The missing reagent: a KMT2A-fusion or wild-type-KMT2A proximity bait in a KMT2A-rearranged leukaemia line - IntAct has curated no KMT2A-bait interaction experiment since April 2015 and no PSIP1-bait since October 2012
also called KMT2A bait BioID · the scaffold-as-bait experiment · TurboID KMT2A::AFF1
Nobody has tagged the protein this project is about and asked what sits next to it in a leukaemia cell for over a decade. Everything currently known about its neighbours comes from other people tagging other proteins and finding it by accident.
The same thing, for a scientist
**The gap this closes, measured rather than asserted.** Across all 376 IntAct records touching KMT2A (Q03164), KMT2A is the experimental bait in 96 rows across **18 publications, the most recent PMID 25751424, Nature Structural & Molecular Biology, April 2015**; across all 114 records touching PSIP1 (O75475), PSIP1 is bait in 11 rows across **4 publications, the most recent PMID 22872640, J Biol Chem, October 2012** [E1]. **Zero of the thirteen never-discussed partners adjudicated by Q129 comes from an experiment in which either protein was the bait** [E1]. And this graph has already recorded that in the only modern proximity map it can reach - the Khoroshkin 50-RBP BioID2 census in K562 - **no super elongation complex component, no KMT2A-complex component and no fusion protein is a bait**; KMT2A, MEN1, AFF1, AFF4, ELL, MLLT1, CDK9, CCNT1 and DOT1L appear only as prey [E2]. **So the single missing reagent for this project's first-priority dimension is a tagged KMT2A or KMT2A-fusion bait, in a cell that has the disease.** Everything else - which partners are real, whether there is a nucleolar pool, whether the ribosomal block is background - is currently being argued from other laboratories' bait choices, and no amount of re-reading IntAct will settle any of it.
Design
Readout. Proximity labelling (TurboID or miniTurbo, chosen over BioID2 for a 10-minute rather than 18-hour labelling window, which matters for a protein whose chromatin residence is dynamic), streptavidin capture, LC-MS/MS.
Baits, in priority order.
- KMT2A::AFF1 - the index fusion, C-terminally tagged. This is the construct nobody has ever used as a proximity bait anywhere.
- Wild-type KMT2A, tagged at the N-terminus and separately at the C-terminus, because taspase1 cleaves the protein into N320 and C180 and UniProt already models the two fragments as separate localisation entities with their own (uncoded) location comments. A single tag reports on one fragment.
- PSIP1/LEDGF, to close the 2012 gap on the tether.
System. A KMT2A-rearranged human leukaemia line (SEM or RS4;11 for the t(4;11) B-lineage index disease; MV4-11 or MOLM-13 if a myeloid arm is wanted), with K562 as the deliberate wild-type comparator, because the existing BioID2 census is in K562 and running the new bait in the same line makes the two datasets directly comparable rather than merely adjacent.
Controls, and these are the part that decides whether the result is usable.
- Unfused TurboID, same line, same biotin pulse - the standard soluble-nuclear background.
- A size-matched irrelevant nuclear bait. This project has already found that KMT2A's apparent proximity promiscuity is largely explained by its 3,969 residues; a control of comparable size is the only way a new dataset avoids rediscovering that.
- No-biotin arm, as Khoroshkin et al. used, for endogenous biotinylation.
- Nuclease and no-nuclease arms. Three of the thirteen declined partners are nucleic-acid-binding proteins and none of the source experiments recorded a nuclease control; one comparison would adjudicate that whole class.
Analysis. Report the raw bait-prey table AND the promiscuity-corrected one, with each prey's baits-per-prey count from the existing K562 census beside it, so that a prey recovered by 25 of 50 unrelated baits is visibly flagged rather than presented as a KMT2A partner.
Pre-specified readouts for this project's open questions: is any ribosome-biogenesis or nucleolar protein recovered above the size-matched control; is any of ABT1, KRR1, UTP3, IRF2, OASL or PRKRA recovered at all; does the fusion's neighbourhood differ from wild-type KMT2A's.
Why this and not the cheaper things first
It is not first. Q129 identified three re-analyses that cost an afternoon each and should precede any wet work: the crosslinked residue positions in Fasci supplemental Tables S1/S2; Supplementary Table 2 of the Hubel ISG paper; and the size-matched promiscuity correction for PSIP1, specified in ana-q129-psip1-within-length-decile-percentile-in-the-k562-bioid2-map.
But none of them can produce a KMT2A-bait row, and that is the structural gap. Re-analysis can only re-read other people's bait choices. The honest position is that this project's interactome layer has been growing for a decade entirely from by-catch, and no amount of careful curation changes what question was asked.
