Oncofusion

start / graph / mechanism-nes-crm1-chromatin-addressing

Mechanismstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q61-export-axis

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

also called pre-bound XPO1 platform · chromatin-bound CRM1 as an address · NES-CRM1 addressing · FG-repeat-CRM1 addressing

Several completely different leukaemia mutations end up switching on the same set of HOX genes. The thing they share is not a DNA-binding protein but a lift operator: the export carrier XPO1 is already parked on those genes, and each lesion carries a different handle that grips it.

The same thing, for a scientist

**The route, stated as a route.** XPO1/CRM1 is pre-bound at HOX cluster chromatin. A lesion that carries an XPO1-gripping module - **FG repeats** (nucleoporin fusions) or a **leucine-rich NES** (NPM1c, CALM) - is thereby delivered to those loci without any DNA-binding domain of its own. The partner half then supplies the chromatin or transcriptional activity. This is **addressing by receptor capture**, and it is categorically different from the CXXC/menin addressing that a KMT2A fusion uses. **The instances, each in the system it was measured in:** | lesion | grip on XPO1 | system measured | evidence | |---|---|---|---| | **NUP98::HOXA9** | FG repeats | mouse embryonic stem cells | [E1] | | **SET::NUP214** | NUP214 FG repeats | human T-ALL line LOUCY, AML line MEGAL | [E2] | | **NPM1c** | C-terminal NES created by the frameshift | human AML lines; OCI-AML3 | [E2] | | **NPM1::MLF1 and NPM1::CCDC28A** | NPM1-encoded, XPO1-dependent | murine bone marrow transplantation | [E3] | | **CALM-AF10** | CALM NES aa 544-553 (human PICALM, UniProt Q13492) | murine bone marrow; a CRM1-AF10 chimera is leukaemogenic | see `mechanism-calm-af10-crm1-nuclear-export-requirement` | | **DEK::NUP214** | NUP214 FG repeats | **57 primary AML samples**; CUT&RUN at FOXC1 and HOXA/HOXB | [E6] | | **UBTF-TD** | **NES motifs created DE NOVO by the duplication** | human cord-blood CD34+ models | [E7] | **Verdict on the King's question.** This is a **shared route, not three coincidences**, and the count has grown from three to seven during this beat alone. The strongest reasons: (i) the addressing was demonstrated first in a system with no leukaemia fusion of this graph's kind at all - mouse ES cells expressing NUP98-HoxA9 [E1]; (ii) the same group's order-of-recruitment test shows XPO1 is upstream, not cargo [E2]; (iii) the CALM-AF10 arm supplies the reciprocal proof, because **fusing CRM1 itself to AF10 causes leukaemia and mutating the CRM1 residues that contact NUP214 abolishes it**; (iv) the grip module is interchangeable - heterologous NES motifs from ABL1, Rev, PKIA and APC substitute for the CALM NES; and (v) **UBTF-TD invents its NES from scratch and loses the chromatin phenotype when that invented NES is mutated** [E7], which removes the last alternative explanation, that the grip module was already doing transport for some other reason. **Two things this route does not establish.** It does not say what puts XPO1 on HOX chromatin, and it does not say the route is fusion-specific: CRM1 signals at HOX are also present in **HEL**, which carries none of these lesions [E2].

Go deeper
Why 'coincidence' is the wrong null, and what would have made it right

A sceptic's null is that CRM1 is an abundant, sticky protein that shows up in any ChIP and any pulldown, and that five groups have each found the same artefact. Three observations break that null:

  1. Selectivity within one genome. In LOUCY the CRM1 signal is not uniform - it is concentrated at HOXA and especially HOXB, and is absent in K562 and HL60 2. A sticky-protein artefact does not respect cluster boundaries or cell identity.
  2. Directionality. Knocking the fusion down leaves CRM1 on chromatin; inhibiting CRM1 removes the fusion 2. An artefact has no direction.
  3. Sufficiency from the other side. A CRM1-AF10 chimera - the receptor replacing the 5' partner entirely - transforms, and it stops transforming when the NUP214-contact residues are mutated (see mechanism-calm-af10-crm1-nuclear-export-requirement). You cannot get that from a contaminant.

