Oncofusion

start / graph / protein-nurd-complex

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

The NuRD complex - two separable modules, three paralogue choices that change what it does, and a graph that carried its ATPase and its shared core but never the scaffolds that make it NuRD

also called NuRD · NURD · NRD complex · Mi-2/NuRD · nucleosome remodelling and deacetylase complex · MeCP1 (the MBD2-containing form)

NuRD is the only chromatin machine that both slides nucleosomes and strips acetyl marks off histones, and it is built from two halves that work on their own. Which version of three swappable parts it contains decides whether it reads methylated DNA, which enzyme it clamps and which gene it silences - so 'NuRD' on its own does not name one object.

The same thing, for a scientist

NuRD (originally NRD/NURD) was identified twice in 1998 as an endogenous human histone-deacetylase complex that also carries ATP-dependent nucleosome remodelling activity supplied by CHD3 and CHD4, and that remodelling is what permits deacetylation of oligonucleosomal histones the deacetylase cannot otherwise reach [E1][E2]. Its composition resolves into two biochemically separable, independently catalytically active sub-modules [E3]: a **deacetylase module** of MTA1/2/3 + HDAC1/2 + RBBP4/7, and a **remodelling module** of CHD3/4/5 + GATAD2A/B + MBD2/3. The HDAC1/2 + RBBP4/7 part of the first module is **not NuRD-specific at all**: it is a core shared with the Sin3-histone deacetylase complex, and in isolation its deacetylase activity is severely compromised, so the MTA scaffold is what makes it work [E4]. Three paralogue choices change the complex's identity rather than merely its parts: MTA1 versus MTA2 versus MTA3 (which transcription factors recruit it and which HDAC clamp forms); **MBD2 versus MBD3, which give mutually exclusive complexes with different biochemical and functional properties** - MBD2/NuRD reads 5-methyl-CpG and carries PRMT5/MEP50, MBD3/NuRD does not read 5mC [E4][E5]; and GATAD2A versus GATAD2B, paralogous p66 proteins that act in synergy but bind MBD2 with different affinities and through different numbers of interaction domains [E6]. GATAD2A is the bridge: its CR1 binds HDAC1, HDAC2, MBD2 and MTA2, and its CR2 binds CHD4 and histone tails, with tail acetylation blocking the latter [E6]. The measurement this entry exists to carry is that in the published pan-MLL-fusion AP-MS network the remodelling module scores **5 of 7** and the deacetylase module **0 of 7** (Fisher two-sided p = 0.0210), and in the index fusion's own column the whole complex is **one protein, CHD4** [E7].

Go deeper
Composition, by module
modulecanonical memberscatalytic activityin the MLL-fusion network
deacetylaseMTA1, MTA2, MTA3, HDAC1, HDAC2, RBBP4, RBBP7histone deacetylation (HDAC1/2)0 of 7
remodellingCHD3, CHD4, CHD5, GATAD2A, GATAD2B, MBD2, MBD3ATP-dependent nucleosome remodelling (CHD3/4/5)5 of 7 — CHD3, CHD4, GATAD2A, GATAD2B, MBD2

Which subunits are NuRD's own and which are borrowed. HDAC1, HDAC2, RBBP4 and RBBP7 are shared with the Sin3/HDAC complex 4 and, on later chemoproteomic evidence, with NCoR, CoREST and MiDAC as well - HDAC1/2 sit in at least five ELM-SANT-scaffolded complexes. So the subunits that make the complex NuRD rather than a generic HDAC assembly are exactly the ones this graph did not carry until this beat: MTA1/2/3, MBD2/3, GATAD2A/B. The ATPase (CHD4) and the shared core (HDAC1, HDAC2, RBBP4, RBBP7) all had entries.

Also variably present in purifications and not scored here: CDK2AP1/DOC-1 (a core subunit of both MBD2/NuRD and MBD3/NuRD 5), ZMYND8, and in the MBD2 form PRMT5 and MEP50.

The two modules are real objects, not a diagram convenience

Purification of endogenous Drosophila NuRD shows a stable core with CHD4 associating only transiently; systematic recombinant reconstitution of all subunit combinations shows sub-modules that retain enzymatic activity; and single-molecule imaging of CHD4 in live mouse embryonic stem cells with and without MBD3 shows how the two modules associate in vivo. The sub-modules are present in both nucleus and cytosol 3.

Three consequences for how this graph should reason.

