Oncofusion

start / graph / hyp-mll-fusions-engage-the-nurd-remodelling-arm-without-its-deacetylase-arm

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q193-the-nurd-subunits

MLL fusions engage NuRD's remodelling arm without its deacetylase arm - and if so, every 'NuRD is repressive' inference in this graph is pointing the wrong way

also called the module-selectivity hypothesis · Q193 hypothesis

NuRD has two halves that work independently: a motor that slides nucleosomes and a set of scissors that removes acetyl marks. The published fishing experiment recovers the motor half and none of the scissors half, which raises the possibility that MLL fusions borrow the motor and leave the scissors behind - the opposite of treating NuRD as a silencing machine.

The same thing, for a scientist

In the Skucha 2018 pan-MLL-fusion AP-MS network the NuRD remodelling module scores 5 of 7 subunits and the deacetylase module 0 of 7 (Fisher two-sided p = 0.0210), and the four subunits shared with Sin3/HDAC score 0 of 4 [E1]. The hypothesis is that this is a property of the association rather than of the assay: that MLL fusions - or the chromatin state they create - recruit or co-occupy with CHD3/CHD4-GATAD2-MBD2 sub-modules that are not carrying MTA-HDAC1/2-RBBP. It is plausible because the sub-modules are genuine, independently catalytically active cellular species present in both nucleus and cytosol [E2], because CHD4 associates with the NuRD core only transiently [E2], and because an MLL fusion is an activator whose programme would be opposed by a deacetylase. **Prior probability 0.25, and the low prior is deliberate:** the competing explanation is mundane and cannot currently be excluded - the absent arm is the more abundant and more promiscuous arm, which is exactly what a bait-specificity filter on a top-300-per-bait list removes first [E1]. The hypothesis is worth writing down because it is cheap to attack: one re-search of public raw mass-spectrometry files distinguishes the two explanations before any wet work.

Go deeper
What would follow if it were true
  1. The repressive reading of CHD4 in the pan-fusion core is wrong. gene-chd4 already flags the tension - a repressive remodeller inside a gain-of-function activator's interactome - and resolves it as unsettled. Module selectivity would resolve it: the fusion would be recruiting remodelling activity, not repression.
  2. HDAC inhibition would be the wrong intervention for any NuRD-directed hypothesis in this disease, because the deacetylase arm would not be the one in contact.
  3. The MBD3 dependency changes meaning. MBD3 couples the two arms; if the fusion engages an uncoupled remodelling module, then MBD3 loss in SEM would not be disabling a fusion-associated complex, and its mechanism would lie elsewhere.
  4. It generalises beyond NuRD. The Sin3 hits in the same dataset are also scaffold-only, with the shared HDAC1/2 core at 0 of 4 - so the hypothesis in its broadest form is that this interactome contains corepressor scaffolds stripped of catalysis.
Why the prior is only 0.25

Three things hold it down, and they are stated here rather than in a caveat because they are most of the reason this is a hypothesis and not a finding.

  • The abundance confound is unexcluded and points the same way as the hypothesis. HDAC1/2 and RBBP4/7 are among the most abundant and most promiscuous nuclear proteins; a model that ranks by bait specificity penalises exactly that profile. Every observation supporting the hypothesis is equally well explained by it.
  • The dataset's negative control is its strongest signal. CCT/TRiC, a canonical affinity-purification contaminant, scores 8 of 9 subunits at p = 2.0e-10 - orders of magnitude above the remodelling module's 4.4e-06 - so complex-level enrichment on this table is demonstrably not evidence of specific binding 1.
  • The index fusion's own column contains one NuRD subunit. The 5-of-7 is a union over seven baits; MLL-AF4 contributes CHD4 alone. The hypothesis is therefore about MLL fusions as a class, and is nearly unconstrained for KMT2A::AFF1 specifically.

A fourth consideration cuts the other way and is why the prior is not lower: the sub-modules are real, they are independently active, and CHD4's association with the core is transient 2. Recovering a remodelling sub-module without a deacetylase module is a normal cellular species, not a broken complex - which makes the observation easy to believe even if it turns out to say nothing about fusions.

