Oncofusion

start / graph / concept-hdac1-2-membership-is-shared-so-an-hdac-signal-names-no-complex

Explainerstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q193-the-nurd-subunits

HDAC1 and HDAC2 belong to at least five different corepressor complexes, so an HDAC1/2 observation names no complex - and this graph carried both enzymes without the scaffolds that would tell them apart

also called NuRD versus Sin3/HDAC · the shared HDAC1/2 core · why an HDAC readout is ambiguous

HDAC1 and HDAC2 are the erasers that rub acetyl marks off histones, and several completely different machines use the same two erasers. So finding HDAC1 somewhere tells you an eraser is present - it does not tell you which machine brought it, and the machines silence different genes and respond differently to drugs.

The same thing, for a scientist

The NuRD and Sin3 histone-deacetylase complexes share a four-protein core - HDAC1, HDAC2, RBBP4 (RbAp48) and RBBP7 (RbAp46) - whose deacetylase activity in isolation is severely compromised; what makes each complex what it is, and what restores activity, is the scaffold around that core [E1]. Later chemoproteomic profiling of sixteen HDAC inhibitors against intact megadalton complexes rather than purified catalytic subunits found HDAC1/2 in **NuRD, Sin3, NCoR, CoREST and a then-novel mitotic deacetylase complex (MiDAC)**, all scaffolded by ELM-SANT-domain subunits, and showed the complexes are **pharmacologically distinguishable** - aminobenzamides bind the HDAC-NCoR complex more strongly than the HDAC-Sin3 complex - concluding that HDAC-inhibitor selectivity must be evaluated in the context of complexes and not catalytic subunits [E2]. This graph's exposure to that ambiguity was structural rather than accidental: it carried `gene-hdac1`, `gene-hdac2`, `gene-rbbp4` and `gene-rbbp7` - all four shared subunits - plus `gene-chd4` described as "the NuRD ATPase", and **none** of the subunits that distinguish the complexes: no MTA, no MBD, no GATAD2, no SIN3A. HDAC1 is named in 60 node files. The distinguishing measurement this beat ran makes the point concretely: in the pan-MLL-fusion AP-MS network the shared HDAC1/2 + RBBP4/7 core is **0 of 4** while Sin3-specific subunits score 4 of 17 and NuRD remodelling-module subunits 5 of 7 [E3] - so in that dataset the scaffolds are informative and the shared enzymes carry no information at all.

Go deeper
The concrete difference: two complexes, one core
NuRDSin3/HDAC
scaffoldMTA1/MTA2/MTA3 — BAH + ELM2 + SANTSIN3A or SIN3B — PAH1-3 + HID
remodelling ATPaseCHD3/CHD4/CHD5none
deacetylaseHDAC1, HDAC2HDAC1, HDAC2
histone chaperoneRBBP4, RBBP7RBBP4, RBBP7
DNA-modification readerMBD2 (5mC); MBD3, which does not read 5mCnone
adaptorsGATAD2A/GATAD2B (p66-alpha/beta), CDK2AP1SAP18, SAP30, SUDS3/SDS3, ARID4A/ARID4B, ING1/ING2
how it is delivered to DNAmethyl-CpG reading and nucleosome remodellingsequence-specific repressors docking the PAH domains

The operational rule. An observation of HDAC1 or HDAC2 - a co-IP band, a mass-spectrometry row, a ChIP peak, an inhibitor response - is consistent with at least five complexes. To name one you need a scaffold: an MTA or a GATAD2 or an MBD for NuRD; SIN3A or SIN3B or SUDS3 or SAP30 for Sin3. The same holds for RBBP4 and RBBP7, which are in PRC2 and CAF-1 as well, and which this graph already carries in a PRC2 folder.

Why a shared core is a trap specifically for this project

This graph's recurring failure mode is the inference that complex membership transfers a property. An existing entry records it in its own bite_back: adopting "complex membership transfers a chromatin readout to the scaffold" as an annotation standard "would credit large parts of the corpus on the epigenetics dimension without any new measurement."

A shared core makes that inference worse in both directions.

