Oncofusion

start / graph / hyp-nup98-partner-obligate-interface

Hypothesisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q11-comparative

NUP98 fusions share the menin dependency (confirmed preclinically) — and the partner-supplied reader/writer module is not why

also called NUP98 menin hypothesis · third genotype hypothesis

KMT2A-rearranged and NPM1-mutant leukaemias already share one drug because both need the same helper protein, menin. NUP98-rearranged leukaemia needs it too — this was shown in mice in 2022 — but through borrowed normal MLL1, not through the NUP98 partner module this entry originally blamed.

The same thing, for a scientist

Menin inhibition works in two genotypes (KMT2A-rearranged and NPM1-mutant) because both conscript the same MLL1-menin chromatin adaptor to maintain a HOXA/MEIS1 program. This entry originally predicted that NUP98 fusions would be menin-MLL1-dependent too, and more generally that the partner-supplied reader/writer module in each NUP98 fusion (NSD1 SET, KDM5A PHD) is an obligate interface. **The first half was already settled when this entry was written and is now recorded as confirmed: Heikamp et al. (Blood 2022, PMID 34582559) showed VTP50469 is active in NUP98-HOXA9 and NUP98-JARID1A cells at IC50s comparable to KMT2A-r and NPM1c, and prolongs survival in NUP98-NSD1 and NUP98-JARID1A PDX models.** The second half is **not** what that result shows. The dependency runs through wild-type MLL1 and menin -- both MLL1 and the NUP98 fusion are evicted from chromatin together -- so the drug prediction was right for a reason other than the one this entry gave. What remains genuinely open is the partner-module claim (is the NSD1 SET or KDM5A PHD an obligate, separately druggable interface?) and the entire clinical question, since the pivotal menin-inhibitor trials enrolled by KMT2A/NPM1 genotype and not by NUP98.

Go deeper
Why the prediction was reasonable (and had already been confirmed in 2022 — see the correction below)1 deeper

The menin dependency is a dependency of a program, not of a genotype. That is the whole content of the NPM1-mutant result: a CRISPR domain screen nominated the menin-binding site in MLL1, and menin-MLL1 inhibition was profoundly active in NPM1-mutant models even though there is no KMT2A rearrangement anywhere in those cells 1. The same chemistry reversed pre-leukaemic self-renewal in NPM1-mutant knock-in mice 2, and one clinical agent produced remissions in both genotypes 3.

NUP98 fusions converge on the same HOXA/MEIS1 output. NUP98-HOXA9 does it by phase-separating into condensates that broaden chromatin occupancy and generate CTCF-independent loops at proto-oncogenes 4. If maintaining HOXA expression in a myeloid progenitor requires wild-type MLL1 and therefore menin — which is what the NPM1-mutant result implies — then NUP98-r should inherit the dependency for exactly the same reason.

The honest status of the evidence

Menin dependency in NUP98-rearranged models has been reported in the preclinical literature and NUP98-rearranged patients have been included in menin-inhibitor trial cohorts, but this Queen did not verify a primary NUP98-specific menin-dependency paper within this cycle, so the node claims the analogy as a hypothesis rather than as established fact and cites only what was checked. Confirming or refuting this with a primary citation is the single highest-value follow-up for the next cycle, because if it is already established the hypothesis collapses into a literature-retrieval failure rather than a scientific gap.

The second, more novel half: the partner module as the interface

Beyond menin, each NUP98 fusion carries a specific partner module whose chromatin-binding function is presumably obligate:

  • NUP98::NSD1 retains the NSD1 SET domain, an H3K36 methyltransferase. That is a genuine catalytic site — unusual in this half of the atlas — and therefore a candidate for conventional inhibitor design. If the SET activity is required, NUP98::NSD1 is the one chromatin fusion in paediatric AML with an enzyme on the fusion itself.
  • NUP98::KDM5A retains the KDM5A PHD finger, a histone-reader module. Reader domains are shallow but have been drugged (BET bromodomains), so this is a plausible if difficult interface.