  1. Finding one module in a pulldown is not finding a truncated complex. It may be finding a normal cellular species. This blunts the mechanistic reading of the module split in the MLL-fusion interactome 7 and the caveat is recorded there.
  2. Losing MBD3 uncouples two active machines rather than switching one off. That is why gene-mbd3's bite-back warns that an MBD3 degrader might release a free, active CHD4 remodeller.
  3. A cytosolic pool exists, so a nuclear-lysate AP-MS is not sampling a single compartment's worth of complex.
Paralogue choice, and what each choice changes

MBD2 versus MBD3 - the sharpest of the three. MBD2 and MBD3 assemble into mutually exclusive NuRD-like complexes with different biochemical and functional properties; PRMT5 and MEP50 are MBD2-specific and PRMT5 methylates MBD2's RG-rich N terminus and is recruited with it to CpG islands in a methylation-dependent manner, with H4R3 methylated at those loci 5. MBD2 binds methylated DNA and directs NuRD to it; MBD3 does not directly bind methylated DNA and instead mediates MTA2's association with the HDAC core 4. So the choice decides whether the complex is a methylation reader at all, and whether an arginine methyltransferase travels with it.

GATAD2A versus GATAD2B (p66-alpha versus p66-beta). Both bind MBD2 and MBD3 and both are potent repressors; p66-alpha binds MBD2 with higher affinity and through two interaction domains where p66-beta uses one, and the two act in synergy such that knocking down either impairs the other's repressive function 6. In the MLL-fusion network GATAD2B appears in five bait columns and GATAD2A in one, and in this project's DepMap scan GATAD2A ranks 866 against GATAD2B's 2,141 - an asymmetry nobody has explained 7.

MTA1 versus MTA2 versus MTA3. All three carry BAH-ELM2-SANT; the ELM2-SANT pair of MTA1 wraps HDAC1 on both sides of the active site with the BAH domain presenting nucleosomes, and inositol tetraphosphate at that interface is a genuine regulator of the enzyme 8. The paralogues differ in their C-terminal halves and in which transcription factors recruit them - BCL11B, for instance, binds MTA1 and MTA2 directly in T lymphocytes. MTA2 has zero experimental structures, so its clamp is known only by homology.

What this complex is doing in a KMT2A-fusion graph, stated conservatively

The honest summary of the evidence, and the order matters because the strength falls off steeply.

  1. CHD4 co-purifies with all seven MLL fusions tested including KMT2A::AFF1 (p = 0.0116, 7|9 spectra, negative control 2|2), and CHD4 is separately and independently implicated in KMT2A::AFF1 biology - its alteration accompanies myeloid lineage switch at relapse. That is gene-chd4's territory and it is the strongest link.
  2. MBD3 is a validated CRISPR vulnerability in KMT2A::AFF1 cell lines (SEM depletes, 697 does not), which is a dependency and not a binding result, and MBD3 is absent from all seven AP-MS columns.
  3. GATAD2A and GATAD2B are in the network but not in the index fusion's column; GATAD2A only under MLL-AF9.
  4. The entire deacetylase module - MTA1/2/3, HDAC1/2, RBBP4/7 - is absent from the whole network.
  5. The word "NuRD" appears zero times in the body of the source paper, a measurement already recorded in gene-chd4. The complex-level reading is this graph's construction from the paper's supplementary table, not the authors' claim.

So: there is no published demonstration that the NuRD complex, as a complex, binds any KMT2A fusion. There is one subunit that co-purifies with all seven, one that is a dependency in the index fusion's lineage, two scaffold subunits under other baits, and an absent catalytic module. An entry, figure or brief that says "NuRD is part of the KMT2A::AFF1 machine" is over-reading every one of those items.

