Oncofusion

start / graph / fusion-nup98-hoxa9

Fusionstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Qn-nup98-family

NUP98::HOXA9, t(7;11)(p15;p15) — the founding NUP98 fusion, and an incomplete oncogene

also called NUP98-HOXA9 · NHA9 · t(7;11)(p15;p15) · NUP98/HOXA9

The first NUP98 fusion ever found, reported by two labs in the same journal issue in 1996. On its own it does not make acute leukaemia: mice first get a slow overgrowth of blood cells and only later, after acquiring more damage, an acute leukaemia.

The same thing, for a scientist

t(7;11)(p15;p15) joins the **GLFG repeat region of NUP98** in frame to the **HOXA9 homeobox**, described simultaneously by two groups in *Nature Genetics* in 1996 [E1][E2]. HOXA9 is **P31269, 272 aa**, with the homeodomain at residues **206–265**; the fusion retains both the **DNA-binding and the PBX-heterodimerisation functions**, and transformation requires both [E3]. The fusion encodes two isoforms carrying either **19 or 37 NUP98 FG repeats**, and those repeats act as a **potent transactivation domain that recruits CBP/p300** and can be functionally replaced by VP16 [E3] — the earliest mechanistic account of why the FG region matters, predating the condensate literature by two decades. The most important qualification is that **NUP98-HOXA9 is not sufficient**: in mice it induces a **myeloproliferative disease** first, with animals only *"eventually succumbing to AML"*, and **co-expression of Meis1 is required** to accelerate the MPD-to-AML transition [E4]. That incompleteness propagates into the menin literature — the in vivo *Mll1*-deletion experiment that underpins the NUP98–MLL1 dependency claim had to be run in a **NUP98-HOXA9 + Meis1a** system for the same reason [E5]. Within the family it is also the **least menin-sensitive** of the well-studied partners: VTP50469 IC50 **517 nM against 84 nM for NUP98-KDM5A** [E6].

Go deeper
Retained domains, and the two isoforms

From the founding reports 12: "Chimaeric messages spliced over the breakpoint fuse the GLFG repeat domains of NUP98 in-frame to the HOXA9 homeobox", described in the companion paper as the "amino terminal half of NUP98 fused in frame to HOXA9".

On the NUP98 side (P52948): FG repeats 1 (1–156), GLEBS (157–213) and FG repeats 2 (214–480) are retained; the autoproteolytic peptidase S59 domain (738–880) and the entire Nup96 chain are lost 7.

On the HOXA9 side (P31269, 272 aa): the homeodomain at 206–265 is retained, together with the PBX-interaction function 38.

The two-isoform observation — 19 versus 37 FG repeats 3 — deserves more attention than it gets. The published valency series for NUP98 condensation finds a threshold between 27 and 36 repeats. If both isoforms are made from the same allele and only one clears that threshold, then a single patient's fusion allele produces one condensation-competent and one condensation-incompetent protein, which is an internal control nobody has used.

The FG repeats as a CBP/p300-recruiting transactivation domain

The 1999 result is the oldest mechanistic account and it is still not superseded 3: the FG repeats bind CREB-binding protein, act as a potent transactivation domain, and can be functionally replaced by VP16 — a heterologous activation domain with no phase-separation pretensions.

That last detail is the one to keep. If VP16 substitutes for the FG repeats and preserves oncogenicity, then whatever the FG repeats contribute is a function VP16 also has, and VP16 is a small acidic activation domain, not a 37-repeat low-complexity region. This is an early and independent argument that the FG region's essential contribution is coactivator recruitment, not bulk condensation — see mechanism-nup98-fg-condensate-assay-audit.

Transformation also required both the DNA-binding and the PBX-heterodimerisation functions of HOXA9 3, so the fusion is not simply an activation domain dragged to chromatin by anything; it needs the partner's specific DNA-binding behaviour.

Incomplete oncogene, and what that costs the field

NUP98-HOXA9 expression in haematopoietic stem cells induces chronic and then acute myeloid leukaemia in mice; the chronic phase comes first and Meis1 co-expression is required to accelerate the transition, with the authors concluding "overlapping yet distinct molecular mechanisms for MPD versus AML" 4.

