ARID4B / SAP180 (Q4LE39) - the Sin3A-complex subunit that shares SIN3A's HDAC1-binding region, is in the MLL-fusion interactome, and is the second of only two epigenetic vulnerabilities a CRISPR screen found in KMT2A-rearranged ALL

This is real experimental data — atoms placed by crystallography or cryo-EM, deposited by the people who did the experiment.

• ARID4B is a 1,312-residue scaffold with one PDB entry and no ligand pocket. Like every other protein in this beat, the subunit that carries the selectivity is the one with no chemical matter, and the subunit with chemical matter - HDAC1 - is shared across five complexes.
• If ARID4B and HDAC1 compete for the same SIN3A surface, then removing ARID4B could INCREASE deacetylase occupancy on the scaffold rather than reduce it. A drug designed on the assumption that ARID4B loss disables Sin3 repression could do the opposite, and nothing has measured which.
• The entire index-fusion case for ARID4B rests on one 2023 paper in a lower-scrutiny venue whose ARID4B-specific data this entry did not read, against a DepMap rank of 11,283 over eleven KMT2A-rearranged lines. That is a thin base for a therapeutic hypothesis and the entry should not be cited as though it were thicker.

1. Do ARID4B and HDAC1 compete for SIN3A 687-829? A co-IP with a SIN3A 687-829 deletion in a t(4;11) line, probing both, would say whether an ARID4B dependency delivers deacetylase to the scaffold or strips it off - and the two answers imply opposite drug strategies.
2. What is ARID4B's actual effect size in SEM and ALL-PO? The screen reports it as validated and tabulates no per-gene score; the number exists in the authors' data.
3. Why do the two epigenetic hits of an unbiased screen in this disease belong to the two different HDAC1/2 corepressor complexes, and neither to the catalytic core? Either the scaffolds carry the selectivity, or the catalytic subunits are too essential to score - and a comparison of the HDAC1 and HDAC2 dropout profiles in the same screen would distinguish those.
4. Is ARID4B nuclear in a leukaemic cell? Its only experimentally coded UniProt localisation is cytoplasmic and its nuclear call is a domain-based rule, which is an odd position for a Sin3 subunit and is one immunofluorescence image away from resolution.

1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

   CRISPR-Cas9 knockout screens with sgRNA libraries against the whole human epigenome and kinome, run in KMT2A-rearranged ALL lines SEM (KMT2A::AFF1), ALL-PO (KMT2A::AFF1) and KOPN-8 (KMT2A::MLLT1) against KMT2A-wildtype lines 697 and NALM-6, identified ARID4B - described by the authors as a component of the SIN3A transcriptional corepressor complex - together with MBD3 and the receptor kinase BMPR2 as novel molecular vulnerabilities and attractive therapeutic targets in KMT2A-rearranged ALL.

   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 MCP TOOL'S RENDERING, NOT PUBMED, DELETES THE SYMBOLS: through that tool the abstract reads 'the epigenetic regulators and, as well as the receptor kinase', but the raw PubMed record contains ARID4B, MBD3 and BMPR2 (efetch XML, 2026-10-03), so an auditor searching title/abstract for ARID4B WILL find it; the symbols were read from the PMC full text and the record's keyword list on 2026-09-24. I READ THE VALIDATION DETAIL FOR MBD3 (GFP competition, SEM depletes, 697 does not) AND NOT FOR ARID4B: no ARID4B-specific effect size, replicate count, statistical test or line-by-line selectivity was verified here, and this item must not be cited for one. The 'component of the SIN3A corepressor complex' statement is the authors' background framing, not a measurement in this paper. Cell lines only - no primary sample, no xenograft, no normal haematopoietic comparator, therefore no therapeutic window. IJMS is a lower-scrutiny venue than the other primary sources in this beat.

   ▸Evidence that argues against this (1)

   • DepMap disagrees, over far more cell lines: ARID4B ranks 11,283 of 17,728 with a KMT2A-r-versus-other difference of +0.0164 - the wrong direction - and near-zero dependency in both arms (-0.147 and -0.164). My reading is that this is two nulls rather than an anti-dependency, but a reader weighting 11 KMT2A-rearranged lines against this paper's three would conclude ARID4B is not a KMT2A-r vulnerability. The lineage split is the likeliest reconciliation and it is untested. DepMap via this repository 2026
2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

   ARID4B is present in the Skucha 2018 960-protein pan-MLL-fusion AP-MS network (UniProt Q4LE39, negative control 0|0) in two of seven bait columns - MLL-AF9 p = 0.0138 with 10|7 spectra and MLL-EEN p = 0.0553 with 9|6 - and is absent from the MLL-AF4 (= KMT2A::AFF1) column; it is one of four Sin3/HDAC subunits recovered anywhere in the network, all four of which are Sin3-specific, while the four subunits shared with NuRD score 0 of 4.