  • Forward: from "HDAC1 is here" to "NuRD is here" is a leap across a four-way ambiguity. HDAC1 is named in 60 node files in this corpus.
  • Backward: from "NuRD is here" to "deacetylation is happening here" fails too, because the isolated shared core has severely compromised activity 1 and because the two NuRD sub-modules are separable and independently active - so a NuRD subunit can be present with no deacetylase module attached 4.

And the measurement this beat ran shows both failures would have fired on real data: in the pan-MLL-fusion network the scaffolds are present and the shared enzymes are not 3.

What follows for a therapeutic argument

Do not treat an HDAC inhibitor as a complex-directed agent unless the proposal says how complex selectivity would be achieved and demonstrated. The chemoproteomic evidence that complexes can be discriminated is a demonstration in cell extracts against an immobilised ligand 2; no clinical HDAC inhibitor is marketed on complex selectivity, and no HDAC inhibitor trial in this graph is designed around one. The destructive half of that paper - that subunit-level selectivity claims are meaningless - is the part safe to rely on.

The constructive corollary is where a real opportunity sits: the scaffolds are what differ, so the scaffolds are where selectivity would have to come from - and they are precisely the proteins with no chemical matter. MTA2 has zero PDB entries, SIN3A has zero PDB entries, and the only published molecule anywhere in this beat is a peptide that disrupts the p66-alpha-MBD2 coiled coil in an erythroid system.

