Five KMT2A-free NUP98 fusions bind MLL1 and the NSL complex — and one unbiased mass-spectrometry study failed to reproduce it
also called NUP98 NSL complex · NUP98 MLL1 interaction · Xu 2016 NUP98 NSL MLL1 · GLFG-dependent MLL1 binding
Five different NUP98 fusions, none containing any piece of MLL1, were shown to physically grab MLL1 and a partner complex called NSL, and deleting MLL1 delayed the resulting leukaemia in mice. A later study that searched for every protein binding these fusions, without deciding in advance what to look for, did not find MLL1 at all.
The same thing, for a scientist
This is the observation that makes NUP98 a candidate second menin-dependent family, and it is **contested at the level of the primary assay**, so the entry states both sides. **NUP98-HOXA9, NUP98-HOXD13, NUP98-NSD1, NUP98-PHF23 and NUP98-TOP1 — five fusions containing no KMT2A sequence — pull down MOF, PHF20, OGT1, HCF1, WDR5 and MLL1**, with CXXC1 and EZH2 negative; the retained N-terminal NUP98 fragment alone (aa 1–469) binds and **full-length wild-type NUP98 does not**; **deleting both GLFG domains abolishes binding while deleting GLEBS has no effect** [E1]. ChIP-seq shows fusion/MLL1 co-occupancy at **37 promoters (p = 3.7e-14)** and **520 gene bodies (p = 1.45e-204)** including the *Hoxa* and *Hoxb* clusters and *Meis1*, with co-enriched H3K4me3 and H4K16ac [E1]. In vivo, a genuine ***Mll1*-floxed conditional knockout** (not shRNA) **delays** NUP98-HOXA9 leukaemogenesis [E1]. **But:** an unbiased AP-MS study of five Strep-HA-tagged NUP98 fusions at near-physiological levels in human HL-60, recovering 616 proteins and 501 high-confidence interactors, states verbatim that *"our AP-MS experiments did not identify the previously reported interaction of NUP98 fusion proteins with the NSL and MLL1 complexes"* [E2]. Two later studies using different methods do recover MLL1 [E3][E4]. The honest state: **three laboratories support a NUP98-fusion–MLL1 association and one unbiased mass-spectrometry study failed to detect it; the NSL complex specifically rests almost entirely on the original paper.**
What the founding assay actually was, and why that matters
It was not co-immunoprecipitation and not mass spectrometry. It was biotin/streptavidin affinity purification — a C-terminal Flag-Avi tag plus BirA ligase — followed by targeted antibody western blot for a pre-chosen list: MOF, PHF20, OGT1, HCF1, WDR5, MLL1, with CXXC1 and EZH2 as negative controls 1. Two systems were used: 293T stable lines, and human U937 cells for NUP98-HOXA9 and NUP98-HOXD13.
A targeted western blot after a stringent affinity pull-down is a legitimate and sensitive way to ask "is protein X there?". It is not a way to ask "what is there?", and it cannot report abundance or stoichiometry. That is precisely the axis on which the later unbiased study disagrees 2, and the original authors' own proposed reconciliation is that the interaction "might occur at low frequencies".
So the two results are not necessarily contradictory. A low-occupancy, functionally important interaction would be visible to a targeted blot and invisible to an unbiased interactome. But "low frequency" is a claim with consequences: a sub-stoichiometric interaction is harder to argue is the load-bearing mechanism, and it is harder to drug.
The domain mapping is the most transferable part
Binding maps to the GLFG repeats, not GLEBS 1:
- the retained N-terminal NUP98 fragment alone (aa 1–469) binds
- full-length wild-type NUP98 does not
- deleting both GLFG domains abolishes binding
- deleting the GLEBS region (the RAE1-interaction module, P52948 residues 157–213) has no effect
This is a clean separation-of-function on the NUP98 side and it explains why the interaction is a property of the fusion rather than of the nucleoporin: in the full-length protein the FG region is presumably occluded or otherwise unavailable, and truncation at the fusion breakpoint exposes it.
It also makes a prediction the field has not tested: NUP98::KMT2A retains the same GLFG repeats and should therefore bind MLL1 by this mechanism — yet it is the one NUP98 fusion that is menin-inhibitor resistant and HOXA-low (fusion-nup98-kmt2a). Either the GLFG–MLL1 interaction is not sufficient for the HOXA programme, or something about carrying KMT2A in cis changes the outcome. Nobody has run the pull-down on that fusion.