Facts and figures
representative subunitQ14839 (CHD4) - named explicitly as REPRESENTATIVE and not as the complex's accession, per this graph's rule that a complex entry must never carry a bare uniprot field.
what makes it NuRD rather than a generic hdac assemblyMTA1/2/3, MBD2/3 and GATAD2A/B. HDAC1, HDAC2, RBBP4 and RBBP7 are shared with Sin3/HDAC (and, on chemoproteomic evidence, with NCoR, CoREST and MiDAC), and CHD3/CHD4 have documented NuRD-independent activity. Before this beat this graph carried the ATPase and all four shared-core subunits and none of the NuRD-specific scaffolds - which is precisely why a reader could not tell a NuRD readout from a Sin3 readout.
shown for fusionNO COMPLEX-LEVEL DEMONSTRATION EXISTS for any KMT2A fusion. Subunit by subunit: CHD4 co-purifies with all seven Skucha baits including MLL-AF4; GATAD2B with five and CHD3 with five, neither including MLL-AF4; GATAD2A and MBD2 with one each, neither MLL-AF4; MBD3, CHD5 and all seven deacetylase-module subunits with none. MBD3 is a validated CRISPR dependency in KMT2A::AFF1 cell lines without any binding evidence. The string 'NuRD' appears zero times in the source paper's body text.
epigenetic readoutTWO ACTIVITIES IN ONE COMPLEX: ATP-dependent nucleosome remodelling (CHD3/4/5) and histone deacetylation (HDAC1/2), with remodelling required for the deacetylase to reach oligonucleosomal substrates and deacetylation stimulated by ATP on nucleosomal templates. The MBD2 form adds 5-methyl-CpG reading and, via PRMT5/MEP50, H4R3 methylation at CpG islands. GATAD2A's histone-tail binding is BLOCKED by tail acetylation, so the complex's own chromatin grip strengthens after it has deacetylated. NONE of these readouts has been measured at a KMT2A fusion in any system.
epigenetic readout absence classnever-measured
localizationNucleus; and the sub-modules are present in the CYTOSOL as well [E3], which matters because every interactome claim in this beat comes from nuclear-lysate affinity purification.
subnuclear localizationNo compartment for the complex. Subunit-level: GATAD2A has a punctate, MBD2-dependent nuclear distribution that UniProt calls 'Nucleus speckle' on evidence that does not include a speckle marker (see gene-gatad2a); MBD3 and MTA2 have no assignment in any system.
absence classmeasured-not-reported
absence class noteFor the seven deacetylase-module subunits' per-bait spectral counts against the index fusion: measured-not-reported, because the AP-MS generated them and the top-~291 table discarded them; measured-not-analysed against the public PXD009338 raw files. For a complex-level experiment - reciprocal co-IP of an MTA or a GATAD2 with a KMT2A fusion, or size-fractionation asking which NuRD species carries the fusion: never-measured.
relationship to existing entriesNEW as a complex entry. gene-chd4 (NuRD's ATPase), gene-hdac1, gene-hdac2, gene-rbbp4 and gene-rbbp7 all existed; gene-mta2, gene-mbd3 and gene-gatad2a were written by this beat. This entry does NOT restate gene-chd4's Skucha row, its lineage-switch evidence or its therapeutic tension - it links to it. It answers gene-chd4's first open question ('Are other NuRD subunits present in the Skucha 960-protein network, and if so does the pan-fusion core contain a NuRD module rather than a lone ATPase?') with: YES, five of fourteen, and they are ALL in one module - so the pan-fusion core contains half a NuRD, and the index fusion's column contains a lone ATPase after all. Conforms to this graph's rule 69b: subunit-level accessions in data.complex_members, no bare uniprot field, one field explicitly named representative_subunit.
measurements
subunits_present0.000subunits_present: 0fisher_two_sided_p0.021fisher_two_sided_p: 0.021subunits_present_in_index…1.000subunits_present_in_index_bait: 1subunits_present5.000subunits_present: 5
complex members
{
  "deacetylase_module": {
    "MTA1": {
      "accession": "Q13330",
      "length_aa": 715,
      "role": "scaffold; ELM2-SANT clamps HDAC1, BAH presents nucleosomes",
      "in_skucha_960": false
    },
    "MTA2": {
      "accession": "O94776",
      "length_aa": 668,
      "role": "scaffold; modulates the HDAC1/2 core's activity",
      "in_skucha_960": false
    },
    "MTA3": {
      "accession": null,
      "length_aa": null,
      "role": "scaffold paralogue; accession NOT fetched in this beat and deliberately left null rather than guessed",
      "in_skucha_960": false
    },
    "HDAC1": {