Facts and figures
the operational ruleTo name a complex from an HDAC1/2 observation you must have observed a SCAFFOLD: an MTA, an MBD or a GATAD2 for NuRD; SIN3A, SIN3B, SUDS3, SAP30 or an ARID4 for Sin3. RBBP4 and RBBP7 do not count - they are in PRC2 and CAF-1 as well. CHD3/CHD4 do not fully count either, because a CHD3/CHD4-family ATPase has documented NuRD-independent activity.
relationship to existing entriesNEW, and it is the entry the beat's dispatch asked for. It does not restate gene-hdac1 or gene-hdac2, which own their own biology; it states what those entries cannot, namely that their subject is shared. It is the concept counterpart of protein-nurd-complex and gene-sin3a, which carry the two complexes as objects. It generalises the caution already recorded in mechanism-q181's bite_back - that complex membership must not transfer a chromatin readout to a scaffold - into the specific case where the SHARED subunit is the one being observed.
shared subunits
[
  "HDAC1",
  "HDAC2",
  "RBBP4",
  "RBBP7"
]
complexes containing hdac1 2
[
  "NuRD",
  "Sin3/HDAC",
  "NCoR/SMRT",
  "CoREST",
  "MiDAC"
]
distinguishing subunits
{
  "NuRD": [
    "MTA1",
    "MTA2",
    "MTA3",
    "MBD2",
    "MBD3",
    "GATAD2A",
    "GATAD2B",
    "CHD3",
    "CHD4",
    "CHD5",
    "CDK2AP1"
  ],
  "Sin3/HDAC": [
    "SIN3A",
    "SIN3B",
    "SUDS3",
    "SAP30",
    "SAP18",
    "ARID4A",
    "ARID4B",
    "ING1",
    "ING2",
    "BRMS1",
    "BRMS1L",
    "PHF12",
    "MORF4L1"
  ]
}
corpus exposure before this beat
{
  "shared_subunits_with_entries": "ALL FOUR - gene-hdac1, gene-hdac2, gene-rbbp4, gene-rbbp7",
  "nurd_distinguishing_subunits_with_entries": "NONE. gene-chd4 existed and is the ATPase, which is NuRD-family but has documented NuRD-INDEPENDENT activity in Drosophila.",
  "sin3_distinguishing_subunits_with_entries": "NONE. gene-ing2 existed but was written as a PHD-finger reader in a writers-and-readers beat, not as a Sin3 subunit.",
  "word_boundary_naming_counts_2026_09_24": "HDAC1 named in 60 node files, HDAC2 in 22, RBBP4 in 17, RBBP7 in 14, CHD4 in 38 - against MTA2 in 1, MBD3 in 1, GATAD2A in 4, SIN3A in 7, all of them incidental. Counted with scripts/beat-inventory.py.",
  "why_this_matters": "Every one of those 60 HDAC1 mentions is, on its own, consistent with five complexes. The graph had maximal exposure to the ambiguity and zero means of resolving it."
}
measurements
[
  {
    "quantity": "subunits_present",
    "subject": "the four subunits shared by NuRD and Sin3/HDAC (HDAC1, HDAC2, RBBP4, RBBP7) in the Skucha pan-MLL-fusion network",
    "value": 0,
    "units": "subunits",
    "n": 4,
    "universe": "960-protein network, union over seven MLL-fusion baits, top ~291-293 preys per bait",
    "null": 0.19,
    "note": "null is 4 x 960 / 20431 against the proteome background; complex-specific scaffolds in the same dataset score 5 of 7 (NuRD remodelling) and 4 of 13 (Sin3-specific)"
  },
  {
    "quantity": "node_files_naming_the_symbol",
    "subject": "HDAC1 versus the NuRD- and Sin3-distinguishing scaffolds in this corpus before this beat",
    "value": 60,
    "units": "node files (word-boundary matched)",
    "n": 3086,
    "universe": "all node files in data/graph/nodes as of 2026-09-24, counted with scripts/beat-inventory.py",
    "null": null,
    "note": "against MTA2 in 1, MBD3 in 1, GATAD2A in 4 and SIN3A in 7 - all incidental. HDAC2 22, RBBP4 17, RBBP7 14, CHD4 38."
  }
]
How this could bite us back
  • Every HDAC-inhibitor proposal in this graph inherits this ambiguity: the drug hits a catalytic subunit shared by at least five corepressor complexes in every cell in the body, so 'targeting NuRD with an HDAC inhibitor' is not a mechanism, it is a hope.
  • The subunits that WOULD give selectivity - MTA2, SIN3A, the MBDs, the GATAD2s - have no chemical matter and, for MTA2 and SIN3A, no experimental structures at all. Selectivity and tractability point in opposite directions here.
  • Reading an HDAC1 signal as NuRD would let this graph credit itself on the epigenetics dimension without any new measurement - the exact error an existing entry's bite_back warns against. This concept exists to make that error visible, and it will only work if entries that mention HDAC1 are actually checked against it.
Still open
  1. How many of this corpus's 60 HDAC1 mentions actually name a complex? A pass over them asking 'is a scaffold subunit named in the same entry?' would turn this concept into a measurement, and would produce a list of entries that need qualifying.
  2. Which HDAC1/2 complex, if any, is engaged at KMT2A::AFF1 target genes? No experiment in this graph distinguishes them at any locus in any KMT2A-rearranged cell, and a scaffold-directed ChIP - an MTA and a SIN3A in the same t(4;11) line - would be the first.
  3. Does complex-selective HDAC inhibition do anything in this disease? The chemoproteomic work says complexes are pharmacologically distinguishable and nobody has used that in a KMT2A-rearranged system; an aminobenzamide-versus-hydroxamate comparison in a t(4;11) line is a cheap first look.
  4. Is MiDAC relevant here? It was novel when discovered and is the least-characterised of the five; this graph has no entry for it and no beat has asked whether it appears in any MLL-fusion dataset.
Every source behind this entry (4)
  1. E1established 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 the NuRD and Sin3-histone deacetylase complexes, and the histone deacetylase activity of that core complex in isolation is severely compromised; the NuRD-specific subunit MTA2 modulates its enzymatic activity.

    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. 'Shared' describes co-purifying polypeptide composition, not one physical particle interconverting between complexes, and the paper gives no estimate of what fraction of cellular HDAC1/2 sits in each. The expected_absent_terms record that this paper names only two complexes - NCoR, CoREST and MiDAC are later additions - so a term-matching auditor should not expect them here.

    Evidence that argues against this (1)
    • This paper's two-complex framing is itself too narrow, which strengthens the concept rather than weakening it: HDAC1/2 are now known to sit in at least five ELM-SANT-scaffolded complexes. So a reader who took 'shared between NuRD and Sin3' as an exhaustive list would still misattribute an HDAC1 signal, just with a 1-in-2 error rate instead of 1-in-5. Bantscheff M et al. 2011
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    A combination of affinity capture and quantitative mass spectrometry profiling sixteen HDAC inhibitors against multiple HDAC complexes scaffolded by ELM-SANT domain subunits - including NuRD, Sin3, NCoR, CoREST and a novel mitotic deacetylase complex (MiDAC) - showed inhibitors cluster by target profile, with aminobenzamides binding the HDAC-NCoR complex more strongly than the HDAC-Sin3 complex, and concluded that HDAC inhibitor selectivity should be evaluated in the context of HDAC complexes and not purified catalytic subunits.

