Oncofusion

start / graph / hyp-menin-primacy-is-geometry-not-biology

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q5d-interactome3

Menin is the best handle on KMT2A::AFF1 because of where the break falls, not because menin is biologically special

also called retained-fragment hypothesis · geometry hypothesis

Menin may not be a uniquely good drug target so much as the only well-studied protein that still has somewhere to hold on to after the KMT2A gene is broken. If that is right, then finding more targets means looking at the small piece that survives the break, not at the famous partners that are cut away.

The same thing, for a scientist

This cycle's systematic enumeration found that the well-established wild-type KMT2A interactions cluster on the C-terminal fragment that every clinical breakpoint discards — HCF-1 at aa 1799-1802, PHD3/CyP33 at approximately 1566-1627, the CBP-binding TAD at 2829-2883, MOF at the C-terminus, and the WRAD subunits near the SET domain — while the interactions that survive cluster in the retained N-terminal 1,406 residues: menin at the MBM motifs, LEDGF through menin, and PAF1C at sequences flanking CxxC. The hypothesis is that **menin's therapeutic primacy is a consequence of breakpoint geometry rather than of menin having a special mechanistic role**. If true, two predictions follow. First, the number of possible handles on the KMT2A half of the fusion is bounded by how much surface aa 1-1406 actually presents, which is small. Second, the search for additional handles should be redirected: exhaustively onto the retained fragment and the AFF1 moiety, and away from the wild-type KMT2A complex literature, most of which describes a protein the leukemia does not make. Prior probability is set at 0.55 — the geometric correlation is strong and the causal claim is not established.

Go deeper
The observation the hypothesis generalises

Yokoyama 2004 reported, in the paper that founded the menin field, that of the components it purified from the 1-MDa MLL complex, only menin coprecipitated with fusion MLL — HCF-1, HCF-2, ASH2L, WDR5 and RBBP5 did not — and that menin binding maps within the first 1,406 residues, 'consistently retained in all MLL fusion proteins associated with human leukemias' 1.

Twenty-two years of subsequent work has treated that as a fact about menin. This hypothesis reads it as a fact about aa 1-1406. Menin was not selected by the biology for a special role; it was selected by the breakpoint for being in the surviving half.

The pattern held across every additional interaction this cycle enumerated. Contacts on the discarded fragment: CyP33/PHD3 2, CBP/TAD, MOF/C-terminus. Contacts on the retained fragment: PAF1C at CxxC-flanking sequences 3, and menin/LEDGF. Nothing enumerated this cycle broke the pattern.

What would follow if it is true

A bound on the target space. The retained KMT2A fragment is roughly a third of the protein and contains the AT-hooks, the CxxC domain, the menin-binding motifs and the PAF1C-binding region. Almost everything else about KMT2A — its PHD fingers, bromodomain, TAD and SET domain — is gone. The number of protein-protein interfaces available on the fusion's KMT2A half is therefore small and countable, and menin plus PAF1C may be most of it.

A redirection. If the wild-type KMT2A interactome is mostly irrelevant to the fusion, then mining that literature for new targets has low expected yield, and the productive directions are (i) exhaustive mapping of what binds aa 1-1406 in a leukemic cell, and (ii) the AFF1 moiety, which contributes the entire activation function since the KMT2A TAD is deleted.

A reframing of the resistance problem. If menin is the only available handle on the retained fragment rather than the best of several, then MEN1 gatekeeper mutations are not a setback to be routed around with a different KMT2A-half target — they may exhaust that half entirely, and the alternative has to come from the partner side or from downstream nodes such as MEIS1 protein stability.

The strongest arguments against

Ascertainment. The wild-type KMT2A interactome was mapped mostly between 2001 and 2012, and the C-terminal domains (PHD, bromo, SET) are the structurally tractable, historically interesting parts. That the literature is richer there may reflect where people looked rather than where interactions are.

Absence of a modern map. No proximity-labelling study of any KMT2A fusion exists 4. The retained fragment could be covered in transient partners that affinity purification of stable complexes never sees, in which case the target space is not small and this hypothesis is wrong for a mundane reason.

Menin may be special after all. Menin is not merely present on the retained fragment; it is required, its interface is a well-formed pocket, and MEN1 itself is a tumour suppressor whose loss in other tissues causes tumours. A geometry-only account does not explain why menin's removal is so consequential when PAF1C's contact, also on the retained fragment, is not a comparable vulnerability. The honest position is that geometry explains availability and something else explains criticality.