What would still refute the route: an XPO1 ChIP in a lesion-free primary haematopoietic cell showing the same HOX occupancy with the same intensity, i.e. that the 'platform' is simply where XPO1 always is and the lesions are irrelevant to it. The HEL observation 2 is a partial version of exactly that and is recorded as counter-evidence, not smoothed away.

The rival reading published in 2025: condensates, not addresses

A 2025 Cell paper reaches the same convergence by a different mechanism. It reports that NPM1c forms nuclear condensates - 'coordinating bodies' (C-bodies) - in human cell lines, mouse models and primary patient samples, that NPM1c phase separation is necessary and sufficient to recruit NUP98 and KMT2A into them, and that nucleoporin and KMT2A fusion proteins form condensates biophysically indistinguishable from NPM1c condensates 4. A commentary on it lists the co-partitioned components as NUP98, KMT2A/MLL1, menin and XPO1/CRM1, and notes that inhibiting either the XPO1 or the menin interaction alters condensate composition and function 5.

These two models are not the same claim. The address model says a pre-existing, locus-restricted XPO1 platform determines where. The C-body model says multivalent condensation determines what assembles, with the locus following. They are compatible - the 2026 perspective explicitly proposes that XPO1 docking lowers the LLPS threshold [see gene-xpo1 E2] - but they make different predictions, and the second one puts KMT2A and menin inside the same object as XPO1, which the address model never required.

This graph's standing caveat on phase-separation assays applies in full to 4 and 5: hexanediol sensitivity and classical FRAP cannot distinguish LLPS from binding to spatially clustered sites, and 'binding to spatially clustered sites' is exactly what the address model predicts.

What this does to the menin-interface hypothesis

Three effects, in increasing order of discomfort for the hypothesis:

  1. Neutral. KMT2A fusions keep CXXC and menin and do not need this route. A second addressing mechanism existing elsewhere does not weaken the first one.
  2. Awkward. The route reaches the same output genes without menin at all, which is the cleanest available demonstration that HOXA activation is not menin's private property.
  3. Directly challenging. A 2026 preprint reports that XPO1 stabilises menin's binding to chromatin and its interactions with KMT2A and KMT2A/MLLT3 in KMT2A-rearranged AML - see mechanism-xpo1-stabilises-menin-chromatin-binding. If that holds, XPO1 is not an alternative to the menin interface; it is upstream of it, and the two addressing systems this graph has been treating as separate are one system with two handles.
Facts and figures
mechanism classchromatin targeting by receptor capture
platformgene-xpo1 (chromatin-bound pool)
order of recruitmentXPO1 first - fusion knockdown does not reduce chromatin-bound XPO1
interchangeability evidenceheterologous NES motifs from ABL1, Rev, PKIA and APC substitute for the CALM NES in transformation assays; and UBTF tandem duplications CREATE an NES de novo, whose mutagenesis abolishes chromatin binding
never measured ininfant B-lineage KMT2A::AFF1, in any system, by any group, as of 2026-09-20
rival modelNPM1c C-body condensates (PMID 41192422), in which convergence is multivalent condensation rather than a locus-restricted pre-bound platform
lesion count7
grip modules
{
  "FG repeats": [
    "NUP98::HOXA9",
    "SET::NUP214",
    "DEK::NUP214",
    "DEK::NUP214"
  ],
  "leucine-rich NES": [
    "NPM1c C-terminal NES",
    "CALM aa 544-553"
  ],
  "receptor itself": "CRM1-AF10 chimera (engineered, leukaemogenic)",
  "de novo NES created by a mutation": [
    "UBTF tandem duplication"
  ]
}
loci reported
[
  "HOXA",
  "HOXB",
  "HOXC (title-only source)",
  "MEIS1 (perspective only)",
  "FOXC1 promoter (DEK::NUP214)"
]
systems
[
  "mouse embryonic stem cells (NUP98-HoxA9)",
  "human T-ALL LOUCY and AML MEGAL (SET-Nup214)",
  "human AML lines (NPM1c)",
  "murine bone marrow transplantation (NPM1 fusions, CALM-AF10, CRM1-AF10)",
  "57 primary AML samples (DEK::NUP214 drug screen and CUT&RUN)",
  "human cord-blood CD34+ (UBTF-TD)"
]
How this could bite us back
  • A shared address is also a shared toxicity: if the platform is a normal feature of HOX regulation, disrupting it hits normal HOX-dependent progenitors, and XPO1 is common-essential besides.
  • The route is defined by a module any protein can carry. There is no lesion-specific surface here to design a selective molecule against - which is precisely why the only agents available are pan-XPO1 inhibitors.
  • If the C-body reading is right, removing XPO1 from the condensate may reorganise rather than dissolve it, and a partially disassembled onco-condensate has never been characterised functionally.
Still open
  1. Is chromatin-bound XPO1 present at HOX in NORMAL haematopoietic progenitors? If it is, the platform is a normal feature the lesions exploit; if it is not, something about the lesion puts it there, and the address model has the causality backwards.
  2. Does KMT2A::AFF1 - which needs none of this - nevertheless sit near chromatin-bound XPO1? Nobody has run an XPO1 ChIP in a KMT2A-rearranged cell, so the question has never been asked in the disease this graph is about.
  3. Address or condensate? The two published models make different predictions about whether the locus determines the assembly or the assembly determines the locus, and the discriminating experiment (locus-resolved occupancy under condensate disruption that is not an XPO1 inhibitor) has not been done.
  4. The grip module is interchangeable in CALM-AF10. Is it interchangeable in SET::NUP214 - would a heterologous NES replacing the FG repeats address the fusion to HOX? That single construct would show whether the FG contact carries information beyond 'bind XPO1'.
  5. Seven lesions now use this route and six of the seven are myeloid. Is that biology - the HOX/MEIS programme is a myeloid-progenitor programme - or is it that nobody has looked in a lymphoid cell?
Every source behind this entry (7)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In mouse embryonic stem cells, NUP98-HoxA9 forms nuclear aggregates and is preferentially targeted and accumulated at Hox cluster regions where the export factor Crm1 is ORIGINALLY PREBOUND; leptomycin B disassembles the nuclear dots, causing loss of NUP98-HoxA9 chromatin binding and loss of its Hox gene activation.