These are mutually exclusive predictions about where the drug should point, and both are testable with existing tools: catalytically dead SET mutants and reader-pocket point mutants in the fusion, scored for transformation.

Correction (cycle 33): this was not a prediction, and the confirmation is not of the mechanism proposed

Withdrawn: the novelty premise. The entry was written as a forward prediction ('we hypothesise that NUP98 fusions are also menin-MLL1-dependent'). It had been demonstrated in 2022 6. Any experiment designed after reading that paper is not a pre-specified test of this hypothesis, and this is the second time in this project that a hypothesis has been framed as novel against literature that already existed (see the breakpoint-position case in the 2026-09-18 counter-sweep memo). The habit is worth naming.

Withdrawn: the inference from the confirmation to the interface logic. The entry's stated mechanism is that each NUP98 fusion's partner-supplied reader/writer module (NSD1 SET, KDM5A PHD) is the obligate interface. Heikamp et al. show the actionable dependency is menin-MLL1, with both MLL1 and the NUP98 fusion evicted from chromatin by the inhibitor 6. The NUP98 fusion is dependent on a borrowed wild-type complex, exactly as KMT2A fusions are, and not on its own partner module. The obligate interface here belongs to the host, not to the partner. That is a materially different claim from the one this entry made, and it weakens the parent hypothesis (hyp-obligate-partner-interface-generalises) rather than supporting it: the successful prediction is another instance of the same menin interface, not an instance of a new partner interface.

What is still open and still worth doing: whether the NSD1 SET or KDM5A PHD module is separately required once menin is blocked (a genuine separation-of-function question, answerable by catalytic-dead or reader-dead knock-ins under menin inhibition), and whether NUP98-rearranged patients respond clinically -- no menin-inhibitor trial has enrolled by NUP98 genotype.

