RUNX1::RUNX1T1 (AML1-ETO), t(8;21)(q22;q22)

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

• Any agent aimed at the Runt-DNA or Runt-CBFB surface will inhibit wild-type RUNX1 and CBFB, which are required for definitive haematopoiesis - predictable on-target marrow failure. This is the same constraint that blocks ETV6::RUNX1, and it is why the NHR2 tetramer (present only in the fusion's ETO moiety) is the strategically better interface.
• HDAC inhibition is the obvious inference from NCOR/SMRT recruitment and has repeatedly failed to deliver a subtype-specific therapy; corepressor complexes are used across the transcriptome, so the therapeutic window is poor. The AML1-ETO9a paradox [E2] additionally suggests corepressor recruitment may not even be the essential activity.
• KIT co-mutation is adverse and the natural response is to add a KIT inhibitor, but KIT inhibition in CBF-AML is co-mutation pharmacology with the usual consequences: on-target resistance mutations, and KIT's essential roles in normal haematopoiesis, mast cells and germ cells.
• MRD-guided transplant is the current answer for the high-risk subset [E4], and transplant in a child carries graft-versus-host disease, infertility, endocrine failure and secondary malignancy. Because the comparison in [E4] was non-randomised and by treatment choice, the true benefit may be smaller than 87.4% versus 61.9% suggests, meaning some children may be transplanted for a gain that is partly selection artefact.
• The leukaemia has a prenatal or early preleukaemic reservoir detectable in cord blood at far above disease incidence (see the stage_of_hematopoiesis note), so as with ETV6::RUNX1 the fusion-positive founder clone can survive therapy and seed a genetically distinct relapse.
• Modelling risk: any preclinical programme that expresses only the canonical full-length AML1-ETO cDNA may be studying a protein that is not sufficient to cause the disease [E2]. Negative results from such models should not be trusted, and this is a general warning for every fusion in this beat where splice isoforms have not been catalogued.
• RUNX2 compensates for RUNX1 loss in this system, so a RUNX1-directed strategy in t(8;21) has a paralogue escape route already documented -- the same failure mode this project recorded for KDM4C in `hyp-paralogue-buffering-hides-dependencies`.

1. Is the NHR2 tetramerisation interface a real drug target? It is required for activity, structurally defined (1WQ6) and stoichiometric rather than diffuse [E3], which makes it far more tractable than the NUP98 FG-repeat disorder. Peptide inhibitors have been reported preclinically but nothing is clinical. Because CBFA2T3/ETO2 carries the same domain and is itself a paediatric fusion partner [E6], an NHR2 inhibitor would be one of very few genuinely cross-fusion agents - worth a dedicated target node.
2. Why does AML1-ETO9a, which LOSES the NHR4/MYND corepressor-binding domain, cause leukaemia when full-length AML1-ETO does not [E2]? If corepressor recruitment were the core mechanism, the truncated isoform should be less oncogenic, not more. Either MYND-mediated repression is not the essential activity, or the full-length protein carries a C-terminal restraint that the truncation removes. Resolving this would change how the whole tf-repression-corruption class is understood.
3. Do human t(8;21) blasts express AML1-ETO9a at a consistent ratio to the full-length transcript, and does that ratio predict MRD kinetics or relapse? Given that MRD by RUNX1-RUNX1T1 transcript level is already the paediatric risk-stratification tool [E4], isoform-resolved MRD is a cheap, immediately clinical question.
4. RUNX1::RUNX1T1 and CBFB::MYH11 attack opposite subunits of the same heterodimer and both give favourable-risk AML, yet only the beta-subunit fusion produced a PPI inhibitor [E5]. Is that because the alpha-subunit fusion's interface is functionally normal, or simply because nobody has run the equivalent screen? A direct attempt to find an AML1-ETO/CBFB interface modulator would settle it.
5. Both this fusion and ETV6::RUNX1 retain a wild-type Runt domain and are detectable prenatally in cord blood, yet one gives myeloid and the other lymphoid disease. What determines that, given an identical DNA-binding module?

