NUP98::KDM5A (NUP98-JARID1A), t(11;12)(p15;p13)

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

• A KDM5 inhibitor would be the intuitive choice and would be the wrong drug: the JmjC catalytic domain it binds (PDB 5CEH) is DELETED from the fusion [E5]. This node exists partly to prevent that error. Worse, inhibiting wild-type KDM5A demethylase activity would RAISE H3K4me3 globally - the very mark the fusion exploits - so the intuitive drug could plausibly make the biology worse.
• Targeting the condensate is not formulable. LLPS runs on hundreds of weak, interchangeable multivalent contacts [E2][E3]; any agent that dissolves this condensate will also dissolve nucleoli, speckles and stress granules, which is general cytotoxicity rather than therapy.
• PHD fingers are small, shallow, zinc-dependent reader surfaces. Even with a structure, achieving selectivity over the dozens of other PHD fingers in the proteome - including those in KMT2A itself, PHF23, and the wild-type KDM5 family - would be extremely hard, and off-target PHD engagement would perturb normal chromatin reading broadly.
• The prognostic evidence is a null result on 11 patients [E4]. Treating this fusion as prognostically neutral within AMKL risks under-stratifying it; the honest position is that the study was underpowered, not that the fusion is benign.
• The fusion locks an active chromatin state that DEFINES the leukaemia stem cell [E1]. That means the therapeutic target is a self-renewing population, and any agent that removes the lock must do so in the stem cell, not just in the bulk blast - the same problem that limits TKIs in CML.
• The FOLR1 precedent in this disease comes with its own warning: FOLR1 is expressed on renal and pulmonary epithelium, and the authors flag that as a toxicity that must be addressed before clinical use [E6]. Fusion-specific transcriptomes yield surface antigens that are fusion-ENRICHED, not fusion-exclusive, and on-target off-tumour toxicity is the predictable consequence.

1. Why megakaryoblastic? NUP98::NSD1 gives monocytic AML and NUP98::KDM5A gives AMKL, with an identical 5-prime partner. The 3-prime chromatin specificity must set lineage, but the mechanism is unknown: does the PHD finger's H3K4me3 preference bias it toward loci already active in a megakaryocyte-erythroid-biased progenitor, or does it reprogramme the lineage outright? Comparative CUT&RUN for the two fusions in the same human progenitor background would answer this and is the highest-value experiment in this node.
2. Is the retained KDM5A C-terminal PHD finger ligandable? It is the only experimentally required, specific interaction in the fusion [E1], but it has no experimental structure [E5] and no chemical probe. A structure of the PHD finger bound to an H3K4me3 peptide would be the enabling step, and a structure-guided peptidomimetic or covalent ligand would be the first genuinely mechanism-matched agent for any NUP98 fusion.
3. Does NUP98::KDM5A form liquid-like condensates at endogenous expression levels in primary AMKL blasts? The family evidence is strong [E2][E3] but this specific fusion has not been demonstrated, and the whole condensate-rewiring classification for it rests on inference.
4. Because the fusion locks H3K4me3 and dominates over Polycomb [E1], would INCREASING Polycomb activity, or reducing H3K4 methylation upstream (for example by inhibiting the KMT2 family or the WDR5 interaction), shift the balance back? That is a counterintuitive therapeutic direction - adding repression rather than removing activation - and no one appears to have tested it.
5. Has anyone applied the fusion-specific surface-antigen discovery approach that produced FOLR1 for CBFA2T3::GLIS2 [E6] to NUP98::KDM5A in the same disease? If not, that is a concrete, tractable and immediately fundable gap.
6. Is NUP98::KDM5A the most menin-sensitive NUP98 partner because its retained PHD finger reads H3K4me3 - the mark menin-MLL1 itself deposits - so that inhibiting the tether removes the fusion's own binding site rather than merely its address? That is a testable feed-forward model and it predicts something the other NUP98 partners should not show: a menin inhibitor should reduce NUP98::KDM5A chromatin occupancy MORE than it reduces NUP98-HOXA9 occupancy, in proportion to the H3K4me3 loss at the same loci. No study has measured fusion occupancy and H3K4me3 together under a menin inhibitor in this fusion.

