63% fusion-positive (n=16) [E5]. Within a paediatric AMKL cohort whose whole-cohort 5-year OS was 42 +/- 6%, it is one of the four defining fusion classes [E4].
uniprot5O75081
uniprot3Q9BZE0
uniprot3 noteGLIS2 is Q9BZE0 (verified: reviewed human entry, 524 aa, 'Zinc finger protein GLIS2'). Note that the accession Q9H4L1 is INACTIVE (merged to Q9NP70) and is not GLIS2 - a trap for anyone working from older annotation.
breakpoints noteExon-level breakpoints for this cryptic inversion are not well catalogued in the sources we verified; treat the exon numbers as unverified conventions. The reliable statements are the cytoband and the domain architecture [E1].
PDB noteStructural coverage is essentially absent. GLIS2 (Q9BZE0) has ZERO PDB entries, so the DNA-binding module of this fusion has no experimental coordinates. 9DE2 is the only CBFA2T3/ETO2 entry: the ETO2 MYND domain bound to a peptide from GATAD2A. The oligomerisation and corepressor structures usually cited for this fusion family (1WQ6, the NHR2 tetramer; 2ODD, MYND-SMRT) are from the PARALOGUE RUNX1T1/ETO and are homology arguments, not structures of this protein - they are deliberately omitted from this list [E6].
mechanism class noteArchitecturally a tf-repression-corruption device (oligomerising ETO-family corepressor scaffold plus an imported zinc-finger DNA-binding domain), but the measured functional output is a GAIN: induced BMP signalling and increased progenitor self-renewal [E1], with an activated surface-antigen programme (CD56-bright, FOLR1) [E3][E5]. Like TCF3::HLF this is really a 'foreign-specificity import' - a developmental (Hedgehog/Gli) DNA-binding specificity dropped into a megakaryoblast - and the vocabulary has no term for it.
druggability classemerging
druggability noteThe fusion is undruggable (no enzyme, no pocket, and GLIS2 has no structure at all). The emerging designation rests entirely on immunotherapy: FOLR1 was identified from the fusion-specific transcriptome and anti-FOLR1 CAR-T cells showed preclinical efficacy, with FOLR1 expression in renal and pulmonary epithelium as the unresolved toxicity [E3]. BMP-pathway inhibition [E1] and targeting the endothelial-niche dependency [E3] are untested hypotheses.
stage of hematopoiesisFetal/infant megakaryocyte-erythroid-biased haematopoietic stem/progenitor cell. The strongest argument is the extreme paediatric skew [E2] plus the direct ontogeny experiment, in which fetal HSCs gave rapid megakaryoblastic leukaemia while adult bone-marrow HSCs gave myeloid-shifted disease after strikingly longer latency - implying a cell of origin that is far more permissive in a developmental window, not one that is absent afterwards. Human transformation required cord-blood HSPCs PLUS an endothelial-cell niche, and that combination reproduced the disease transcriptome, morphology and immunophenotype [E3]
age peakinfants and early childhood (primarily under 2-3 y); vanishingly rare in adults, with a single reported adult case
developmental order rank1
diseases
[
  "pediatric non-Down-syndrome acute megakaryoblastic leukemia (AMKL, FAB M7)",
  "pediatric non-megakaryoblastic AML (minority)"
]
frequency
{
  "pediatric_non_ds_amkl_pct": 27,
  "pediatric_amkl_independent_cohort_pct": 12.4,
  "adult_amkl_pct": null
}
breakpoints
[
  {
    "id": "bp-cbfa2t3-nhr-glis2-zinc-fingers",
    "gene5_exon_last": 11,
    "gene3_exon_first": 3,
    "frequency_pct": 100,
    "population": "pediatric non-Down-syndrome AMKL",
    "domains_retained_5": [
      "NHR1 (nervy homology region 1, TAF110-like)",
      "NHR2 (oligomerisation domain - tetramerises in the ETO paralogue)",
      "NHR3"
    ],
    "domains_lost_5": [
      "part or all of the C-terminal NHR4 / MYND domain (NCOR/SMRT corepressor binding), depending on junction"
    ],
    "domains_retained_3": [
      "five C2H2 zinc fingers (GLIS2 DNA-binding domain)",
      "GLIS2 C-terminus"
    ],
    "domains_lost_3": [
      "GLIS2 N-terminal region"
    ],
    "phenotype_notes": "The fusion is an oligomerising ETO-family corepressor scaffold bolted onto a Gli-family zinc-finger DNA-binding domain. Exon numbers are convention labels inferred from the domain architecture and were NOT verified against a breakpoint-sequence dataset; the domain-level description follows [E1]. Whether NHR4/MYND is retained varies in the literature and we do not assert a single answer - which matters, because MYND is the corepressor-recruiting surface."
  }
]
PDB IDS
[
  "9DE2"
]
mechanism class
[
  "tf-repression-corruption"
]