Facts and figures
statementMenin's status as the druggable handle on KMT2A::AFF1 follows primarily from the position of the breakpoint — menin binds within the retained N-terminal 1,406 residues while most other established KMT2A partners bind past the break — rather than from menin having a mechanistically privileged role. The corollary is that the fusion's KMT2A-side interaction surface is small and nearly enumerated, and that further handles must come from the retained fragment or from the AFF1 moiety.
originliterature-gap
prior probability0.55
predictions
[
  "An unbiased proximity-labelling map of KMT2A::AFF1 will recover menin, LEDGF, PAF1C subunits and SEC/AEP components, and will NOT recover WRAD subunits, HCF-1, CyP33, CBP or MOF.",
  "No new high-confidence KMT2A-side interactor will map C-terminal to residue ~1450.",
  "Interactors of the AFF1 moiety will outnumber new KMT2A-side interactors in any modern unbiased map, because the AFF1 moiety supplies the activation function the deleted KMT2A TAD used to supply.",
  "Handles discovered on the retained fragment other than menin (PAF1C is the existing example) will tend to be non-selective dependencies, because the retained fragment's partners are shared with wild-type KMT2A."
]
experiments
[
  "exp-turboid-map-of-kmt2a-aff1"
]
analyses
[
  "ana-kmt2a-interactome-systematic-enumeration"
]
falsifiers
[
  "A proximity-labelling or modern AP-MS map of KMT2A::AFF1 that recovers WRAD subunits, HCF-1, CyP33, CBP or MOF as fusion-associated would show the geometric exclusion is wrong and that the 2004 negatives were sensitivity limits.",
  "Discovery of a druggable, KMT2A-selective interaction mapping C-terminal to residue 1450 on the fusion would refute the geometric bound directly.",
  "A demonstration that menin's criticality is separable from its position — for example an engineered fusion in which the menin-binding motifs are moved past the break and menin dependence persists through a second surface — would show the primacy is biological rather than geometric.",
  "If a systematic map of the retained fragment finds many more interactors than menin and PAF1C, the 'small and nearly enumerated' claim fails even if the geometric correlation holds."
]
How this could bite us back
  • If menin is the only handle on the retained fragment rather than the best of several, then MEN1 gatekeeper mutations exhaust that half of the fusion and there is no second KMT2A-side shot.
  • A small interaction surface is good for specificity and bad for redundancy: one interface means one resistance mechanism suffices.
  • The hypothesis could be self-fulfilling for the field — if people stop looking at the discarded fragment, they will not find the exception that disproves it.
Still open
  1. How much solvent-accessible protein-protein interaction surface does the retained KMT2A fragment actually present? A structure-informed estimate would turn 'small' into a number.
  2. Is PAF1C's failure to be a selective vulnerability, despite binding the retained fragment, evidence against the hypothesis or simply evidence that availability and criticality are different properties?
  3. Does the AFF1 moiety present more druggable surface than the KMT2A moiety? If the activation function is entirely partner-supplied, the partner is where the leverage should be.
  4. Do other KMT2A fusions with different breakpoints (exon 8 vs exon 11) retain measurably different interactomes? The hypothesis predicts small, breakpoint-dependent differences that nobody has looked for.
Every source behind this entry (4)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    Menin, but not HCF-1, HCF-2, ASH2L, WDR5 or RBBP5, coprecipitates with fusion MLL proteins; menin interaction maps within the first 1,406 residues of MLL-N, consistently retained in all leukemia-associated MLL fusions; HCF binding requires MLL aa 1799-1802.

    Leukemia proto-oncoprotein MLL forms a SET1-like histone methyltransferase complex with menin to regulate Hox gene expression. · Yokoyama A, Wang Z, Wysocka J, Sanyal M, Aufiero DJ, Kitabayashi I, Herr W, Cleary ML · Molecular and Cellular Biology 2004
    PMID 15199122DOI
    What this evidence cannot show

    Full text read via PMC. Fusions tested are KMT2A::EP300 in 293 cells and endogenous fusion KMT2A in the HB line, not KMT2A::AFF1. 2004 mass-spectrometry depth means the negative for the other components is weaker than the positive for menin.