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

   Fusing an H3K4me3/me2-binding PHD finger (from PHF23 or from JARID1A/KDM5A) to NUP98 generates potent oncoproteins that arrest haematopoietic differentiation and induce AML in mice; PHD-finger mutations abrogating H3K4me3 binding abolish leukaemic transformation; and the fusion prevents differentiation-associated removal of H3K4me3 at HOX, GATA3, MEIS1, EYA1 and PBX1, acting as a chromatin boundary factor that dominates over Polycomb silencing to lock these loci in an active state defining the leukaemia stem cell.

   Haematopoietic malignancies caused by dysregulation of a chromatin-binding PHD finger · Wang GG, Song J, Wang Z, Dormann HL, Casadio F, Li H, Luo JL, Patel DJ, Allis CD · Nature 2009
   PMID 19430464DOI
   ▸What this evidence cannot show

   Murine retroviral models with ectopic expression; the two fusions studied (NUP98-PHF23 and NUP98-JARID1A) were largely modelled in mouse haematopoietic stem/progenitor cells rather than in human megakaryoblasts, so the mechanism is established for the NUP98-PHD class but the megakaryocytic lineage specificity of the human disease is not explained by these experiments. The 'chromatin boundary factor' model is an interpretation of mark distributions rather than a directly measured competition with PRC2.

   ▸Evidence that argues against this (2)

   • The chromatin-reader model is necessary but not sufficient, and what is missing is a partner complex the model does not predict. Xu et al. showed that NUP98 fusions including NUP98-PHF23 - one of the two fusions in this very claim - physically interact with MLL1 and with the non-specific-lethal (NSL) histone-modifying complex; that NUP98-HOXA9 and MLL1 co-localise on chromatin at Hox promoters; that MLL1 is REQUIRED for proliferation of NUP98-fusion cells in vitro and in vivo; and that inactivating MLL1 decreases expression of fusion-bound genes and reverses the expression signature of NUP98-rearranged human leukaemias. A PHD finger that cannot bind H3K4me3 abolishes transformation - and so does removing MLL1. A 'chromatin boundary factor that dominates over Polycomb silencing' is a cis-acting description of a lesion that also has an obligatory trans-acting complex requirement. Xu H et al. 2016
   • The cytogenetic designation this entry carried until 2026-09-20 was wrong and was wrong in its source: t(11;15)(p15;q35) belongs to NUP98::NSD1. KDM5A is 12p13.33, so NUP98::KDM5A is t(11;12)(p15;p13). Any statement in this graph that keys a fusion to a band taken from an abstract should be checked against both partners' gene records. National Center for Biotechnology Information 2026
2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   Across five different NUP98 fusion proteins expressed in human leukaemia cells, shared interactors are enriched for biomolecular-condensation proteins that colocalise with the fusions in nuclear puncta; NUP98 fusions alter the global composition of cellular condensates, and an artificial FG-repeat-containing fusion protein phenocopies their nuclear localisation and their capacity to drive oncogenic gene expression.

   Biomolecular condensation of NUP98 fusion proteins drives leukemogenic gene expression · Terlecki-Zaniewicz S, Humer T, Eder T, Schmoellerl J, Heyes E, Manhart G, Kuchynka N, Parapatics K, Liberante FG, Muller AC, Tomazou EM, Grebien F · Nat Struct Mol Biol 2021
   PMID 33479542DOI
   ▸What this evidence cannot show

   Ectopic expression in cell lines rather than patient blasts; nuclear puncta by confocal imaging are consistent with condensates but liquid-like behaviour is not demonstrated for each fusion by FRAP or in vitro reconstitution. Whether NUP98::KDM5A specifically was among the five fusions assayed should be confirmed before citing this for KDM5A alone; we cite it here as family-level evidence.

