EWSR1::FLI1 — the comparator that shows condensate formation has an optimum, and that forcing a fusion into the nucleolus switches it off
also called EWS-FLI1 · prion-like domain BAF · GGAA microsatellite · LCD hub · nucleolar sequestration
In Ewing sarcoma the fusion protein's sticky disordered tail recruits a chromatin-remodelling machine to repetitive GGAA stretches, and that tail's phase behaviour is required. But the same group that imaged it found that pushing the stickiness higher switches the target genes off — and that deliberately dragging the fusion into the nucleolus, a natural droplet, stops it driving cancer at all.
The same thing, for a scientist
The Ewing comparator is the best-controlled fusion-condensate system and it delivers two results that cut against the naive 'more condensate, more oncogenic' reading. (1) **The prion-like domain retargets a chromatin remodeller.** Boulay et al. showed the BAF (mSWI/SNF) complex interacts with EWSR1 and is recruited by EWS-FLI1 to tumour-specific enhancers at GGAA microsatellites, contributing to target activation; this is a neomorphic property of the fusion relative to wild-type FLI1 and **depends on tyrosine residues necessary for phase transitions of the EWSR1 prion-like domain**, with short EWSR1 fragments fused to FLI1 sufficient to recapitulate BAF retargeting [E1]. (2) **The mechanism at physiological levels is a dynamic hub, not a demixed phase.** Chong et al.'s live-cell single-molecule imaging showed transcription-factor low-complexity domains forming local high-concentration hubs that stabilise DNA binding, recruit Pol II and activate transcription, with selective, highly dynamic LCD-LCD interactions and **no detectable phase separation under physiological conditions** [E2]. (3) **There is an optimum, and the nucleolus is past it.** Chong et al. 2022 found EWS::FLI1 requires a narrow optimum of LCD-LCD interactions to activate GGAA-microsatellite target genes; increasing interactions toward putative LLPS **represses** those genes in patient-derived cells; and ectopically sequestering EWS::FLI1 into the nucleolus, a well-documented LLPS compartment, **inhibits EWS::FLI1-driven transcription and oncogenic transformation** [E3]. For this graph the third result is the important one, because the one compartment in which a KMT2A fusion has been photographed is the nucleolus. On the Ewing precedent, nucleolar partition of MLL-AF4 is at least as likely to be an inactive sink as an active site — and, read the other way, it suggests a therapeutic strategy nobody has tried for KMT2A: not blocking the fusion, but relocalising it.
Three findings, in the order they should be read1 deeper
Phase behaviour matters — Boulay 2017. BAF is mutated in over 20% of human tumours; in Ewing it is recruited by EWS-FLI1 to tumour-specific enhancers at GGAA microsatellites. The requirement maps to tyrosines needed for EWSR1 prion-like-domain phase transitions, and short EWSR1 fragments fused to FLI1 suffice 1. Note the causal structure: the phase-competent residues are required for retargeting a remodeller, i.e. for composition, not for the existence of a droplet.
But the physiological object is a hub — Chong 2018. Single-molecule live imaging of LCD hubs at synthetic and endogenous loci: they stabilise DNA binding, recruit Pol II, activate transcription, are dynamic and partner-selective, and occur without detectable phase separation at physiological levels 2. The same observable, a bright transcriptionally active focus, is produced without demixing.
And there is a maximum — Chong 2022. Quantitative single-cell and single-molecule imaging in patient-derived Ewing cells: EWS::FLI1 needs a narrow optimum of LCD-LCD interaction to activate GGAA-microsatellite targets; pushing past it toward LLPS represses them; and forced nucleolar sequestration inhibits both transcription and transformation 3. The authors present the last as a candidate therapeutic strategy for disease-causing transcription factors.
The transfer to KMT2A, stated carefully
KMT2A::AFF1 is not EWS::FLI1. FLI1 is a sequence-specific ETS transcription factor whose DNA-binding is weak and whose oncogenic gain comes from the EWSR1 LCD; KMT2A brings its own chromatin-targeting apparatus (CXXC, AT-hooks, the menin-LEDGF tether) and the 3' partner brings elongation machinery rather than DNA-binding rescue. The architectures are inverted.
What transfers is the shape of the dose-response, which is a property of multivalent interaction physics rather than of any particular protein: too little interaction, no hub; too much, a sequestering phase. If AFF1's IDR contributes multivalency to a chromatin-tethered fusion, the same non-monotonic curve should apply, and every intervention that changes the fusion's effective valency — including overexpression in a model system — moves along it in a direction the experimenter has not controlled. That is a concrete reason the graph's overexpression-based nucleolar observation may be an artefact of a specific kind: pushing an IDR-containing fusion above its optimum drives it into the nucleolus and switches it off, which is exactly the experiment Chong et al. did deliberately.
