Oncofusion

start / graph / exp-turboid-map-of-kmt2a-aff1

Experimentmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q5d-interactome3

Proximity-labelling map of KMT2A::AFF1 — the experiment nobody has run in twenty years of interactome work

also called TurboID KMT2A::AFF1 · proximity interactome of MLL-AF4

Tag the fusion protein with an enzyme that sprays a chemical label on anything that comes near it, then collect everything that got labelled. It would give the first list of what is actually next to the fusion inside a living leukemia cell, which no one has ever produced.

The same thing, for a scientist

Knock a TurboID (or miniTurbo) cassette in-frame into the endogenous *KMT2A::AFF1* allele at the 3' end of the fusion open reading frame in a t(4;11) B-ALL line (SEM or RS4;11), with a parallel line carrying the tag on the retained wild-type *KMT2A* allele as the discriminating control. Label with biotin, capture on streptavidin, quantify by TMT or label-free MS against three controls: untagged parental, wild-type-KMT2A-tagged, and nuclear-localised free TurboID. The readout is the set of proteins enriched on the fusion and not on wild-type KMT2A, and the set enriched on wild-type KMT2A and not on the fusion. As of 2026-09-18 no proximity-labelling study of KMT2A or any KMT2A fusion is indexed in PubMed, so every composition claim in this field rests on affinity purification of stable complexes, most of it done between 2004 and 2012.

Go deeper
Why this specific design

Endogenous tagging, not overexpression. The fusion is expressed from its own promoter at its own stoichiometry. Overexpression is the single most common way to manufacture false interactions in this field, and the existing literature has plenty of it.

The wild-type allele as the control. KMT2A-rearranged cells retain one normal KMT2A allele. Tagging it in a sister line gives a matched comparator that shares cell type, chromatin state and the N-terminal 1,406 residues, and differs only in what is fused past the break. A protein enriched on both is a retained-fragment partner; a protein enriched only on wild-type is a discarded-fragment partner; a protein enriched only on the fusion is a partner-supplied contact. That three-way partition is exactly what the graph's interactome entries currently have to infer from breakpoint arithmetic.

C-terminal tag. Placing TurboID at the 3' end of the fusion ORF avoids disturbing the menin-binding motifs at the extreme N-terminus, which are the interaction this project most needs undisturbed.

Proximity, not binding. The result is a proximity list, and the entry that consumes it must say so. A BioID/TurboID hit means 'was within roughly 10 nm for the labelling window', not 'binds'. Anything interesting from the screen needs orthogonal confirmation by co-IP or a structural method before it enters the graph as an interaction.

What it would settle
  • Whether WRAD subunits (WDR5, RBBP5, ASH2L, DPY30) are near KMT2A::AFF1 at all. The graph asserts they are interactors; the founding biochemistry says they do not coprecipitate with fusion MLL 1. This is the cleanest available adjudication.
  • Whether the retained-fragment interaction surface is as sparse as the geometric hypothesis predicts, or is covered in transient partners that stable-complex purification never saw.
  • Which SEC species — SEC, SEC-L2, SEC-L3 — is actually adjacent to the fusion in a B-lymphoid blast, a question that fifteen years of generic 'an AFF-family scaffold' statements have left open.
  • Whether DotCom components are near KMT2A::AFF1 at all, given that reconstitution says DOT1 does not associate with AF9-AF4-P-TEFb complexes 2.
  • Whether anything nobody has thought of is there. This is the only design on the table with a chance of returning a genuinely new name.
Cost, risk and what would make it fail

Cost tier: medium. Two knock-in lines, biotin labelling, streptavidin capture, one TMT MS run with replicates. Within reach of a single well-equipped lab over roughly six to nine months.

Risks. (i) The knock-in may perturb fusion function — mitigated by checking HOXA9/MEIS1 expression and proliferation against parental. (ii) Nuclear proximity labelling produces long lists dominated by abundant chromatin proteins — mitigated by the free-TurboID and wild-type-allele controls, which is why both are specified. (iii) t(4;11) lines are notoriously few; SEM and RS4;11 are the realistic options and neither is an infant primary sample. (iv) A negative result — nothing fusion-specific beyond what is known — is a publishable and useful outcome for this graph, and should be pre-committed to as such.