1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

   t(8;21)(q22;q22) juxtaposes 5-prime sequences of AML1/RUNX1 on chromosome 21 to 3-prime sequences of ETO on chromosome 8, producing a fusion transcript; AML1 shares 69% amino-acid identity with the Drosophila segmentation gene runt over the DNA-binding region.

   Identification of breakpoints in t(8;21) acute myelogenous leukemia and isolation of a fusion transcript, AML1/ETO, with similarity to Drosophila segmentation gene, runt · Erickson P, Gao J, Chang KS, Look T, Whisenant E, Raimondi S, Lasher R, Trujillo J, Rowley J, Drabkin H · Blood 1992
   PMID 1391946
   ▸What this evidence cannot show

   Nine t(8;21) cases analysed by restriction mapping and cDNA cloning; the protein product's function was inferred from the runt homology rather than measured, and no DOI is registered for this 1992 Blood article so the PubMed record is the citable anchor.

   ▸Evidence that argues against this (1)

   • Detecting the fusion transcript is not detecting the disease, and this is the clearest published demonstration of it for t(8;21). Miyamoto et al. found AML1/ETO mRNA by RT-PCR in the marrow and/or blood of ALL 18 patients in continuous complete remission for 12-150 months (median 45) after chemotherapy or peripheral-blood stem-cell transplantation, and 51 of 2469 colonies (2.1%) grown from their clonogenic progenitors expressed it - across granulocyte-macrophage, mixed, erythroid AND megakaryocyte colonies - carrying the same X-inactivation allele as the diagnostic blasts, while constitutive haematopoiesis was polyclonal. Only the four allogeneic BMT recipients were negative. The translocation therefore sits in a multipotent progenitor that can persist benignly for over a decade in cured people; the fusion is a clonal marker, not a disease state. Miyamoto T et al. 1996
2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

   Full-length AML1-ETO is a 752-residue protein containing the RUNX1 N-terminus including its DNA-binding domain plus almost the entire ETO protein; t(8;21) occurs in 15% of all AML and 40-50% of FAB M2; full-length AML1-ETO expression does NOT cause leukaemia, whereas the alternatively spliced C-terminally truncated 575-residue AML1-ETO9a isoform causes rapid leukaemia in mice, and co-expression of both accelerates AML onset and blocks differentiation at a more immature stage.

   A previously unidentified alternatively spliced isoform of t(8;21) transcript promotes leukemogenesis · Yan M, Kanbe E, Peterson LF, Boyapati A, Miao Y, Wang Y, Chen IM, Chen Z, Rowley JD, Willman CL, Zhang DE · Nat Med 2006
   PMID 16892037DOI
   ▸What this evidence cannot show

   The leukaemogenicity comparison is in a murine retroviral transduction-transplantation model with non-physiological expression levels, so the 'full-length does not cause leukaemia' statement is model-specific - other groups have produced leukaemia from full-length AML1-ETO with added cooperating mutations or different expression systems. The relative abundance of AML1-ETO9a versus full-length transcript in human patient blasts, and whether the ratio predicts anything clinically, is not established here.

   ▸Evidence that argues against this (1)

   • The differentiation-block framing needs the native factor put back in. t(8;21) Kasumi-1 cells REQUIRE wild-type RUNX1: knockdown attenuates the mitotic checkpoint and causes apoptosis, knockdown of AML1-ETO in those cells rescues them, and an active RUNX1 allele is usually preserved in t(8;21) patients. The fusion and the native factor compete at common sites and the malignant state depends on the ratio. Ben-Ami O et al. 2013
3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

   The fusion retains the RUNX1 Runt domain, whose complex with CBFB on DNA functions as a DNA-induced molecular clamp (PDB 1H9D, 1E50, 1IO4; 1CMO is a solution-NMR structure of the isolated Runt domain, not of the clamp), and adds the ETO NHR2 oligomerisation domain, whose tetramer structure is critical for AML1-ETO activity (PDB 1WQ6), and the NHR4/MYND corepressor domain, solved in complex with SMRT (PDB 2ODD).

