Oncofusion

start / graph / mechanism-condensate-transcriptional-rewiring

Mechanismmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q11-comparative

Mechanism class 4 — condensate-driven transcriptional rewiring (NUP98-r, EWSR1::FLI1)

also called phase separation class · prion-like domain class · IDR condensate rewiring

Some fusions bring along a floppy, sticky stretch of protein that makes many copies clump together into droplets on the DNA. Those droplets concentrate the machinery that switches genes on, creating brand-new super-switches that the normal cell never had.

The same thing, for a scientist

Here the fusion acquires a low-complexity intrinsically disordered region — nucleoporin FG repeats, or the EWSR1 prion-like domain — whose multivalent weak interactions drive liquid-liquid phase separation. The condensate is not incidental: it increases the residence time and apparent avidity of the chimeric transcription factor on chromatin, converts point binding into broad super-enhancer-like occupancy, recruits coactivator and remodelling machinery, and in the NUP98 case generates CTCF-independent chromatin loops enriched at proto-oncogenes. The DNA-binding specificity comes from the other half (HOXA9 homeodomain, KDM5A/NSD1 reader-writer modules, FLI1 ETS domain binding GGAA microsatellites), so this class explains how a weak or ordinary DNA-binding module becomes a dominant oncogenic driver.

Go deeper
The NUP98 case: phase behaviour is necessary, and it is transferable

NUP98-HOXA9 forms nuclear puncta that depend on the NUP98 FG repeats. Deleting the FG repeats abolishes both puncta and leukemic transformation; replacing them with an unrelated LLPS-competent IDR from FUS restores genome-wide binding breadth and target-gene activation. Deeply sequenced Hi-C showed that the phase-separated chimera induces CTCF-independent chromatin loops enriched at proto-oncogenes 1.

Two things follow. First, the oncogenic property is a material property (multivalency, phase behaviour), not a sequence-specific binding event — which is a genuinely new kind of drug target and a genuinely hard one. Second, because the IDR is interchangeable, there is nothing chemically specific about NUP98 to grab hold of; a drug would have to perturb condensate physics, not a binding site.

The Ewing case: a prion-like domain that relocates a chromatin remodeller1 deeper

EWS-FLI1 uses the EWSR1 prion-like domain (PrLD) to recruit the BAF (mSWI/SNF) chromatin-remodelling complex to tumour-specific enhancers at GGAA microsatellites. This is neomorphic — wild-type FLI1 does not do it — and it depends on the tyrosine residues required for the PrLD phase transition. The step from 'depends on residues needed for the phase transition' to 'works by phase separation' does not follow, and at endogenous abundance it appears to be wrong: EWS::FLI1 needs a narrow optimum of low-complexity-domain interaction, and driving it further toward LLPS represses its own target genes and blocks transformation (counter-evidence on 2). Fusing short EWSR1 fragments to FLI1 is sufficient to recapitulate BAF retargeting and EWS-FLI1 activity 2.

This is the mechanistically closest solid-tumour analogue of KMT2A::AFF1: a DNA-targeting module fused to a module whose job is to bring a large chromatin machine to the wrong places. It is also the clearest demonstration of the strategic problem, because after two decades Ewing sarcoma still has no approved targeted agent against the fusion.

Does KMT2A::AFF1 belong here too?

Partially, and this is an open question worth the lab's attention. AFF1 and AFF4 contain long low-complexity regions and the SEC has been described as a phase-separating assembly; the KMT2A fusion program has the same super-enhancer-like breadth that condensate mechanisms produce. But unlike NUP98-HOXA9 there is no published IDR-deletion or IDR-swap experiment showing that KMT2A::AFF1 transformation requires phase behaviour. The graph therefore records the condensate assignment for KMT2A::AFF1 as speculative and proposes the experiment in hyp-obligate-partner-interface-generalises.

Therapeutic implication — stated plainly

This class has yielded no drugs. Not one approved agent targets a condensate-forming fusion oncoprotein through its condensate. The reasons are structural, not neglect: an IDR has no pocket, no fixed conformation, and no unique sequence to recognise; the same physics is used by dozens of essential normal assemblies (nucleoli, speckles, Pol II clusters), so selectivity is very hard; and cellular assays for 'condensate inhibition' are still immature.