    PMID 26740045DOI
    What this evidence cannot show

    MOUSE EMBRYONIC STEM CELLS with stable expression of a human fusion - not a leukaemia, not a haematopoietic cell, not a patient. Abstract-level for this beat. Leptomycin B is a covalent tool compound and not a clinical agent. This is the founding observation for the whole platform model and it comes from a non-haematopoietic system, which cuts both ways: it removes leukaemia-specific confounding and it removes leukaemia-specific relevance.

    Evidence that argues against this (1)
    • MOUSE EMBRYONIC STEM CELLS ARE NOT A LEUKAEMIA, AND LEPTOMYCIN B IS NOT A SELECTIVE PROBE OF DOCKING. mESCs have an unusually open, hyperdynamic chromatin state in which targeting experiments generalise poorly, and leptomycin B blocks XPO1-mediated export globally - so disassembly after treatment is consistent with docking being disrupted AND with the cell's whole nucleocytoplasmic economy being disrupted. Separating the two needs a docking-deficient XPO1 or an NES mutant, not a pan-inhibitor. Oncofusion knowledge graph (internal entry) 2026
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Chromatin-bound CRM1 recruits both SET-Nup214 and NPM1c onto HOX clusters in human leukaemia cell lines; CRM1 binding is selective for HOXA and especially HOXB in LOUCY and absent in K562 and HL60; SET-Nup214 knockdown lowers HOX expression without significantly reducing chromatin-bound CRM1; and KPT-330 removes both CRM1 and Nup214 peaks and lowers HOX expression.

    PMID 31755865DOI
    What this evidence cannot show

    Human leukaemia CELL LINES only; no patient sample, no xenograft, no infant B-lineage KMT2A::AFF1. Same group and reagents as [E1], so the platform model and this test are not independent. KPT-330 at 1,000 nM for 24 h is a high non-clinical exposure and also degrades SET-Nup214, confounding disassembly with protein loss.