      "accession": null,
      "length_aa": null,
      "role": "catalytic; SHARED with Sin3/HDAC, NCoR, CoREST, MiDAC. Accession not re-fetched here - gene-hdac1 exists and owns it.",
      "in_skucha_960": false
    },
    "HDAC2": {
      "accession": null,
      "length_aa": null,
      "role": "catalytic; SHARED. gene-hdac2 exists and owns the accession.",
      "in_skucha_960": false
    },
    "RBBP4": {
      "accession": null,
      "length_aa": null,
      "role": "histone chaperone; SHARED with Sin3/HDAC, PRC2, CAF-1. gene-rbbp4 exists.",
      "in_skucha_960": false
    },
    "RBBP7": {
      "accession": null,
      "length_aa": null,
      "role": "histone chaperone; SHARED. gene-rbbp7 exists.",
      "in_skucha_960": false
    }
  },
  "remodelling_module": {
    "CHD3": {
      "accession": "Q12873",
      "length_aa": null,
      "role": "ATPase paralogue",
      "in_skucha_960": true,
      "skucha_baits": 5,
      "in_index_bait": false
    },
    "CHD4": {
      "accession": "Q14839",
      "length_aa": 1912,
      "role": "ATPase; the pan-fusion co-purifier",
      "in_skucha_960": true,
      "skucha_baits": 7,
      "in_index_bait": true,
      "index_bait_p": 0.0116,
      "index_bait_spectra": "7|9",
      "neg_ctrl": "2|2"
    },
    "CHD5": {
      "accession": null,
      "length_aa": null,
      "role": "ATPase paralogue; accession not fetched, left null rather than guessed",
      "in_skucha_960": false
    },
    "GATAD2A": {
      "accession": "Q86YP4",
      "length_aa": 633,
      "role": "BRIDGE - CR1 binds HDAC1/HDAC2/MBD2/MTA2, CR2 binds CHD4 and histone tails",
      "in_skucha_960": true,
      "skucha_baits": 1,
      "in_index_bait": false,
      "bait": "MLL-AF9 only, p = 0.0361, 10|12"
    },
    "GATAD2B": {
      "accession": "Q8WXI9",
      "length_aa": 593,
      "role": "bridge paralogue",
      "in_skucha_960": true,
      "skucha_baits": 5,
      "in_index_bait": false
    },
    "MBD2": {
      "accession": "Q9UBB5",
      "length_aa": 411,
      "role": "reads 5-methyl-CpG and directs NuRD to methylated DNA; carries PRMT5/MEP50",
      "in_skucha_960": true,
      "skucha_baits": 1,
      "in_index_bait": false,
      "bait": "MLL-EEN only, p = 0.0827 - ABOVE 0.05"
    },
    "MBD3": {
      "accession": "O95983",
      "length_aa": 291,
      "role": "does NOT read 5mC; mediates MTA2's association with the core; the KMT2A::AFF1 CRISPR dependency",
      "in_skucha_960": false
    }
  }
}
paralogue choices
{
  "MTA1_vs_MTA2_vs_MTA3": "All carry BAH-ELM2-SANT. Differ in C-terminal halves and in recruiting transcription factors (e.g. BCL11B binds MTA1 and MTA2 directly in T lymphocytes). Only MTA1 has a structure with HDAC1; MTA2 has ZERO PDB entries. All three absent from the MLL-fusion network.",
  "MBD2_vs_MBD3": "MUTUALLY EXCLUSIVE COMPLEXES. MBD2/NuRD reads 5-methyl-CpG and carries PRMT5/MEP50 (H4R3 methylation at CpG islands); MBD3/NuRD does not read 5mC and instead mediates MTA2's attachment to the core. This is the choice that decides whether the complex is a DNA-methylation reader at all.",
  "GATAD2A_vs_GATAD2B": "Both bind MBD2 and MBD3; p66-alpha binds MBD2 at higher affinity through two interaction domains against p66-beta's one; they act in synergy so knockdown of either impairs the other. In the MLL-fusion network GATAD2B is in five bait columns and GATAD2A in one; in this project's DepMap scan GATAD2A ranks 866 and GATAD2B 2,141. The asymmetry is unexplained.",
  "CHD3_vs_CHD4_vs_CHD5": "CHD4 is in all seven bait columns and CHD3 in five (not the index one); CHD5 is absent. CHD4 is also the only subunit with independent KMT2A::AFF1 evidence, via lineage switching at relapse."
}
measurements — the numbers behind the chart
[
  {
    "quantity": "subunits_present",
    "subject": "NuRD remodelling module in the Skucha 2018 pan-MLL-fusion network",
    "value": 5,
    "units": "subunits",
    "n": 7,
    "universe": "960-protein network, union over seven MLL-fusion baits, top ~291-293 preys per bait",
    "null": 0.33,
    "note": "CHD3, CHD4, GATAD2A, GATAD2B, MBD2; null is 7 x 960 / 20431 against the proteome background"
  },
  {
    "quantity": "subunits_present",
    "subject": "NuRD deacetylase module in the Skucha 2018 pan-MLL-fusion network",
    "value": 0,
    "units": "subunits",
    "n": 7,
    "universe": "960-protein network, union over seven MLL-fusion baits, top ~291-293 preys per bait",
    "null": 0.33,