Two consequences:

  1. Almost every in vivo NUP98-HOXA9 experiment is really a NUP98-HOXA9 + Meis1 experiment. That includes the Mll1-deletion arm behind the MLL1-dependency claim 5. Since MEIS1 re-expression is precisely what abrogates menin-inhibitor sensitivity in NUP98-fusion cells 6, the standard model for this fusion is built on the very gene that confers drug resistance in the same system. Nobody has commented on this and it deserves to be checked rather than assumed harmless.
  2. Human progression is also multi-hit. A t(7;11) patient acquired a second fusion, ETV6-LPXN, at relapse 9.
Clinical picture

AML, predominantly FAB M2 and M4 in the founding series 2. Unlike NUP98::NSD1 and NUP98::KDM5A — which are cryptic and paediatric-skewed — NUP98::HOXA9 is visible on karyotype and is one of the two commonest NUP98 partners in adult AML 10.

Outcome differs from NSD1 in a specific and clinically relevant way: NUP98::HOXA9 and other non-NSD1 partners showed higher remission rates but frequent relapse, whereas NSD1 is characterised by induction failure 10. That is the same split seen between NSD1 and KDM5A in the paediatric cohort (epi-nup98-rearranged-paediatric-aml-outcome), and it recurs often enough across datasets to be worth treating as a real axis: NSD1 fails at induction, the rest relapse.

No verified paediatric-specific NUP98::HOXA9 frequency or EFS/OS figure was obtained in this session, and none is asserted.

Facts and figures
gene5NUP98
gene3HOXA9
cytobandt(7;11)(p15;p15)
partner classhomeodomain
uniprot5P52948
uniprot3P31269
length AA 51817
length AA 3272
isoforms19 or 37 NUP98 FG repeats
fg repeat function earliest accountpotent transactivation domain recruiting CBP/p300; functionally replaceable by VP16
sufficiencyNOT sufficient for AML; induces MPD first, Meis1 co-expression required to accelerate transformation
menin inhibitor ic50 vtp50469 NM517
menin inhibitor ic50 comparator nup98 kdm5a NM84
adult vs paediatricone of the two commonest NUP98 partners in adult AML; visible on karyotype, unlike cryptic NSD1 and KDM5A
prognosishigher remission rates than NUP98::NSD1 but frequent relapse (adult cohort); no verified paediatric-specific EFS/OS obtained
diseases
[
  "AML (FAB M2, M4)",
  "myeloproliferative disease in mouse models"
]
breakpoints
[
  {
    "id": "bp-nup98-hoxa9-t7-11",
    "domains_retained_5": [
      "FG repeats 1 (1-156)",
      "GLEBS (157-213)",
      "FG repeats 2 (214-480)"
    ],
    "domains_lost_5": [
      "peptidase S59 (738-880)",
      "Nup96 chain (881-1817)"
    ],
    "domains_retained_3": [
      "Homeodomain"
    ],
    "domains_lost_3": [
      "N-terminal HOXA9 sequence upstream of the breakpoint"
    ],
    "phenotype_notes": "Two isoforms with 19 or 37 NUP98 FG repeats; both DNA binding and PBX interaction required for transformation.",
    "functions_retained_3": [
      "PBX-heterodimerisation function (not a drawable domain; required for transformation)"
    ],
    "gene3_breakpoint_residue": "UniProt P31269 annotates the NUP98-HOXA9 translocation breakpoint at HOXA9 residues 163/164, so the fusion retains HOXA9 164-272 including the homeodomain (206-265)."
  }
]
How this could bite us back
  • The fusion is an incomplete oncogene, so a drug that reverses the fusion's own programme leaves whatever second hit drove acute transformation untouched - which is what a differentiating agent with a short remission looks like clinically.
  • The workhorse in vivo model carries Meis1, the very gene whose overexpression abrogates menin-inhibitor sensitivity; preclinical efficacy estimates from that model should be treated as suspect in both directions until a Meis1-free arm exists.
  • NUP98::HOXA9 is the least menin-sensitive well-characterised partner and is common in adults, so an adult-weighted trial cohort is enriched for the partner least likely to respond.
Still open
  1. Do the 19-repeat and 37-repeat isoforms differ in condensation and in oncogenicity? One allele makes both, and the published valency threshold sits between them, which makes this a free internal control nobody has used.
  2. Does a Meis1-free NUP98-HOXA9 in vivo model respond to menin inhibition? Because MEIS1 abrogates sensitivity, the standard model may systematically under-report the drug effect - or the standard model's sensitivity may be an artefact of the Meis1 it carries.
  3. Why is NUP98-HOXA9 six-fold less menin-sensitive than NUP98-KDM5A? An isogenic comparison would show whether this is the fusion or the cell line.
  4. If VP16 can replace the FG repeats, what exactly is the condensate contributing that a small acidic activation domain does not?
Every source behind this entry (10)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    t(7;11)(p15;p15) fuses NUP98 to HOXA9 in human myeloid leukaemia, joining the amino-terminal half of NUP98 in frame to HOXA9.