Facts and figures
designCRISPR knock-in of TurboID (or miniTurbo) in-frame at the 3' end of the endogenous KMT2A::AFF1 ORF in SEM and RS4;11; a parallel line with TurboID knocked into the retained wild-type KMT2A allele; biotin pulse labelling (10-30 min for miniTurbo, 1-3 h for TurboID); streptavidin capture; quantitative MS (TMT or label-free) with four arms: fusion-tagged, wild-type-KMT2A-tagged, nuclear free TurboID, untagged parental. Three biological replicates per arm.
readoutProteins significantly enriched on the fusion relative to each control, partitioned three ways: shared with wild-type KMT2A (retained-fragment partners), unique to wild-type KMT2A (discarded-fragment partners), unique to the fusion (partner-supplied contacts).
controlsUntagged parental line; nuclear-localised free TurboID for abundance/compartment bias; wild-type-KMT2A-tagged line as the discriminating control; functional checks (HOXA9/MEIS1 expression, proliferation) that the knock-in does not perturb fusion function.
cost tiermed
timeline6-9 months for two knock-in lines, labelling optimisation, and one quantitative MS campaign with replicates
prespecified interpretationRecovery of WRAD subunits, HCF-1, CyP33, CBP or MOF on the fusion would falsify the breakpoint-geometry hypothesis. Recovery of only menin, LEDGF, PAF1C and SEC/AEP components would support it. Any novel hit is a proximity call and requires orthogonal co-IP confirmation before entering the graph as an interaction.
limitationsSEM and RS4;11 are cell lines, not infant primary blasts; proximity is not binding; nuclear proximity labelling is biased toward abundant chromatin proteins even with controls.
tests hypotheses
[
  "hyp-menin-primacy-is-geometry-not-biology"
]
literature recheck
{
  "date": "2026-09-19",
  "counts": {
    "BioID AND KMT2A": 0,
    "TurboID AND KMT2A": 0,
    "BioID AND MLL": 0,
    "(BioID OR TurboID) AND (KMT2A OR MLL)": 0,
    "BioID (alone)": 853,
    "TurboID (alone)": 501,
    "KMT2A (alone)": 3216,
    "BioID AND leukemia": 14,
    "TurboID AND leukemia": 12
  },
  "false_positives_checked": {
    "16433901": "3D-FISH gene positioning, 2006, not proximity labelling",
    "26836506": "LC-HRMS methods validation, unrelated"
  },
  "method_note": "Single-term counts were run in the same session as the combination queries specifically to prove the index was reachable, because a silent query failure and a true zero look identical."
}
How this could bite us back
  • A long proximity list invites over-interpretation; without the wild-type-allele control the experiment would generate more graph entries than knowledge.
  • If the experiment returns nothing beyond the known partners, the field loses its last easy hope of a second handle, and the menin-resistance problem becomes sharper rather than softer.
Still open
  1. Would a fetal-liver-derived KMT2A::AFF1 model, rather than an established line, give a different proximity interactome? The cell of origin matters to this project and cell lines may not represent it.
  2. Should the reciprocal AFF1::KMT2A product be tagged in the same campaign? It would cost little extra and would settle whether RUNX1 and SIAH contacts are real in cells.
  3. Is the labelling radius of TurboID small enough to discriminate 'in the complex' from 'in the same chromatin neighbourhood' at a transcriptionally active locus?
Every source behind this entry (5)
  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, and menin is the only identified MLL/HCF complex component interacting with both wild-type and fusion MLL.

    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 were KMT2A::EP300 and endogenous fusion KMT2A in the HB line, not KMT2A::AFF1, which is precisely why this experiment is needed.

