Oncofusion

start / graph / absence-no-experiment-has-ever-asked-whether-a-nurd-scaffold-subunit-binds-a-kmt2a-fusion

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q193-the-nurd-subunits

No experiment has ever ASKED whether MTA2, MBD3, GATAD2A or SIN3A binds a KMT2A fusion - four of them were nonetheless in the lysate of a seven-fusion pulldown, and the class of that absence is measured-not-reported, not never-measured

also called the NuRD scaffold absence · Q193 bounded absence

Nobody has ever done a targeted experiment to see whether these four scaffold proteins stick to a KMT2A fusion protein. But they were all present in the cells of one published fishing experiment that would have caught them, so the numbers exist in a public file and just were not written down.

The same thing, for a scientist

This entry bounds one absence per subunit and, more usefully, states which instrument would have found each and why it did not report. **The literature bound**: zero records under both current and retired symbols tie MTA2 or GATAD2A to a KMT2A fusion; MBD3 returns exactly one record, which is a CRISPR dependency screen and not a binding experiment; SIN3A returns four records of which three fail a title check and the fourth is haematopoietic differentiation biology with no fusion in it [E1]. Positive control MEN1[tiab] AND MLL[tiab] = 55. **The instrument bound**, which is the part a PubMed count cannot give: all four proteins were in the nuclear lysate of the Skucha 2018 affinity purification against seven MLL fusions including KMT2A::AFF1, an experiment whose design could have recovered any of them, and which did recover CHD3, CHD4, GATAD2A, GATAD2B and MBD2 somewhere in the panel [E2]. GATAD2A and SIN3A were recovered under other baits and not the index one; MTA2 and MBD3 were recovered under none. Because the published table keeps only the top ~291-293 preys per bait, **a subunit's absence is a truncated rank, not a zero spectral count** - so the correct `absence_class` is `measured-not-reported`, and because the raw files are deposited at ProteomeXchange **PXD009338** it upgrades to `measured-not-analysed`: closable by compute, not by wet work. Calling any of these `never-measured` would overstate the cost of closing it by years, which is exactly the error this corpus's absence-class convention exists to prevent.

Go deeper
Per subunit: what was asked, what was in the lysate, what class the absence is
subunittargeted experiment in any KMT2A fusionin the Skucha lysaterecovered, and under which baitabsence_class for the index fusion
MTA2none, under MTA2 or MTA1L1yesno baitmeasured-not-reported -> measured-not-analysed
MBD3none binding; ONE dependency screen in SEM/ALL-POyesno baitmeasured-not-reported -> measured-not-analysed
GATAD2Anone, under GATAD2A or p66alphayesMLL-AF9 only, p = 0.0361, 10|12measured-not-reported -> measured-not-analysed
SIN3AnoneyesMLL-AF9, MLL-EEN, MLL-ENL; neg ctrl 1|2measured-not-reported -> measured-not-analysed

The distinction this table is making. "Nobody has asked" and "the answer is not in the literature" are both true here, and neither means nobody measured. A mass spectrometer generated a spectral count for every one of these proteins in every one of the seven pulldowns. The paper reported the top ~300 per bait. The rest are in PXD009338.

Which experiments would have found it, and why each did not report
  1. Skucha 2018 AP-MS (PMID 29777171), seven Strep-HA MLL fusions in Jurkat, nuclear lysate, LC-MS/MS. Could have found all four. Did not report them for the index bait because of a top-300-per-bait rank truncation, not a threshold. measured-not-reported; raw data public.
  2. Skucha's own negative-selection RNAi screen, 128 conserved interactors at 6 shRNAs each in MLL-AF9 MOLM-13. Could not have found any of them: the library IS the interactor list, and none of the four is on it. never-measured for that screen specifically.
  3. Schneider 2023 CRISPR epigenome + kinome libraries in SEM and ALL-PO (both KMT2A::AFF1) and KOPN-8, against 697 and NALM-6 (PMID 37686014). Almost certainly screened all four, being epigenome-wide. Reported three hits - ARID4B, MBD3, BMPR2 - and tabulated no per-gene scores. measured-not-reported for the other three; closable by opening the supplementary data. NOTE that this screen is a DEPENDENCY instrument and would not answer the binding question for any of them.
  4. DepMap, which screens all four across its panel and yields ranks (GATAD2A 866, MTA2 1,037, MBD3 13,274, SIN3A 17,096). A real measurement, of dependency and not of binding.