The therapeutic idea the graph does not currently contain
Every KMT2A intervention in this graph works by blocking an interface (menin-MBM1, DOT1L, ENL YEATS, CBX8 chromodomain) or by degrading a protein. Chong et al. demonstrate a third modality: relocalise the oncoprotein into a compartment where it cannot work 3. They did it by engineering an LCD-LCD interaction that drags EWS::FLI1 into the nucleolus, and transformation stopped.
A chemically induced proximity approach — a bifunctional molecule that ties the fusion to a nucleolar or other sequestering scaffold rather than to an E3 ligase — is technically adjacent to the degrader chemistry already being developed for this target and is not in anyone's KMT2A pipeline. The KMT2A-specific handle would be the menin-binding motif or the CXXC domain, both of which have ligands or ligandable structures in this graph. The risk is obvious and should be stated: sequestration is reversible in a way degradation is not, so the pharmacology would demand continuous occupancy.
| compartment | LCD hub at GGAA microsatellites; nucleolus as an engineered sink |
|---|---|
| fusion studied | fusion-ewsr1-fli1 |
| retained module | EWSR1 prion-like / low-complexity domain (tyrosine-rich) |
| dose response shape | non-monotonic — a narrow optimum of LCD-LCD interaction, with excess repressing |
| forced nucleolar sequestration effect | inhibits transcription and oncogenic transformation |
| evidence class for that fusion | strong: patient-derived cells, live single-molecule imaging, engineered gain and loss of interaction |
| evidence class for kmt2a | none — no equivalent measurement exists |
| novel modality implied | chemically induced relocalisation of the oncoprotein rather than inhibition or degradation |
- A relocalisation drug is inherently reversible: unlike a degrader, it requires continuous occupancy, so intermittent exposure would restore the oncoprotein to its working location immediately.
- If condensate formation has an optimum, a partial condensate-disrupting drug could move cells from the repressive over-interacting regime back into the optimum and thereby increase oncogenic transcription — an on-target, dose-dependent worsening that a standard dose-escalation design would misread as lack of efficacy.
- The nucleolus is essential for ribosome biogenesis in every proliferating cell; dumping a large chromatin protein into it as a therapeutic strategy risks nucleolar stress toxicity in normal tissue, which the graph already flags for the MLL-AF4 nucleolar-stress finding.
- Does KMT2A::AFF1 show a non-monotonic dependence on AFF1 IDR valency? A titratable-valency series in an isogenic line would test whether the optimum principle generalises to a chromatin-tethered elongation fusion, and would reinterpret every overexpression result in this beat.
- Is the reported nucleolar localisation of MLL-AF4 an active state or an inactivated sink? The Ewing result makes the second hypothesis at least as probable, and the two predict opposite signs for any intervention that changes nucleolar partition.
- Could a chemically induced proximity agent sequester KMT2A::AFF1 into the nucleolus using its menin-binding motif or CXXC domain as the handle? This is a third modality alongside interface blockade and degradation, and no group is pursuing it in leukemia.
- Does AFF1's IDR recruit a chromatin remodeller the way the EWSR1 prion-like domain recruits BAF? The AFF1 interactome has been mapped for elongation factors; the remodeller question has not been asked with IDR mutants.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The BAF (mSWI/SNF) chromatin remodelling complex interacts with EWSR1 and is recruited by EWS-FLI1 to tumour-specific enhancers at GGAA microsatellites where it contributes to target gene activation; this is a neomorphic property of EWS-FLI1 relative to wild-type FLI1 and depends on tyrosine residues necessary for phase transitions of the EWSR1 prion-like domain, and fusion of short EWSR1 fragments to FLI1 is sufficient to recapitulate BAF retargeting and EWS-FLI1 activities.
Cancer-Specific Retargeting of BAF Complexes by a Prion-like Domain · Boulay G, Sandoval GJ, Riggi N, Iyer S, Buisson R, Naigles B, Awad ME, Rengarajan S, Volorio A, McBride MJ, Broye LC, Zou L, Stamenkovic I, Kadoch C, Rivera MN · Cell 2017What this evidence cannot show
Ewing sarcoma cells and mesenchymal stem cells with expressed constructs; the tyrosine mutants that block phase transition also change the protein's chemistry generally, so 'depends on phase transition' is inferred from the residue class rather than demonstrated by an orthogonal condensate perturbation.
Evidence that argues against this (1)
- 'Depends on tyrosines required for phase transitions' is an inference from residue class, and the same laboratory system that defined those transitions later measured the physiological behaviour and did not find a phase. Chong et al. showed by live-cell single-molecule imaging that transcription-factor low-complexity domains - the class the EWSR1 prion-like domain belongs to - form local high-concentration hubs at genomic loci that stabilise DNA binding and recruit Pol II, that the LCD-LCD interactions are dynamic and partner-selective, and that at physiological levels these interactions occur WITHOUT detectable phase separation, while remaining differentially hexanediol-sensitive. Tyrosine mutants therefore remove multivalent aromatic contacts that drive hub formation and phase separation alike, so BAF retargeting by EWS-FLI1 is equally well explained as avidity at GGAA microsatellites, with no demixed phase required. Chong S et al. 2018
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Live-cell single-molecule imaging shows transcription-factor low-complexity domains forming local high-concentration interaction hubs at synthetic and endogenous genomic loci that stabilise DNA binding, recruit RNA Pol II and activate transcription, with dynamic, partner-selective LCD-LCD interactions occurring under physiological conditions without detectable phase separation.