How this could bite us back
  • FOLR1 is fusion-enriched, not fusion-exclusive. The authors explicitly flag expression in renal and pulmonary epithelium as a toxicity that must be addressed before clinical use [E3]. Anti-FOLR1 CAR-T in an infant therefore risks on-target off-tumour renal and pulmonary injury in an organ system with little reserve - and CAR-T toxicity, unlike a small molecule, cannot simply be stopped.
  • CD56 is a trap if used alone. The same antigen appears in three immunophenotypic clusters with 5-year EFS of 69%, 39% and 19% [E5]. A centre that risk-stratifies on CD56 positivity rather than the full multidimensional phenotype will mix favourable t(8;21) AML with this disease.
  • The 'inv(16)' name collision is a live clinical hazard. inv(16)(p13.3q24.3)/CBFA2T3::GLIS2 has 15-30% survival [E2]; inv(16)(p13;q22)/CBFB::MYH11 is favourable-risk core-binding-factor AML. A reporting or database error in either direction leads to under-treatment or unnecessary transplant.
  • The lesion is cytogenetically cryptic, so it is invisible to karyotype-only workup [E1]. Diagnostic under-ascertainment is itself a driver of the poor outcomes attributed to the biology.
  • If the endothelial niche promotes transformation and maintenance [E3], the niche is also a protected reservoir: therapy that clears circulating and marrow-space blasts may leave niche-resident cells, which is a mechanism for the relapse pattern this disease shows.
  • BMP-pathway inhibition is the obvious mechanism-derived small-molecule idea [E1], but BMP signalling is required for skeletal, cardiac and haematopoietic development. Inhibiting it in an infant is a developmental on-target risk of the same character as inhibiting NSD1 or CBF.
Still open
  1. Why is this fusion so nearly absent from adult leukaemia [E2]? The adult-versus-fetal HSC experiment has now been done in an inducible mouse model and gave a graded answer - transformation still occurs from adult bone-marrow HSCs, but with myeloid-shifted phenotype and much longer latency. The open question is therefore narrower and better: is the residual adult permissiveness enough to explain the observed adult case rate, or does the human endothelial niche [E3] contribute a second, independent developmental restriction? Running the adult-derived versus cord-blood HSPC comparison in the same human niche system is still the clean experiment.
  2. Is the endothelial-niche requirement [E3] a modelling artefact or a genuine dependency of the established leukaemia? If the latter, the niche interaction is a therapeutic target and also a sanctuary site for residual disease - and it would explain why this leukaemia is so refractory.
  3. Is NHR4/MYND retained in the fusion, and does it matter? The literature is not consistent on whether the corepressor-binding MYND domain survives the junction, and since MYND is the NCOR/SMRT-recruiting surface [E6], the answer determines whether this fusion is a corepressor device at all or purely an oligomerising DNA-binding scaffold. This is a resolvable ambiguity that the graph should close.
  4. Does the NHR2 oligomerisation domain tetramerise in CBFA2T3::GLIS2 as it does in AML1-ETO [E6], and is oligomerisation required for transformation? If so, the NHR2 interface becomes a shared target across two paediatric AML fusions from the same corepressor family - one of the few genuinely cross-fusion targets in this beat.
  5. The frequency estimates differ two-fold between cohorts (27% versus 12.4%) [E1][E4]. Is that ascertainment, AMKL case definition, or geography? It matters for trial design in a disease this rare.
  6. How many children with CBFA2T3::GLIS2 were risk-stratified as favourable because a karyotype or a report said inv(16)? The fusion is cryptic and the other inv(16) is favourable-risk, so the misclassification is mechanically possible and its direction is de-escalation; no source in this graph quantifies it. A retrospective re-analysis of banked paediatric AMKL and non-DS AML samples originally reported as inv(16)-positive, by fusion-capable RNA sequencing, would give the number and is an archive question rather than a laboratory one.
Every source behind this entry (7)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Transcriptome sequencing of non-Down-syndrome AMKL identified a cryptic inv(16)(p13.3q24.3) encoding a CBFA2T3-GLIS2 fusion protein in 27% of paediatric cases; expression in Drosophila and murine haematopoietic cells induced BMP signalling and markedly increased haematopoietic progenitor self-renewal.