Facts and figures
statementMLL fusions, or the chromatin state they create, associate preferentially with NuRD's remodelling sub-module (CHD3/CHD4 + GATAD2A/GATAD2B + MBD2) and not with its deacetylase sub-module (MTA1/2/3 + HDAC1/2 + RBBP4/7), so the NuRD species in the MLL-fusion interactome carries nucleosome-remodelling activity without histone-deacetylase activity.
origindata-pattern
prior probability0.25
what would make it boring even if trueIf module-selective association is simply what every AP-MS of every CHD4-containing complex looks like - because the sub-modules are separable and CHD4's core association is transient - then the hypothesis is true of NuRD generally and says nothing about MLL fusions. The third falsifier above, and the fourth prediction, are designed to catch that: the claim only earns its keep if the asymmetry is LARGER at a fusion than at a matched non-fusion bait.
predictions
[
  "A re-search of the public PXD009338 raw files for MTA1, MTA2, MTA3, HDAC1, HDAC2, RBBP4 and RBBP7 peptides in the MLL-AF4 and MLL-AF9 runs returns spectral counts that are ZERO or at negative-control level, rather than substantial counts that the top-300 rank filter discarded.",
  "A CHD4 immunoprecipitation from a t(4;11) B-ALL line (SEM or RS4;11) recovers GATAD2A and GATAD2B at readily detectable levels and recovers MTA2 and HDAC1 at markedly lower stoichiometry than a CHD4 IP from a KMT2A-wildtype B-ALL line such as 697 or NALM-6.",
  "H3K27ac at KMT2A::AFF1-bound promoters does not fall when CHD4 is degraded, because the deacetylase arm is not present to be lost - whereas nucleosome positioning or accessibility at those promoters does change.",
  "The same module asymmetry appears in any other MLL-fusion interactome dataset generated in a different host, i.e. it is not a property of Jurkat."
]
experiments
[
  "exp-q193-scaffold-directed-co-ip-of-nurd-modules-in-a-t411-line"
]
analyses
[
  "ana-the-mll-fusion-interactome-recovers-corepressor-scaffolds-without-their-hdac1-2-catalytic-core",
  "ana-q193-planned-re-search-of-pxd009338-for-the-nurd-deacetylase-module"
]
falsifiers
[
  "SCOREABLE FORM: In a re-search of the PXD009338 raw files, MTA2 or HDAC1 peptides in the MLL-AF4 pulldown at a spectral count within 2-fold of CHD4's 7|9 - i.e. roughly 4 or more spectra per replicate - with negative-control counts at or below CHD4's 2|2. That result would show the deacetylase module was present and merely rank-truncated, and the hypothesis would be REFUTED with the artefact explanation confirmed.",
  "SCOREABLE FORM: In a CHD4 co-IP from SEM (human t(4;11) KMT2A::AFF1 B-ALL), recovery of MTA2 and HDAC1 at a band intensity ratio to CHD4 within 2-fold of that seen in a CHD4 co-IP from 697 (KMT2A-wildtype B-ALL), run side by side on the same blot. Equal module stoichiometry in the two lineages would show that no fusion-associated module selection is occurring.",
  "SCOREABLE FORM: A matched-abundance, matched-complex-promiscuity control set of 7 nuclear proteins with no expected MLL-fusion relationship, scored against the same Skucha table, returning 0 of 7. That would show the filter alone produces this result and the module split carries no information.",
  "SCOREABLE FORM: H3K27ac at the 4,646-promoter set this project already uses falling by more than 1.5-fold at KMT2A::AFF1-bound promoters within 24 h of CHD4 degradation in a t(4;11) line. A deacetylase-dependent acetylation change at fusion-bound loci would place the deacetylase arm in the complex after all."
]
measurements
[
  {
    "quantity": "prior_probability",
    "subject": "MLL fusions associate with NuRD's remodelling module and not its deacetylase module",
    "value": 0.25,
    "units": null,
    "n": null,
    "universe": "subjective prior set by this beat, held low because the abundance/promiscuity artefact explains every supporting observation equally well",
    "null": null,
    "note": "the observation it rests on is Fisher two-sided p = 0.0210 on 5/7 versus 0/7, in a dataset whose strongest complex signal is a known contaminant at p = 2.0e-10"
  }
]
How this could bite us back
  • If this hypothesis is right, then every therapeutic argument in this graph that reaches for an HDAC inhibitor on NuRD grounds is aimed at the arm that is not there - and the arm that is there has no chemical matter at all.
  • If it is wrong in the artefact direction, the graph will have learned that its pan-fusion complex-level reasoning is filtered by protein abundance, which would require re-reading every complex-level row in the corpus, not only the NuRD one.
  • The hypothesis is attractive precisely because it tidies an existing awkwardness - a repressive complex in an activator's interactome. That is a reason to distrust it: a hypothesis that resolves a tension the graph already found uncomfortable is the kind this project has been wrong about before.
Still open
  1. Is there any independent MLL-fusion interactome dataset, in a non-Jurkat host, in which the same fourteen NuRD subunits can be scored? A second dataset is the difference between a hypothesis about MLL fusions and a hypothesis about one experiment.
  2. Does the same module asymmetry hold for BAF/SWI-SNF, which the earlier corpus analysis scored at 6 of 17 without splitting it by module? If several multi-module complexes show the same pattern, the filter explanation strengthens sharply.
  3. If the remodelling module really is the engaged one, what is it doing at fusion-bound loci - opening chromatin for elongation, or positioning nucleosomes at promoters? Those are different experiments and neither has been done in a KMT2A-rearranged cell.
Every source behind this entry (3)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    In the Skucha 2018 960-protein pan-MLL-fusion AP-MS network the NuRD remodelling module scores 5 of 7 subunits and the deacetylase module 0 of 7 (Fisher two-sided p = 0.0210, background-free), the four subunits shared with Sin3/HDAC score 0 of 4, the Sin3-specific subunits score 4 of 13, and the CCT/TRiC contaminant control scores 8 of 9 at p = 2.0e-10.