    PMID 8563753DOI
    What this evidence cannot show

    A 1996 cytogenetic-molecular report; domain boundaries in the paper are described at the level of 'amino terminal half' rather than residues, and the residue numbers used in this entry come from UniProt, not from this paper.

    Evidence that argues against this (1)
    • Identifying the fusion says nothing about whether it is sufficient to cause acute leukaemia; the mouse work five years later showed it is not, producing a myeloproliferative disease that required Meis1 to progress. Kroon E et al. 2001
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    The companion report describes chimaeric messages spliced over the breakpoint fusing the GLFG repeat domains of NUP98 in frame to the HOXA9 homeobox, in AML predominantly of FAB types M2 and M4.

    PMID 8563754DOI
    What this evidence cannot show

    Small original case series; FAB subtype distribution from a 1996 cohort predates molecular classification and should not be quoted as a modern frequency.

    Evidence that argues against this (1)
    • Later large cohorts find NUP98::NSD1 and NUP98::KDM5A far commoner than HOXA9 in paediatric AML, so the founding partner is not the representative one and conclusions drawn from it may not generalise to the family. Bertrums EJM et al. 2023
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The NUP98-HOXA9 fusion encodes isoforms with either 19 or 37 NUP98 FG repeats fused to the DNA-binding and PBX-heterodimerisation domains of HOXA9; the FG repeats bind CREB-binding protein, act as a potent transactivation domain and can be functionally replaced by VP16; and transformation requires both the DNA-binding and the PBX-interaction functions.

    PMID 9858599DOI
    What this evidence cannot show

    1999 biochemistry in reporter and transformation assays, largely in overexpression. 'Functionally replaced by VP16' is a statement about an assay readout, not proof that the endogenous mechanisms are the same.

    Evidence that argues against this (1)
    • The modern account attributes the FG repeats' function to biomolecular condensation rather than to coactivator recruitment per se, and reports valency thresholds that a small acidic activation domain like VP16 could not satisfy - so either the two accounts describe different aspects of the same thing, or the VP16 substitution result is telling us the condensation is not the essential part. Ahn JH et al. 2021
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    NUP98-HOXA9 expression in haematopoietic stem cells induces a myeloproliferative disease from which mice only eventually succumb to AML, with Meis1 co-expression required to accelerate the transition; the authors conclude that overlapping yet distinct mechanisms drive MPD versus AML.

    PMID 11157742DOI
    What this evidence cannot show

    Retroviral overexpression in mouse marrow, not a knock-in, so expression level and timing are not physiological. Long latency means secondary events accumulate and are largely uncharacterised in this model.

    Evidence that argues against this (1)
    • Because Meis1 co-expression became the standard way to make this model tractable, and because MEIS1 re-expression is exactly what abrogates menin-inhibitor sensitivity in NUP98-fusion cells, the workhorse NUP98-HOXA9 in vivo model may systematically misreport drug sensitivity. Nobody has tested a Meis1-free in vivo arm against a menin inhibitor. Heikamp EB et al. 2022
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The in vivo Mll1 conditional-deletion experiment supporting the NUP98-MLL1 dependency was performed in a NUP98-HOXA9 plus Meis1a system, and delays rather than abolishes leukaemogenesis, with the leukaemias that form arising from cells that escaped Mll1 deletion.

    PMID 27889185DOI
    What this evidence cannot show

    Read via a delegated full-text pass in this session. The Meis1a requirement is a property of the model, not necessarily of the human disease, and the escaper problem is common to conditional-deletion leukaemia experiments generally.