    An Inv(16)(p13.3q24.3)-encoded CBFA2T3-GLIS2 fusion protein defines an aggressive subtype of pediatric acute megakaryoblastic leukemia · Gruber TA, Larson Gedman A, Zhang J, Koss CS, Marada S, Ta HQ, Chen SC, Su X, Ogden SK, Dang J, Wu G, Gupta V, Andersson AK, Pounds S, Shi L, Easton J, Barbato MI, Mulder HL, Manne J, Wang J, Rusch M, Ranade S, Ganti R, Parker M, Ma J, Radtke I, Ding L, Cazzaniga G, Biondi A, Kornblau SM, Ravandi F, Kantarjian H, Nimer SD, Dohner K, Dohner H, Ley TJ, Ballerini P, Shurtleff S, Tomizawa D, Adachi S, Hayashi Y, Tawa A, Shih LY, Liang DC, Rubnitz JE, Pui CH, Mardis ER, Wilson RK, Downing JR · Cancer Cell 2012
    PMID 23153540DOI
    What this evidence cannot show

    The 27% figure comes from a discovery cohort of only 14 transcriptome-sequenced diagnostic samples, with a separate recurrency/validation cohort; the functional work is in Drosophila and murine cells with ectopic expression, so the BMP induction is demonstrated in model systems rather than in human blasts, and the causal chain from BMP signalling to leukaemia is not dissected.

    Evidence that argues against this (2)
    • The 27% does not survive a larger, unselected series. Central pathology review of Children's Oncology Group trials AAML0531 and AAML1031 identified 107 confirmed non-Down-syndrome AMkL cases (5.5% of trial AML), in which CBFA2T3::GLIS2 accounted for 16% - second to RBM15::MRTFA at 20%, ahead of NUP98 fusions at 10% and KMT2A rearrangements at 7%. The discovery estimate came from 14 diagnostic transcriptomes plus a validation set; it is the small-cohort end of a range whose larger-cohort end is 12-16%. The node already records the 12.4% European figure [E4]; the COG series is a third independent cohort landing in the same lower band, which makes 27% the outlier rather than the consensus. Chisholm KM et al. 2023
    • The mechanistic half is partly superseded, and the substitution is not cosmetic. Thirant et al. dissected the fusion in human and mouse cells and located megakaryocytic identity in the GLIS2 moiety while showing that BOTH ETO2 and GLIS2 domains are required for increased self-renewal; the fusion binds DNA directly and controls transcription by upregulating and interacting with the ETS-related factor ERG at ENHANCER elements, and specific interference with ETO2-GLIS2 OLIGOMERISATION reverses transcriptional activation at those enhancers and promotes megakaryocytic differentiation. BMP signalling, the axis reported from Drosophila and murine progenitors, is not the operative mechanism in that account. The self-renewal half of the claim stands; the BMP half is a species- and system-specific finding that the human work replaced with an enhancer-and-oligomerisation model. Thirant C et al. 2017
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.review

    CBFA2T3-GLIS2 is the most frequent chimeric oncogene identified to date in non-Down-syndrome AMKL, is overwhelmingly a paediatric lesion (a single adult case has been reported), is not exclusively restricted to the FAB M7 subgroup, and is associated with overall survival of 15-30% in almost all reported case series.

    CBFA2T3-GLIS2-positive acute myeloid leukaemia. A peculiar paediatric entity · Masetti R, Bertuccio SN, Pession A, Locatelli F · Br J Haematol 2018
    PMID 30592296DOI
    What this evidence cannot show

    A narrative review, so the 15-30% survival range is a summary across heterogeneous small case series with differing therapy and eras rather than a pooled analysis; the review's 'never identified in adults' was an absence claim that depended on how systematically adult AMKL had been screened with a cryptic-fusion-capable assay; it has since been falsified by a published adult case and is withdrawn from this entry (see counter_evidence and the correction section dated 2026-09-19).