   ▸Evidence that argues against this (1)

   • The graph's condensate standard applies here too, and the failure mode that bites this design is measurement at non-physiological abundance. McSwiggen, Mir, Darzacq and Tjian surveyed the in-vivo LLPS evidence across cellular processes and concluded that it is often phenomenological and inadequate to discriminate phase separation from other possible mechanisms, that the causal relationship and functional consequences are more elusive still, and that the necessary measurements are of proteins at endogenous abundance in their native state. 'Five different NUP98 fusion proteins EXPRESSED in human leukaemia cells', with an ARTIFICIAL FG-repeat-containing fusion protein as the phenocopy, is an ectopic-expression design throughout; the puncta it reports and the interactome enrichment it reports are equally consistent with a stoichiometric multiprotein assembly. The claim survives as evidence of shared partners. It does not survive as evidence that condensation is the mechanism. McSwiggen DT et al. 2019
3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   The NUP98 FG-repeat IDR is essential for liquid-liquid phase separation and leukaemic transformation by NUP98-HOXA9; LLPS promotes chromatin occupancy, is required for broad super-enhancer-like binding at leukaemogenic genes, and induces CTCF-independent chromatin loops enriched at proto-oncogenes; substituting the unrelated LLPS-forming FUS IDR reproduces the effect.

   Phase separation drives aberrant chromatin looping and cancer development · Ahn JH, Davis ES, Daugird TA, Zhao S, Quiroga IY, Uryu H, Li J, Storey AJ, Tsai YH, Keeley DP, Mackintosh SG, Edmondson RD, Byrum SD, Cai L, Tackett AJ, Zheng D, Legant WR, Phanstiel DH, Wang GG · Nature 2021
   PMID 34163069DOI
   ▸What this evidence cannot show

   The mechanistic work is on NUP98::HOXA9, not NUP98::KDM5A, so this is a family-level inference. Ectopic expression in HEK293/HeLa and murine systems; LLPS at endogenous fusion concentration in patient blasts is not shown. The IDR-swap experiment demonstrates sufficiency of an IDR in these assays but does not exclude additional specific functions of the native FG repeats.

   ▸Evidence that argues against this (1)

   • General caution rather than specific refutation: transcriptional-condensate claims across the field have repeatedly been challenged as artefacts of overexpression concentration. The IDR-swap sufficiency result makes the NUP98 case stronger than most, but no study has demonstrated NUP98::KDM5A condensates at endogenous levels in primary AMKL blasts. We found no paper specifically refuting NUP98-fusion LLPS. The first half of this note is not merely a general caution and is now cited: at physiological expression, low-complexity-domain interactions between transcription factors and the Pol II machinery form dynamic hubs WITHOUT detectable phase separation (Chong et al., Science 2018, PMID 29930090), and 1,6-hexanediol and classical FRAP, the two assays most condensate claims rest on, cannot distinguish phase separation from binding to spatially clustered sites (Muzzopappa et al., Nat Commun 2022, PMID 36526633). The second half, that no study has shown NUP98::KDM5A condensates at endogenous levels in primary AMKL blasts, remains a bounded absence with nothing to cite. Chong S et al. 2018
4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   NUP98/JARID1A is a cytogenetically cryptic recurrent fusion in 11/105 (10.5%) of paediatric acute megakaryoblastic leukaemia, characterised by HOXA/B gene overexpression, with no significant difference from other AMKL in sex, age or white cell count, and it was NOT an independent prognostic factor for 5-year OS or EFS although the whole AMKL cohort had a poor 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

   Only 11 fusion-positive cases, so the negative prognostic finding is an underpowered null result, not evidence of prognostic neutrality; absence of an independent effect within an already-dismal AMKL cohort is also hard to interpret because there is little outcome headroom to detect a difference. Cases were assembled retrospectively across multiple international study groups with heterogeneous therapy.