    Evidence that argues against this (1)
    • The paper reports KPT-330-sensitive CRM1 binding at HOX regions in HEL cells, which carry none of these lesions, and says the biological significance is undetermined. If XPO1 sits on HOX chromatin in cells with no lesion, 'pre-bound platform' may describe a normal state that the lesions merely exploit - which is the interesting version - or an antibody artefact, which nothing in the paper excludes. Oka M et al. 2019
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In mouse bone-marrow transplantation assays, the rare paediatric AML fusions NPM1::MLF1 and NPM1::CCDC28A both bind the HOX gene cluster and cause aberrant HOX up-regulation IN COOPERATION WITH XPO1, and the XPO1 inhibitor selinexor suppresses HOX activation and colony formation driven by both fusions; NPM1::CCDC28A cells were also sensitive to menin inhibition.

    PMID 39443736DOI
    What this evidence cannot show

    MURINE bone-marrow transduction and transplantation with human fusions; rare paediatric AML fusions, not KMT2A-rearranged disease and not B-lineage. Abstract-level for this beat. This is the FOURTH and FIFTH lesion in the route and is the one that most directly supports generality, because neither fusion was in anyone's platform model when the work began. Note that the same paper reports menin-inhibitor sensitivity for NPM1::CCDC28A, so XPO1 and menin dependence are not mutually exclusive in one lesion.

    Evidence that argues against this (1)
    • 'IN COOPERATION WITH XPO1' IN A TRANSPLANT ASSAY IS A GENETIC INTERACTION, NOT A DEMONSTRATION OF CHROMATIN DOCKING. Murine bone-marrow transplantation reports whether leukaemia arises, and two rare paediatric fusions requiring XPO1 to up-regulate HOX is compatible with XPO1 acting anywhere upstream - including on the localisation of the fusion itself rather than on its residence at the locus. These are also rare lesions with few independent reports. Oncofusion knowledge graph (internal entry) 2026
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Mutant NPM1 (NPM1c) forms nuclear condensates in human cell lines, mouse models and primary patient samples; NPM1c phase separation is necessary and sufficient to recruit NUP98 and KMT2A into those condensates; the condensates are necessary for regulating gene expression, for in vivo leukaemic expansion and for maintaining the undifferentiated state; and nucleoporin and KMT2A fusion proteins form condensates biophysically indistinguishable from NPM1c condensates.

    Disparate leukemia mutations converge on nuclear phase-separated condensates. · Datar GK, Khabusheva E, Anand A, et al. · Cell 2025
    PMID 41192422DOI
    What this evidence cannot show

    Abstract-level for this beat. This graph's standing rule is that 1,6-hexanediol sensitivity and classical FRAP cannot distinguish LLPS from binding to spatially clustered sites, and that hexanediol itself condenses chromatin in living cells; 'pharmacological destabilization of phase separation' in the abstract is not further specified there, so the discriminating-assay question is open. 'Biophysically indistinguishable' is a similarity claim and not a shared-mechanism claim. The systems include primary patient samples, which is more than any chromatin-bound-CRM1 paper offers.

    Evidence that argues against this (1)
    • PHASE SEPARATION IS THE CLAIM CLASS THIS GRAPH HOLDS TO THE HIGHEST BAR, AND 'NECESSARY AND SUFFICIENT TO RECRUIT' IS A STRONG FORM OF IT. Condensate necessity is usually established with 1,6-hexanediol or with mutants that also change charge, folding and binding, so the separation-of-function is rarely clean. This graph's own standards entry requires endogenous tagging, concentration-matched controls and a mutant that abolishes condensation without abolishing binding before a condensate claim carries mechanistic weight. Oncofusion knowledge graph (internal entry) 2026
  5. E5weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.review

    A commentary on the C-body work states that NPM1c condensates show co-partitioning of NUP98, KMT2A/MLL1, menin and XPO1/CRM1, and that blockade of co-partitioned components - the XPO1/CRM1 or the menin interaction, by inhibitors - significantly alters condensate composition and functionality.

    A Unified Model: Chromatin-Bound Multicomponent Condensates. · Lau BK, Haarer EL, Ahn JH, et al. · Cancer Research 2026
    PMID 41481253DOI
    What this evidence cannot show

    A COMMENTARY, not a measurement, and read at abstract level. Its statement that menin and XPO1 co-partition in the same condensate is the single most consequential sentence in this beat for the menin-interface hypothesis, and it is a secondary source describing another group's figures. It is carried at weak confidence and must be checked against the primary paper's own figures before anything is built on it.