    "note": "MTA1, MTA2, MTA3, HDAC1, HDAC2, RBBP4, RBBP7 all absent"
  },
  {
    "quantity": "fisher_two_sided_p",
    "subject": "NuRD remodelling (5/7) versus deacetylase (0/7) module presence",
    "value": 0.021,
    "units": null,
    "n": 14,
    "universe": "the 14 canonical NuRD subunits, seven per module, scored in one dataset; background-free by construction",
    "null": 1,
    "note": "one-sided p = 0.0105. Reassigning MBD2/MBD3 to the deacetylase module weakens this to 3/5 versus 0/7, same direction."
  },
  {
    "quantity": "subunits_present_in_index_bait",
    "subject": "NuRD, all 14 subunits, in the MLL-AF4 (= KMT2A::AFF1) column",
    "value": 1,
    "units": "subunits",
    "n": 14,
    "universe": "the 291 proteins retained in the MLL-AF4 column",
    "null": null,
    "note": "CHD4 alone, p = 0.0116, 7|9 spectra, negative control 2|2"
  }
]
How this could bite us back
  • The two modules are independently active, so removing the bridge between them does not switch the complex off - it releases a free, active remodeller and a free, active deacetylase. Any degrader aimed at MBD3 or a GATAD2 has to be evaluated against that, and no experiment in this graph has measured which way it goes.
  • The only druggable part of this complex is HDAC1/2, and HDAC1/2 belong to at least five different corepressor complexes in every cell in the body. A NuRD-directed proposal that reaches for an HDAC inhibitor has no complex selectivity, and the module this graph's own data supports is the one WITHOUT the enzyme.
  • CHD4 disruption in MLL/AF4 models induces myeloid gene expression and accompanies immunotherapy-resistant myeloid relapse in human patients. NuRD is a repressive complex in a disease whose escape route is de-repression of an alternative lineage programme, so complex-directed inhibition carries a lineage-switch risk as a primary, not secondary, concern.
  • The complex's own grip on chromatin strengthens after deacetylation - GATAD2A's histone-tail binding is blocked by acetylation. So an HDAC inhibitor would EVICT the scaffold, and two agents aimed at the same complex would oppose each other.
  • There is no published demonstration that NuRD as a complex binds any KMT2A fusion, and the word does not appear in the source paper's body text. If this entry is ever cited for 'NuRD is in the pan-fusion core', it will have been misread: the pan-fusion core contains five of fourteen subunits across seven baits, and the index fusion's column contains one.
Still open
  1. Which NuRD species, if any, carries a KMT2A fusion? A size-fractionation of a t(4;11) nuclear extract followed by a fusion immunoblot across the fractions would say whether the fusion travels with a holo-complex, with a remodelling sub-module, or with free CHD4 - and no experiment in this graph currently distinguishes those three.
  2. Is the deacetylase module's absence from the interactome real or a filter artefact? The test is a matched-abundance, matched-promiscuity control set scored against the same table: if a set of equally abundant, equally shared nuclear proteins with no expected fusion relationship is also 0 of 7, the filter explains it.
  3. Which MBD does the fusion's chromatin carry? MBD2/NuRD and MBD3/NuRD are mutually exclusive, MBD2 is the one in the interactome and MBD3 is the one that is a dependency in KMT2A::AFF1 lines. A single MBD2 and MBD3 co-IP in SEM would tell the graph which complex it has been talking about.
  4. Does the GATAD2A/GATAD2B asymmetry in the bait columns survive a re-search of the raw data? Five columns versus one is a large difference for a synergistic paralogue pair and the public PXD009338 files can settle whether it is detectability.
  5. Does inositol tetraphosphate regulation give a handle nobody has tried? It is a bona fide regulator of class I HDAC complexes at the MTA1-HDAC1 interface, which is an allosteric site inside a complex rather than an active site shared by five - and nothing in this graph has looked at it.
Every source behind this entry (8)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    CHD3 and CHD4 supply a novel ATP-dependent nucleosome remodelling activity that is part of an endogenous human histone deacetylase complex - the nucleosome remodelling and deacetylating (NRD) complex - and that activity facilitates the deacetylation of oligonucleosomal histones, which histone deacetylases are otherwise unable to deacetylate in vitro.