What this class does deliver is indirect vulnerabilities. Because condensates concentrate coactivators, condensate-driven tumours are often hypersensitive to inhibiting a recruited component rather than the fusion — BAF/SMARCA4 in Ewing, BRD4 in fusion-positive rhabdomyosarcoma, menin in NUP98-r. That is the same 'attack the obligate partner' logic the lab is pursuing for KMT2A, and this class is the strongest argument that the logic generalises beyond menin.

Scope correction 2026-09-19 — what this class does and does not cover

Three bounds were added to this entry by the Q11 counter-argument sweep, and the reader should carry them forward.

1. Condensate formation is necessary in the NUP98 case but is not sufficient. Inhibiting the condensate constituent SMARCA5 destroys NUP98-NSD1 transformation while the condensates remain intact, so what matters is which proteins the condensate concentrates, not that it forms. The drug target implied by this class is therefore a constituent, which has pockets, rather than the phase behaviour, which does not.

2. More phase separation is not more oncogenic. In patient-derived Ewing cells the relationship is non-monotonic and pushing EWS::FLI1 toward LLPS switches its program off. Any therapeutic reasoning of the form 'dissolve the condensate to stop the oncogene' assumes a monotonic relationship that has been measured and is not there.

3. The KMT2A sentence in this entry was a cross-family extension, not a finding. SEC phase separation is a normal feature of rapid transcriptional activation in unmodified cells, driven by ENL's IDR and fluidised by AFF4; and the cytoplasmic KMT2A partners transform through coiled-coil oligomerisation with no SEC and no elongation function at all. The claim that a condensate account 'covers the KMT2A super elongation complex' holds for the elongation-class partners and is not established for the class as a whole. The open question at the foot of this entry - whether KMT2A::AFF1 transformation requires phase behaviour - is still open, and the graph tracks it as hyp-aff1-condensates-dispensable-for-kmt2a-fusion-function.

Facts and figures
mechanism classcondensate-rewiring
invariant modulean acquired low-complexity intrinsically disordered region with multivalent weak interactions (NUP98 FG repeats, EWSR1 prion-like domain)
variable modulethe DNA-targeting module that sets the address (HOXA9 homeodomain, FLI1 ETS domain at GGAA microsatellites, PAX3 paired/homeodomain)
druggable nodeno direct node; indirectly the recruited complex (BAF/SMARCA4, BRD4, menin) or the concentration-dependent coactivator the condensate depends on
druggability classundrugged
why not druggedAn IDR has no pocket and no fixed conformation, the underlying physics is shared with essential normal nuclear bodies so selectivity is very hard, and cellular assays for condensate perturbation are immature.
strategic valueIt is the strongest independent argument that attacking an obligate recruited partner, rather than the fusion, is the general solution for non-enzymatic fusions.
member fusions
[
  "fusion-nup98-nsd1",
  "fusion-nup98-kdm5a",
  "fusion-ewsr1-fli1",
  "fusion-pax3-foxo1"
]
candidate members unproven
[
  "fusion-kmt2a-aff1"
]
drugs yielded
[]
drugs failed or stalled
[
  "BET inhibitors in fusion-positive rhabdomyosarcoma",
  "condensate-modulating chemistry generally (no clinical agent)"
]
How this could bite us back
  • Condensate physics is used by essential normal assemblies, so any agent that genuinely changes phase behaviour is likely to be broadly toxic rather than selective.
  • Partner-complex dependencies identified in condensate-driven tumours have repeatedly failed to translate (BET inhibitors in rhabdomyosarcoma), so in-vitro dependence in this class is a weak predictor of clinical benefit.
  • If condensate multivalency buffers the fusion complex, partial inhibition of any single recruited partner may be absorbed with no phenotype, forcing near-complete target engagement and a narrow window.
  • The intuitive intervention may be backwards. Because the LCD-interaction/transcription relationship is non-monotonic in the one system where it has been measured at endogenous abundance, an agent that partially dissolves a fusion condensate could land on the optimum rather than past it and increase oncogenic output before it decreases it. Dose-response in this class must be measured, not assumed.
Still open
  1. Does KMT2A::AFF1 transformation actually require phase behaviour, i.e. does deleting or swapping the AFF1 low-complexity region kill transformation the way the NUP98 FG-repeat deletion does?
  2. Is there a partner whose recruitment is concentration-dependent in a condensate but dispensable in normal diffuse transcription, giving a genuinely fusion-selective window?
  3. Can condensate material state (viscosity, exchange rate) be measured in primary patient blasts well enough to serve as a pharmacodynamic biomarker, or is this class untestable in patients with current tools?
  4. Is the non-monotonic LCD-interaction optimum found for EWS::FLI1 a general property of condensate-forming fusions, and where on that curve does an untreated NUP98-NSD1 or KMT2A::AFF1 blast actually sit?
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The NUP98 FG-repeat IDR is required for chimeric-oncoprotein puncta, broad super-enhancer-like chromatin occupancy, CTCF-independent looping at proto-oncogenes and leukemic transformation, and can be functionally replaced by an unrelated LLPS-competent IDR.