   ▸Evidence that argues against this (1)

   • A larger and differently constructed cohort reaches the opposite conclusion about whether this fusion class separates prognostically. de Rooij et al. performed RNA and exome sequencing on 99 non-DS-AMKL specimens (75 paediatric, 24 adult) and resolved paediatric non-DS-AMKL into SEVEN subgroups with VARYING outcomes, each defined by a chimeric oncogene with cooperating mutations in epigenetic and kinase-signalling genes, with NUP98-KDM5A among the named recurrent fusions and 30-40% of cases previously of unknown aetiology. 'Not an independent prognostic factor' in a 105-case screen is a statement about power and about the comparator: when the comparator is decomposed into its own genomic subgroups instead of being treated as one uniformly poor-prognosis block, the fusion classes do separate. The 42 +/- 6% whole-cohort survival quoted here is an average over groups now known to differ. de Rooij JD et al. 2017
5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

   The NUP98 C-terminal domain lost in the fusion is the autoproteolytic / NUP96-binding module that anchors NUP98 in the nuclear pore complex; and the KDM5A catalytic JmjC domain retained in wild-type KDM5A but LOST in the fusion is the site targeted by published KDM5 inhibitors (PDB 5CEH), whereas the fusion retains only the C-terminal PHD finger.

   PDB entries 1KO6 and 2Q5X (NUP98 C-terminal autoproteolytic domain) and 5CEH (Structure of histone lysine demethylase KDM5A in complex with selective inhibitor) · RCSB Protein Data Bank · RCSB PDB 2015
   ▸What this evidence cannot show

   Verified by fetching the RCSB entry records. The structures establish domain identity and inhibitor site; the therapeutic inference (that a KDM5 inhibitor would miss the fusion because the JmjC domain is deleted) follows from the domain architecture in [E1] and is a deduction, not an experimental result. We could find no experimental structure of the KDM5A C-terminal PHD finger in the RCSB entries returned for P29375.

   ▸Evidence that argues against this (1)

   • Read 5CEH's contents before leaning on it, because they cut both ways. Fetched from RCSB on 2026-09-19: 'Structure of histone lysine demethylase KDM5A in complex with selective inhibitor', X-ray at 3.14 A, with TWO polymer entities - a 790-residue KDM5A entity (UniProt P29375) and a 10-residue chain deposited as 'Unknown Peptide' - plus three heteroatom components (the inhibitor, Ni and Zn). The 790-residue entity is an engineered linked construct, not full-length KDM5A, and 3.14 A is coarse for side-chain-level reasoning about a drug pocket. The wider point strengthens the entry rather than weakening it: 5CEH is one of FORTY-SIX PDB entries mapped to P29375, so the KDM5 catalytic site is among the best-covered pockets in this node - and it sits entirely on the half of KDM5A that NUP98::KDM5A discards. RCSB Protein Data Bank 2016
6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   In the same paediatric AMKL population, a fusion-specific transcriptome approach yielded a validated surface target (FOLR1) and functional CAR-T cells for CBFA2T3-GLIS2, demonstrating that antigen discovery from fusion-driven expression programmes is a workable route for otherwise undruggable fusions.

   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

   This is evidence about a DIFFERENT fusion (CBFA2T3::GLIS2) in the same disease, cited here only to support the strategic claim that antigen discovery is a viable route. It is preclinical (in vitro plus xenograft) with no clinical data, and the authors themselves flag FOLR1 expression in renal and pulmonary epithelium as a toxicity concern that must be solved before clinical use. Nothing analogous has been done for NUP98::KDM5A.