Facts and figures
statementNUP98-rearranged leukaemia is menin-MLL1 dependent for the same programmatic reason as NPM1-mutant AML, and in addition each NUP98 fusion depends on its partner-supplied chromatin module (NSD1 SET catalytic activity, KDM5A PHD reader function) as an obligate, separately druggable interface.
originliterature-gap (mis-scoped: the menin half was already answered in the literature at the time of writing)
prior probability0.85
clinical prizeNUP98::NSD1 and NUP98::KDM5A are among the worst-prognosis paediatric AML subtypes with no targeted therapy; a positive result repurposes an already-approved drug class immediately rather than requiring new chemistry.
verification gapThis Queen did not verify a primary NUP98-specific menin-dependency publication in this cycle, so the hypothesis may already be partly answered in the literature; resolving that is the first task of the next cycle.
prior probability noteRaised for the menin-dependency half, which is confirmed preclinically. The partner-reader/writer-module half is separately unsupported and should be scored on its own; treat this number as applying only to the drug prediction.
predictions
[
  "CONFIRMED PRECLINICALLY BEFORE THIS ENTRY WAS WRITTEN (Heikamp 2022, PMID 34582559): menin-MLL1 inhibition reduces HOXA-cluster and Meis1 expression, induces differentiation markers and extends survival in NUP98-NSD1 and NUP98-JARID1A/KDM5A models at IC50s comparable to KMT2A-rearranged models. Retained here as a record of what the entry predicted, not as an open prediction.",
  "A catalytically dead NSD1 SET mutant within NUP98::NSD1 will fail to transform, making NUP98::NSD1 the only paediatric chromatin fusion with a conventional enzyme target on the fusion itself.",
  "A KDM5A PHD reader-pocket point mutant within NUP98::KDM5A will fail to transform, nominating the reader pocket as the interface.",
  "NUP98-rearranged patients treated with a menin inhibitor will show differentiation responses and, if they relapse, MEN1 interface mutations indistinguishable from those seen in KMT2A-rearranged patients. STILL OPEN: no menin-inhibitor trial has enrolled by NUP98 genotype.",
  "If the KDM5A arm is real, a ligand directed at the KDM5A PHD reader function will abolish NUP98-KDM5A transformation while leaving FG-repeat condensate behaviour and menin docking intact. No such ligand has been reported; a PubMed search for NSD1 SET-domain inhibition in NUP98-NSD1 leukaemia returns 1 title-checked record and the KDM5A equivalent returns none."
]
experiments
[
  "exp-cross-family-interface-disruption-panel"
]
analyses
[
  "ana-comparative-interface-druggability-scan"
]
falsifiers
[
  "Menin inhibition has no effect on HOXA9/MEIS1 expression, differentiation or survival in validated NUP98::NSD1 and NUP98::KDM5A models, showing the dependency does not track the program after all.",
  "CRISPR knockout of MEN1 is tolerated by NUP98-fusion-transformed cells while killing KMT2A-rearranged controls in the same screen.",
  "A catalytically dead NSD1 SET mutant and a reader-dead KDM5A PHD mutant both transform normally, showing the partner modules are dispensable scaffolds rather than obligate interfaces.",
  "RUN IN CYCLE 61, DOES NOT FIRE AND IS NOT CASHED. Of 57 menin-inhibitor trials, 23 name NUP98 and 22 carry it in eligibility criteria, but ALL 23 return hasResults false - so no response rate exists on either side of the comparison. One complete remission is published (PMID 39633542, Haematologica 2025) and a case report is not a rate. See data.FALSIFIER_ADJUDICATION_cycle_61. Original text: NUP98-rearranged patients enrolled on menin-inhibitor trials show a response rate indistinguishable from unselected refractory AML, i.e. no genotype-specific signal.",
  "ALREADY PARTLY RESOLVED IN THE WRONG DIRECTION FOR THE MECHANISM: menin-MLL1 inhibition works in NUP98-fusion models by evicting wild-type MLL1 and the NUP98 fusion from chromatin (PMID 34582559), not by disrupting the partner-supplied NSD1 SET or KDM5A PHD module. A reader-dead or catalytic-dead partner-module knock-in that changes nothing under menin inhibition would complete the refutation of the partner-interface half.",
  "RUN IN CYCLE 61 AND CONTRADICTED. Enrolment is broad and genotype-directed: 22 of 57 menin-inhibitor trials carry NUP98 in eligibility, NCT07686107 is a Phase 2 of revumenib plus FLA for NUP98-rearranged children alone at minimum age 1 year, and the one published outcome is a complete remission rather than an absence of response. See data.FALSIFIER_ADJUDICATION_cycle_61. Original text: A menin-inhibitor trial that enrols NUP98-rearranged patients and shows no responses, despite the preclinical result, would show that the mouse and PDX dependency does not survive human disease context.",