So the honest formulation is: no binding experiment has ever been aimed at these proteins in a KMT2A-rearranged cell; one binding experiment that could have seen them reported a truncated list; and two dependency instruments have measured all four without addressing binding at all.

The archive check, run and recorded

This corpus's rule is that an absence bounded by literature search alone is the weak kind, and that one query in the matching archive upgrades it. Run on 2026-09-24:

  • ProteomeXchange / PRIDE: PXD009338 is the deposited raw dataset behind the seven-fusion AP-MS whose processed table this beat read. It exists, it is public, and it has not been re-searched for MTA2, MBD3, GATAD2A or SIN3A peptides by this project, and no publication found in E1 reports such a re-search. This is the single named step that closes four absences at once.
  • GEO: NOT RUN for occupancy of any of the four. No ChIP-seq absence is claimed in this entry for that reason, and an occupancy claim must not be read into it.
  • DepMap: run, and it is the reason none of these can be called never-measured without qualification.

Recording the GEO check as not run is deliberate: a beat that lists two archives and silently skips a third has produced a stronger-looking bound than it earned.

Facts and figures
scriptanalyses/q193-nurd-module-split/run.py
statusrun
absence classmeasured-not-reported
absence class notemeasured-not-reported for the AP-MS spectral counts and for the CRISPR per-gene scores; UPGRADES TO measured-not-analysed because the AP-MS raw files are public at PXD009338, so the binding half is closable by compute rather than wet work. It is NOT never-measured for any of the four. What IS never-measured, and is recorded separately in each gene entry rather than here: any targeted binding experiment, any occupancy experiment, any imaging and any histone-mark readout for any of the four in a KMT2A-rearranged cell.
result summaryZero targeted binding experiments for MTA2, MBD3, GATAD2A or SIN3A against any KMT2A fusion, under current and retired symbols, with a working positive control. All four were in the lysate of a seven-fusion AP-MS; GATAD2A scored under MLL-AF9 and SIN3A under three baits, neither under MLL-AF4; MTA2 and MBD3 scored under none. Absence class measured-not-reported, upgradeable to measured-not-analysed against PXD009338.
shown for fusionFor the binding question: NO in the strict sense for all four - nobody has looked in any fusion with a targeted experiment. Partial by co-purification: GATAD2A in MLL-AF9, SIN3A in MLL-AF9/MLL-EEN/MLL-ENL, neither in KMT2A::AFF1; MTA2 and MBD3 in none. For the DEPENDENCY question MBD3 is YES in KMT2A::AFF1 (SEM, and ALL-PO screened) and the other three are unreported.
relationship to existing entriesThis entry states the absence ONCE, at the level of the question, so the four gene entries do not each have to argue it. gene-mta2, gene-mbd3, gene-gatad2a and gene-sin3a carry their own per-subunit absence_class fields and their own never-measured flags for localisation and epigenetic readout; this entry carries the shared instrument analysis and the archive checks. It does not duplicate ana-the-mll-fusion-interactome-recovers-corepressor-scaffolds-without-their-hdac1-2-catalytic-core, which is about the module SPLIT; this is about what was never asked.
archive checks
{
  "ProteomeXchange_PXD009338": "EXISTS, public, is the raw data behind the seven-fusion AP-MS, and has NOT been re-searched for these four proteins' peptides by this project or by any publication found in E1. This is the named closing step.",
  "DepMap": "RUN. All four screened; ranks GATAD2A 866, MTA2 1,037, MBD3 13,274, SIN3A 17,096 of 17,728. Dependency, not binding.",
  "GEO": "NOT RUN. No occupancy absence is claimed here and none may be inferred from this entry."
}
measurements
[
  {