Imaging dynamic and selective low-complexity domain interactions that control gene transcription · Chong S, Dugast-Darzacq C, Liu Z, Dong P, Dailey GM, Cattoglio C, Heckert A, Banala S, Lavis L, Darzacq X, Tjian R · Science 2018What this evidence cannot show
Model transcription factors in cultured cells; absence of detectable phase separation is bounded by imaging sensitivity. Neither KMT2A nor AFF1 was tested.
Evidence that argues against this (1)
- The hub-not-condensate framing that this entry borrows from Chong et al. 2018 was contested in the same year by imaging of similar power. Cho et al. showed by live-cell superresolution and light-sheet microscopy in mouse embryonic stem cells that Mediator and Pol II form large stable clusters that colocalise, associate with chromatin, have properties of phase-separated condensates and are sensitive to transcriptional inhibitors. 'Dynamic hub' and 'condensate' are not cleanly separable by the measurements available, and a comparator entry should say so rather than adopt one camp's vocabulary. Read: abstract only. Cho WK et al. 2018
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
EWS::FLI1 requires a narrow optimum of low-complexity-domain interactions to activate its GGAA-microsatellite-associated target genes; increasing LCD-LCD interactions toward putative liquid-liquid phase separation represses transcription of those genes in patient-derived cells; and ectopically creating LCD-LCD interactions to sequester EWS::FLI1 into the nucleolus, a well-documented LLPS compartment, inhibits EWS::FLI1-driven transcription and oncogenic transformation.
Tuning levels of low-complexity domain interactions to modulate endogenous oncogenic transcription · Chong S, Graham TGW, Dugast-Darzacq C, Dailey GM, Darzacq X, Tjian R · Molecular Cell 2022What this evidence cannot show
The sequestration is engineered, so it demonstrates what forced nucleolar partition does rather than what any fusion does spontaneously; the tuning is achieved by altering LCD valency, which could have effects beyond the amount of demixing. Ewing sarcoma, not leukemia.
Evidence that argues against this (1)
- The sign of the LCD/condensate effect on FET-fusion transcription is disputed, and this comparator entry currently reports only one side. Zuo et al. developed an in-vitro single-molecule assay to visualise phase separation directly on DNA and found FET fusion proteins phase-separate AT their target binding loci, that the resulting condensates recruit RNA polymerase II and ENHANCE gene transcription, and that there is a threshold number of fusion-binding DNA elements above which condensation is favoured - offering a mechanism for why GGAA microsatellite repeat number matters. Against Chong et al. 2022, in which increasing LCD-LCD interaction toward LLPS represses EWS::FLI1 targets, the two results disagree on whether more demixing means more or less transcription. Zuo is reconstituted and Chong is in patient-derived cells, so Chong has the better system, but the disagreement is real and the comparator should not be exported to KMT2A as a settled 'more condensate equals off'. Read: abstract only. Zuo L et al. 2021
- related to →The nucleolus — the one compartment with direct published images of a KMT2A fusion inside it
The sign of the condensate effect is disputed within the FET literature (Zuo 2021, PMID 33674598 finds loci-specific LLPS enhances transcription), so the EWS::FLI1 precedent cannot be used to argue that MLL-AF4's reported nucleolar partition is an off state
- related to →EWSR1::FLI1 (EWS-FLI1), t(11;22)(q24;q12)
prion-like domain retargets BAF; LCD interaction has a narrow optimum
- refutes →The nucleolus — the one compartment with direct published images of a KMT2A fusion inside it
forced nucleolar sequestration inactivates a fusion TF, so nucleolar partition may be an off state
the tyrosine-rich prion-like domain is the phase-competent module
supplies the hub-without-phase-separation alternative and the non-monotonic dose-response
a disordered low-complexity module welded to a DNA-binding factor - the same architecture, a different family
- ← related toTAF15 (RBP56, TAF2N) - the FET-family, prion-like partner that donates a low-complexity N-terminus to ZNF384
the comparator that would have to be repeated for TAF15::ZNF384; no condensate experiment exists for any ZNF384 fusion
a FET N-terminus on a DNA-binding partner; every condensate experiment that exists for EWSR1::FLI1 is absent here
imports the hub-imaging, hexanediol and tyrosine-mutant assays already defined for EWSR1::FLI1, and its warning that the dose-response is non-monotone
a second system in which forcing a driver into the nucleolus switches it off