  "The second half of the statement is a category error as written: the NSD1 SET domain is an autoinhibited enzyme active site, opened by a covalent fragment-derived ligand (BT5, PMID 32868895), not a protein-protein interface. If 'obligate partner interface' is allowed to include catalytic domains, the hypothesis stops distinguishing this class from the kinase class and loses its content."
]
status of the prediction
{
  "menin_dependency_in_nup98_fusions": "CONFIRMED PRECLINICALLY (Heikamp 2022, PMID 34582559) -- and already published before this entry was written",
  "partner_module_as_obligate_interface": "NOT shown; the confirmed dependency runs through wild-type MLL1/menin, with MLL1 and the NUP98 fusion evicted together",
  "clinical": "untested -- pivotal menin-inhibitor trials enrolled by KMT2A/NPM1 genotype, not NUP98"
}
FALSIFIER ADJUDICATION cycle 61
{
  "why": "the cycle-61 sweep (analyses/the-unrun-falsifier-queue/) found this the only confidence-strong entry at the top of the unrecorded-falsifier queue, with seven falsifiers and no recorded outcomes. Falsifiers [3] and [5] are the same clinical question and were answerable from public sources.",
  "[3] response rate indistinguishable from unselected refractory AML": "DOES NOT FIRE, AND IS NOT CASHED. No trial among the 23 that name NUP98 has posted results, so no rate exists on either side of the comparison. One complete remission is published (PMID 39633542, Haematologica 2025) and a case report is not a rate.",
  "[5] a trial enrolling NUP98 patients that shows no responses": "CONTRADICTED. 22 of 57 menin-inhibitor trials carry NUP98 in their eligibility criteria, NCT07686107 is a Phase 2 for NUP98-rearranged children alone at minimum age 1 year, and the one published outcome is a complete remission.",
  "AND AN UNEXPECTED SUPPORT FOR THE CORRECTED MECHANISM": "NSD1 and KDM5A appear in ZERO of the 57 trial records while NUP98 appears in 23. Trials enrol on NUP98 rearrangement as a class and never stratify by partner - independent support, from clinical practice rather than mechanism, for this entry's corrected position that the dependency does not run through the partner-supplied module. The counter-reading is in the analysis entry: partner identification needs RNA sequencing many centres do not run at relapse, so an untestable criterion would not be written regardless of whether the partner matters.",
  "effect_on_the_prior": "none. 0.85 unchanged - neither falsifier fired and neither is efficacy data.",
  "see": "ana-nup98-is-enrolled-in-22-menin-trials-and-the-partner-gene-in-none",
  "adjudicated_by": "king, cycle 61, 2026-09-22"
}
How this could bite us back
  • NSD1 and KDM5A are broadly expressed chromatin enzymes with essential normal functions, so partner-module inhibition risks toxicity well beyond the leukaemic clone.
  • If NUP98 fusions prove menin-dependent, they will also inherit MEN1 interface resistance mutations, and paediatric AML has far less salvage capacity than adult disease.
  • Condensate-driven local concentration of the fusion complex could require near-complete target occupancy for any phenotype, collapsing the therapeutic window even where the dependency is real.
  • If the NUP98 dependency is another instance of the same menin interface rather than a new one, then the portfolio of menin-interface drugs is less diversified than the cross-genotype story suggests: one pocket now carries three diseases, and one resistance mechanism threatens all three at once.
Still open
  1. Has NUP98-fusion menin dependency already been demonstrated in a primary publication, and if so with which fusions and at what potency relative to KMT2A-rearranged controls?
  2. Is NSD1 SET catalytic activity required for NUP98::NSD1 transformation, which would make it the rare chromatin fusion with an enzyme target on the fusion itself?
  3. Do the FG-repeat condensate properties change the pharmacology of an adaptor-interface drug, for example by locally concentrating the target so that higher occupancy is needed?
  4. Would combining a menin inhibitor with a partner-module inhibitor (NSD1 or KDM5A) pre-empt the interface escape mutations that defeat menin inhibitors alone?
  5. Is the partner-supplied module (NSD1 SET, KDM5A PHD) required at all once menin is inhibited? A catalytic-dead SET or reader-dead PHD knock-in tested under menin inhibition separates the partner-interface claim from the host-adaptor claim, and nobody has run it.
  6. Why has no menin-inhibitor trial enrolled by NUP98 genotype four years after the preclinical result, given that NUP98-NSD1 and NUP98-KDM5A are among the worst-prognosis paediatric AML subtypes? The answer is probably regulatory and commercial rather than scientific, and it is checkable in the trial registry.
Every source behind this entry (6)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    NPM1-mutant AML, which has no KMT2A rearrangement, is nonetheless exceptionally dependent on the menin-binding site in MLL1, showing the menin dependency tracks the HOXA/MEIS1 program rather than the genotype.

