Oncofusion

start / graph / hyp-xpo1-is-a-shared-address-for-hoxa-driven-leukaemia

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q61-export-axis

Chromatin-bound XPO1 as a shared, druggable address for HOXA-driven leukaemia - and the objection that it may be a general chromatin cofactor rather than an address at all

also called the XPO1 address hypothesis · one platform, many handles

The idea is that a single protein parked on the HOX genes acts as a docking station that many different leukaemia mutations grab onto, and that blocking it could treat all of them at once. The uncomfortable extension is that KMT2A-rearranged leukaemia, which was supposed to use a completely different route, might be using it too.

The same thing, for a scientist

**Statement.** A locus-restricted pool of chromatin-bound XPO1 at the HOXA/HOXB clusters (and possibly MEIS1 and FOXC1) is a shared addressing platform. Seven leukaemia lesions dock onto it through two interchangeable grip modules - FG repeats or a leucine-rich NES - and their common output, aberrant HOX expression with a differentiation block, is a property of the platform rather than of any one lesion. **Therefore XPO1 inhibition is a genotype-spanning intervention for HOXA-driven leukaemia**, and the correct patient selection is by *address usage*, not by fusion identity. **The extension, which is the part worth testing.** Two 2026 sources - a preprint and a commentary - place **menin and KMT2A in the same assembly as XPO1**. If that is right, the platform is not KMT2A-free after all, and the menin interface and the XPO1 address are two handles on one object rather than two competing addressing systems. **Prior.** Moderate for the core statement (0.6) - the seven-lesion convergence with a de novo-NES necessity experiment is strong. **Low for the extension (0.2)** - one preprint and one commentary, both unreplicated. **What is already true and needs no test:** the platform exists in LOUCY, MEGAL, mouse ES cells, t(6;9) primary AML and cord-blood UBTF-TD models; XPO1 inhibitors displace the lesions from it; and XPO1 inhibition has been tested in four preclinical KMT2A-rearranged studies, all combinations, all myeloid. **What is not true and is often assumed:** that XPO1 is a *selective* dependency. It is not - DepMap mean gene effect **-2.14** across 1,208 lines, KMT2A-selectivity rank 2,228 of 17,728, and the near-zero difference sits in the floor-compression regime where selectivity cannot be measured. **THE PREMISE IS NOW CONTESTED, 2026-09-21 (Q68).** The statement above opens with 'a LOCUS-RESTRICTED pool of chromatin-bound XPO1'. Locus-restriction is what makes this an address rather than a background, and it rests on ChIP-seq from one laboratory in LOUCY and MEGAL. Chen et al. (PMID 38917791, Cell Chem Biol 2024), using a molecular-glue degrader and an XPO1-C528S resistance-allele control, report that **XPO1 and p300 co-occupy hundreds of chromatin loci** and that degrading XPO1 reduces p300's occupancy so that the degrader phenocopies p300 inhibition. If chromatin-bound XPO1 is that widespread, then (a) finding it at HOXA in a HOX-driven line is the expected observation and discriminates nothing, (b) the seven lesions may be docking onto a background rather than sharing an address, and (c) removing it moves many programmes at once, which is a different therapeutic proposition from a genotype-spanning targeted intervention. **This is the single strongest published objection to the hypothesis and it is not answered by any evidence item here.** It is falsifiable cheaply: an XPO1 CUT&RUN with a matched p300 track in one KMT2A-rearranged line.