Cost tier: high, but bounded - this is one well-controlled proximity experiment with four bait constructs, not a programme.
| design | TurboID or miniTurbo proximity labelling with streptavidin capture and LC-MS/MS, in a KMT2A-rearranged human leukaemia line (SEM or RS4;11 for the t(4;11) B-lineage index disease; MV4-11 or MOLM-13 for a myeloid arm) plus K562 as a deliberate wild-type comparator matched to the existing 50-bait BioID2 census. Baits in priority order: (1) KMT2A::AFF1, C-terminally tagged; (2) wild-type KMT2A tagged separately at N- and C-termini, because taspase1 cleavage produces N320 and C180 and a single tag reports on only one fragment; (3) PSIP1/LEDGF. |
|---|---|
| readout | Bait-prey protein lists with quantification, reported BOTH raw and promiscuity-corrected, with each prey's baits-per-prey count from the existing K562 BioID2 census printed alongside so that a generic prey is visibly flagged rather than presented as a KMT2A partner. Pre-specified questions: is any nucleolar or ribosome-biogenesis protein recovered above the size-matched control; is any of ABT1, KRR1, UTP3, IRF2, OASL or PRKRA recovered at all; does the fusion's neighbourhood differ from wild-type KMT2A's. |
| controls | Unfused TurboID in the same line and biotin pulse; a SIZE-MATCHED irrelevant nuclear bait, because this project has already shown KMT2A's apparent promiscuity is largely explained by its 3,969 residues and an unfused control cannot correct for that; a no-biotin arm for endogenous biotinylation, as Khoroshkin et al. used; and paired nuclease / no-nuclease arms, because three of the thirteen partners declined in this beat are nucleic-acid-binding proteins and none of the source experiments recorded a nuclease control. |
| cost tier | high |
| timeline | 6-9 months for construct generation, line derivation, labelling optimisation and mass spectrometry, assuming an existing proximity-proteomics collaboration; the constructs are the long pole. |
| absence class | never-measured |
| absence class note | No instrument has been pointed at the question 'what is near a KMT2A fusion in a cell that has the disease'. No archive substitute exists: the deposited proximity data this project can reach (PXD041608, the 50-RBP BioID2 map) contains no chromatin-scaffold bait by design. This is wet work and years, not a supplementary table - which is precisely why the three cheaper prerequisites above come first. |
prerequisites
[ "Run a PubMed search for KMT2A-bait and PSIP1-bait interaction experiments published since 2015 and 2012. If modern pull-downs exist and are merely uncurated, the reagent is not missing and this experiment should not be commissioned as written.", "Open Fasci supplemental Tables S1/S2 and Hubel Supplementary Table 2 - two afternoons that adjudicate six of the thirteen declined rows without any wet work.", "Run ana-q129-psip1-within-length-decile-percentile-in-the-k562-bioid2-map, which is compute-only and determines whether PSIP1 needs its own size-matched control here." ]
- If this experiment is run and returns a large, generic neighbourhood, it will be tempting to report the novel-looking members. The promiscuity-corrected table and the size-matched control must be pre-specified as the primary readout, or the new dataset will simply regenerate the class of row this beat spent a cycle declining.
- A fusion-specific neighbour recovered here would immediately become an intervention candidate, and the same objections that sank the partners in this beat would apply to it: one experiment, one line, one method, no reciprocal arm. The experiment should be designed to produce its own confirmation - reciprocal tagging of the top hits - rather than a hit list.
- Every line available is a cell line. Even SEM and RS4;11 are decades-old cultures, and this project's recurring documented failure is generalising from a model to infant B-lineage KMT2A::AFF1 in patients. A primary-sample arm would be far harder and is the only version that answers the question that matters.
- Does a KMT2A or KMT2A-fusion proximity dataset already exist unpublished or uncurated? The Human Cell Map (192 BioID baits in HEK293) was noted in an earlier cycle as sitting behind a JavaScript application with no REST endpoint found, and it has never been checked for KMT2A as a bait.
- Should the first bait be the fusion or wild-type KMT2A? The fusion is the disease-relevant molecule but lacks the C-terminal SET domain and everything downstream of the breakpoint, so a wild-type arm is needed to tell fusion-specific neighbours from KMT2A neighbours - which doubles the cost and is the main design decision.