    PMID 27535106DOI
    What this evidence cannot show

    Cell lines and mouse models; the generalisation from NPM1-mutant to any HOXA-high leukaemia is an inference this hypothesis is designed to test, not a result of this paper.

    Evidence that argues against this (1)
    • The inference drawn here - that menin dependency tracks the HOXA/MEIS1 program rather than the KMT2A rearrangement - is correct and is exactly what undermines the 'partner-supplied obligate interface' half of this hypothesis. The same dependency is now described for entities with no fusion and no partner at all, including AML with UBTF tandem duplications alongside NPM1-mutant AML and NUP98-rearranged disease, all grouped as HOX-dependent. If a lesion with no partner shares the dependency, then menin sensitivity in NUP98-rearranged leukaemia is evidence about the transcriptional state, not evidence that the NUP98 partner supplies an obligate interface. Cited as a map of the current dependency landscape rather than as primary evidence for any single entity. Bruzzese A et al. 2026
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Menin-MLL1 inhibition reversed extended myeloid progenitor self-renewal in NPM1-mutant knock-in mice, establishing the dependency in a non-KMT2A genetic background in vivo.

    PMID 32001657DOI
    What this evidence cannot show

    A single cooperating genotype in mice; durability after withdrawal was limited and no NUP98 model was tested.

    Evidence that argues against this (1)
    • A dependency established in a clean knock-in mouse at the preleukaemic stage is the most favourable setting available, and it does not transfer to the population the drug is used in. Serial samples from multiply relapsed KMT2A::AFF1 patients were markedly less sensitive to menin inhibition than earlier samples from the same individuals, with the resistance acquired in the complete absence of menin-inhibitor exposure and with on-target repression of the canonical target genes fully preserved. If this hypothesis is going to predict that NUP98-rearranged patients respond, it has to predict which ones, and prior therapy is a stronger determinant than genotype in the only serial data available. Mahdavi L et al. 2025
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    One menin inhibitor produced clinical remissions across two genetically unrelated leukaemia genotypes in the same trial, which is the precedent for expecting cross-genotype coverage.

    The menin inhibitor revumenib in KMT2A-rearranged or NPM1-mutant leukaemia · Issa GC, Aldoss I, DiPersio J, et al. · Nature 2023
    PMID 36922593DOI
    What this evidence cannot show

    Single-arm phase 1; NUP98-rearranged disease is not a reported enrolment genotype in this publication, so the extension is untested here.

    Evidence that argues against this (1)
    • The precedent is real but shallow, and its depth is the quantity this hypothesis is implicitly borrowing. In the NPM1-mutant arm of that same trial, the registration analysis gave CR+CRh in 23.4% of 64 efficacy-evaluable patients, a median duration of CR+CRh of 4.7 months, and 5 of 30 responders proceeding to transplant. 'Clinical remissions across two genetically unrelated genotypes' is accurate; what it licenses for a third genotype is a similar expectation - roughly a quarter of patients, for under five months - not a demonstration that the dependency is obligate. Arellano ML et al. 2025
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    NUP98 fusions drive a broad, super-enhancer-like HOXA-cluster program through FG-repeat-dependent phase separation and CTCF-independent chromatin looping, converging on the same transcriptional output as KMT2A-rearranged and NPM1-mutant disease.

    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

    Centres on NUP98-HOXA9 rather than the clinically common NUP98::NSD1 and NUP98::KDM5A, and does not examine menin dependency at all.