    Evidence that argues against this (1)
    • A COMMENTARY DESCRIBING ANOTHER GROUP'S FIGURES IS THE WEAKEST SOURCE CLASS IN THIS GRAPH, AND CO-PARTITIONING IS NOT THE CLAIM THE AXIS NEEDS. That menin and XPO1/CRM1 appear in the same NPM1c condensate says they are in one compartment; it does not say XPO1 positions menin, nor that either is required for the other's chromatin residence. Conflating co-partitioning with stabilisation would be exactly the overreach the companion entry explicitly refuses. Oncofusion knowledge graph (internal entry) 2026
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In 57 primary AML samples screened against more than 500 compounds, t(6;9)/DEK::NUP214 cases showed a selective response to selinexor and eltanexor; CUT&RUN showed DEK::NUP214 bound directly at the FOXC1 promoter and the HOXA/HOXB clusters; and both expression and fusion binding at those regions were selectively reduced by XPO1 inhibition.

    DEK::NUP214 acts as an XPO1-dependent transcriptional activator of essential leukemia genes. · Kaya F, Bewicke-Copley F, Miettinen JJ, et al. · Leukemia 2025
    PMID 40204893DOI
    What this evidence cannot show

    Abstract-level for this beat. Younger-adult AML; the number of t(6;9) cases within the 57 is not stated and could be very small. Recorded in full in `mechanism-dek-nup214-xpo1-dependent-transcriptional-activator`. This is the only instance of the route with primary patient samples AND an occupancy readout that moves under the drug. SCOPE CORRECTED 2026-09-21. The abstract of PMID 40204893 was re-read in full. The 57 primary AML samples are the MULTI-OMICS AND DRUG-SCREEN arm only - whole genome sequencing, targeted sequencing, transcriptomics and >500-compound screening - and that arm is what shows the selective t(6;9) response to selinexor and eltanexor. The CUT&RUN is a SEPARATE experiment with no n stated in the abstract, and the key result that DEK::NUP214 binding falls under XPO1 inhibition is stated as occurring 'in t(6;9) cells', not in the 57 primary samples. Attributing the occupancy readout to 57 primary patient samples makes it a much stronger claim than the paper supports. The conflation entered via a hand-back summary and was then repeated in a dispatch brief before being caught by the beat that received it.

    Evidence that argues against this (1)
    • TWO EXPERIMENTS, TWO POPULATIONS - AND THIS ITEM READS AS ONE. The 57 primary AML samples are the multi-omics and >500-compound drug-screen arm; the CUT&RUN is separate, with no n stated in the abstract, and the binding-falls-under-XPO1-inhibition result is stated as occurring 'in t(6;9) cells'. The occupancy readout is therefore not a 57-patient result. **This conflation travelled through a hand-back summary into a dispatch brief before a receiving beat caught it**, which is why it is recorded here rather than quietly fixed. Oncofusion knowledge graph (internal entry) 2026
  7. E7strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    UBTF tandem duplications create nuclear export signal motifs that interact directly with XPO1; in human cord-blood CD34+ models these interactions drive aberrant chromatin binding and transcriptional activation, mutagenesis of the NES motifs abolishes chromatin localisation and the downstream phenotypes, and XPO1 inhibition disrupts UBTF-TD chromatin localisation in preclinical human models.

    PMID 41182175DOI
    What this evidence cannot show

    Abstract-level for this beat. Paediatric and AYA myeloid disease; engineered cord-blood CD34+ models rather than patient blasts. The abstract does not name the target loci, so whether UBTF-TD uses the SAME loci as the other six lesions is unknown and is not asserted. Recorded in full in `mechanism-ubtf-td-de-novo-nes-xpo1-chromatin-binding`.

    Evidence that argues against this (1)
    • CORD-BLOOD CD34+ MODELS ARE THE RIGHT HUMAN SYSTEM AND STILL NOT THIS DISEASE, AND 'INTERACT DIRECTLY WITH XPO1' IS AN IN-CELL INTERACTION CLAIM RATHER THAN A STRUCTURE. UBTF-TD is the strongest argument in this entry that these contacts are docking rather than transport, because the NES motifs are created de novo - but it is one lesion, one group, and a myeloid model, while the index disease here is infant B-lineage KMT2A::AFF1, which is not among the seven lesions this entry generalises over. Oncofusion knowledge graph (internal entry) 2026
Connected to