Go deeper
Predictions, in descending order of how cheap they are to check
  1. An XPO1 CUT&RUN in a KMT2A-rearranged B-lineage line will show either no locus-restricted signal (core hypothesis holds, extension dies) or a HOXA-restricted one (extension lives). exp-xpo1-chip-in-kmt2a-rearranged-with-menin-inhibitor.
  2. MEN1 pocket resistance mutations will NOT confer cross-resistance to selinexor. If they do, the reported menin-XPO1 link is two independent cytotoxicities.
  3. The focal HOXA H3K79me2 peak in CALM-AF10 cells will collapse under an XPO1 inhibitor while the global H3K79 deficit is unchanged. exp-locus-resolved-h3k79me2-under-xpo1-inhibition.
  4. A heterologous NES fused in place of the NUP214 FG repeats will address SET::NUP214 to HOXA. The reciprocal of the CALM-AF10 substitution experiment, never run for a nucleoporin fusion.
  5. FOXC1 will be an output of the SET::NUP214 and NPM1c addresses too, not only DEK::NUP214. If it is, the axis is a FOXC1/HOX programme.
  6. Chromatin-bound XPO1 will be present at HOX in normal haematopoietic progenitors. If it is absent there and present in leukaemia, the platform is induced and the whole 'pre-bound' framing is backwards.
What would kill it, stated so it can actually be killed
  • A chromatin fractionation showing the XPO1 ChIP signal is RanGTP- and cargo-loaded, i.e. a transport intermediate passing close to DNA rather than a resident platform. That reduces the whole axis to a crosslinking artefact with a consistent bias.
  • An XPO1 ChIP in a lesion-free primary progenitor showing the same HOX occupancy at the same intensity, with the lesions changing nothing about it. The HEL observation in PMID 31755865 is a partial version of this and is already on record as counter-evidence.
  • Locus-restricted XPO1 occupancy failing to reproduce outside the Oka laboratory. Two of the four foundational chromatin papers are from one group; DEK::NUP214 and UBTF-TD are independent and are the reason the hypothesis survives that concern at all.
  • Selinexor displacing the fusion from HOXA without changing HOX expression in any system, which would sever the occupancy-to-output link the therapeutic argument needs.
  • For the extension specifically: menin occupancy unchanged under selinexor in a KMT2A-rearranged cell, in a peer-reviewed experiment.
What it would mean if it held

Patient selection by address usage rather than by fusion. The candidate group is larger than any single fusion: SET::NUP214 and DEK::NUP214 T-ALL and AML, NUP98-fusion AML, NPM1-mutant AML (the largest genotype in adult AML), NPM1-rearranged paediatric AML, UBTF-TD AML, PICALM::MLLT10 - and, if the extension holds, KMT2A-rearranged disease as well.

And there is an existing assay for enrolment. XPO1 mRNA rises 2-fold or more under selinexor as a pharmacodynamic marker of target engagement, demonstrated in 15 of 16 children in the paediatric phase 1. That is a target-engagement readout, not an address-usage readout - and the gap between those two is exactly the biomarker this hypothesis lacks.