    Evidence that argues against this (2)
    • FALSIFIED, in the strict sense. Huang et al. report an adult patient with acute leukaemia carrying concurrent BCR::ABL1 and CBFA2T3::GLIS2 fusions (Lancet Haematology, September 2026). It is a single case with no abstract, so it establishes existence and nothing about frequency - but 'has never been identified in adults' is an existence claim, and one case retires it. The absence was an absence of ascertainment rather than of biology: adult AMKL is rare, the inv(16)(p13.3q24.3) is cytogenetically cryptic and invisible to karyotype, and adult series were not routinely screened with fusion-capable RNA sequencing. The entry's paediatric-exclusivity statement has been withdrawn throughout and replaced with 'overwhelmingly paediatric; isolated adult cases reported'. Huang J et al. 2026
    • The biology never required paediatric exclusivity either, and this was shown before the adult case appeared. In an inducible transgenic model, CBFA2T3-GLIS2 expression in FETAL haematopoietic stem cells induced rapid megakaryoblastic leukaemia, whereas expression in ADULT bone-marrow haematopoietic stem cells produced a shift toward myeloid transformation with strikingly delayed leukaemogenic potential - transformation still occurred. Chromatin-accessibility and single-cell transcriptome analysis attributed the difference to ontogeny-dependent activity of ERG, SPI1, GATA1 and CEBPA. Adult cells are therefore LESS permissive, not non-permissive, and the expected clinical consequence is exactly what was eventually observed: adult cases that are rare, late and phenotypically shifted away from FAB M7. The 'developmental window' reading in this node is a statement about probability and phenotype, not about possibility. Lopez CK et al. 2019
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Expressing CBFA2T3-GLIS2 in human cord-blood HSPCs on an endothelial-cell co-culture niche recapitulates the transcriptome, morphology and immunophenotype of the human disease and gives aggressive xenograft leukaemia; interrogating the resulting transcriptome identified FOLR1 as a fusion-specific target, and anti-FOLR1 CAR-T cells showed preclinical efficacy, with the caveat that FOLR1 is also expressed in renal and pulmonary epithelium.

    CBFA2T3-GLIS2 model of pediatric acute megakaryoblastic leukemia identifies FOLR1 as a CAR T cell target · Le Q, Hadland B, Smith JL, Leonti A, Huang BJ, Ries R, Hylkema TA, Castro S, Tang TT, McKay CN, Perkins L, Pardo L, Sarthy J, Beckman AK, Williams R, Idemmili R, Furlan S, Ishida T, Call L, Srivastava S, Loeb AM, Milano F, Imren S, Morris SM, Pakiam F, Olson JM, Loken MR, Brodersen L, Riddell SR, Tarlock K, Bernstein ID, Loeb KR, Meshinchi S · J Clin Invest 2022
    PMID 36136600DOI
    What this evidence cannot show

    Entirely preclinical: in vitro co-culture and xenograft models, no clinical data. The transformation model uses transduced cord-blood cells rather than the patient's own cell of origin, and the endothelial niche is an engineered co-culture. The authors themselves flag FOLR1 expression in renal and pulmonary epithelium as an unresolved toxicity problem, so the target is fusion-enriched rather than fusion-exclusive.

    Evidence that argues against this (1)
    • Folate-receptor-alpha-directed T cells have been given to patients once, and the result was negative on every axis. In a phase 1 study, 14 women with metastatic ovarian cancer received autologous T cells carrying an anti-FRalpha single-chain antibody fused to the FcRgamma signalling domain. There was NO reduction in tumour burden in any patient; 111In-labelled cells showed a lack of specific localisation to tumour except for faint signal in one peritoneal deposit; gene-modified cells were abundant in the circulation for the first two days and barely detectable at one month in most patients; and an inhibitory serum factor developed in three of six patients tested that significantly reduced the cells' response to FRalpha-positive tumour cells. The construct generation, the co-stimulation and the disease all differ from the paediatric AMKL programme, but this is the only FRalpha-directed CAR read-out in humans that exists, and it bounds how far a preclinical xenograft result licenses the word 'functional'. Kershaw MH et al. 2006
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In an independent screen of 105 paediatric AMKL cases, CBFA2T3/GLIS2 was present in 13 (12.4%), alongside RBM15/MKL1 (n=16), KMT2A rearrangements (n=13) and NUP98/JARID1A (n=11), in a cohort with 5-year overall survival of 42 +/- 6%.