The in vivo claim: 'delays', never 'abolishes'
The dispatch brief for this beat described the result as "MLL1 is required for NUP98-HOXA9 proliferation in vitro and in vivo". The in vivo half needs qualifying in three ways 1:
- The model requires Meis1a co-expression to make NUP98-HOXA9 leukaemogenesis fast enough to assay. So the in vivo requirement is measured in a two-oncogene system.
- At death, Mll1-deleted mice's leukaemic cells were "no longer completely excised... the leukemias that eventually form are caused by cells that have escaped Mll1 deletion". Every treated mouse still died of leukaemia, from Mll1-intact escapers.
- Independent papers restate it correctly as "delays onset of disease". This entry uses delays, and any downstream use should too.
What is genuinely strong is the specificity control: Mll1 deletion hit NUP98 fusions much harder than MLL-AF9, and had no effect on HOXA9+MEIS1-transformed cells 1. And the rescue is informative about the node: Hoxa9+Meis1 co-overexpression restored clonogenicity and leukaemogenicity, and a Hoxb gene alone also rescued.
The HOXB arm, which is the part that is genuinely NUP98-specific
NUP98-HOXA9 and NUP98-HOXD13 bind the Hoxb clusters, which MLL-AF9 does not 1. Mll1 deletion downregulated Hoxa and Hoxb plus Meis1, and overexpression of a single Hoxb gene rescued the clonogenic defect — the functional proof that the HOXB arm carries weight rather than being a bystander.
The original authors argue this is why NUP98 fusions are more MLL1-dependent than MLL-AF9 is. If that is right it inverts the usual framing of this beat: NUP98 would not merely be a second family that happens to share the KMT2A target, it would be a family that depends on wild-type MLL1 more than the KMT2A-fusion family does, because it needs MLL1 at a locus the KMT2A fusion reaches by itself.
That is a testable and unusually specific claim, and it has not been revisited in the decade since.
What the failed replication found instead
The AP-MS study's positive results are worth as much as its negative one 2. Across five Strep-HA-tagged fusions (NUP98-KDM5A, -NSD1, -HOXA9, -DDX10, -PSIP1) expressed at near-physiological levels in human HL-60, it recovered 616 proteins, 501 high-confidence interactors, 27 shared by all five — and the shared core is RNA-metabolism proteins (DDX24, RAE1), not chromatin complexes.
That is a substantively different model of what a NUP98 fusion is: an RNA-associated assembly rather than a chromatin-complex recruiter. It has not been reconciled with the chromatin model and both are in the literature simultaneously.
Two later studies do recover MLL1 by other routes: BioID of NUP98-PHF23 recovers KMT2A/MLL1, KMT2B, WDR5, RBBP5, EP300, BRD4, EP400 and SWI/SNF, with a LacO-array tethering reconstitution showing FG-dependent MLL1 recruitment 3; and NUP98::HOXA9 nuclear bodies are reported to "promote the condensation of MLL1" 4. Proximity labelling and targeted blotting both detect transient and sub-stoichiometric associations that co-purification loses, so the pattern of which methods see it and which do not is itself consistent with a real but low-occupancy interaction.
| n fusions | 5 |
|---|---|
| none contain kmt2a | true |
| binding assay | biotin/streptavidin affinity purification (Flag-Avi tag + BirA) followed by TARGETED antibody western blot |
| domain requirement | GLFG repeats required; GLEBS dispensable; N-NUP98 (aa 1-469) alone binds; full-length wild-type NUP98 does not |
| chipseq overlap promoters | 37 |
| chipseq overlap promoters p | 3.7e-14 |
| chipseq overlap gene bodies | 520 |
| chipseq overlap gene bodies p | 1.45e-204 |
| in vivo tool | Mll1-floxed conditional knockout x UBC-CreER (genetic, not shRNA) |
| in vivo verb | DELAYS leukaemogenesis; every treated mouse still died, from Mll1-intact escapers |
| in vivo caveat | model required Meis1a co-expression; excision incomplete at death |
| rescue | Hoxa9+Meis1 co-overexpression restored clonogenicity and leukaemogenicity; a single Hoxb gene alone also rescued |
| hoxb specificity | NUP98 fusions bind Hoxb clusters, which MLL-AF9 does not |
| replication status | 3 labs support a NUP98-fusion/MLL1 association (targeted blot, BioID, imaging); 1 unbiased AP-MS study failed to detect it; the NSL complex specifically rests almost entirely on the original paper |
| untested prediction | NUP98::KMT2A retains the same GLFG repeats and should bind MLL1 by this mechanism, yet is HOXA-low and menin-inhibitor resistant; the pull-down has never been run on it |
fusions tested for mll1 nsl binding
[ "NUP98-HOXA9", "NUP98-HOXD13", "NUP98-NSD1", "NUP98-PHF23", "NUP98-TOP1" ]
proteins blotted positive
[ "MOF", "PHF20", "OGT1", "HCF1", "WDR5", "MLL1" ]
negative controls
[ "CXXC1", "EZH2" ]
- If the MLL1 association is sub-stoichiometric, a menin inhibitor is being asked to disrupt a minority complex, and the achievable depth of response in NUP98-rearranged disease may be intrinsically lower than in KMT2A-rearranged disease - which is what the clinical durability data show.