    Chromatin deacetylation by an ATP-dependent nucleosome remodelling complex. · Tong JK, Hassig CA, Schnitzler GR, Kingston RE, Schreiber SL · Nature 1998
    PMID 9804427DOI
    What this evidence cannot show

    1998 biochemical purification from human cells, read from the abstract. Note the expected_absent_terms: this paper calls the complex NRD, not NuRD - the name in the literature comes from the companion paper - so a term-matching auditor should not expect 'NuRD' here. The complex is now known to be compositionally heterogeneous and this paper resolves no subtype. This entry is already cited for the same claim inside gene-chd4 and is repeated here because a complex entry must state its own founding composition.

    Evidence that argues against this (1)
    • In Drosophila, Mi-2 - the single fly orthologue of CHD3/CHD4 - has activity INDEPENDENT of the NuRD complex, shown during neuronal maturation. So the presence of a CHD3/CHD4-family ATPase does not establish that NuRD is present, which is the single most important caution for any entry in this graph that reads a CHD4 signal as a NuRD signal. Invertebrate and neuronal, cited only for the separability point. Aughey GN et al. 2023
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    NURD is a human complex containing both ATP-dependent nucleosome disruption activity and histone deacetylase activity, in which deacetylation is stimulated by ATP on nucleosomal templates - suggesting nucleosome disruption aids the deacetylase to access its substrates - and one of its subunits was identified as MTA1, a metastasis-associated protein with a region similar to the nuclear receptor corepressor N-CoR; antibodies against NURD partially relieve transcriptional repression by thyroid hormone receptor.

    NURD, a novel complex with both ATP-dependent chromatin-remodeling and histone deacetylase activities. · Xue Y, Wong J, Moreno GT, Young MK, Cote J, Wang W · Molecular Cell 1998
    PMID 9885572DOI
    What this evidence cannot show

    1998 biochemical purification from HeLa and Jurkat extracts, read from the abstract. THE MTA IDENTIFIED HERE IS MTA1, NOT MTA2 - MTA2 was found the following year - and the expected_absent_terms record that. The N-CoR similarity noted here is the ELM2-SANT region, which is exactly the shared feature that later chemoproteomics used to define the family of ELM-SANT-scaffolded HDAC complexes. Nothing about leukaemia or fusions.

    Evidence that argues against this (1)
    • The ATP-stimulation-of-deacetylation result, which is the functional argument for coupling the two modules, sits awkwardly with the later finding that the two sub-modules are independently catalytically active and that CHD4 associates only transiently with the core. If the deacetylase module works without the remodeller, then coupling is a regulated option rather than an architectural requirement - and an AP-MS that recovers only the remodelling arm is not recovering a broken complex. Zhang W et al. 2016
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    NuRD assembles from preformed, catalytically active sub-modules: purification of endogenous Drosophila NuRD shows a stable core of subunits with the chromatin remodeller CHD4 associating transiently; systematic recombinant production of all subunit combinations identifies stable sub-modules that retain enzymatic activity; single-molecule imaging of CHD4 in live mouse embryonic stem cells with and without MBD3 shows how the deacetylase and remodelling sub-modules associate in vivo; and the sub-modules are present in both the nucleus and the cytosol.

    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, Watson AA, Lando D, Lee SF, Palayret M, Kloet SL, Smits AH, Deery MJ, Vermeulen M, Hendrich B, Klenerman D, Schaffitzel C, Berger I, Laue ED · Journal of Molecular Biology 2016
    PMID 27117189DOI
    What this evidence cannot show

    THREE SYSTEMS, NONE HUMAN HAEMATOPOIETIC: endogenous purification in DROSOPHILA, reconstitution in recombinant MultiBac insect cells, imaging in MOUSE embryonic stem cells. Read from the abstract. The exact module boundaries are not enumerated in the abstract; the seven-and-seven split scored in E7 is this project's curation informed by this architecture, not a list this paper publishes. No fusion protein anywhere in it.