    Phase separation drives aberrant chromatin looping and cancer development · Ahn JH, Davis ES, Daugird TA, et al. · Nature 2021
    PMID 34163069DOI
    What this evidence cannot show

    Mostly NUP98-HOXA9 in cell lines and mouse transplant, often at non-physiological expression where phase separation is easier to observe; whether clinically common NUP98::NSD1 and NUP98::KDM5A depend on LLPS to the same degree is not directly shown.

    Evidence that argues against this (3)
    • A general concern across the field is that overexpression-driven puncta can be artefacts and that many reported oncogenic condensates have not been confirmed at endogenous levels; this is a live methodological dispute rather than a single contrary paper. (Citation title repaired 2026-09-19: an explanatory parenthetical had been spliced onto the paper's real title. The point it was making stands - Boulay 2017 is cited here as the standard of evidence this class should be held to, because it used defined domain-swap controls rather than deletion alone.) Boulay G et al. 2017
    • Necessity was shown; sufficiency was not, and the difference decides whether the condensate is the drug target. In NUP98-NSD1/FLT3-ITD immortalised primary murine HSPC, the condensate constituent SMARCA5 is required for transformation, but inhibiting SMARCA5 leaves the NUP98-NSD1 condensates fully intact while the transformed phenotype collapses. The authors conclude that condensate formation is not sufficient for maintenance of transformation and that the therapeutic handle is the condensate's constituents, not its phase behaviour. Applied to this entry, 'the oncogenic property is a material property' is at best incomplete: a cell can have the material property and not the disease state. Jevtic Z et al. 2022
    • The evidence class itself is contested on method. A systematic appraisal of in vivo LLPS evidence across cellular processes concludes that the evidence is often phenomenological and inadequate to discriminate phase separation from other self-organising mechanisms, that causal relationships and functional consequences in vivo are more elusive still, and that the decisive measurements are quantitative ones made on proteins at endogenous abundance. Puncta imaged from transfected constructs, FRAP recovery, and 1,6-hexanediol sensitivity - the standard triad - do not individually establish LLPS. This does not overturn the NUP98 result; it bounds what any experiment of that design can conclude, and this graph already holds that standard as `localization-condensate-evidence-standards`. McSwiggen DT et al. 2019
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The EWSR1 prion-like domain is required for EWS-FLI1 to retarget BAF (mSWI/SNF) complexes to tumour-specific GGAA-microsatellite enhancers, a neomorphic activity absent from wild-type FLI1 and dependent on residues needed for the domain phase transition.

    Cancer-specific retargeting of BAF complexes by a prion-like domain · Boulay G, Sandoval GJ, Riggi N, et al. · Cell 2017
    PMID 28844694DOI
    What this evidence cannot show

    Ewing cell lines and engineered constructs; tyrosine-mutant phenotypes conflate phase behaviour with specific protein contacts, so 'phase transition is the mechanism' is an inference rather than a direct measurement in tumours.