- The NSL half of this claim would, if taken at face value, nominate MOF as a target; the DepMap data in this beat say MOF has no fusion specificity, so acting on the NSL half would burn a programme on a lineage requirement.
- A published failure to replicate sitting unreconciled for five years is a risk to any programme built on the original result. Before a NUP98-directed menin trial is designed on mechanistic grounds, somebody should repeat the pull-down with quantitative mass spectrometry.
- Is the NUP98-fusion/MLL1 interaction sub-stoichiometric? A quantitative measurement - absolute stoichiometry rather than presence - would reconcile the targeted-blot and AP-MS results and decide whether it can carry the mechanism.
- Does NUP98::KMT2A bind MLL1 through its GLFG repeats? It retains them and is menin-resistant, so the answer separates the GLFG-MLL1 interaction from the HOXA programme. The experiment is a single pull-down.
- Is the NSL complex genuinely engaged, or is MOF recovered because MOF is everywhere? The NSL claim rests on one study's targeted blot and the DepMap data show KAT8 has no fusion-specific requirement at all.
- Is the HOXB dependence real and NUP98-specific? A single Hoxb gene rescued the Mll1-deletion phenotype a decade ago and nobody has followed it up.
- Which is upstream: CRM1 pre-binding at Hox clusters, or MLL1 association? Both are published as the targeting mechanism for the same fusion and they have never been tested against each other.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
Five NUP98 fusions containing no KMT2A sequence (NUP98-HOXA9, -HOXD13, -NSD1, -PHF23, -TOP1) bind MOF, PHF20, OGT1, HCF1, WDR5 and MLL1 by biotin affinity purification and targeted western blot; binding requires the GLFG repeats and not GLEBS; the retained N-terminal fragment alone binds while full-length NUP98 does not; fusion and MLL1 co-occupy 37 promoters and 520 gene bodies including Hoxa, Hoxb and Meis1; and conditional Mll1 deletion delays NUP98-HOXA9 leukaemogenesis while having no effect on HOXA9+MEIS1-transformed cells.
NUP98 Fusion Proteins Interact with the NSL and MLL1 Complexes to Drive Leukemogenesis · Xu H, et al. · Cancer Cell 30(6):863-878 2016What this evidence cannot show
The binding assay is a TARGETED western blot after affinity purification, not an unbiased interactome, so it can confirm presence but not abundance, stoichiometry or completeness. Binding was assayed in 293T stables and in U937 for only two of the five fusions. The in vivo arm required Meis1a co-expression and ended with incomplete Cre excision, so the correct verb is 'delays', not 'abolishes' - every treated mouse still died of leukaemia from Mll1-intact escapers. Confidence is 'moderate' rather than 'strong' because of the published failure to replicate below.
Evidence that argues against this (2)
- An unbiased AP-MS study of five NUP98 fusions at near-physiological levels in human HL-60 explicitly failed to detect this interaction, stating that 'our AP-MS experiments did not identify the previously reported interaction of NUP98 fusion proteins with the NSL and MLL1 complexes', and reports RNA-metabolism proteins as the shared core interactome instead. Terlecki-Zaniewicz S et al. 2021
- An entirely MLL1-independent targeting mechanism has been reported for the same fusion: NUP98-HoxA9 accumulates preferentially where CRM1/XPO1 is pre-bound at Hox clusters, and leptomycin B disassembles the dots and abolishes chromatin binding and Hox activation. If CRM1 pre-binding is the primary targeting determinant, MLL1 association could be a consequence of co-localisation rather than a cause of it. Oka M et al. 2016
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
An unbiased AP-MS interactome of five Strep-HA-tagged NUP98 fusions at near-physiological levels in HL-60 recovered 616 proteins and 501 high-confidence interactors, of which 27 were shared by all five and the shared core was RNA-metabolism proteins, and it did not detect the previously reported NSL and MLL1 complex interactions.