Facts and figures
statementA locus-restricted pool of chromatin-bound XPO1 at HOXA/HOXB (and possibly MEIS1 and FOXC1) is a shared addressing platform used by at least seven leukaemia lesions through interchangeable FG-repeat or NES grip modules, making XPO1 inhibition a genotype-spanning intervention for HOXA-driven leukaemia; the untested extension is that KMT2A-rearranged disease uses the same platform via menin.
originliterature-gap
prior probability0.6
prior probability extension0.2
candidate patient group if trueSET::NUP214 and DEK::NUP214 T-ALL and AML, NUP98-fusion AML, NPM1-mutant AML, NPM1-rearranged paediatric AML, UBTF-TD AML, PICALM::MLLT10 - plus KMT2A-rearranged disease only if the extension holds
missing biomarkerXPO1 mRNA induction measures target engagement, not address usage; no assay exists for whether a patient's lesion is using the chromatin platform
predictions
[
  "XPO1 CUT&RUN in a KMT2A-rearranged B-lineage line gives either no locus-restricted signal (core holds, extension dies) or a HOXA-restricted one (extension lives)",
  "MEN1 pocket resistance mutations do NOT cross-resist selinexor",
  "the focal HOXA H3K79me2 peak in CALM-AF10 cells collapses under XPO1 inhibition while the global H3K79 deficit is unchanged",
  "a heterologous NES replacing the NUP214 FG repeats still addresses SET::NUP214 to HOXA",
  "FOXC1 is an output of the SET::NUP214 and NPM1c addresses, not only DEK::NUP214",
  "chromatin-bound XPO1 is present at HOX in normal haematopoietic progenitors"
]
experiments
[
  "exp-xpo1-chip-in-kmt2a-rearranged-with-menin-inhibitor",
  "exp-locus-resolved-h3k79me2-under-xpo1-inhibition"
]
analyses
[
  "ana-xpo1-inhibition-in-kmt2a-rearranged-bounded",
  "ana-xpo1-chromatin-measurements-never-made",
  "ana-xpo1-e571k-across-leukaemia-cohorts"
]
falsifiers
[
  "chromatin fractionation showing the XPO1 ChIP signal is RanGTP- and cargo-loaded, i.e. a transport intermediate rather than a resident platform",
  "an XPO1 ChIP in a lesion-free primary haematopoietic progenitor showing the same HOX occupancy at the same intensity, with the lesions changing nothing",
  "locus-restricted XPO1 occupancy failing to reproduce outside the originating laboratory in a nucleoporin-fusion system",
  "selinexor displacing a fusion from HOXA without changing HOX expression in any system",
  "for the extension: menin occupancy unchanged under selinexor in a KMT2A-rearranged cell, peer reviewed"
]
lesions on the route
[
  "NUP98::HOXA9",
  "SET::NUP214",
  "DEK::NUP214",
  "NPM1c",
  "NPM1::MLF1 and NPM1::CCDC28A",
  "CALM-AF10 / PICALM::MLLT10",
  "UBTF-TD"
]
How this could bite us back
  • XPO1 is common-essential. A genotype-spanning intervention whose target every cell needs is a genotype-spanning toxicity as much as a genotype-spanning therapy.
  • The only randomised test of selinexor in acute leukaemia was negative and numerically unfavourable (NCT02088541, median OS 3.2 vs 5.6 months), and the only paediatric trial found dose-limiting cerebellar toxicity. A hypothesis this attractive has to be read against that.
  • XPO1 inhibition induces P2RY2-AKT signalling in AML, an adaptive escape route present from the first dose.
  • If the extension holds, menin inhibition and XPO1 inhibition are not orthogonal, and a resistance mechanism that loosens the shared assembly would reduce sensitivity to both - the opposite of the usual case for combining them.
  • Selecting patients by 'address usage' requires an assay nobody has built, so in practice enrolment would default to fusion lists, which is what the hypothesis claims to improve on.
Still open
  1. What is the biomarker for address usage? Without one, a genotype-spanning trial has no enrolment criterion beyond the fusion list it was supposed to replace.
  2. Six of the seven lesions are myeloid. Is the platform a myeloid-progenitor feature, and does that alone exclude infant B-lineage disease from the hypothesis?
  3. If the extension holds and menin and XPO1 are one assembly, what is the right sequencing - simultaneous, or XPO1 inhibition at menin-inhibitor relapse?
  4. Can the platform be targeted without inhibiting export? Every existing agent hits Cys528 in the NES groove, and a chromatin-selective XPO1 ligand does not exist even as a concept.
  5. Is the chromatin-bound XPO1 pool locus-restricted at all? PMID 38917791 reports XPO1 co-occupying hundreds of loci with p300 in human cells; if that is what is at HOXA too, the 'address' is a background and this hypothesis fails on its first word. One XPO1 CUT&RUN with a matched p300 track settles it.
  6. Does any published XPO1-inhibitor chromatin result carry an XPO1-C528S resistance-allele control? None cited in this beat does, and PMID 38917791 shows the control is available and decisive.
Every source behind this entry (5)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Chromatin-bound CRM1 recruits SET-Nup214 and NPM1c onto HOX clusters in human leukaemia cells and is upstream of the fusion, since SET-Nup214 knockdown lowers HOX expression without significantly reducing chromatin-bound CRM1.

    PMID 31755865DOI
    What this evidence cannot show

    Human leukaemia cell lines only; one laboratory; the same paper reports CRM1 at HOX in HEL, which carries none of these lesions, and calls the significance undetermined.

    Evidence that argues against this (1)
    • The HEL observation in the same paper is the strongest published argument against locus-restricted XPO1 being lesion-related at all. Oka M et al. 2019
  2. E2strong 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 and drive aberrant chromatin binding in human cord-blood CD34+ models; mutagenesis of those 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; one group; paediatric and AYA myeloid disease. This is the single strongest support for the hypothesis because the grip module is created de novo, removing the 'the NES was already doing transport' alternative. The abstract does not name the target loci, so whether UBTF-TD uses the SAME address is not established.