    NUP98/JARID1A is a novel recurrent abnormality in pediatric acute megakaryoblastic leukemia with a distinct HOX gene expression pattern · de Rooij JDE, Hollink IHIM, Arentsen-Peters STCJM, van Galen JF, Berna Beverloo H, Baruchel A, Trka J, Reinhardt D, Sonneveld E, Zimmermann M, Alonzo TA, Pieters R, Meshinchi S, van den Heuvel-Eibrink MM, Zwaan CM · Leukemia 2013
    PMID 23531517DOI
    What this evidence cannot show

    This study's primary focus was NUP98/JARID1A, so the CBFA2T3/GLIS2 count is incidental and the paper does not report a fusion-specific survival analysis for it. Retrospective multi-group cohort with heterogeneous therapy. The 12.4% differs from the 27% in [E1], most likely because of ascertainment and cohort composition; the true frequency in non-DS-AMKL is probably between the two.

    Evidence that argues against this (1)
    • The frequency estimates disagree substantially: 27% of paediatric non-DS-AMKL in the discovery cohort [E1] versus 13/105 (12.4%) in this independent European screen. The discrepancy is likely driven by the very small discovery cohort (14 sequenced cases) and by differences in AMKL case definition and Down-syndrome exclusion, and it means the frequency field in this node carries real uncertainty. Gruber TA et al. 2012
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Unsupervised clustering of immunophenotypic expression profiles in 769 children with de novo AML identified three distinct CD56-positive clusters: one enriched for t(8;21) (72%, 5-year EFS 69%), one for 11q23/KMT2A (69%, EFS 39%), and the RAM phenotype (n=16) enriched for CBFA2T3-GLIS2 (63%, EFS 19%); the RAM phenotype is characterised in part by high-intensity CD56 expression.

    Deciphering the Significance of CD56 Expression in Pediatric Acute Myeloid Leukemia: A Report from the Children's Oncology Group · Pardo LM, Voigt AP, Alonzo TA, Wilson ER, Gerbing RB, Paine DJ, Dai F, Menssen AJ, Raimondi SC, Hirsch BA, Gamis AS, Meshinchi S, Wells DA, Brodersen LE, Loken MR · Cytometry B Clin Cytom 2019
    PMID 31294507DOI
    What this evidence cannot show

    The RAM cluster contains only 16 patients, so the 19% EFS estimate is imprecise and 63% fusion-positivity means over a third of RAM cases are something else - the phenotype is a strong flag, not a diagnostic test. Clustering was unsupervised and the cluster boundaries are method-dependent; results come from a single trial (AAML0531) with a single reference flow laboratory, so reproducibility across centres and instruments is not established here.

    Evidence that argues against this (1)
    • The RAM-to-fusion mapping is looser outside COG. In the second published RAM-AML series, 11 of 1102 paediatric AML patients (1%) had the RAM phenotype and RNA sequencing found CBFA2T3-GLIS2 in only 3 of 7 sequenced cases (42.8%), below the 63% of the COG cluster, while none of 21 contemporaneous non-RAM AMKL patients carried it. Cytogenetic and molecular studies showed no recurrent abnormality in RAM-AML at all, and the authors had to ADD a marker - CD36, consistently negative in RAM-AML and moderate-to-bright in non-RAM AMKL (p<0.0001) - before the clusters separated cleanly on principal-component analysis. A day-one flow pattern that is 43-63% concordant with the fusion, in series of 16 and 11 patients, is a trigger for sequencing rather than a substitute for it, and the 19% EFS attached to n=16 carries a confidence interval wide enough to overlap the other two clusters. Panda D et al. 2020
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    Structural coverage of this fusion is almost nil: GLIS2 (Q9BZE0) has zero PDB entries and CBFA2T3 (O75081) has one (9DE2, the ETO2 MYND domain bound to a GATAD2A peptide); the relevant oligomerisation and corepressor structures come from the paralogue RUNX1T1/ETO (1WQ6, the NHR2 tetramer critical for AML1-ETO activity; 2ODD, the MYND domain with SMRT).