    Evidence that argues against this (10)
    • The pillar of this hypothesis is a 2004 affinity-purification NEGATIVE, and a direct head-to-head shows that negatives from that method class do not generalise. Lambert et al. compared BioID with an AP-MS protocol optimised for chromatin-associated complexes on the same baits and found the two 'identified few shared interaction partners and enriched for distinct biological processes', with BioID recovering proteins of lower cellular abundance that AP-MS missed. If two contemporaneous methods on the same bait share few partners, a 2004 co-IP that failed to see HCF-1, ASH2L, WDR5 or RBBP5 on a fusion establishes that they are not in a stable, detergent-resistant complex with it, not that they are absent from its neighbourhood. Read: abstract in full, not full text. Lambert JP et al. 2014
    • The hypothesis's operational conclusion - redirect away from the wild-type KMT2A complex literature because it 'describes a protein the leukemia does not make' - is contradicted by genetics. Thiel et al. show menin recruits BOTH wild-type MLL and MLL-AF9 to HOX loci, that wild-type MLL controls MLL-AF9-induced histone methylation, and that the wild-type Mll allele is required for MLL-AF9-induced leukemogenesis AND for maintenance of MLL-AF9-transformed cells. Every KMT2A-rearranged patient keeps one intact allele. Partners that bind only past the breakpoint therefore remain functionally required in the disease, in trans, and are not off the therapeutic map. Read: abstract in full, not full text. Thiel AT et al. 2010
    • A second genetic result makes the trans-dependency mechanistic and specific to the discarded fragment. Milne et al. identify an essential MLL1 recruitment domain spanning the CXXC domain AND the PHD fingers, controlled by direct PAF1C contact and by H3K4me2/3 reading, and report that 'MLL1 fusion proteins lack the PHD fingers and require prebinding of a wild-type MLL1 complex and CXXC domain recognition of DNA for stable HoxA9 association'. On this account the fusion is not geometrically self-sufficient on aa 1-1406 at all: it borrows the discarded fragment's function from the wild-type allele. That reframes the target space from 'what binds the retained fragment' to 'what the fusion needs, wherever it sits'. Read: abstract in full, not full text. Milne TA et al. 2010
    • The strongest empirical counter is that a discarded-fragment contact has already yielded a KMT2A-selective agent. MM-401 blocks the MLL1-WDR5 interaction - WDR5 binds the Win motif adjacent to the SET domain, i.e. as far past the breakpoint as a KMT2A contact gets - and Cao et al. report that it 'specifically blocks proliferation of MLL cells by inducing cell-cycle arrest, apoptosis, and myeloid differentiation without general toxicity to normal bone marrow cells or non-MLL cells', without affecting other MLL-family HMTs. Selectivity for KMT2A-rearranged disease was therefore obtained from exactly the literature this hypothesis proposes to stop mining. Read: abstract in full, not full text. Cao F et al. 2014
    • Counter to the counter, and the reason the WDR5 result does not settle the matter: Mishra et al. deleted the MLL1 SET domain and found the H3K4 methyltransferase activity of MLL1 dispensable both for maintaining HSCs and for supporting MLL-AF9-driven leukemogenesis, with the relevant output instead being recruited MOF/H4K16 acetylation opposed by SIRT1. So the enzymatic activity MM-401 blocks is genetically dispensable in the same model, and the MM-401 phenotype may run through complex assembly or a non-catalytic route. The honest position is that the discarded fragment is therapeutically live but mechanistically unresolved - which is still fatal to a flat 'low expected yield' claim. Read: abstract in full, not full text. Mishra BP et al. 2014