   PDB entries 1H9D (AML1/CBF-beta/DNA complex, 'DNA-Induced Molecular Clamp'), 1IO4 (Runt-CBFbeta heterodimer with C/EBPbeta on the CSF1R promoter), 1WQ6 ('The tetramer structure of the Nervy homology two domain, NHR2, is critical for AML1/ETO's activity') and 2ODD ('Solution structure of the MYND domain from AML1-ETO complexed with SMRT') · RCSB Protein Data Bank · RCSB PDB 2006
   ▸What this evidence cannot show

   Verified by fetching each RCSB entry record. Every entry is an isolated domain or domain pair - there is no structure of the intact AML1-ETO protein, and no structure showing the Runt domain and the NHR2 tetramer in the same molecule, so the assembled picture is a composite model. The claim that the NHR2 tetramer is critical for activity is quoted from the entry's own primary-citation title and should ideally be traced to the functional data in that paper.

   ▸Evidence that argues against this (2)

   • One of the four listed clamp structures is not a clamp structure. Fetched from RCSB on 2026-09-19, 1CMO is a SOLUTION NMR structure with a single 127-residue polymer entity mapping to UniProt Q01196 (RUNX1) and zero non-polymer entities: no CBFB chain, no DNA. It shows the isolated Runt domain in solution and cannot evidence a DNA-induced heterodimeric clamp. The clamp claim survives on 1H9D, 1E50 and 1IO4, which are genuine crystal structures of the assembly; what is withdrawn is the count and the implied method. The claim text and the PDB note have been corrected, and the same error in fusion-cbfb-myh11#E5 is corrected there. RCSB Protein Data Bank 1999
   • The NHR2 oligomerisation interface has produced peptides and no drug-like molecule in nearly two decades, and the one clinically used agent anywhere that acts at a fusion oligomerisation module (arsenic at the PML RBCC) is escaped by point mutation at its binding surface with the oncoprotein intact. The NHR2 tetramer being structurally solved is not evidence that it is chemically addressable. Goto E et al. 2011
4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

   In 129 paediatric t(8;21) AML patients stratified by RUNX1-RUNX1T1 transcript level after two consolidation cycles, allogeneic transplant improved 5-year relapse-free survival in high-risk patients (87.4% vs 61.9% with chemotherapy, p=0.026; 5-year OS 82.8% vs 71.4%, p=0.26); extramedullary infiltration at diagnosis predicted relapse (50% vs 18.4%, p=0.004); and c-KIT-mutant high-risk patients trended better with transplant (82.9% vs 75%, p=0.40).

   Allogeneic hematopoietic stem cell transplantation can improve the prognosis of high-risk pediatric t(8;21) acute myeloid leukemia in first remission based on MRD-guided treatment · Hu GH, Cheng YF, Lu AD, Wang Y, Zuo YX, Yan CH, Wu J, Sun YQ, Suo P, Chen YH, Chen H, Jia YP, Liu KY, Han W, Xu LP, Zhang LP, Huang XJ · BMC Cancer 2020
   PMID 32539815DOI
   ▸What this evidence cannot show

   Not randomised: transplant versus chemotherapy in the high-risk group was by treatment choice, so selection bias is substantial and the 87.4% versus 61.9% difference cannot be attributed to transplant with confidence. Single-centre Chinese cohort; the KIT subgroup comparison is explicitly non-significant and underpowered. The MRD threshold used to define high risk is protocol-specific and not internationally standardised.

   ▸Evidence that argues against this (1)

   • The c-KIT limb does not replicate, in a larger series from the same country. Qin et al. screened 351 core-binding-factor AML patients at diagnosis for c-KIT exon 17 and exon 8 mutations and followed 250: 36.5% were mutated, and c-KIT was an independent adverse factor for relapse, disease-free and overall survival in ADULT t(8;21) (all p<=0.016) but showed no significant effect in PAEDIATRIC t(8;21) or in adult inv(16) (all p>0.05). If c-KIT carries no prognostic weight in children with t(8;21), a 'c-KIT-mutant high-risk' subgroup trending toward transplant benefit at 82.9% versus 75% (p=0.40) in a 129-patient cohort is describing noise. The transplant comparison itself was not randomised: allocation to transplant requires surviving to transplant, and in the entry's own numbers the relapse-free survival gap (87.4% vs 61.9%, p=0.026) does not carry through to overall survival (82.8% vs 71.4%, p=0.26) - the signature of selection rather than of a treatment effect. Qin YZ et al. 2014
5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   CBFB::MYH11, the fusion that corrupts the OTHER subunit of the same core-binding-factor heterodimer, outcompetes wild-type CBFB for binding to RUNX1 and has yielded a protein-protein-interaction inhibitor (AI-10-49) that disrupts the fusion-RUNX1 interaction and restores RUNX1 transcriptional activity - demonstrating that the CBF interface is druggable in principle when the pathology is an altered affinity at that interface.