    Evidence that argues against this (2)
    • The design's discriminating control is the weakest part of it, and two papers say so. The experiment proposes tagging the retained wild-type KMT2A allele in a sister line so that a three-way partition (shared / wild-type-only / fusion-only) can be read off. Thiel et al. show menin recruits BOTH wild-type MLL and the MLL-AF9 fusion to the same HOX loci, that wild-type MLL controls the fusion's histone methylation output, and that the wild-type allele is required for leukemogenesis and maintenance. A control that co-occupies the bait's own target loci and is functionally coupled to it is not a matched negative; both tagged lines will biotinylate the same chromatin neighbourhood, and the 'fusion-only' set will be small and dominated by the partner moiety by construction. Read: abstract in full, not full text. Thiel AT et al. 2010
    • Worse, the coupling is physical. Milne et al. report that MLL1 fusion proteins 'require prebinding of a wild-type MLL1 complex' for stable HoxA9 association. If the wild-type complex is prebound at the loci the fusion occupies, then a TurboID on either allele labels the other allele's complex, and a labelling radius of roughly 10 nm cannot separate them. The pre-specified interpretation in data.prespecified_interpretation - that recovering WRAD, HCF-1, CyP33, CBP or MOF on the fusion would falsify the geometry hypothesis - is not safe: those proteins would be expected on the prebound wild-type complex regardless. The experiment needs a wild-type-allele knockout arm, or an allele-resolved orthogonal test, before its falsifier means what it says. Read: abstract in full, not full text. Milne TA et al. 2010
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In reconstituted defined complexes, DOT1 forms a stable complex with AF9 and does not interact with AF9-AF4-P-TEFb complexes.

    Function of leukemogenic mixed lineage leukemia 1 (MLL) fusion proteins through distinct partner protein complexes. · Biswas D, Milne TA, Basrur V, Kim J, Elenitoba-Johnson KSJ, Allis CD, Roeder RG · Proceedings of the National Academy of Sciences USA 2011
    PMID 21896721DOI
    What this evidence cannot show

    Abstract read in full; not full text. In vitro reconstitution with recombinant proteins, which says what can assemble rather than what does assemble in a nucleus — again the reason for an in-cell experiment.

    Evidence that argues against this (2)
    • Published failure mode, bystander class, on a chromatin fusion oncoprotein. Elzi et al. applied BioID to EWS-FLI1 and the protein they identified 'as a protein located in the vicinity of EWS-Fli-1 within a cell' and carried forward was the cation-independent mannose 6-phosphate receptor, a trans-Golgi-network-to-endosome cargo receptor. The finding was real - it led to a lysosomal turnover mechanism - but it is not a transcription-complex partner, and it illustrates that a nuclear fusion's proximity list is not automatically a chromatin list. The listed risk (ii) 'nuclear proximity labelling produces long lists dominated by abundant chromatin proteins' understates the problem: the lists are not confined to chromatin proteins. Read: abstract in full, not full text. Elzi DJ et al. 2014
    • Published failure mode, method-discordance class. Lambert et al. benchmarked BioID against a chromatin-optimised AP-MS protocol on the same baits and found the two 'identified few shared interaction partners and enriched for distinct biological processes'. A TurboID map of KMT2A::AFF1 will therefore not be commensurable with the 2004-2012 affinity-purification literature it is meant to adjudicate: agreement and disagreement with Yokoyama 2004 would both be expected from method difference alone. The experiment is worth running as a discovery tool and is a poor instrument for settling a twenty-year-old co-IP negative. Read: abstract in full, not full text. Lambert JP et al. 2014
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    No proximity-labelling study of KMT2A or any KMT2A fusion protein in leukemia is indexed in PubMed as of 2026-09-18.

    What this evidence cannot show

    Bounded by query wording and indexing; preprints not covered. A study using APEX2 or unusual phrasing could be missed. Read as 'not findable by these queries'.