   ▸Evidence that argues against this (1)

   • The route was demonstrated for one fusion, and the AMkL data say these subgroups are not interchangeable. In 107 centrally reviewed non-Down-syndrome AMkL cases from COG AAML0531 and AAML1031, CBFA2T3::GLIS2 (16%) was enriched in trisomy 3 (p=.015) and in the RAM phenotype with high CD56 (p<.001), while NUP98 fusions (10%) were enriched in trisomy 6 (p<.001), monosomy 13/del(13q) (p<.001), trisomy 21 (p=.026) and complex karyotypes (p=.026), with separately poor outcomes in AAML1031. Different cytogenetic company, different surface phenotype, different outcome profile. A surface antigen read out of the CBFA2T3-GLIS2 expression programme is evidence that the METHOD can work; it is not evidence that FOLR1 or any other antigen carries over to NUP98::KDM5A, for which no validated fusion-specific surface target has been reported. This is the cross-family generalisation the brief for this beat names as the most common defect on this material. Chisholm KM et al. 2023
7. E7strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

   The menin-MLL1 dependency shared across NUP98 fusions is graded by the 3-prime partner, and NUP98::KDM5A is the most sensitive partner measured: in murine NUP98-fusion leukaemia cells the VTP50469 IC50 was 84 nM for NUP98-KDM5A against 517 nM for NUP98-HOXA9, the leukaemias depended genetically on Men1 and Kmt2a at a level comparable with the essential-gene control Rpa3, menin inhibition evicted both MLL1 and the NUP98 fusion protein itself from target genes, MEIS1 re-expression abrogated sensitivity, and 2 of 3 NUP98-rearranged patient-derived xenografts responded.

   The menin-MLL1 interaction is a molecular dependency in NUP98-rearranged AML. · Heikamp EB, Henrich JA, Perner F, Wong EM, Hatton C, Wen Y, Barwe SP, Gopalakrishnapillai A, Xu H, Uckelmann HJ, Takao S, Kazansky Y, Pikman Y, McGeehan GM, Kolb EA, Kentsis A, Armstrong SA · Blood 2022
   PMID 34582559DOI
   ▸What this evidence cannot show

   The 84 nM against 517 nM comparison is across separate murine transformation models, not an isogenic partner swap in one background, so partner identity is confounded with model, expression level and passage. The graph's own reading is that the load-bearing downstream node is MEIS1 rather than HOXA occupancy, since NUP98-HOXD13 cells lost no significant Hoxa occupancy. Human NUP98::KDM5A primary material was not the system in which the IC50 values were measured.

   ▸Evidence that argues against this (1)

   • The shared FG-repeat module cannot be what confers menin sensitivity, because one NUP98 fusion has it and is resistant. NUP98::KMT2A retains the same N-terminal NUP98 IDR but deletes KMT2A exon 1, removing the menin-binding motif from the MLL1 moiety it supplies; it expresses reduced rather than elevated HOXA genes and is resistant to MI-2-2 at 3-12 micromolar and to JQ1 at doses where KMT2A-AF9 cells arrest. So partner-graded sensitivity is not a family property carried by the 5-prime module; it tracks the availability of an intact menin-MLL1 tether, and a partner that removes it exits the class. Fisher JN et al. 2020

• ← part ofNUP98 — domain architecture and residue coordinates (with the 1817/920 numbering trap resolved)

  same NUP98 FG-repeat contribution with a different 3-prime partner

• ← supportsRepairing 85 defective citation blocks, and the finding that the queue's own numbers were wrong in three directions at once

  that entry's uncited condensate-artefact caution now carries PMID 29930090 and names PMID 36526633; the residual NUP98::KDM5A endogenous-level claim stays a bounded absence

• ← related toNUP98 fusions share the menin dependency (confirmed preclinically) — and the partner-supplied reader/writer module is not why

  Same result in the KDM5A/JARID1A arm: VTP50469 killed NUP98-JARID1A cells at KMT2A-r-comparable IC50 and prolonged PDX survival (PMID 34582559). The KDM5A PHD reader module was not the interface disrupted.