- Is a size-matched irrelevant nuclear bait actually obtainable? Few nuclear proteins are near 3,969 aa and inert; the practical alternative may be a tagged, catalytically dead or chromatin-binding-deficient KMT2A, which controls for size but not for localisation.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
Across all 376 IntAct REST records touching KMT2A (Q03164), KMT2A is the experimental bait in 96 rows spanning 18 publications, the most recent being PMID 25751424 (Nat Struct Mol Biol, April 2015), and prey in 165 rows spanning 40 publications. Across all 114 records touching PSIP1 (O75475), PSIP1 is bait in 11 rows spanning 4 publications, the most recent being PMID 22872640 (J Biol Chem, October 2012), and prey in 91 rows spanning 29 publications. Of the thirteen IntAct-curated human partners of these two proteins that are never discussed anywhere in this corpus, zero come from an experiment in which KMT2A or PSIP1 was the bait: eight are prey-side spoke-expanded rows and five come from a bait-free crosslinking dataset.
IntAct molecular interaction database, REST findInteractions on Q03164 and O75475, experimentalRole counted across all 376 and 114 records, with publication years fetched from NCBI esummary, retrieved 2026-09-23 by Q129 · IntAct (EMBL-EBI) 2026What this evidence cannot show
This bounds IntAct's curation, NOT the literature. A KMT2A-bait or PSIP1-bait experiment published after those dates and not curated by IntAct would be invisible, and no PubMed search was run to check - which matters, because 'the field stopped asking' and 'IntAct stopped curating' justify completely different responses, and only the first justifies this experiment. Running that search is a prerequisite to commissioning the work.
Evidence that argues against this (1)
- Both proteins have been studied intensively since those dates by ChIP-seq, cryo-EM, degron and CRISPR methods, none of which IntAct curates as interaction experiments. It is entirely possible that modern KMT2A pull-downs exist in the literature and simply have not been curated, in which case the reagent is not missing and the gap is a database gap. That possibility has not been excluded. IntAct (EMBL-EBI) 2026
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Khoroshkin et al. fused BioID2 to 50 human RBPs in K562 cells and captured each neighbourhood by streptavidin pull-down and mass spectrometry with matched no-biotin controls. This graph's entry ana-rbp-bioid2-proximity-map-k562 records that no super elongation complex component, no KMT2A-complex component and no fusion protein is among the 50 baits, and that AFF1, AFF4, ELL, MLLT1, CDK9, CCNT1, KMT2A, MEN1 and DOT1L appear only in the prey column. Independently verified here: the 50 baits are AGGF1, APOBEC3C, BUD13, CPSF6, DDX21, DDX52, DDX55, DDX6, DGCR8, EFTUD2, EIF3G, EXOSC5, FASTKD2, FXR2, GNL3, GPKOW, HNRNPC, HNRNPA1, ILF3, LARP4, LARP7, LIN28B, NOLC1, NONO, NPM1, NSUN2, PPIL4, QKI, RBM15, RO60, RPS11, RPS3, SBDS, SF3B4, SMNDC1, SRSF1, SRSF7, SUPV3L1, TAF15, TBRG4, TRA2A, U2AF1, U2AF2, UTP3, XRCC6, YWHAG, ZC3H11A, ZC3H8, ZNF800 and ZRANB2.
Systematic identification of post-transcriptional regulatory modules. · Khoroshkin M, Buyan A, Dodel M, Navickas A, Yu J, Trejo F, et al. (22 authors), Goodarzi H · Nature communications 2024What this evidence cannot show
Full text read 2026-09-23 via Europe PMC fullTextXML (PMC11385195, 144,676 stripped characters); the bait list was enumerated from IntAct's curation of the deposition rather than from the paper's own supplementary table, which was not opened. The baits were selected on three criteria the authors state - ENCODE eCLIP availability, presence in the ORFeome clone library, and diversity of RNA-metabolic process - so the absence of chromatin scaffolds is a consequence of the study's design and not an oversight.
Evidence that argues against this (1)
- This was an RBP study and correctly contains no chromatin baits, so citing it as evidence of a gap in KMT2A proximity data is slightly unfair to it. The stronger form of the argument is that NO published proximity map of any kind uses a KMT2A-complex component as bait - and this beat verified that only for IntAct's curation, not for the literature or for unindexed resources such as the Human Cell Map, which ana-rbp-bioid2-proximity-map-k562 notes sits behind a JavaScript application with no REST endpoint found. Oncofusion Knowledge Graph (internal) 2026
the experiment this audit implies: the missing reagent is a tagged KMT2A or KMT2A-fusion bait in a cell that has the disease
designed against that census as the wild-type K562 comparator; none of its 50 baits is a chromatin scaffold, by the study's own design criteria
would produce the first KMT2A-bait proximity dataset in a KMT2A-rearranged cell; pre-specified readouts include whether any nucleolar protein survives a size-matched control
the experiment that would supply a KMT2A-bait dataset and let the ratio be measured directly instead of inferred from two studies