    Evidence that argues against this (1)
    • PUBLISHED CONFLICTING RESULT on the scope of the lesion this item calls the hypothesis's single strongest support. Kaburagi et al. detected UBTF internal tandem duplication in 1.2% (6 of 503) of Japanese paediatric de novo AML, in 0 of 175 ADULT AML patients, and in 0 of 65 cell lines - including 15 AML, 39 ACUTE LYMPHOBLASTIC LEUKAEMIA, five CML and six neuroblastoma lines. Five of the six UBTF-ITD cases also carried FLT3-ITD and all six had high PRDM16 expression. A lesion found in six children, absent from every one of 39 lymphoid lines tested, and co-occurring with FLT3-ITD in five of six cases, is a narrow and confounded base for a hypothesis that has to reach infant B-lineage KMT2A::AFF1 - which is not among the seven lesions this entry generalises over either. Kaburagi T et al. 2023
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In 57 primary AML samples, t(6;9)/DEK::NUP214 cases showed a selective response to selinexor and eltanexor, and CUT&RUN showed both DEK::NUP214 binding at FOXC1 and HOXA/HOXB and the selective reduction of that binding under 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. The only instance of the route with primary patient samples and an occupancy readout that moves under the drug, and the independent replication that keeps the hypothesis from resting on one laboratory. The number of t(6;9) cases within the 57 is not stated. 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)
    • NAMED METHODOLOGICAL DEFEATER: the occupancy half of this item has no drug-resistant-allele control. Chen et al. establish the standard directly for XPO1 - every phenotype they attribute to XPO1 degradation is shown to be abrogated in cells expressing the known XPO1-C528S drug-resistance allele - and in the same work they show that reducing XPO1 pulls p300 off hundreds of loci. A CUT&RUN showing that DEK::NUP214 occupancy falls under a SINE therefore has two readings: the fusion is held there by XPO1, or the SINE has broadly disturbed chromatin occupancy at active loci. A C528S arm separates them and was not run. Chen YF et al. 2024
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    XPO1 is common-essential with a DepMap mean gene effect of -2.142 across 1,208 lines, and shows a KMT2A-selectivity difference of only -0.070 (rank 2,228 of 17,728) between 11 KMT2A-rearranged and 43 other leukaemia lines.

    DepMap KMT2A-selective dependency scan (full genome), analyses/depmap-kmt2a-selective-scan-full · Oncofusion Knowledge Graph project · internal analysis over DepMap CRISPR gene effect 2026
    What this evidence cannot show

    Read from this graph's own scan on 2026-09-20. Floor compression makes the selectivity number uninterpretable rather than null. Cited here as the constraint on the hypothesis, not as support: whatever is true of the address, the target has no intrinsic therapeutic index.

    Evidence that argues against this (1)
    • PUBLISHED CONFLICTING RESULT that measures the therapeutic index directly rather than inferring it from a screen, and it is worse than the DepMap number implies. Leptomycin B - elactocin, the prototype global XPO1 inhibitor - reached a human phase 1. Across five schedules the dose-limiting toxicity was profound anorexia and malaise, there were no partial or complete responses, and the authors recommended against further trials. That is an in-human measurement of what blocking XPO1 non-selectively does, and it does not depend on floor compression, on Cellosaurus lineage labels or on how a Chronos score scales. The hypothesis's constraint is not that XPO1 has no measured therapeutic index; it is that the one that was measured was unacceptable. Newlands ES et al. 1996
  5. E5weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.preprint

    A 2026 preprint reports that XPO1 stabilises menin's binding to chromatin and its interactions with KMT2A and KMT2A/MLLT3, and that ziftomenib plus selinexor is synergistic in KMT2A-rearranged and NPM1-mutant AML models including primary CD34+ progenitors and patient-derived xenografts.

    PMID 41959134DOI
    What this evidence cannot show

    PREPRINT, not peer reviewed, abstract-level, single group, myeloid models only. This is the sole basis for the hypothesis's extension into KMT2A-rearranged disease and the extension's prior is set at 0.2 because of it.

    Evidence that argues against this (1)
    • AUTHORS' OWN STATED LIMITATION in the work the extension's prior rests on next to. The synergy the preprint reports is a combination-index measurement on viability, and the most relevant published constraint on that class of result comes from the authors of the schedule work: selinexor given once or twice weekly produced only a TEMPORARY disruption of the XPO1-NPM1c interaction, which they give as the reason it did not translate into clinical benefit. A synergistic CI in a 72-hour culture does not speak to whether the interaction is disrupted for long enough in a patient, and the preprint's xenograft arm does not report interaction duration either. The extension's prior of 0.2 is generous for a single unreviewed group reporting a viability endpoint on the one question the same class of agent has already failed clinically. Pianigiani G et al. 2022
Connected to