Oncofusion

start / graph / gene-rbbp4

Genestrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q41-prc2-arm

RBBP4 - the shared histone-binding subunit that every leukaemia line needs, and whose identical-length paralogue cannot replace it

also called RBBP4 · RBAP48 · RbAp48 · CAF-1 subunit C · chromatin assembly factor 1 subunit C · retinoblastoma-binding protein 4 · NURF55

RBBP4 is a small disc-shaped protein that grips histones and is shared by several different chromatin machines, Polycomb included. Every leukaemia cell line tested needs it badly - and its near-identical twin RBBP7 is present in larger amounts yet cannot stand in for it.

The same thing, for a scientist

RBBP4 (UniProt Q09028, 425 aa) is a WD40 histone-binding subunit shared by PRC2, NuRD, Sin3, CoREST and CAF-1. In mouse embryonic stem cells, *Rbbp4* mutation disrupts **PRC2 assembly** and H3K27me3 establishment on target chromatin and reshapes H3K27ac genome-wide, especially at putative enhancers [E1] - so its PRC2 role is assembly-level and not accessory. In this project's DepMap panel RBBP4 is the most depleted PRC2 gene by a wide margin, **index mean -1.486 against -1.462 in comparator leukaemia lines and -1.347 across all 1,208 lines** [E2]. **That is a pan-essential gene at the Chronos floor and carries no selectivity claim**: its difference of -0.024 (rank 6,113 of 17,728) is distance-to-floor, and the myeloid and lymphoid halves of the index arm are indistinguishable (-1.484 and -1.489). **The paralogue question is the interesting one, and for once it resolves cleanly.** This project's standing rule is that a paralogue split is only real if the silent paralogue is expressed. RBBP7 is not silent: it is detected in **7/7** panel lines at median **139.0 nTPM, higher than RBBP4's 77.5**, while its gene effect is **-0.013** [E2]. **An abundantly transcribed, 425-residue, 92%-similar paralogue does not rescue RBBP4 loss in any leukaemia line.** The developmental literature says the redundancy is real but window-specific: double knockdown of *Rbbp4* and *Rbbp7*, **but not either individually**, is lethal at the morula-to-blastocyst transition [E3], while *Rbbp4* loss alone is lethal slightly later at implantation [E4]. So 'RBBP4 and RBBP7 are redundant' is true in one developmental window and false in leukaemia lines, and this entry records which.

Go deeper
The paralogue control, run and passed

analyses/expression-control-for-paralogue-splits/ exists in this repository because two earlier findings assumed a low-scoring paralogue was present and not needed when the alternative - not transcribed - produces identical data. This beat ran the same control on the PRC2 pair 2:

geneHPA median nTPM (7 panel lines)lines detectedDepMap index meanpan-essentiality
RBBP477.57/7-1.486-1.347
RBBP7139.07/7-0.013-0.155

Per-line RBBP7 nTPM: SEM 226.4, RS4;11 120.5, MV4-11 137.3, THP-1 139.0, NOMO-1 433.3, MOLM-13 153.4, OCI-AML-2 115.9.

The split passes. RBBP7 is not merely expressed, it is the more abundant transcript in all seven lines, and it still does not buffer RBBP4 loss. This is the first paralogue split in this project where the expression control came back in favour of the finding rather than collapsing it.

What the control does not establish is protein level, complex occupancy, or whether the two paralogues partition between complexes (RBBP7 is the HAT1 type-B subunit; RBBP4 is the CAF-1 p48 subunit). A functional non-redundancy driven by complex partitioning is a different mechanism from one driven by intrinsic activity, and nothing here distinguishes them.

Floor compression: what this entry is careful NOT to claim

RBBP4's pan-essentiality is -1.347 across 1,208 DepMap lines. A gene depleted almost everywhere has very little dynamic range left in which selectivity could appear, so its index-versus-comparator difference of -0.024 is uninterpretable rather than small - the same caution this graph already records for HCFC1 and KAT8.

So:

  • Claimed: RBBP4 is a strong, essentially universal dependency of leukaemia cell lines, present in the PRC2 assembly step, and not buffered by RBBP7.
  • Not claimed: that RBBP4 is a KMT2A-selective dependency, that the difference between index and comparator means anything, or that RBBP4 is a therapeutic target in this disease. A gene every cell needs is not a selective target however it scores.

The literature agrees with the non-selectivity: (RBBP4 OR RBBP7) AND (KMT2A OR MLL) AND leukemia returns 2 PubMed records and title-checking finds neither on topic - an AML RNF5/HDAC-inhibitor paper (PMID 34518534) and a dendritic-cell miRNA paper (PMID 24699235) 5. Per Rule 64 the complex form was run and is larger but recovers no RBBP-specific leukaemia study.