    • A second discarded-fragment contact that is a validated, druggable dependency. MOF/KAT8 binds the MLL1 C-terminal region (Dou 2005), yet Valerio et al. show conditional Mof deletion reduces tumour burden and prolongs survival in an MLL-AF9 model, that MOF HAT activity is required for maintenance by catalytically dead rescue, and that the MYST HAT inhibitor MG149 is antiproliferative against the human KMT2A-rearranged lines MOLM13 (IC50 25.4 uM) and NOMO1 (50.9 uM). Two caveats read from the full text and material to this graph: Mof loss does NOT downregulate the MA9 target genes Hoxa9/Meis1 (the mechanism appears to be DNA damage and global H4K16ac loss), and Kasumi-1, U937 and K562 are also MG149-sensitive - so MOF is a real requirement of KMT2A-rearranged leukemia but not a fusion-selective handle. Read: FULL TEXT via PMC5501293. Valerio DG et al. 2017
    • LEDGF/PSIP1 is the strongest candidate for a second druggable handle on the retained fragment, and the case FOR it is stronger than this hypothesis allows. Mereau et al. show that expressing LEDGF(325-530) impairs clonogenic growth of MLL-fusion-transformed human and murine haematopoietic cells without affecting FLT3-ITD-immortalised or normal lineage-depleted marrow cells, that two 12-residue peptides (LEDGF 424-435 and 375-386) phenocopy this, that all of them disrupt the LEDGF/p75-MLL interaction, and that expression of LEDGF(325-530) or LEDGF(375-386) lengthened latency in a mouse MLL-AF9 bone-marrow-transplant model. They conclude there is 'a direct rationale for the design of small molecule inhibitors targeting this interaction'. A second retained-fragment interface has been disrupted selectively and has produced an in vivo effect, which is more than PAF1C has done. Read: abstract in full, not full text. Mereau H et al. 2013
    • And the therapeutic-window half of the LEDGF case, which is what usually kills chromatin targets. El Ashkar et al. made a conditional Psip1 knockout in the haematopoietic compartment: knockout mice are viable, and LEDGF/p75 is 'dispensable for steady-state hematopoiesis but essential for the initiation of MLL-mediated leukemia'. Two qualifications the entry should carry if it uses this: the knockouts did show several haematopoietic defects, reduced in vitro colony-forming activity and decreased HOX-gene expression in HSCs, so 'dispensable' means tolerated rather than inert; and the demonstration is for INITIATION, not for maintenance of established disease, which is the clinically relevant setting. Read: abstract in full, not full text. El Ashkar S et al. 2018
    • The case AGAINST LEDGF as a druggable second handle, stated by the same laboratory group. Tesina et al. show that the LEDGF/p75 integrase-binding domain engages JPO2, PogZ, IWS1, MED1, HIV integrase and MLL1 through a single conserved, intrinsically disordered IBD-binding motif, that the PogZ interaction is 'nearly identical' to the MLL1 interaction, and they conclude explicitly that 'the similar binding modes of LEDGF/p75 interaction partners represent a new challenge for the development of selective interaction inhibitors'. The MLL-LEDGF interface is a disordered-peptide groove shared across a family of cellular partners - the opposite of the well-formed menin pocket. That is a concrete, structural reason why LEDGF is not simply menin's equal on the retained fragment. Read: abstract in full, not full text. Tesina P et al. 2015
    • A further complication for any static LEDGF-interface inhibitor, and the beat's best example of a multivalent, regulated contact on the retained fragment. Sharma et al. show the IBD-binding motifs of LEDGF's cellular partners are switched between low- and high-affinity states by phosphorylation, and that eliminating the IBM phosphorylation sites on MLL1 disrupts the oncogenic potential of primary MLL1-rearranged leukemic cells - which they frame as 'a previously unknown oncogenic dependency'. The affinity of the KMT2A-LEDGF contact is therefore a kinase-set variable, so both the target occupancy an inhibitor must achieve and the kinase upstream of it are part of the target. It also nominates a route that is neither menin nor PAF1C, which is a direct counter to the hypothesis's 'menin may be most of it'. Read: abstract in full, not full text. Sharma S et al. 2018
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    MLL PHD3 is lost in all MLL translocation products and reinsertion of PHD3 into MLL fusion proteins abrogates their transforming activity; PHD3 binds H3K4me3 and the CYP33 RRM at distinct, mutually inhibitory sites.