• ← related toMechanism class 1 — chromatin and elongation hijack (KMT2A-r, NUP98-r)

  Attribution correction (Ahn 2025, PMID 39922194): in NUP98-PHF23 the partner PHD finger targets chromatin and the FG repeats select which coactivators partition in, so chromatin occupancy is partner-driven, not FG-driven, in exactly the reader-domain NUP98 fusions this class names.

• ← related toCBFA2T3::GLIS2, inv(16)(p13.3q24.3)

  Frequency correction (Chisholm 2023, PMID 36789545): in 107 centrally reviewed COG non-DS AMkL cases CBFA2T3::GLIS2 was 16% and NUP98 fusions 10%, so the 27% discovery figure is the outlier against 12-16% in larger cohorts.

• related to →Phase-2 dossier: the menin MBM1 pocket - the only candidate here that is already de-risked, and therefore the only one whose gate is clinical

  Trans-acting requirement (Xu 2016, PMID 27889185): NUP98-PHF23 and other NUP98 fusions interact with MLL1 and the NSL complex and REQUIRE MLL1 for proliferation in vivo, which the cis-acting chromatin-boundary model does not predict and which extends the MLL1/menin dependency beyond KMT2A fusions.

• related to →Condensates in KMT2A-fusion leukemia — an honest verdict: not established for AFF1/AFF4/AF9, load-bearing only in ENL-mutant and NUP98 fusions

  Condensate-standard bound (McSwiggen 2019, PMID 31594803): in-vivo LLPS evidence is often phenomenological and cannot discriminate phase separation from other mechanisms without endogenous-abundance measurement, and this claim rests entirely on ectopic expression.

• related to →Bayesian borrowing: how many infants a prior is worth, and when it lies

  Comparator decomposition (de Rooij 2017, PMID 28112737): RNA and exome sequencing of 99 non-DS-AMKL specimens resolved seven genomic subgroups with varying outcomes, so 'not an independent prognostic factor' against a pooled AMKL block is a power-and-comparator statement.

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

  Structural read-through (RCSB 5CEH, fetched 2026-09-19): a 3.14 A structure of a 790-residue engineered KDM5A construct plus an 'Unknown Peptide', one of 46 entries mapped to P29375 - the best-covered pocket in this node sits on the discarded half of the protein.

• related to →Myeloid-directed antigens (CD33, CD123) — the only immunotherapy axis that survives lineage switch

  Cross-family bound (Chisholm 2023, PMID 36789545): CBFA2T3::GLIS2 and NUP98-fusion AMkL differ in cytogenetic associations and surface phenotype, so a FOLR1-style antigen from one programme does not transfer to NUP98::KDM5A.

• ← related toDocking onto pre-bound XPO1 is a general route to leukaemic chromatin, used by at least seven different lesions - not three coincidences

  NUP98 FG repeats are the grip module in the founding NUP98::HOXA9 experiment

• depends on →Mechanism class 4 — condensate-driven transcriptional rewiring (NUP98-r, EWSR1::FLI1)
• ← part ofKDM5A (JARID1A / RBP2 H3K4 demethylase) domain architecture

  reference domain coordinates for the KDM5A side of this fusion

• ← related toThe nuclear pore and NUP98 fusions — a fusion that takes a pore protein's disordered repeats off the pore and builds nuclear droplets with them

  NUP98 fusions as a class form FG-driven nuclear puncta

• ← treated bySAVE (NCT05360160): the first prospective menin-inhibitor trial to report a NUP98-rearranged subgroup outcome

  NUP98r subgroup is genotype-pooled and not resolved by partner

• ← regulatesWhy a nucleoporin fusion needs menin-MLL1: the second menin-dependent fusion family, and what it borrows rather than carries

  NUP98-JARID1A models sensitive to VTP50469 in vitro and in PDX

• ← related toNUP98 FG condensates, audited assay by assay: no half-bleach has ever been done and no NUP98 fusion has ever been endogenously tagged

  the PHD-finger reader is the clustered-binding alternative to bulk partitioning

• ← related toNUP98 rearrangements are 7.2% of paediatric AML and 1.6% of adult, with 35% five-year overall survival — and the partner changes everything