Accession resolution - the case where the length check cannot help at all

gene_exact:RBBP4 AND organism_id:9606 AND reviewed:true returns exactly one reviewed human record, Q09028, primary gene name RBBP4, synonym RBAP48, 425 aa 6.

RBBP7 (Q16576) is also exactly 425 aa. Rule 65 prescribes asserting both accession and length so that verify-accessions.py can catch a swap that reading the protein name could not. For this pair the length check is blind: a swapped accession would pass any length comparison. The protein names are also nearly interchangeable in free text - Histone-binding protein RBBP4 / RBBP7, Retinoblastoma-binding protein p48 / p46, RBAP48 / RBAP46 - so the discriminator is the accession and nothing else.

This is a stronger version of the Rule 65 hazard than HDAC7/HDAC9, where 1011 aa against 952 aa at least gave a check. It is recorded here so a future beat does not assume the length assertion is doing work it cannot do.

Facts and figures
symbolRBBP4
uniprotQ09028
length AA425
uniprot primary gene nameRBBP4
uniprot resolution notegene_exact:RBBP4 AND organism_id:9606 AND reviewed:true returned exactly one row; primary gene name RBBP4 matched; length 425 asserted. WARNING: the paralogue RBBP7 (Q16576) is ALSO 425 aa, so the length check cannot discriminate this pair and the accession is the only safe identifier. Resolved 2026-09-20.
paraloguegene-rbbp7
complexprotein-prc2-complex
expression hpa nTPM median kmt2a r panel77.5
other complexes
[
  "NuRD",
  "Sin3",
  "CoREST",
  "CAF-1 (p48 subunit)",
  "NURF"
]
essentiality
{
  "depmap_kmt2a_r_mean": -1.4856,
  "depmap_other_leukaemia_mean": -1.4619,
  "depmap_difference": -0.0236,
  "depmap_rank_of_17728": 6113,
  "depmap_pan_essentiality_all_lines": -1.3471,
  "depmap_index_myeloid_mean": -1.4836,
  "depmap_index_lymphoid_mean": -1.4891,
  "n_index_lines": 11,
  "n_comparator_lines": 43,
  "source": "analyses/prc2-dependency-lineage-split/results.json",
  "interpretation": "pan-essential and floor-compressed; strong dependency, NO selectivity claim"
}
paralogue expression control
{
  "rule": "a paralogue split is only real if the silent paralogue is expressed (analyses/expression-control-for-paralogue-splits)",
  "rbbp7_median_nTPM": 139,
  "rbbp7_lines_detected": "7/7",
  "rbbp7_depmap_index_mean": -0.0131,
  "verdict": "PASSES - RBBP7 is more abundantly transcribed than RBBP4 in every panel line and still does not buffer RBBP4 loss, so the asymmetry is functional rather than an expression artefact"
}
depmap kmt2a selectivity
{
  "chronos_mean_kmt2a_rearranged": -1.4856,
  "chronos_mean_other_leukaemia": -1.4619,
  "difference": -0.0236,
  "pan_essentiality_all_lines": -1.3471,
  "selectivity_rank": 6113,
  "rank_denominator": 17728,
  "provenance": "Joined 2026-09-21 from analyses/depmap-kmt2a-selective-scan-full (generated 2026-09-19T10:12:06Z), the project's own genome-wide scan over 11 KMT2A-rearranged and 43 other leukaemia DepMap lines. NOT re-fetched and NOT independent of that scan: every caveat it carries applies here, including that the KMT2A-rearranged arm is a small number of lines drawn from nine individuals, that the rank denominator is 17,728 genes, and that a difference is not a dependency. Added because only 26 of 308 gene entries carried this number, which made any KMT2A-selectivity question answerable for 8% of the corpus.",
  "how_to_read": "A negative difference means MORE required in KMT2A-rearranged lines than in other leukaemia lines. A rank near 17,728 is ANTI-selective. Neither is evidence of a fusion-driven dependency on its own: check pan_essentiality_all_lines before calling anything selective, and note that a paralogue-redundant gene can score near zero for reasons that have nothing to do with this disease - see ana-the-paralogue-redundancy-excuse-is-not-tested-and-my-proxy-decided-the-answer."
}
How this could bite us back
  • RBBP4 is at the Chronos floor in 1,208 cell lines and is required in mouse early embryogenesis. Any agent that removes it removes it from every dividing cell, so it is an essential-function liability rather than a target, and its high score in this disease is a property of all cells rather than of the fusion.
  • RBBP4 is shared across PRC2, NuRD, Sin3, CoREST and CAF-1, so an RBBP4-directed intervention is simultaneously an H3K27 methylation intervention, a deacetylase intervention and a replication-coupled chromatin-assembly intervention. The bite-back is not one pathway but four, and the acetylation arm moves in the opposite direction from the methylation arm at the same residue.
  • RBBP7 will not rescue. In leukaemia lines the more abundant paralogue provides no buffering, so there is no compensatory reserve protecting normal haematopoiesis either.
Still open
  1. Is RBBP7 excluded from PRC2 in leukaemia cells, or present in PRC2 and functionally inert? Bulk RNA cannot distinguish these, and they imply different interventions: if RBBP7 is excluded, forcing its incorporation is a route to buffering normal tissue against an RBBP4-directed agent; if it is present and inert, it is not.
  2. Does the leukaemia RBBP4 dependency run through PRC2, through the HDAC1/2-containing complexes, or through CAF-1? An RBBP4 separation-of-function allele that binds one complex and not the others does not exist and would answer it - this is the same 'shared subunit' problem the graph already records for WDR5 and DPY30.
  3. Could the identical 425-residue length of RBBP4 and RBBP7 have already produced a silent swap somewhere in this project? No check in this repository can detect one, and the two earlier paralogue incidents (CSNK2A1/CSNK2A2, HDAC7/HDAC9) were both found only by a length or a direction looking wrong.
Every source behind this entry (6)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    CRISPR-Cas9 Rbbp4 mutant alleles in mouse embryonic stem cells disrupted Polycomb repressive complex 2 assembly and H3K27me3 establishment on target chromatin, altered H3K27 acetylation genome-wide - particularly at putative enhancers - and produced dramatic changes in transcriptional profiles closely tied to the deregulation of H3K27ac.