   Chemical biology. A small-molecule inhibitor of the aberrant transcription factor CBFbeta-SMMHC delays leukemia in mice · Illendula A, Pulikkan JA, Zong H, Grembecka J, Xue L, Sen S, Zhou Y, Boulton A, Kuntimaddi A, Gao Y, Rajewski RA, Guzman ML, Castilla LH, Bushweller JH · Science 2015
   PMID 25678665DOI
   ▸What this evidence cannot show

   This is evidence about a DIFFERENT fusion (CBFB::MYH11), cited here for the comparative argument about the CBF interface. It is preclinical: mouse survival delay plus selective cell death in primary patient blasts, no clinical trial data, and AI-10-49 has not advanced to an approved agent. The inference that no analogous approach works for AML1-ETO because its interface is functionally normal is our reasoning, not a published result.

   ▸Evidence that argues against this (1)

   • The AI-10-49 precedent is weaker as evidence for interface druggability than it reads. A ClinicalTrials.gov query in September 2026 finds no registered interventional study of any CBFbeta-SMMHC:RUNX1 inhibitor, eleven years after publication. 'Druggable in principle' has not converted to a trial. US National Library of Medicine 2026
6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   CBFA2T3/ETO2, a paralogue of RUNX1T1/ETO with the same NHR2 and MYND architecture, is itself a recurrent paediatric AML fusion partner (in CBFA2T3::GLIS2), so the ETO-family oligomerisation and corepressor modules appear in at least two distinct paediatric leukaemia fusions.

   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, et al., Mardis ER, Wilson RK, Downing JR · Cancer Cell 2012
   PMID 23153540DOI
   ▸What this evidence cannot show

   Cited for the paralogue relationship and the existence of a second ETO-family paediatric fusion, not for any statement about t(8;21). Whether CBFA2T3::GLIS2 actually tetramerises through NHR2 as AML1-ETO does is a homology inference and has not, to our knowledge, been demonstrated.

   ▸Evidence that argues against this (1)

   • The same modules are core machinery in normal blood, which changes what 'appears in two paediatric fusions' licenses. Goardon et al. identified ETO2/CBFA2T3 as a component of the TAL-1/SCL complex in erythroid cells, recruited through E2A/HEB alongside GATA-1, LMO2 and Ldb1; ETO2 actively represses erythroid TAL-1 target genes and governs the expansion of erythroid progenitors, and a change in ETO2 stoichiometry WITHIN that complex is what licenses terminal erythroid maturation, including induction of Gfi-1b and p21(Cip). An oligomerisation-and-corepressor module shared by two leukaemia fusions and by the normal erythroid master complex is a recurrence of physiology, not a leukaemia-specific signature; a drug against it should be expected to hit erythro-megakaryopoiesis, which is exactly the lineage these two fusions arise in. Goardon N et al. 2006
7. E7strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   t(8;21) cells require native RUNX1 for survival, and RUNX1 is required for AML1-ETO cell survival through BCL2 with RUNX2 able to compensate.

   Transcription factor RUNX1 promotes survival of acute myeloid leukemia cells. · Goyama S, Schibler J, Cunningham L, et al. · Journal of Clinical Investigation 2013
   PMID 23979164DOI
   ▸What this evidence cannot show

   Abstract only. Human cord blood transduction and mouse genetics; RUNX1 overexpression conversely inhibits normal cord blood growth, so the effect is dose-dependent and no therapeutic window is demonstrated. Concordant with PMID 24055056.