    Evidence that argues against this (1)
    • The dose-response is not monotonic, which inverts the therapeutic reading of this whole class. Quantitative single-cell and single-molecule imaging in patient-derived Ewing cells shows EWS::FLI1 requires a NARROW OPTIMUM of low-complexity-domain interactions to activate its GGAA-microsatellite target genes: increasing LCD-LCD interactions toward putative LLPS REPRESSES those genes, and ectopically driving EWS::FLI1 into a bona fide LLPS compartment (the nucleolus) inhibits both its transcription program and oncogenic transformation. If more phase separation switches the oncogene off, then 'the prion-like domain works by phase transition' is the wrong causal statement, and a drug that promoted condensation would be a plausible therapy rather than a paradox. Chong S et al. 2022
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Condensate-driven fusions create dependence on a recruited partner complex that is more druggable than the fusion itself, as shown by the complete BRD4 dependence and BET-inhibitor susceptibility that PAX3-FOXO1 super enhancers confer in alveolar rhabdomyosarcoma.

    PAX3-FOXO1 establishes myogenic super enhancers and confers BET bromodomain vulnerability · Gryder BE, Yohe ME, Chou HC, et al. · Cancer Discovery 2017
    PMID 28446439DOI
    What this evidence cannot show

    Cell lines and xenografts; BET inhibitors have since underperformed clinically in fusion-positive rhabdomyosarcoma, so partner-complex dependence measured in vitro has not translated to patient benefit here.

    Evidence that argues against this (1)
    • The 'recruited partner is more druggable than the fusion' inference has been tested clinically in the hardest possible test case and it did not deliver. NUT carcinoma is driven by BRD4-NUT, a fusion in which the BET protein IS the oncoprotein, so BET dependence is not inferred from a super-enhancer argument but built into the lesion. In the first-in-human study of molibresib (65 patients), among 19 patients with NUT carcinoma only four achieved a confirmed OR UNCONFIRMED partial response, and dose was capped by haematological toxicity: thrombocytopenia in 51% of all patients with six grade 4 events driving the dose-limiting toxicities at 60-100 mg. If maximal BET dependence yields roughly one in five unconfirmed responses at a platelet-limited dose, a BET dependency inferred from super-enhancer occupancy in PAX3-FOXO1 rhabdomyosarcoma is a much weaker basis for calling the partner 'more druggable'. Piha-Paul SA et al. 2019
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The same class logic covers the KMT2A super elongation complex, whose AFF1/AFF4 scaffold assembles ELL and P-TEFb into the complex that all KMT2A chimeras share.

    PMID 20159561DOI
    What this evidence cannot show

    This paper establishes complex composition, not phase behaviour; the claim that KMT2A::AFF1 is a condensate-driven fusion in the same sense as NUP98-HOXA9 is not supported by it and is flagged speculative in this node.

    Evidence that argues against this (2)
    • This is the entry's cross-family generalisation and it does two things the cited paper does not license. First, SEC phase separation is a property of NORMAL transcription, not a neomorphic property of the fusion: ENL and its IDR are sufficient to initiate liquid droplet formation of the SEC in unmodified cells, AFF4 fluidises those droplets, and the droplets form within minutes of serum exposure as part of ordinary rapid transcriptional induction. MLL fusion boosts SEC phase separation on top of a condensate the healthy cell already makes. So condensate formation cannot by itself be what distinguishes the leukaemic state, and this class label does not separate KMT2A-fusion transcription from the serum response of a normal cell. Guo C et al. 2020
    • Second, 'all KMT2A chimeras' is the recurring over-reach of this literature and it is falsified by the cytoplasmic partners. AF1p/EPS15 and GAS7 are cytoplasmic proteins with no elongation function; their coiled-coil oligomerisation domains alone are necessary AND sufficient for transformation by the corresponding MLL fusions, and synthetic dimerisation modules with no biology at all reproduce Hox activation and enhanced progenitor self-renewal. Forced self-association, not delivery of the AFF1/AFF4-nucleated SEC, is the description that covers every partner. The SEC statement is true of the elongation-class partners on which it was demonstrated. So CW et al. 2003
Connected to