    Evidence that argues against this (1)
    • The specificity control in the same work is strong in the opposite direction: Mll1 deletion had no effect on HOXA9+MEIS1-transformed cells, so the requirement is not simply a consequence of the HOXA9/MEIS1 programme being present. Xu H et al. 2016
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    NUP98-HOXA9 is the least menin-sensitive of the well-studied NUP98 partners, with a VTP50469 IC50 of 517 nM against 84 nM for NUP98-KDM5A, and MEIS1 re-expression abrogates sensitivity.

    The menin-MLL1 interaction is a molecular dependency in NUP98-rearranged AML · Heikamp EB, et al. · Blood 139(6):894-906 2022
    PMID 34582559DOI
    What this evidence cannot show

    A six-fold IC50 difference between two mouse cell lines is within the range that expression level, growth rate or clone-to-clone variation can produce; the comparison is not isogenic. Read via a delegated full-text pass in this session.

    Evidence that argues against this (1)
    • If NUP98-HOXA9 carries HOXA9 in the fusion itself, it may be partly independent of the endogenous HOXA programme that menin-MLL1 maintains, which would explain a higher IC50 without any difference in the menin interaction. That is a structural explanation for the gradient that has nothing to do with drug sensitivity per se. Deshpande AJ et al. 2014
  7. E7established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    NUP98 is P52948, 1817 aa, with FG repeats 1 at 1-156, GLEBS at 157-213, FG repeats 2 at 214-480 and the peptidase S59 autoproteolytic domain at 738-880.

    What this evidence cannot show

    Primary gene name and length read back from the fetched record. UniProt does not annotate an RRM in P52948, so no RNA-binding-domain boundaries are given anywhere in this entry despite the leukaemia literature describing a C-terminal RNA-binding region.

    Evidence that argues against this (1)
    • The exact NUP98 residue endpoint of this specific patient fusion was not verified; the frequently quoted 'aa 1-469' is the N-NUP98 construct used in an interaction study, not a patient breakpoint, and conflating the two would put a construct boundary into a clinical entry. Xu H et al. 2016
  8. E8established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    HOXA9 is P31269, 272 aa, with the homeodomain at residues 206-265.

    UniProtKB entry P31269, Homeobox protein Hox-A9 (HOXA9) · UniProt Consortium · UniProtKB 2026
    What this evidence cannot show

    Primary gene name asserted from the fetched record rather than from a gene_exact search, which matches synonyms. The only UniProt synonym is HOX1G.

    Evidence that argues against this (1)
    • Which HOXA9 residues the fusion actually retains depends on the breakpoint, and the founding papers describe the junction at the level of 'the homeobox' rather than a residue, so the retained segment may extend beyond the annotated homeodomain. Borrow J et al. 1996
  9. E9moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    A patient with t(7;11) NUP98-HOXA9 AML acquired a second fusion, ETV6-LPXN, at relapse, indicating that progression is driven by acquired secondary lesions.

    PMID 26542893DOI
    What this evidence cannot show

    A single case. A second fusion at relapse could be a passenger of genomic instability rather than a driver of progression, and one patient cannot distinguish the two.

    Evidence that argues against this (1)
    • Single-case acquisition of a second lesion is unremarkable in relapsed AML generally and is not specific to NUP98-HOXA9; treating it as evidence for a multi-hit requirement leans on the mouse data rather than on this observation. Kroon E et al. 2001
  10. E10moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    In adult AML, NUP98::NSD1 is associated with poor induction response whereas NUP98::HOXA9 and NUP98 fusions with other partners show higher remission rates but frequent relapse.

    PMID 42120929DOI
    What this evidence cannot show

    Adult cohort of 95 NUP98-rearranged patients; the partner-level subgroups are small and were not the primary endpoint. This paper's details were obtained in a delegated research pass in this session and the primary figures were not read.

    Evidence that argues against this (1)
    • The largest paediatric cohort concludes that NUP98 fusions should be reclassified as high-risk REGARDLESS of partner, which is the opposite emphasis. Whether partner-specific outcome differences are actionable or are small-subgroup noise is genuinely unsettled. Bertrums EJM et al. 2023
Connected to