   ▸Evidence that argues against this (1)

   • The requirement is not specific to this fusion, which turns a putative vulnerability into an essentiality confound. In the very paper cited, a level of RUNX1 activity was required for the growth of MLL-AF9 cells as well as AML1-ETO cells; combined Runx1/Cbfb loss inhibited MLL-AF9 leukaemia development in mice; and the BCL2-mediated survival effect was demonstrated in MLL-fusion leukaemia. The same work shows RUNX1 overexpression INHIBITS normal cord-blood cell growth by inducing myeloid differentiation, so the dependency is dose- and context-dependent in both directions. RUNX1 dependence therefore spans at least two unrelated fusion classes plus a dose window in normal cells; read as a t(8;21)-selective target it is the lineage-essentiality confound this graph tracks elsewhere. Goyama S et al. 2013

• ← part ofRUNX1 — domain architecture and residue coordinates (with the RUNX1b/RUNX1c +27 offset)

  5-prime partner in RUNX1::RUNX1T1, t(8;21); same Runt domain, opposite orientation

• ← related toMechanism class 2 — transcription-factor corruption (RUNX1::RUNX1T1, PML::RARA, ETV6::RUNX1, CBFB::MYH11) — a balance, not a pure dominant negative

  Class correction (Ben-Ami 2013, PMID 24055056; Goyama 2013, PMID 23979164): 'dominant negative' is withdrawn for the core-binding-factor members. t(8;21) and inv(16) cells REQUIRE native RUNX1 -- knockdown kills them, AML1-ETO knockdown rescues, an active RUNX1 allele is preserved in patients -- so the pathology is a maintained balance, and both source papers propose inhibiting rather than restoring RUNX1.

• related to →The dependencies this project keeps failing to see are the buffered ones: paralogue redundancy and guide placement, not absence of biology

  Published instance of the paralogue-buffering failure mode inside core-binding-factor disease (Goyama 2013, PMID 23979164): RUNX2 compensates for RUNX1 loss in AML1-ETO and MLL-AF9 cells, so a single-gene CRISPR score understates the dependency.

• ← related toMechanism class 2 — transcription-factor corruption (RUNX1::RUNX1T1, PML::RARA, ETV6::RUNX1, CBFB::MYH11) — a balance, not a pure dominant negative

  Delivery-assay artefact (Vives 2003, PMID 12678851): apparent energy-independent TAT entry came from fixation redistribution and ionic surface binding; uptake is endocytic, so a TAT-NHR2 competitor is largely endosomal, not nucleoplasmic.

• ← related toHypothesis — forced-oligomerisation interfaces are the mutation-resistant drug targets the fusion field has been ignoring

  Delivery artefact in the NHR2 peptide evidence (Vives 2003, PMID 12678851): apparent energy-independent TAT entry came from fixation redistribution and surface ionic binding; uptake is endocytic.

• related to →Fusion-positive clones that never become leukaemia: persistence in remission, spontaneous clearance, and silent rearrangements in healthy people

  Marker-versus-disease bound (Miyamoto 1996, PMID 8639850): AML1/ETO transcripts persisted in all 18 t(8;21) patients in remission for 12-150 months, in 2.1% of clonogenic colonies across four lineages, so the fusion transcript is a clonal marker rather than a disease state.

• related to →Nobody needs to enrol anyone: the PTH question is a regression on data that already exists

  Subgroup replication failure (Qin 2014, PMID 25449688): in 351 CBF-AML patients c-KIT mutation was adverse only in ADULT t(8;21) and had no significant effect in paediatric t(8;21) (p>0.05), undercutting the 129-patient c-KIT/transplant subgroup.

• related to →Is the selectivity just pan-essentiality? No: it explains 0.4% of the variance

  Physiology confound (Goardon 2006, PMID 16407974): ETO2/CBFA2T3 is a core component of the normal erythroid TAL-1/SCL complex whose stoichiometry gates terminal maturation, so the shared ETO module is recurrent physiology, not a leukaemia-specific signature.

