Oncofusion

start / graph / localization-nuclear-pore-nup98-fg-comparator

Mechanismstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q4b-localization

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

also called NUP98 FG repeats · nucleoporin fusion · NUP98-HOXA9 condensate · NUP98-PHF23 · pore-derived IDR

NUP98 is normally part of the nuclear pore, the gate in the nuclear envelope, and the sticky repetitive stretch it uses to make the gate selective ends up in leukemia fusion proteins. In the fusion that stretch no longer sits at the pore: it builds droplets inside the nucleus that stick the fusion to DNA over broad regions and pull in the machinery that switches genes on, including KMT2A itself.

The same thing, for a scientist

NUP98 fusions are the cleanest published case of a fusion repurposing a compartment-defining module. The retained element is the nucleoporin FG (phenylalanine-glycine) repeat IDR, which in the intact nucleoporin creates the selective permeability barrier of the nuclear pore. In NUP98-HOXA9 the FG IDR is essential both for LLPS puncta and for leukemic transformation; phase separation promotes chromatin occupancy, generates a broad super-enhancer-like binding pattern at leukaemogenic genes, and induces **CTCF-independent chromatin loops** enriched at proto-oncogenes; replacing the FG repeats with an unrelated LLPS-forming FUS IDR reproduces the enhanced genome-wide binding and target activation [E1]. In NUP98-NSD1, core interactome binding is largely dependent on the FG repeats that mediate condensates, SMARCA5 and BPTF are condensate constituents and SMARCA5 is functionally essential — but SMARCA5 inhibition does not affect the condensates, so **functional activity rather than condensate formation per se is what maintains transformation** [E2]. In NUP98-PHF23 a mutagenesis-proteomics-genomics-reconstitution study separates the two halves cleanly: the PHD finger targets the condensate to H3K4me3/2-marked developmental genes while the FG repeats determine condensate **composition** and gene activation, selectively partitioning transcriptional regulators — notably the **KMT2/MLL H3K4 methyltransferases**, histone acetyltransferases and BRD4 — and FG repeats tethered to a locus are sufficient to partition those regulators and activate a reporter [E3]. Two consequences for KMT2A. First, the NUP98 condensate concentrates KMT2/MLL, which makes the two fusion families biochemically adjacent rather than merely analogous. Second, menin is a dependency in NUP98-rearranged AML and the inhibitor evicts both MLL1 and the NUP98 fusion protein itself from chromatin [E4] — the only published measurement of a *fusion protein* being evicted by a menin inhibitor, and it is in the other family.

Go deeper
What is and is not a 'pore' claim1 deeper

The compartment in the title is misleading if read literally, and the graph should be precise. NUP98 fusions are not reported at the nuclear pore. The oncogenic species is nucleoplasmic: FG-repeat-driven puncta inside the nucleus, on chromatin. The pore is the origin of the module, not its destination. What the fusion inherits is a piece of protein evolved to create a selective phase — the FG mesh whose whole job is to admit some molecules and exclude others — and it uses that selectivity inside the nucleus to choose which coactivators to concentrate 3.

This makes NUP98 fusions a different mechanistic class from PML::RARA, which disrupts an existing compartment, and from KMT2A fusions, which as far as anyone has shown do not build or destroy a compartment at all. Three fusion families, three relationships to nuclear organisation: build (NUP98), break (PML), neither established (KMT2A).

The 'necessary but not sufficient' result that keeps the field honest

Jevtic et al. found that NUP98-NSD1 interactome binding depends on the FG repeats that mediate condensate formation, identified SMARCA5 and BPTF as condensate constituents, validated SMARCA5 in NUP98-NSD1 patient cells, and showed SMARCA5 is essential for transformation. But SMARCA5 inhibition did not affect the condensates, from which the authors conclude that functional activity rather than condensate formation per se is crucial for maintaining the transformed phenotype 2.

That is the model the whole condensate field should be held to: the condensate is a delivery mechanism for an activity, and removing the activity kills the cell while leaving the droplet intact. Applied to KMT2A, it predicts that even if a KMT2A-fusion condensate were found, the druggable node would be a constituent's catalytic or binding function, not the droplet.

Why NUP98 is the most informative comparator for this lab, not EWSR1

Three specific connections, none of which hold for the Ewing sarcoma comparator:

  1. The NUP98 condensate concentrates KMT2/MLL methyltransferases 3. The two fusion families are not just analogous; one of them partitions the other's protein.
  2. Menin dependency is shared. The menin-MLL1 interaction is a molecular dependency in NUP98-rearranged AML, and VTP-50469 evicts MLL1 and the NUP98 fusion protein itself from chromatin 4. This is the single published instance of the fusion being shown to leave chromatin under menin inhibition, and it is direct evidence that the menin interface governs fusion positioning beyond the KMT2A family.
  3. Same patients, same clinic. NUP98-NSD1 and NUP98-KDM5A are paediatric AML fusions already in this graph, treated in the same trials with the same menin inhibitors.