    RBBP4 dysfunction reshapes the genomic landscape of H3K27 methylation and acetylation and disrupts gene expression. · Mu W, Murcia NS, Smith KN, Menon DU, Yee D, Magnuson T · G3 (Bethesda) 2022
    PMID 35416979DOI
    What this evidence cannot show

    Abstract read from the eutils record; full text not read. Mouse embryonic stem cells, not haematopoietic and not leukaemic. The alleles are described as 'viable Rbbp4 mutant alleles' rather than nulls, so the phenotype is hypomorphic and the magnitude is not transferable. RBBP4 sits in HDAC1/2-containing complexes as well as PRC2, so the H3K27ac effect cannot be attributed to PRC2 loss and the two effects cannot be separated in this design.

    Evidence that argues against this (1)
    • Because RBBP4 is shared between PRC2 and the HDAC1/2-containing complexes, any RBBP4 perturbation changes methylation and acetylation at the same residue simultaneously. That makes it a poor reagent for asking what PRC2 alone does, and it means a leukaemia phenotype from RBBP4 loss cannot be attributed to Polycomb without a complex-specific separation-of-function allele. Zhan Y et al. 2024
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    In 11 KMT2A-rearranged DepMap leukaemia lines RBBP4 gene effect is -1.486 against -1.462 in 43 other leukaemia lines (difference -0.024, rank 6,113 of 17,728) with pan-essentiality -1.347 across 1,208 lines and no myeloid/lymphoid difference (-1.484 vs -1.489), while RBBP7 gene effect is -0.013 despite being transcribed in 7/7 Human Protein Atlas panel lines at a higher median (139.0 nTPM) than RBBP4 (77.5 nTPM).

    What this evidence cannot show

    RBBP4's pan-essentiality of -1.347 places it near the Chronos floor, so the index-versus-comparator difference is distance-to-floor and is NOT interpretable as selectivity. nTPM is bulk RNA from immortalised lines and is neither protein level nor complex occupancy; RBBP7 could be abundant and yet excluded from PRC2. Human Protein Atlas carries 7 of the 11 index lines. Raw effects, no z, no permutation null. DepMap release version not stated by the Open Targets API.