• related to →Ranking every KMT2A::AFF1 partner on dependency versus essentiality

  Essentiality confound (Goyama 2013, PMID 23979164): RUNX1 is also required for MLL-AF9 cell growth and for MLL-AF9 leukaemia development in mice, so native-RUNX1 dependence is not a t(8;21)-selective vulnerability.

• related to →Dataset: PDB and AlphaFold - the structural census, including the two absences that shape every plan

  Structural correction (RCSB 1CMO, fetched 2026-09-19): 1CMO is a SOLUTION NMR structure of a single 127-residue RUNX1 Runt domain with no CBFB and no DNA, so it cannot evidence the DNA-induced clamp. Entry corrected.

• ← related toAnother fusion family did the experiment this beat wants: in FUS::ERG and CBFA2T3::GLIS2 AML, MHC class I and II are absent at diagnosis, EZH2 drives it, and an EZH2 inhibitor restores presentation

  significantly worse outcomes when EZH2 overexpression co-occurs

• ← related toRUNX1 — activated by KMT2A::AFF1 and then bound by the reciprocal fusion, in the one published two-oncoprotein circuit in t(4;11)

  same protein, opposite role: 5' partner of a different fusion, and a target-programme node in KMT2A-r

• analogous to →CBFB::MYH11 (CBFbeta-SMMHC), inv(16)(p13;q22)

  THE TWO CORE-BINDING-FACTOR LEUKAEMIAS, disrupting the same transcription factor complex from its two subunits. The canonical example in this graph of one complex broken two ways giving one disease class, which is the comparison the KMT2A partner families are measured against.

• depends on →Mechanism class 2 — transcription-factor corruption (RUNX1::RUNX1T1, PML::RARA, ETV6::RUNX1, CBFB::MYH11) — a balance, not a pure dominant negative
• ← part ofRUNX1T1 (ETO / MTG8) domain architecture

  reference domain coordinates for the RUNX1T1 side of this fusion

• ← analogous toBCAT1 in core-binding-factor AML - the closest analogue in another fusion family, where TWO different fusions converge on one metabolic enzyme by rewiring the chromatin at its locus

  RUNX1::ETO converges on the same metabolic enzyme in a RUNX1-dependent manner

• ← related toKMT2A::MAML2, inv(11)(q21q23) - a therapy-related fusion that also causes a solid tumour

  one reported patient developed KMT2A::MAML2 T-lymphoblastic lymphoma after RUNX1::RUNX1T1 AML and myeloid sarcoma

• ← related toThe fusion-oncoprotein TurboID experiment this graph proposes has already been published - for PML::RARA, RUNX1::RUNX1T1, CBFB::MYH11, NUP98::HOXA9 and SET::NUP214, and not for any KMT2A fusion

  RUNX1::RUNX1T1-TurboID labelled nuclear repressor complexes

• ← related toA genuine small-molecule disruptor of an oncogenic oligomer DOES exist - 7.44 against the RUNX1::RUNX1T1 NHR2 tetramer - which corrects this graph's record; and its eighteen-year potency curve says the class is a chemistry problem, which is what this graph concluded for the wrong reason

  the family whose NHR2 tetramer has the only small-molecule oncogenic-oligomer disruptor

• analogous to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)

  partner-contributed self-association module as the obligate, mutation-resistant target - established in one family, hypothesised in the other

• ← part ofRUNX1 K43 acetylation - a retained-portion site that AML1-ETO genuinely requires, in the 5' partner, with the 3' partner supplying the acetyltransferase in trans

  RUNX1 K43 acetylation in the retained 5 prime portion

• ← related toGFI1 — the repressor that LSD1 inhibitors actually work through, and the scan hit whose literature points the opposite way

  CROSS-FAMILY (PI dimension 5): GFI1's subnuclear partner ETO/MTG8 is RUNX1T1, the 3' partner of t(8;21); GFI1's only compartment is one it shares with a different fusion family's partner protein (Q89 2026-09-21)

• regulates →ZEB2 — an EMT transcription factor that turns out to be a leukemia dependency and a guardian of lineage fidelity

  AML1-ETO upregulates Zeb2 in the pre-leukemic HSC compartment; Zeb2 knockdown delays leukemia onset