    Evidence that argues against this (1)
    • The module boundary this entry uses is not the only defensible one. MBD2 and MBD3 are assigned here to the remodelling module because they bridge GATAD2 to the CHD ATPase, but MBD3's documented job is to mediate MTA2's association with the HDAC core - a deacetylase-module function - so a reader could reasonably put MBD2/3 in the core instead. That reassignment changes the interactome score from 5-of-7-versus-0-of-7 to 3-of-5-versus-0-of-7, which is the same direction and a weaker test. Zhang Y et al. 1999
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    HDAC1, HDAC2 and the histone-binding proteins RbAp48 (RBBP4) and RbAp46 (RBBP7) form a core complex shared between NuRD and Sin3-histone deacetylase complexes whose deacetylase activity in isolation is severely compromised; MTA2 is a NuRD subunit that modulates the activity of that core; MBD3 mediates the association of MTA2 with it; MBD3 does not directly bind methylated DNA; and MBD2, which does, interacts with NuRD and directs it to methylated DNA.

    Analysis of the NuRD subunits reveals a histone deacetylase core complex and a connection with DNA methylation. · Zhang Y, Ng HH, Erdjument-Bromage H, Tempst P, Bird A, Reinberg D · Genes & Development 1999
    PMID 10444591DOI
    What this evidence cannot show

    1999 biochemistry from HeLa nuclear extract, read from the abstract. The single most-used citation in this beat and the only source for four separate structural facts, which concentrates risk: if any part of it has been superseded, several entries move together. 'Shared core' describes co-purifying composition, not one particle interconverting between complexes, and the paper gives no estimate of what fraction of cellular HDAC1/2 is in each complex.

    Evidence that argues against this (1)
    • The sharing is broader than this paper's two-complex framing. Chemoproteomic profiling of sixteen HDAC inhibitors against intact megadalton complexes found HDAC1/2 in NuRD, Sin3, NCoR, CoREST and a novel mitotic deacetylase complex, all scaffolded by ELM-SANT subunits, and showed they are pharmacologically distinguishable. So an HDAC1/2 observation identifies one of at least five machines, and NuRD-versus-Sin3 is itself a simplification. Bantscheff M et al. 2011
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    MBD2 and MBD3 assemble into mutually exclusive, distinct Mi-2/NuRD-like complexes - MBD2/NuRD and MBD3/NuRD - with different biochemical and functional properties; DOC-1 (CDK2AP1) is a core subunit of both; PRMT5 and its cofactor MEP50 are specific MBD2/NuRD interactors; PRMT5 stably associates with and methylates the RG-rich N terminus of MBD2; and PRMT5 and MBD2 are recruited to CpG islands in a methylation-dependent manner in vivo, where H4R3 is methylated.

    MBD2/NuRD and MBD3/NuRD, two distinct complexes with different biochemical and functional properties. · Le Guezennec X, Vermeulen M, Brinkman AB, Hoeijmakers WA, Cohen A, Lasonder E, Stunnenberg HG · Molecular and Cellular Biology 2006
    PMID 16428440DOI
    What this evidence cannot show

    Tagged-protein mass spectrometry in a human cell line not named in the abstract; read from the abstract. 'Mutually exclusive' describes the purified complexes recovered and is not proof that no particle contains both. This graph already carries gene-prmt5, so the MBD2-side consequence of the paralogue choice is visible in the corpus - but PRMT5's entry was not written about NuRD and should not be read as a NuRD entry.

    Evidence that argues against this (1)
    • Mutual exclusivity makes the MLL-fusion interactome data harder to read, not easier: MBD2 is in the network (MLL-EEN only, p = 0.0827, ABOVE 0.05) and MBD3 is entirely absent, yet MBD3 is the CRISPR dependency in KMT2A::AFF1 lines. If the two forms are separate complexes then the interactome signal and the dependency signal concern DIFFERENT complexes, and no entry in this graph may fuse them. Schneider P et al. 2023
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The paralogous p66 proteins GATAD2A and GATAD2B both bind MBD2 and MBD3 directly and are potent transcriptional repressors, with p66-alpha binding at higher affinity via two interaction domains where p66-beta uses one; they function in synergy, with knockdown of either impairing the other's repressive function; both bind the tails of all octamer histones in vitro and histone-tail ACETYLATION interferes with that binding; and in the human feature table p66-alpha's CR1 (144-178) mediates interaction with HDAC1, HDAC2, MBD2 and MTA2 while CR2 (340-480) is histone-tail-binding and mediates interaction with CHD4 and CDK2AP1 - making GATAD2A the physical bridge between the two modules.