    Evidence that argues against this (2)
    • 'The experiment nobody has run in twenty years' is accurate for KMT2A and wrong as a statement about chromatin oncofusions. Kitagawa et al. applied BioID to EWS-ETS fusion transcription factors in Ewing sarcoma cells explicitly 'to identify the transcriptional complexes of EWS-ETS fusion transcription factors', and recovered AHDC1 - a protein absent from the prior interactome - whose knockdown reduced EWS-FLI1 transcriptional activity and lowered BRD4 and BRG1. The known partners BRD4 and BRG1 were reached as downstream consequences, not as proximity hits. That is a fair forecast of what a KMT2A::AFF1 map returns: one or two genuinely new adjacent proteins, and no clean verdict on the canonical partners. Read: abstract in full, not full text. Kitagawa T et al. 2022
    • The experiment's third risk - that the knock-in may perturb fusion function - is mitigated here only by post-hoc checks on HOXA9/MEIS1 and proliferation. The AFF1 moiety retained in KMT2A::AFF1 carries the C-terminal homology domain (aa 949-1209 of AFF1), which Chen and Cramer show is a folded eight-helix domain mediating AFF homo- and heterodimerisation. A C-terminal TurboID cassette, as specified in data.design, is appended directly onto that dimerisation surface. Steric interference with AFF-family dimerisation would not be detected by a HOXA9/MEIS1 readout if dimerisation is partially redundant, and would silently deplete exactly the AFF-scaffold partners the experiment is meant to enumerate. An N-terminal-internal or internal tag, or a dimerisation assay as a release criterion, should be pre-specified. Read: abstract in full, not full text. Chen Y et al. 2019
  4. E_recheck_2026_09_19strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    Re-checked 2026-09-19: PubMed still returns ZERO records for proximity labelling of any KMT2A fusion. 'BioID AND KMT2A' = 0, 'TurboID AND KMT2A' = 0, 'BioID AND MLL' = 0, '(BioID OR TurboID) AND (KMT2A OR MLL)' = 0. The search method was sanity-checked against the same index in the same session - BioID alone returns 853, TurboID 501 and KMT2A 3,216 - so the zeros are real absence, not a failed query. The only two records that surface on looser terms are false positives: PMID 16433901 is 3D-FISH gene positioning (spatial proximity, not proximity labelling, and published in 2006 before BioID existed), and PMID 26836506 is a liquid-chromatography mass-spectrometry methods paper. Both were fetched and their titles read.

    What this evidence cannot show

    A PubMed title/abstract search cannot see a proximity map published inside a paper about something else, deposited only in a repository such as PRIDE, or described with vocabulary the query does not cover (split-BioID, miniTurbo, BASU, uncommon spellings). It also cannot see preprints outside PubMed. This bounds the claim to 'not published as a findable, indexed result', which is weaker than 'has never been done'.

    Evidence that argues against this (1)
    • Absence of an indexed result is the weakest kind of evidence and this project has been wrong on exactly this pattern before: 'nobody has looked' at the endocrine endpoints was false (an FDA post-marketing requirement and the REVUFORJ label both existed), and 'never measured for KMT2A' was false (Yamamoto 1998, PMID 9737688, screened 65 cord bloods). The honest reading is that the experiment is not findable in the literature, which is a reason to ask people in the field rather than to assert priority. NCBI PubMed 2026
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    This experiment is not a new method, it is an unused protocol. TurboID has already been fused **in frame to oncofusion cDNAs and expressed in primary murine haematopoietic cells**: PML::RARA, RUNX1::RUNX1T1 and CBFB::MYH11 were each mapped this way, recovering common and unique nuclear repressor complexes for the first two and a largely cytoplasmic interactome for CBFB::MYH11 through its MYH11 myosin domain. BioID has separately been applied to NUP98::HOXA9 and SET::NUP214. **No KMT2A fusion has been mapped this way, and the reason is a selection criterion rather than a technical barrier: the TurboID study chose the three favourable-risk AML oncofusions, which excludes KMT2A by definition.**

    PMID 38061017DOI
    What this evidence cannot show

    Both abstracts fetched and read by the coordinator on 2026-09-20; neither full text was read, so the labelling radius, expression level relative to endogenous, and the controls used are not recorded here. The TurboID work is murine and myeloid; a KMT2A::AFF1 version would need a B-lymphoid or primary infant context that no published protocol covers.

    Evidence that argues against this (2)
    • An in-frame TurboID tag on a 400 kDa KMT2A fusion is not obviously the same proposition as tagging PML::RARA. The KAT6A modelling literature resorted to domain-focused truncation precisely because fusion size impeded construct work, and KMT2A::AFF1 is larger still. 'The protocol exists' understates the construct problem. Cells 2020
    • Proximity labelling of the fusion would answer a different question from the one this graph keeps asking. It reports what is near the fusion, not what binds it, and this beat's own concept entry records that proximity and affinity purification recover few shared partners by design. A TurboID map would not settle the WRAD or PAF1c membership questions it is often hoped to settle. Journal of Clinical Investigation 2023
Connected to