    What this evidence cannot show

    The module lists are my curation and the MBD2/MBD3 assignment is the contestable boundary; reassigning them to the deacetylase module gives 3/5 versus 0/7, same direction, weaker test. 'Absent' means outside the top ~291-293 preys per bait, not a zero spectral count. Jurkat T-ALL host, doxycycline-induced Strep-HA bait, nuclear lysate - so co-purification is not evidence of a direct contact for any recovered subunit either.

    Evidence that argues against this (1)
    • THE COMPETING EXPLANATION, WHICH I CANNOT EXCLUDE AND WHICH WOULD ACCOUNT FOR EVERY OBSERVATION HERE: the absent arm is the abundant, promiscuous arm. HDAC1/2 belong to at least five ELM-SANT-scaffolded corepressor complexes and RBBP4/7 to four, so a bait-specificity model removes them first regardless of the biology. The hypothesis and the artefact make identical predictions about this dataset, which is why the test has to be a re-search of the raw files rather than more analysis of the processed table. Bantscheff M et al. 2011
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    NuRD's deacetylase and remodelling sub-modules are preformed, independently catalytically active species present in both nucleus and cytosol, and the chromatin remodeller CHD4 associates with the stable core only transiently - established by purification of endogenous Drosophila NuRD, systematic recombinant reconstitution, and single-molecule imaging of CHD4 in live mouse embryonic stem cells with and without MBD3.

    The Nucleosome Remodeling and Deacetylase Complex NuRD Is Built from Preformed Catalytically Active Sub-modules. · Zhang W, Aubert A, Gomez de Segura JM, Karuppasamy M, Basu S, Murthy AS, Diamante A, Drury TA, Balmer J, Cramard J, et al. · Journal of Molecular Biology 2016
    PMID 27117189DOI
    What this evidence cannot show

    Drosophila, recombinant insect cells and mouse ES cells - no human haematopoietic system and no fusion protein anywhere in it. Read from the abstract. It licenses the CLAIM THAT MODULE-SELECTIVE ASSOCIATION IS POSSIBLE, which is what a hypothesis needs, and establishes nothing about MLL fusions.

    Evidence that argues against this (1)
    • THIS PAPER IS ALSO THE HYPOTHESIS'S BIGGEST PROBLEM. If sub-modules are ordinary, abundant, independently active cellular species and CHD4's core association is transient, then finding a remodelling module without a deacetylase module in ANY pulldown is unremarkable and carries no information about what the bait was doing. The observation would then be a property of NuRD, not of MLL fusions - and the hypothesis would be true in a trivial and uninteresting sense. Zhang W et al. 2016
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    CHD4 co-purifies with all seven Strep-HA-tagged MLL fusions in Jurkat, with MLL-AF4 (= KMT2A::AFF1) at combined p = 0.0116 and 7|9 spectral counts against a negative control of 2|2, and the string 'NuRD' appears zero times in the body of the source paper, so the paper itself makes no complex-level claim.

    MLL-fusion-driven leukemia requires SETD2 to safeguard genomic integrity. · Skucha A, Ebner J, Schmoellerl J, Roth M, Eder T, Cesar-Razquin A, Stukalov A, Vittori S, Muhar M, Lu B, Aichinger M, Jude J, Mueller AC, Gyorffy B, Vakoc CR, Valent P, Bennett KL, Zuber J, Superti-Furga G, Grebien F · Nature Communications 2018
    PMID 29777171DOI
    What this evidence cannot show

    THE EXPECTED_ABSENT_TERM IS THE POINT: 'NuRD' does not appear in this paper's body text - a measurement recorded in gene-chd4 by a previous beat, which I did not re-run - so any complex-level reading of this dataset is this graph's construction. The CHD4 numbers are from Supplementary Table 5, not the body text, hence support_location 'supplement'. CHD4's negative control of 2|2 is non-zero, making its MLL-AF4 call roughly 3.5-fold over background and dependent on the generalised linear model rather than on absence from control.

    Evidence that argues against this (1)
    • CHD4 has documented NuRD-INDEPENDENT activity: in Drosophila, Mi-2, the single orthologue of CHD3/CHD4, represses ectopic gene expression during neuronal maturation independently of NuRD. So a CHD4 co-purification is not necessarily a NuRD co-purification at all, and the hypothesis's central observation could be about a free ATPase rather than about a module. Aughey GN et al. 2023
Connected to