    Evidence that argues against this (1)
    • The developmental literature reaches the opposite conclusion about redundancy in a different window: double knockdown of Rbbp4 and Rbbp7, but NOT either individually, causes embryonic lethality at the morula-to-blastocyst transition, with compensatory behaviour explicitly demonstrated. Whether RBBP7 can substitute is therefore a property of the cell state, not of the proteins, and the leukaemia-line result should not be generalised. Xiao L et al. 2022
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Double knockdown of Rbbp4 and Rbbp7, but not knockdown of either individually, causes embryonic lethality during the morula-to-blastocyst transition in mice, with dysregulation of cell-cycle, lineage-development and transcription-regulation genes, cell-cycle block, disrupted lineage specification and a dramatic increase in H3.3 and H3K27ac abundance - demonstrating a compensatory relationship between the two paralogues in that window.

    PMID 34709113DOI
    What this evidence cannot show

    Abstract read from the eutils record; full text not read. Mouse preimplantation embryos with siRNA/knockdown rather than knockout, so residual protein remains and 'not individually' is a statement about a partial depletion. The complexes named are SIN3A, NuRD and CoREST - PRC2 is not named in the abstract - so this is evidence about the paralogue pair, not specifically about PRC2.

    Evidence that argues against this (1)
    • A separate mouse study reports that loss of RBBP4 ALONE is lethal, with mutant blastocysts failing to hatch or arresting after hatching and severe apoptosis and histone hyperacetylation - so in an overlapping developmental window the single loss is not compensated. The two reports differ in method (knockdown versus mutant allele) and in the exact stage examined, and this entry does not attempt to adjudicate between them. Miao X et al. 2020
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Rbbp4 mutant mouse embryos appear morphologically normal at the E3.5 blastocyst stage but cannot be recovered at E7.5, mutant blastocysts fail to hatch from the zona or arrest after hatching, apoptosis and histone acetylation are significantly increased, and epiblast and primitive endoderm lineages are compromised while trophoblast is properly specified - establishing RBBP4 as essential in early mammalian embryogenesis.

    PMID 32285100DOI
    What this evidence cannot show

    Abstract read from the eutils record; full text not read. Mouse embryogenesis, not haematopoiesis. This is a constitutive developmental requirement and, per the scope caution this project applies to gnomAD constraint, it is not a safety signal for partial reversible pharmacological inhibition in an adult or child - it says the gene matters in development.

    Evidence that argues against this (1)
    • The hyperacetylation phenotype points to the HDAC-containing complexes rather than to PRC2, so this essentiality is not evidence that RBBP4's Polycomb role is the essential one. Attributing an RBBP4 dependency in leukaemia to Polycomb would be the same error. Mu W et al. 2022
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    PubMed `(RBBP4 OR RBBP7) AND (KMT2A OR MLL) AND leukemia` returns 2 records and title-checking finds neither on topic: an AML RNF5/HDAC-inhibitor study (PMID 34518534) and a dendritic-cell miRNA study (PMID 24699235). Neither RBBP4 nor RBBP7 has been studied in KMT2A-rearranged leukaemia under these symbols.

    What this evidence cannot show

    The symbol set is narrow. RBAP48, RBAP46, RbAp48, RbAp46 and NURF55 are all in active use for these proteins and were NOT included in this query; this graph's own retired-symbol rule says a count keyed to current symbols can under-count tenfold. The complex form was run per Rule 64 and recovers the PRC2 leukaemia literature but no RBBP-specific study. This is an absence of published study under these symbols, not evidence that the proteins are irrelevant.

    Evidence that argues against this (1)
    • The graph's own histone-chaperone beat already scored RBBP4 in its lineage-split panel, so the gene is not absent from this project even though it is absent from the KMT2A literature - and that entry reached the same pan-essentiality conclusion independently. An absence in PubMed is not an absence in the project's own analyses, and both should be checked. Q37-histone-chaperones 2026
  6. E6established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    RBBP4 resolves to a single reviewed human UniProt record, Q09028 (RBBP4_HUMAN), primary gene name RBBP4, synonym RBAP48, 425 amino acids; its paralogue RBBP7 is Q16576 (RBBP7_HUMAN), primary gene name RBBP7, synonym RBAP46, also 425 amino acids.

    What this evidence cannot show

    Resolved 2026-09-20. **Both paralogues are 425 aa**, so the length assertion Rule 65 prescribes cannot detect a swap between them; only the accession can. Their protein-name strings are also near-interchangeable, which is precisely the Rule 65 failure mode where reading the name back does not save you.

    Evidence that argues against this (1)
    • This graph has twice logged a lookup returning the wrong paralogue of the same family with nothing about the record reading as an error (CSNK2A1/CSNK2A2 via a curated alias, HDAC7/HDAC9 via a synonym plus a colliding protein name). RBBP4/RBBP7 is the same shape with the discriminating field removed, and no automated check in this repository can catch a swap between them. UniProt 2026
Connected to