    "quantity": "targeted_binding_experiments",
    "subject": "MTA2, MBD3, GATAD2A and SIN3A against any KMT2A fusion",
    "value": 0,
    "units": "published experiments",
    "n": 4,
    "universe": "PubMed title/abstract, 2026-09-24, current and retired symbols, one tag per term; positive control MEN1[tiab] AND MLL[tiab] = 55",
    "null": null,
    "note": "rule of three on 0 of 4 subunits is uninformative at this n; the meaningful statement is the per-subunit zero with its name list, not a rate"
  },
  {
    "quantity": "subunits_recovered_in_the_index_bait",
    "subject": "MTA2, MBD3, GATAD2A, SIN3A in the MLL-AF4 column of the Skucha table",
    "value": 0,
    "units": "subunits",
    "n": 4,
    "universe": "the 291 proteins retained in the MLL-AF4 column of Skucha Supplementary Table 5",
    "null": null,
    "note": "GATAD2A and SIN3A ARE recovered under other baits, so the index-bait zero is not a detectability zero for those two"
  }
]
How this could bite us back
  • This entry is the kind that gets cited as though it said more than it does. It does NOT say NuRD is absent from the KMT2A::AFF1 complex; it says nobody asked, and the one experiment that could have answered reported a truncated list. A future beat that cites it for 'NuRD does not bind the index fusion' will have inverted it.
  • An absence that is closable by compute and stays open is worse than one that needs a mouse, because it looks like a finding while being an unopened file. Four of the absences here have a named ProteomeXchange accession and a named analysis; if they are still open next cycle, that is a choice.
Still open
  1. Re-search PXD009338 for MTA2, MBD3, GATAD2A, SIN3A, HDAC1, HDAC2, RBBP4 and RBBP7 peptides in the MLL-AF4 run. One compute job closes eight measured-not-reported absences and would say whether the deacetylase module's absence is biology or rank truncation.
  2. Open the Schneider 2023 supplementary data for per-gene scores on MTA2, GATAD2A, SIN3A and the rest of the NuRD and Sin3 subunits in SEM and ALL-PO. An afternoon, and it would turn a screen with three reported hits into a complete dependency profile of both complexes in the index fusion's own lineage.
  3. Is there a KMT2A::AFF1 AP-MS in a B-lineage host at all? Every binding statement in this beat inherits a Jurkat T-ALL proteome, and a single SEM or RS4;11 pulldown with a fusion-specific antibody would replace the entire inference chain.
  4. Has any NuRD or Sin3 subunit ever been used as the BAIT against a KMT2A fusion? Every result here is prey-side. A reciprocal pulldown is a different experiment and this absence does not cover it.
Every source behind this entry (3)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    Bounded literature queries run on 2026-09-24 with one field tag per term and a working positive control return zero records tying MTA2 or GATAD2A to a KMT2A fusion under either their current or their retired symbols, one record for MBD3 (a CRISPR dependency screen, not a binding experiment) and four for SIN3A of which three fail a title check.

    What this evidence cannot show

    NAMES TRIED, which a reader needs to judge any of these zeros: MTA2 and MTA1L1; MBD3; GATAD2A and p66alpha; SIN3A. NOT tried: 'PID' (a three-letter MTA2 alias matching pelvic inflammatory disease and much else, excluded on precision grounds) and 'mSin3A' as a bare token. One tag per term throughout - never the grouped (A OR B)[tiab] form, which silently expands to All Fields - with the querytranslation recorded for every count, and run through pubmed_safe so an empty translation raises rather than returning a fabricated zero. THE COMPLEX-LEVEL PAIR IS THE WARNING IN MY OWN DATA: NuRD with KMT2A returns 0 and NuRD with MLL returns 6, so the two symbol spaces are disjoint here and a single-symbol search would have produced a false absence.