    p66alpha and p66beta of the Mi-2/NuRD complex mediate MBD2 and histone interaction. · Brackertz M, Gong Z, Leers J, Renkawitz R · Nucleic Acids Research 2006
    PMID 16415179DOI
    What this evidence cannot show

    SPLIT SOURCING, stated so a reader can separate it: the synergy, MBD2-dependence, histone-tail binding and acetylation-sensitivity are from this paper's abstract; the two-versus-one interaction-domain count is from its companion PMID 12183469; and the CR1/CR2 interaction partners and the residue ranges 144-178 and 340-480 are from the HUMAN UniProt Q86YP4 feature table fetched 2026-09-24, with boundaries confirmed by string match against the downloaded 633-aa sequence. Cell lines are human and mouse and not haematopoietic. The histone-tail work is in vitro with recombinant tails.

    Evidence that argues against this (1)
    • If GATAD2A's chromatin binding requires DEACETYLATED histone tails, then a fusion-associated GATAD2A with no HDAC1/2 in the same complex - which is what the MLL-fusion interactome shows - would not be stably chromatin-bound, and the interactome signal would be a soluble-pool interaction rather than a chromatin one. The paralogue synergy cuts the same way: a GATAD2A-specific claim in a cell that also expresses GATAD2B is underdetermined, and in the interactome it is GATAD2B, not GATAD2A, that appears in five of seven baits. This repository 2026
  7. E7moderate 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 (CHD3, CHD4, GATAD2A, GATAD2B, MBD2) and the deacetylase module 0 of 7 (MTA1, MTA2, MTA3, HDAC1, HDAC2, RBBP4, RBBP7 all absent), Fisher two-sided p = 0.0210; in the MLL-AF4 (= KMT2A::AFF1) column the whole 14-subunit complex is represented by CHD4 alone; GATAD2B is in five bait columns and GATAD2A in one (MLL-AF9).

    What this evidence cannot show

    The module lists are my curation and MBD2/MBD3's assignment is the contestable boundary (see E3's counter-evidence). 'Absent' means outside the top ~291-293 preys retained per bait, not a zero spectral count - measured-not-reported, upgradeable to measured-not-analysed against the public PXD009338 raw files. The absent arm is also the ABUNDANT, PROMISCUOUS arm, which a bait-specificity filter removes first, so the split has a competing explanation I cannot exclude. Jurkat T-ALL host, doxycycline-induced Strep-HA bait.

    Evidence that argues against this (1)
    • The decisive objection is the dataset's own negative control: CCT/TRiC, a canonical affinity-purification contaminant, scores 8 of 9 subunits at p = 2.0e-10 with 8 of 8 in the MLL-AF4 column - far stronger than the remodelling module's 4.4e-06 and 1 of 7. On a table where the strongest complex signal is an artefact, complex-wide enrichment is not evidence of specific binding, and that caps what the module split can mean. Separately, the word 'NuRD' appears ZERO times in the source paper's body text, so the complex-level reading is this graph's construction and not the authors' claim. Skucha A et al. 2018
  8. E8established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    The crystal structure of HDAC1 bound to MTA1 from the NuRD complex shows the ELM2-SANT domains of MTA1 wrapping completely around HDAC1 and occupying both sides of the active site, with the adjacent BAH domain ideally positioned to recruit nucleosomes to the active site, and establishes inositol-tetraphosphate as a bona fide conserved regulator of class I HDAC complexes rather than a passive structural cofactor.

    Class I HDACs share a common mechanism of regulation by inositol phosphates. · Millard CJ, Watson PJ, Celardo I, Gordiyenko Y, Cowley SM, Robinson CV, Fairall L, Schwabe JWR · Molecular Cell 2013
    PMID 23791785DOI
    What this evidence cannot show

    MTA1 ONLY. MTA2 has ZERO PDB entries and MTA3 none cited here, so the clamp architecture of the paralogue this beat is actually about is a HOMOLOGY INFERENCE - the expected_absent_terms make that auditable. A domain-pair construct from a 715-residue protein is not the full scaffold on a nucleosome. Read from the abstract; resolution and construct boundaries not verified.

    Evidence that argues against this (1)
    • A structure existing is not a structure suitable for design, and paralogues with identical domain architecture routinely behave differently - this graph's own scans put EP300 at rank 108 against CREBBP at 16,950 and GATAD2A at 866 against GATAD2B at 2,141. So the MTA1 clamp should not be used to predict MTA2's behaviour, only to describe what an MTA clamp is. DepMap via this repository 2026
Connected to