Biomolecular condensation of NUP98 fusion proteins drives leukemogenic gene expression · Terlecki-Zaniewicz S, et al. · Nature Structural & Molecular Biology 28(2):190-201 2021What this evidence cannot show
A failure to detect is not a demonstration of absence. AP-MS loses transient and sub-stoichiometric interactions by design, and the cell system (HL-60), the tag (Strep-HA) and the expression level all differ from the original study. The original authors' own suggested reconciliation - that the interaction occurs at low frequency - is compatible with both datasets.
Evidence that argues against this (1)
- Proximity labelling of NUP98-PHF23 by BioID does recover KMT2A/MLL1 along with KMT2B, WDR5, RBBP5, EP300, BRD4 and SWI/SNF, and a LacO-array tethering reconstitution shows FG-dependent MLL1 recruitment. Proximity labelling captures transient associations that co-purification loses, so the two methods disagreeing is the expected pattern for a real low-occupancy interaction. Ahn JH et al. 2025
- E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
BioID proximity labelling of NUP98-PHF23 recovers KMT2A/MLL1, KMT2B, WDR5, RBBP5, EP300, BRD4, EP400 and SWI/SNF components, and a LacO-array tethering reconstitution shows FG-repeat-dependent MLL1 recruitment.
The phenylalanine-and-glycine repeats of NUP98 oncofusions form condensates that selectively partition transcriptional coactivators · Ahn JH, et al. · Molecular Cell 85(4):708-725.e9 2025What this evidence cannot show
Proximity labelling reports neighbourhood, not binding, and a LacO array is a synthetic high-local-concentration system in which recruitment can occur without a physiological interaction. Only one fusion (NUP98-PHF23) was profiled. Details were read via a delegated research pass, not by the author of this entry.
Evidence that argues against this (1)
- If MLL1 partitions into an FG condensate because condensates partition transcriptional coactivators generally, then MLL1 recruitment is a property of the condensate rather than a specific interaction, and the same result would be obtained for any coactivator. The paper's own framing - selective partitioning of coactivators - is closer to this reading than to a specific NUP98-MLL1 interface. Xu H et al. 2016
- E4weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.paper
Nuclear bodies formed by nucleoporin fusions are reported to promote condensation of MLL1 and CRM1 and to rearrange 3D genome structure.
Phase-separated nuclear bodies of nucleoporin fusions promote condensation of MLL1/CRM1 and rearrangement of 3D genome structure · Oka M, et al. · Cell Reports 42(8):112884 2023What this evidence cannot show
Metadata verified from PubMed; the full text was NOT accessible in this session (not in PMC; publisher and Europe PMC both returned HTTP 403). The FRAP geometry, whether 1,6-hexanediol was used, whether in vitro droplets were made and the tagging strategy are all UNVERIFIED and are deliberately not asserted anywhere in this graph. Confidence is 'weak' for that reason alone, not because of any judgement on the work.
Evidence that argues against this (1)
- This graph's standing rule is that puncta, full-bleach FRAP and 1,6-hexanediol cannot distinguish liquid-liquid phase separation from clustered binding to a chromatin template. Because the assays behind this title could not be read, the entry cannot say which model the data support, and the title's use of 'phase-separated' should not be taken at face value. Muzzopappa F et al. 2022
GLFG-dependent binding of wild-type MLL1 by five KMT2A-free NUP98 fusions
MOF recovered in the NSL pull-down (targeted blot)
WDR5 recovered in the NUP98-fusion pull-down
the physical basis for the borrowed-cofactor dependency
coactivator partitioning versus site-specific recruitment are the two readings of the same data
one of the five KMT2A-free fusions in the MLL1/NSL pull-down
among the five fusions whose unbiased AP-MS failed to recover MLL1
one of the five KMT2A-free fusions in the MLL1/NSL pull-down
- ← related toTOP1 domain architecture (UniProt P11387)
NUP98-TOP1 was one of the five fusions in the MLL1/NSL pull-down
The graph already held post-2004 evidence that WRAD subunits are recovered around a fusion oncoprotein - WDR5 and RBBP5 in NUP98-fusion pulldowns and BioID - without connecting it to the KMT2A WRAD question.
The only precedent: BioID of NUP98-PHF23 recovers MLL1, KMT2B, WDR5 and RBBP5 together, with WRAD arriving on a recruited methyltransferase.