    The PHD3 domain of MLL acts as a CYP33-regulated switch between MLL-mediated activation and repression. · Park S, Osmers U, Raman G, Schwantes RH, Diaz MO, Bushweller JH · Biochemistry 2010
    PMID 20677832DOI
    What this evidence cannot show

    Abstract read in full; not full text. This citation supports the premise that a discarded contact can carry real function, which cuts against a purely passive reading of breakpoint geometry: the break is not only removing partners, it is removing a brake.

    Evidence that argues against this (1)
    • PHD3 reinsertion abrogating transformation is a strong sufficiency result and it argues for a specific LOST function rather than for geometry: if restoring a domain in cis kills transformation, what the fusion needs is the absence of PHD3's repressive switch, which is biology and not position. Cell Reports 2014
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    MLL interacts directly with PAF1C through sequences flanking the CxxC domain, and MLL deletions that abolish the PAF1C interaction eliminate MLL-AF9-mediated immortalisation.

    The PAF complex synergizes with MLL fusion proteins at HOX loci to promote leukemogenesis. · Muntean AG, Tan J, Sitwala K, Huang Y, Bronstein J, Connelly JA, Basrur V, Elenitoba-Johnson KSJ, Hess JL · Cancer Cell 2010
    PMID 20541477DOI
    What this evidence cannot show

    Abstract read in full; not full text. The fusion tested is MLL-AF9 in murine immortalisation assays. PAF1C is the second retained-fragment interaction and is therefore the natural test case for whether availability implies criticality — it appears not to, which weakens the hypothesis.

    Evidence that argues against this (2)
    • PAF1C is the hypothesis's own test case for 'available but not critical', and the directionality of interfering with it is not merely unimpressive, it may be backwards. Chen et al. show that reducing PAF1 releases paused Pol II into gene bodies at thousands of genes, increasing nascent and mature transcripts and Ser2 phosphorylation, and that the increase is explained by recruitment of the super elongation complex. KMT2A::AFF1 drives its targets through SEC. A PAF1C-directed agent could therefore raise SEC-dependent output at fusion targets rather than lower it, which is the opposite of the therapeutic direction and a reason PAF1C is a poor second handle independent of any selectivity argument. Read: abstract in full, not full text. Chen FX et al. 2015
    • And the field does not agree on that directionality, which is itself the point. Four months later Yu et al. reported in Science that PAF1 is 'a critical regulator of paused Pol II release', with P-TEFb recruiting PAF1C and PAF1C in turn required for CDK12 recruitment - i.e. PAF1 loss should impair, not enhance, pause release. Two 2015 papers from major labs draw opposite arrows on the same protein. A hypothesis that uses PAF1C as the calibrating example of a non-critical retained-fragment contact is resting on a literature that cannot yet say which way the effect points. Read: abstract in full, not full text. Yu M et al. 2015
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    A PubMed search on 2026-09-18 for proximity-labelling studies of KMT2A or MLL in leukemia returned zero indexed articles.

    What this evidence cannot show

    Bounded by query wording and PubMed indexing; preprints excluded. 'Not findable by these queries' rather than 'does not exist'.

    Evidence that argues against this (3)
    • The null search is real for KMT2A but the framing 'no proximity-labelling study of any KMT2A fusion exists, therefore the transient interactome of chromatin fusions is unexplored' overstates it: proximity labelling HAS been applied to a chromatin-bound fusion oncoprotein in another disease, and what it returned is instructive. Elzi et al. ran BioID on EWS-FLI1 and the hit they carried forward was the cation-independent mannose 6-phosphate receptor, a trans-Golgi-to-endosome cargo receptor, leading to a lysosomal-turnover mechanism rather than to a chromatin partner. So the expected yield of a proximity map on a nuclear fusion is not obviously a list of missed chromatin partners; compartment and abundance bystanders can dominate the shortlist. Read: abstract in full, not full text. Elzi DJ et al. 2014
    • A second precedent, this one on transcription-complex composition. Kitagawa et al. applied BioID to EWS-ETS fusion transcription factors in Ewing sarcoma cells specifically 'to identify the transcriptional complexes' and recovered AHDC1, a protein not previously in the EWS-FLI1 interactome, whose knockdown reduced EWS-FLI1 transcriptional activity and lowered BRD4 and BRG1 protein levels. The known direct partners were reached indirectly. This is evidence both that the missing KMT2A experiment is feasible and that its output is likely to be a novel adjacent protein rather than a clean adjudication of the 2004 WRAD negative. Read: abstract in full, not full text. Kitagawa T et al. 2022
    • Search extended for this repair beat on 2026-09-18 with four further query forms against the NCBI PubMed E-utilities - (BioID OR TurboID OR APEX2 OR 'proximity labeling' OR 'proximity labelling' OR 'proximity-dependent biotinylation') AND (KMT2A OR MLL1 OR 'mixed lineage leukemia'); 'proximity labeling fusion oncoprotein interactome leukemia biotin ligase'; 'TurboID leukemia fusion protein nuclear interactome'; 'BioID menin interactome' - and all four returned zero indexed articles. The beat's negative survives a harder search than the one that generated it, so it should be read as robust within PubMed indexing rather than as an artefact of query wording. It remains bounded by PubMed indexing and excludes preprints. Oncofusion Knowledge Graph project et al. 2026
Connected to