    Evidence that argues against this (1)
    • THE BOUND IS ON REPORTING AND I CAN DEMONSTRATE IT UNDERSTATES MEASUREMENT BY A LARGE MARGIN. All four proteins are screened genome-wide in DepMap with published ranks, and all four were in the lysate of an AP-MS run against seven MLL fusions including the index one - two of them scored in it. None of those measurements appears in any of the literature records counted above. Anyone reading these zeros as 'never tested' would be wrong about all four. This repository 2026
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    The Skucha 2018 affinity purification against seven Strep-HA-tagged MLL fusions in Jurkat recovered CHD3, CHD4, GATAD2A, GATAD2B and MBD2 from the NuRD remodelling module and SIN3A, SUDS3, ARID4B and ING2 from the Sin3/HDAC complex somewhere in its seven-bait panel, while recovering MTA1, MTA2, MTA3, MBD3, CHD5, HDAC1, HDAC2, RBBP4, RBBP7, SIN3B, SAP30 and SAP18 in no bait column at all; the table retains only the ~291-293 most significant preys per bait, so those absences are rank truncations rather than zero spectral counts.

    What this evidence cannot show

    The claim that a protein was 'in the lysate' is an inference from the experiment's design - whole nuclear lysate from a human cell line in which all four genes are expressed - and not a measurement I made. What I measured is presence or absence in the published table. The host is JURKAT, a T-ALL line, and the bait is a doxycycline-induced single-copy transgene, so a partner requiring a B-lineage or infant proteome could not appear regardless of rank. The subunit lists are my curation, written out in the script.

    Evidence that argues against this (1)
    • The same table's negative control caps how much any absence from it can mean: CCT/TRiC, a canonical affinity-purification contaminant, is the strongest complex signal in the dataset at 8 of 9 subunits and p = 2.0e-10. A table whose top complex-level result is an artefact is not a table whose silence is strong evidence either. That argues for treating even these measured-not-reported absences as weak, and for actually re-searching PXD009338 rather than reasoning from the processed file. Skucha A et al. 2018
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    An epigenome-wide and kinome-wide CRISPR-Cas9 screen in KMT2A-rearranged ALL lines including two KMT2A::AFF1 lines reported exactly three validated vulnerabilities - ARID4B, MBD3 and BMPR2 - and tabulated no per-gene score, so for every other epigenetic gene in those libraries, MTA2 and GATAD2A and SIN3A included, a dependency measurement was generated and not reported.

    CRISPR-Cas9 Library Screening Identifies Novel Molecular Vulnerabilities in KMT2A-Rearranged Acute Lymphoblastic Leukemia. · Schneider P, Wander P, Arentsen-Peters STCJM, Vrenken KS, Rockx-Brouwer D, Adriaanse FRS, Hoeve V, Paassen I, Drost J, Pieters R, Stam RW · International Journal of Molecular Sciences 2023
    PMID 37686014DOI
    What this evidence cannot show

    THE INFERENCE THAT MTA2, GATAD2A AND SIN3A WERE IN THE LIBRARY IS NOT VERIFIED - the library composition is described as 'the entire human epigenome' and I did not obtain the gene list. The expected_absent_terms say so explicitly: those symbols do NOT appear in the paper, which is the point of the claim and also its limit. The MCP PubMed tool's rendering of this abstract deletes italicised gene symbols (QUEEN_BRIEF, cycle 46); PubMed itself does not - the raw efetch XML fetched 2026-10-03 contains them, and MBD3[tiab] AND ARID4B[tiab] AND BMPR2[tiab] returns PMID 37686014 alone. ARID4B, MBD3 and BMPR2 were read from the PMC full text and keyword list on 2026-09-24, and are present in the raw abstract too.. This is a DEPENDENCY instrument and cannot answer a binding question for any subunit.

    Evidence that argues against this (1)
    • A screen reporting three hits is not a screen that measured every gene well. Library coverage, sgRNA efficiency and dropout power vary per gene, and a gene with poor guides produces a non-result rather than a negative. So 'measured and not reported' may be too generous for some subunits here, and the honest position for any individual gene in this library is unknown-until-the-supplement-is-opened. Schneider P et al. 2023
Connected to