The transfer that does not work: NUP98 fusions retain a professional phase-separating module with tandem FG repeats. KMT2A retains AT-hooks, a CXXC domain and the menin-binding motifs — no comparable IDR on the 5' side. If KMT2A fusions phase-separate it has to come from the 3' partner, which is why AFF1's IDR, and not KMT2A, is the right place to look.

Facts and figures
compartmentnuclear pore complex (origin) / nucleoplasmic FG condensate (destination)
retained moduleNUP98 phenylalanine-glycine (FG) repeat intrinsically disordered region
evidence class for that familystrong: IDR necessity, IDR-swap sufficiency, Hi-C looping, interactome
evidence class for kmt2anone — KMT2A retains no comparable IDR on the 5' side
condensate partitions kmt2 mlltrue
only published fusion protein eviction by menin inhibitorNUP98 fusions (PMID 34582559)
marker proteins
[
  "NUP98",
  "NUP214",
  "FG-repeat nucleoporins",
  "mAb414"
]
fusions studied
[
  "fusion-nup98-nsd1",
  "fusion-nup98-kdm5a"
]
How this could bite us back
  • If condensate composition rather than condensate existence is what transforms, as the NUP98-NSD1 result argues, then any KMT2A programme that screens for droplet dissolution is screening for the wrong endpoint and will produce compounds that dissolve droplets without killing cells.
  • NUP98 fusions and KMT2A fusions share menin dependency and are treated with the same inhibitors, so resistance mechanisms will cross-pollinate clinically: MEN1 interface mutations selected in one family will be immediately relevant to the other.
Still open
  1. The NUP98-PHF23 condensate partitions KMT2/MLL methyltransferases. Does it partition a KMT2A **fusion** protein, in a cell that carries both lesions? Rare, but NUP98 and KMT2A rearrangements do co-occur, and the answer would say whether the two families' mechanisms are additive or competitive.
  2. Heikamp et al. show the NUP98 fusion protein itself is evicted from chromatin by a menin inhibitor. Nobody has run the equivalent measurement for the KMT2A fusion protein rather than for MLL1; is the KMT2A::AFF1 protein evicted, and with what kinetics?
  3. If FG repeats determine condensate composition and PHD determines targeting, what is the equivalent division of labour in KMT2A fusions — is CXXC the targeting module and the AFF1 IDR the composition module? Stating the two families in the same grammar makes the KMT2A gap precisely defined.
  4. Does replacing the AFF1 C-terminal region of KMT2A::AFF1 with the FUS IDR — the exact swap that worked for NUP98-HOXA9 — produce a transforming protein? A positive result would prove the 3' partner's only essential contribution is a generic IDR; a negative would prove AFF1 contributes specific chemistry, which is the more interesting answer for interface-directed drugs.
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The nucleoporin FG-repeat intrinsically disordered regions of NUP98-HOXA9 are essential for establishing liquid-liquid phase separated puncta and for inducing leukemic transformation; phase separation promotes chromatin occupancy and is required for a broad super-enhancer-like binding pattern at leukaemogenic genes; replacing the FG repeats with the unrelated LLPS-forming FUS IDR recapitulates enhanced genome-wide binding and target activation; and deeply sequenced Hi-C shows phase-separated NUP98-HOXA9 induces CTCF-independent chromatin loops enriched at proto-oncogenes.

    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

    Largely HEK293, HeLa and mouse haematopoietic models with expressed chimera constructs; the IDR-swap experiment is the strongest element because it is a gain-of-function with an unrelated sequence, but neither the puncta nor the looping were measured at endogenous fusion levels in a patient-derived NUP98-r AML line.

    Evidence that argues against this (1)
    • Qualifies what the NUP98 result can be exported as. Ahn et al. themselves show that replacing the NUP98 FG repeats with an unrelated LLPS-competent FUS IDR reproduces the enhanced genome-wide binding and target activation, and Zuo et al. show independently that FET-family IDRs phase-separate at their own target loci, recruit Pol II and enhance transcription with a DNA-element threshold. The demixing property is therefore generic across unrelated IDRs, not a specific property of FG repeats. Two consequences for this graph: (i) 'the FG repeats are essential' means 'an IDR is essential', which is a much weaker statement about druggability, since there is no FG-specific interface to target; (ii) the comparator cannot be used to argue that a KMT2A fusion needs a condensate, because KMT2A::AFF1 would need its own IDR shown to be doing this job, not merely to possess one. Read: abstract only. Zuo L et al. 2021
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    NUP98-NSD1 core interactome binding is largely dependent on the NUP98 FG repeats that mediate formation of liquid-like phase-separated nuclear condensates; SMARCA5 and BPTF are condensate constituents, SMARCA5 is essential for NUP98-NSD1 transformation in haematopoietic cells, but SMARCA5 inhibition does not affect the condensates, indicating that functional activity rather than condensate formation per se maintains the transformed phenotype.