    What this evidence cannot show

    Verified by querying the RCSB search API on the reference-sequence accession field and by fetching individual entry records. 1WQ6 and 2ODD are the PARALOGUE RUNX1T1/ETO, not CBFA2T3/ETO2; using them to infer CBFA2T3 oligomerisation and corepressor binding in this fusion is a homology argument, not direct evidence. Absence of GLIS2 structures means the DNA-binding module of this fusion has no experimental coordinates at all.

    Evidence that argues against this (2)
    • Confirmed, and the one CBFA2T3 entry should be read more carefully than 'the ETO2 MYND domain bound to a GATAD2A peptide'. UniProt cross-reference queries run on 2026-09-19 return ZERO PDB entries for GLIS2 (Q9BZE0) and exactly ONE for CBFA2T3 (O75081). That one, 9DE2, is a SOLUTION NMR structure with a single 71-residue polymer entity that RCSB describes as a 'GATAD2A-MPPL motif and ETO2 NHR4 domain fusion protein' with one bound zinc - a covalently tethered single-chain construct, not two molecules in complex. Nothing about the domain's fold is thereby in doubt; what is in doubt is any inference about binding affinity, stoichiometry or the free peptide's conformation, none of which a tethered NMR construct measures. RCSB Protein Data Bank 2024
    • The GLIS2 gap is real but bounded by homology, and stating the bound changes what 'almost nil' implies for tractability. What would close it is a zinc-finger-DNA structure, and the template exists: Pavletich and Pabo solved the five zinc fingers of human GLI1 bound to a high-affinity site at 2.6 A, showing finger 1 making no DNA contact while fingers 2-5 wrap the major groove in a non-periodic arrangement, with fingers 4 and 5 making extensive base contacts over a conserved nine-base-pair region in DNA of intermediate B/A conformation. GLIS2 is a GLI-similar zinc-finger protein; its DNA-binding module is modellable on that template even though it is unsolved. The structural blindness in this node is specific to the ETO2 oligomerisation and corepressor surfaces and to the fusion junction - not to GLIS2's DNA-reading fold. Pavletich NP et al. 1993
  7. E7established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.review

    The lesion this fusion is most easily confused with sits at the opposite end of the risk scale: in the 2022 ELN genetic risk classification, AML with inv(16)(p13.1q22) or t(16;16)(p13.1;q22) / CBFB::MYH11 is placed in the FAVOURABLE risk category, whereas CBFA2T3::GLIS2 - inv(16)(p13.3q24.3) - carries reported overall survival of 15-30%.

    Diagnosis and management of AML in adults: 2022 recommendations from an international expert panel on behalf of the ELN. · Döhner H, Wei AH, Appelbaum FR, Craddock C, DiNardo CD, Dombret H, Ebert BL, Fenaux P, Godley LA, Hasserjian RP, Larson RA, Levine RL, Miyazaki Y, Niederwieser D, Ossenkoppele G, Röllig C, Sierra J, Stein EM, Tallman MS, Tien HF, Wang J, Wierzbowska A, Löwenberg B · Blood 2022
    PMID 35797463DOI
    What this evidence cannot show

    ELN 2022 is an ADULT AML classification and explicitly so; CBFA2T3::GLIS2 is an overwhelmingly paediatric disease and is not a category in it, so this citation establishes only the risk assignment of the lesion on the other side of the confusion, not a head-to-head comparison. ELN is an expert-panel consensus document, cited here as the reference standard for what the label inv(16) means in practice rather than as a primary outcome measurement.

    Evidence that argues against this (1)
    • The favourable label on the other inv(16) is not uniform either, so the contrast is between a heterogeneous good-risk group and a uniformly bad one rather than between two homogeneous categories. In 88 CBFB::MYH11 patients undergoing transplant, the D816V KIT mutation was significantly associated with post-transplant relapse - in contrast to other KIT mutation types - and remained independently associated with relapse and survival in multivariable analysis alongside CBFB::MYH11 MRD, with the combination of D816V KIT status and pre-transplant MRD stratifying risk further. A pipeline that resolved the two inv(16) lesions correctly would still be wrong to treat CBFB::MYH11 as a single favourable class. Cho BS et al. 2021
Connected to