    Chemoproteomics profiling of HDAC inhibitors reveals selective targeting of HDAC complexes. · Bantscheff M, Hopf C, Savitski MM, Dittmann A, Grandi P, Michon AM, Schlegl J, Abraham Y, Becher I, Bergamini G, Boesche M, Delling M, Dumpelfeld B, Eberhard D, Huthmacher C, Mathieson T, Poeckel D, Reader V, Strunk K, Sweetman G, Kruse U, Neubauer G, Ramsden NG, Drewes G · Nature Biotechnology 2011
    PMID 21258344DOI
    What this evidence cannot show

    Affinity-matrix chemoproteomics in cell extracts, 2011; read from the abstract. Cell systems are not named in the abstract and none is stated to be leukaemic - the expected_absent_terms record that nothing here concerns KMT2A, MLL or leukaemia and that this item must not be cited for anything about this disease. An immobilised-ligand pulldown reports what binds a matrix, which is not target engagement in a cell.

    Evidence that argues against this (1)
    • That complexes are distinguishable in an extract does not mean any approved HDAC inhibitor achieves complex selectivity at a tolerated dose in a patient, and none is marketed on that basis. The constructive half of this paper therefore remains a demonstration rather than a therapeutic route; only the destructive half - that catalytic-subunit selectivity claims are uninformative - is safe to build on, and this concept entry relies only on that half. Bantscheff M et al. 2011
  3. E3moderate 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 four subunits shared between NuRD and Sin3/HDAC - HDAC1, HDAC2, RBBP4, RBBP7 - score 0 of 4 across all seven bait columns, while Sin3-specific subunits score 4 of 17 (SIN3A, SUDS3, ARID4B, ING2) and the NuRD remodelling module scores 5 of 7; so in this dataset the complex-distinguishing scaffolds carry all of the signal and the shared catalytic core carries none.

    What this evidence cannot show

    '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 files. AND THE CAVEAT CUTS BOTH WAYS FOR THIS CONCEPT: a bait-specificity filter removes abundant promiscuous proteins first, and 'promiscuous' is exactly what the shared core IS, so the 0 of 4 is as consistent with the filter working as with the biology. That does not weaken the concept - a subunit whose promiscuity makes it statistically invisible is a subunit that names no complex either way.

    Evidence that argues against this (1)
    • This dataset's own negative control caps what any of its complex-level rows can mean: CCT/TRiC, a canonical affinity-purification contaminant, is the STRONGEST complex signal in it at 8 of 9 subunits, p = 2.0e-10. So the contrast between scaffold rows and shared-core rows is a contrast between two kinds of weak evidence, and the concept's argument rests on the published biochemistry in E1 and E2, not on this measurement. Skucha A et al. 2018
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The backward inference - from a NuRD subunit's presence to deacetylation happening - also fails, because NuRD assembles from preformed sub-modules that are independently catalytically active and present in both nucleus and cytosol, with CHD4 associating only transiently with the stable core.

    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, et al. · Journal of Molecular Biology 2016
    PMID 27117189DOI
    What this evidence cannot show

    Drosophila purification, recombinant reconstitution and mouse ES-cell imaging - none of it human haematopoietic and none of it involving a fusion protein. Read from the abstract. The expected_absent_terms note that this paper is about NuRD only and says nothing about Sin3, so it supports the backward-inference half of this concept and not the shared-core half.

    Evidence that argues against this (1)
    • Cut the other way, this paper also weakens the concept's practical bite: if sub-modules are normal, abundant cellular species, then a pulldown recovering only part of a complex is unremarkable, and an analyst has no way to tell a biologically meaningful partial complex from an ordinary one. The concept's advice therefore reduces to 'name the scaffold you saw', which is a discipline about reporting rather than a route to identifying complexes. Zhang W et al. 2016
Connected to