    SMARCA5 interacts with NUP98-NSD1 oncofusion protein and sustains hematopoietic cells transformation · Jevtic Z, Matafora V, Casagrande F, Santoro F, Minucci S, Garre M, Rasouli M, Heidenreich O, Musco G, Schwaller J, Bachi A · Journal of Experimental & Clinical Cancer Research 2022
    PMID 35073946DOI
    What this evidence cannot show

    Immortalised primary murine HSPC with inducible knockdown plus validation in NUP98-NSD1 patient cells; FRAP and b-isoxazole are the phase-transition readouts, both indirect. The negative (SMARCA5 inhibition does not dissolve condensates) depends on the inhibitor's on-target potency in that system.

    Evidence that argues against this (1)
    • Methodological counter that applies across the NUP98 condensate literature. Much of the in-cell evidence that these puncta are condensates and that the condensates matter rests on 1,6-hexanediol dispersal. Itoh et al. showed by live-cell single-nucleosome imaging that 1,6-hexanediol drastically suppresses chromatin motion and hyper-condenses chromatin in human cells in a dose-dependent way, with chromatin frozen at 5 per cent and above, and that this action is distinct from droplet melting and is reproduced in vitro as cation-dependent chromatin condensation. For a chromatin-bound fusion the reagent changes the chromatin substrate at the same time as the putative droplet, so loss of signal or of transcription after treatment does not identify which was responsible. This does not touch the FG-deletion genetics, which is why the entry's strong grade survives - but it does mean the chemical arm of the case carries no weight. Read: abstract only. Itoh Y et al. 2021
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In NUP98-PHF23 the PHD finger targets the condensate to H3K4me3/2-demarcated developmental genes while the FG repeats determine condensate composition and gene activation; FG repeats are necessary to partition a specific set of transcriptional regulators, notably the KMT2/MLL H3K4 methyltransferases, histone acetyltransferases and BRD4, and are sufficient to partition those regulators and activate a reporter when tethered to a genomic locus.

    The phenylalanine-and-glycine repeats of NUP98 oncofusions form condensates that selectively partition transcriptional coactivators · Ahn JH, Guo Y, Lyons H, Mackintosh SG, Lau BK, Edmondson RD, Byrum SD, Storey AJ, Tackett AJ, Cai L, Sabari BR, Wang GG · Molecular Cell 2025
    PMID 39922194DOI
    What this evidence cannot show

    Integrated mutagenesis, proteomics, genomics and reconstitution, largely in engineered cell systems with tethering assays; NUP98-PHF23 is a rare fusion, and the selective-partitioning result is a composition measurement rather than a demonstration that partitioning is what transforms.

    Evidence that argues against this (1)
    • The partitioning experiments here diagnose a condensate with tools that were benchmarked and found unable to make that call. Muzzopappa et al. tested candidate LLPS assays against well-controlled model systems in vitro and in living cells and found that 1,6-hexanediol treatment and classical FRAP fail to distinguish liquid-liquid phase separation from molecules binding to spatially clustered binding sites without phase-separating, while preferential internal mixing in half-bleach experiments does distinguish them - the basis of their MOCHA-FRAP measurement of interfacial barrier strength, which differs markedly between real nuclear bodies. Selective partitioning of KMT2/MLL methyltransferases, HATs and BRD4 by FG repeats is equally well described as multivalent FG-mediated BINDING, which is what FG repeats do at the pore, and which requires no phase. Muzzopappa F et al. 2022
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The menin-MLL1 interaction is a molecular dependency in NUP98-rearranged AML, and menin inhibition with VTP-50469 evicts MLL1 and the NUP98 fusion protein itself from chromatin.

    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

    Eviction is a chromatin-occupancy readout (ChIP/CUT&RUN class) in mouse NUP98-fusion lines and patient-derived xenografts, not an imaging or protein-stability measurement; it says the fusion leaves chromatin, not where it goes. NUP98 fusions, not KMT2A.

    Evidence that argues against this (1)
    • The clinical durability of this dependency is the part the comparator does not carry over, and in KMT2A-rearranged disease it is now known to be poor in pretreated patients. Mahdavi et al. found that patient-derived xenografts from three KMT2A-rearranged AMLs with high-risk partners (MLLT10 and MLLT4/AFDN) and two infant ALL samples were sensitive to menin inhibition, but that serial samples from two multiply relapsed KMT2A::AFF1 ALL patients had become much less sensitive than earlier samples from the same patients WITHOUT any menin-inhibitor exposure, while retaining on-target transcriptional responses. A menin dependency demonstrated in NUP98-rearranged models, like one demonstrated in treatment-naive KMT2A models, therefore predicts drug activity only in the treatment-naive setting, which is not where second-line agents are used. Mahdavi L et al. 2025
Connected to