    Evidence that argues against this (2)
    • The hypothesis's novelty premise is false: the prediction had already been tested and confirmed four years before this entry was written. VTP50469 kills NUP98-HOXA9 and NUP98-JARID1A/KDM5A leukaemia cells with an IC50 comparable to KMT2A-rearranged and NPM1c cells, upregulates CD11b, downregulates Meis1, and prolongs survival of mice engrafted with NUP98-NSD1 and NUP98-JARID1A patient-derived xenografts. The entry should be reported as CONFIRMED PRECLINICALLY, not as a prediction awaiting test. Heikamp EB et al. 2022
    • And the confirmation bypasses this entry's own interface logic. The dependency runs through MLL1 and menin -- both MLL1 and the NUP98 fusion are evicted from chromatin at a critical set of genes -- not through the partner-supplied reader/writer module (NSD1 SET, KDM5A PHD) that the hypothesis nominates as the obligate interface. So the drug prediction is right and the mechanistic reasoning offered for it is not the reason it is right. Heikamp EB et al. 2022
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Where a chromatin fusion depends on an adaptor interface, resistance emerges at the drug-binding surface rather than by loss of the dependency, which sets the expected failure mode if this hypothesis is confirmed and acted on.

    MEN1 mutations mediate clinical resistance to menin inhibition · Perner F, Stein EM, Wenge DV, et al. · Nature 2023
    PMID 36922589DOI
    What this evidence cannot show

    Observed in KMT2A-rearranged and NPM1-mutant patients; extrapolation to a NUP98-rearranged population is an assumption.

    Evidence that argues against this (1)
    • Setting the expected failure mode as 'resistance at the drug-binding surface' makes the problem look narrower than it is, and this matters most for a genotype where no clinical resistance series exists yet. Menin-inhibitor-resistant AML generated by repeated drug exposure carries an altered epigenome, transcriptome and proteome with NO menin mutation, and the resistant state is defined by new dependencies on BRD4, SMARCA4 and CREBBP; patient-derived MLL1-rearranged and NPM1-mutant cells with ex vivo resistance behaved the same way. For NUP98-rearranged disease the prediction should therefore be two failure modes, and the non-genetic one will be invisible to the MEN1 sequencing that would be done to look for it. Fiskus W et al. 2026
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Menin-MLL1 inhibition with VTP50469 is active in NUP98-fusion leukaemia at an IC50 comparable to KMT2A-rearranged and NPM1c cells, evicts both MLL1 and the NUP98 fusion protein from chromatin, downregulates the HOXA cluster and Meis1, induces differentiation markers, and prolongs survival in NUP98-NSD1 and NUP98-JARID1A patient-derived xenografts.

    The menin-MLL1 interaction is a molecular dependency in NUP98-rearranged AML. · Heikamp EB, Henrich JA, Perner F, et al. · Blood 2022
    PMID 34582559DOI
    What this evidence cannot show

    Abstract only. Mouse cell lines driven by NUP98-HOXA9 and NUP98-JARID1A plus two PDX models; a tool compound rather than an approved agent; survival prolongation, not cure. No clinical menin-inhibitor efficacy data in NUP98-rearranged patients is cited here, and the pivotal menin-inhibitor trials enrolled by KMT2A/NPM1 genotype rather than NUP98.

    Evidence that argues against this (1)
    • This item supports the first half of the hypothesis and is silent on the second, and the second half is where the chemistry claim lives. The one chemical series that exists against a NUP98 partner module reclassifies it: the NSD1 SET domain is an autoinhibited catalytic domain, not a protein-protein interface, and it took a fragment-based covalent strategy to open it - the crystal structure of NSD1 with the covalently bound ligand shows the autoinhibitory loop must move to form a channel-like pocket. The lead, BT5, is irreversible, is explicitly described as first-in-class, and its reported activity is inhibition of H3K36me2, downregulation of target genes, and impaired colony formation from one NUP98-NSD1 patient sample. Calling the NSD1 SET domain an 'obligate, separately druggable interface' merges an enzyme active site with a partner interface, which are the two categories this graph's mechanism classes exist to keep apart. For the KDM5A arm there is no equivalent at all: a PubMed search for NSD1 SET-domain inhibitors in NUP98-NSD1 leukaemia returns exactly 1 record (title-checked: this paper), and what would falsify the KDM5A half is a PHD-reader-directed ligand that loses NUP98-KDM5A transformation while sparing the FG-repeat and menin arms. Nobody has reported one. Huang H et al. 2020
Connected to