  32 of 160; CR rate 81% but relapse risk 68%; median age 2.7 y

• ← part ofNUP98 - the mobile nucleoporin that leaves the pore to put H3K4me3 on haematopoietic promoters, and the shared 5' donor of seven fusions in this graph

  contributes the 5' half: the FG/GLFG region and the off-pore chromatin-facing module

• ← related toAn H3K4me3-reading PHD finger fused to NUP98 makes a potent leukaemia oncoprotein; the H3K4me3-reading PHD finger of KMT2A is deleted from every KMT2A fusion and reinstating it abolishes transformation. Same module class, opposite sign - and the one NUP98 fusion that carries KMT2A's PHD cassette is the weakest member of its family.

  the family where an H3K4me3-reading PHD finger is required for transformation

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

  same H3K4me3-reading PHD module class, required in one family and incompatible with transformation in the other

• ← part ofKDM5A / JARID1A / RBP2 (P29375, 1,690 aa) - already in this graph as the 3' half of NUP98::KDM5A and never written as a gene: the H3K4 demethylase CTCF recruits to a break, the only ERASER in the licensing cascade, and the one cascade member with no blood-lineage ChIP-seq anywhere

  3' partner of the paediatric AML fusion already held in this graph

• ← related toLMO2 (rhombotin-2) - a bridging LIM-only protein, higher in pro-B and MLL-rearranged B-ALL, anti-selective in KMT2A-rearranged lines, and a physical partner of the KDM5A module that NUP98::KDM5A retains

  UNTESTED LEAD: the curated LMO2-interaction region of KDM5A (1623-1690) overlaps the PHD3 module (1607-1661) that this fusion retains

• ← refutesWhy this one rare leukaemia was thought to decide an argument about a drug used in a commoner one - and why it decides less than the title says (corrected 2026-09-21)

  NUP98-JARID1A/KDM5A leukaemia responds to VTP50469 at an IC50 the authors report as similar to KMT2A-rearranged and NPM1c cells

• ← analogous toNUP98::NSD1, t(5;11)(q35;p15.5)

  AXIS: SHARED 5-PRIME PARTNER, OPPOSITE 3-PRIME MODULE CLASS. Identical NUP98 FG-repeat IDR; NSD1 contributes a WRITER whose SET catalysis is required for transformation, KDM5A contributes a READER whose catalytic demethylase domain is discarded at the breakpoint and whose H3K4me3-binding PHD finger is required instead. Same condensate physics, same HOXA output, and different diseases - monocytic cytogenetically normal AML against megakaryoblastic AML - so lineage is set by the 3-prime partner and not by the FG repeats.

• analogous to →CBFA2T3::GLIS2, inv(16)(p13.3q24.3)

  AXIS: SHARED CELL OF ORIGIN, NOT SHARED MECHANISM. The two commonest cryptic fusions of paediatric non-Down-syndrome acute megakaryoblastic leukaemia - NUP98::KDM5A in ~10.5% and CBFA2T3::GLIS2 in 12-27% - both invisible to routine karyotyping, mechanistically unrelated (FG-repeat condensate chromatin-boundary factor against an ETO-family corepressor fused to a GLI-similar zinc finger driving BMP signalling). DISANALOGY AND THE POINT OF THE PAIR: CBFA2T3::GLIS2 is dismal wherever measured, while NUP98::JARID1A was NOT an independent prognostic factor within the same already-poor AMKL cohort - so much of the outcome belongs to the lineage and the age rather than to the driver.

• supports →Menin dependency tracks the availability of an intact menin-binding motif, in cis or in trans — not the fusion family

  The most menin-sensitive NUP98 partner measured (VTP50469 IC50 84 nM, against 517 nM for NUP98-HOXA9), carrying no menin-binding motif of its own and therefore depending on endogenous MBM-bearing MLL1 in trans.