   NuRD sub-module split and Sin3/HDAC scoring over Skucha 2018 Supplementary Table 5 (PMID 29777171), run 2026-09-24: analyses/q193-nurd-module-split/run.py, per-member accessions, bait columns, p-values, spectral counts and negative-control counts persisted in results.json. · This repository (script and results.json committed) 2026
   ▸What this evidence cannot show

   The accession in the table, Q4LE39, matches the one UniProt returns live for gene_exact:ARID4B AND organism_id:9606 AND reviewed:true on 2026-09-24. ONE OF THE TWO CALLS IS ABOVE 0.05: MLL-EEN at p = 0.0553 is retained only because the paper kept the top ~300 preys per bait rather than applying a threshold, so ARID4B's real support is one bait at p = 0.0138. Absence from MLL-AF4 is a rank truncation, not a zero spectral count - measured-not-reported, upgradeable to measured-not-analysed against PXD009338. Jurkat T-ALL host, doxycycline-induced Strep-HA bait. The 17-member Sin3 list is my curation.

   ▸Evidence that argues against this (1)

   • Two bait columns with one call above 0.05 is a marginal entry on a rank-truncated list, and the dataset's own negative control - CCT/TRiC, a canonical contaminant, at 8 of 9 subunits and p = 2.0e-10 - is orders of magnitude stronger than anything in the Sin3 row. A 1,312-residue nuclear scaffold is also a plausible affinity-purification passenger even with a 0|0 control, because a control of two replicates has limited power to exclude low-level background. Skucha A et al. 2018
3. E3weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.database

   In this project's genome-wide DepMap KMT2A-selectivity scan ARID4B ranks 11,283 of 17,728 with a difference of +0.0164, mean CHRONOS -0.1472 in KMT2A-rearranged lines against -0.1635 in other leukaemia lines and a pan-essentiality of -0.1548; all three values sit within noise of zero, so the rank orders two null effects rather than establishing an anti-dependency.

   DepMap KMT2A-selective dependency scan over 17,728 genes: analyses/depmap-kmt2a-selective-scan-full/results.json, generated 2026-09-19. ARID4B rank 11,283, difference +0.0164, mean_kmt2a_r -0.1472, mean_other -0.1635, pan_essentiality -0.1548. · DepMap via this repository 2026
   ▸What this evidence cannot show

   SAME SHAPE AS MBD3's RANK AND IT MUST BE READ THE SAME WAY: both arm means are near zero, so this is not an anti-dependency, it is an absence of effect in either arm of this panel. Standing caveats apply in full - the scan fixes both arms from the first gene with data and drops 5 of 16 eligible KMT2A-rearranged lines from every comparison; it reports a uniform n_all_lines of 1,208 when DepMap screens each gene in a gene-specific set; and it records no DepMap release, only a generation date. Re-read from results.json on 2026-09-24. Rank above the median is a 50% null by construction and percentile 63.6 is not far from it.

   ▸Evidence that argues against this (1)

   • An unbiased CRISPR screen in the relevant lineage reached the opposite conclusion and validated it, and DepMap's KMT2A-rearranged arm in this scan is 11 predominantly myeloid, adult-derived lines against the screen's three infant/paediatric B-ALL lines. If ARID4B's dependency is B-lineage- or infant-specific, DepMap's panel could not see it and this rank is uninformative for the disease in question. Schneider P et al. 2023
4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

   A bounded literature search on 2026-09-24 across four names - ARID4B, SAP180, BRCAA1 and the complex-level pairing with NuRD - returns zero records tying ARID4B to MLL, one to KMT2A (the CRISPR screen itself), three to leukemia and two to SIN3A, against a positive control of 55 for MEN1 with MLL.

   NCBI esearch via scripts/pubmed_safe.py on 2026-09-24. ARID4B[tiab] AND MLL[tiab] -> 0 (translation "ARID4B"[Title/Abstract] AND "MLL"[Title/Abstract]). ARID4B[tiab] AND KMT2A[tiab] -> 1 (37686014). ARID4B[tiab] AND leukemia[tiab] -> 3 (37686014, 26189108, 18728284). ARID4B[tiab] AND SIN3A[tiab] -> 2 (33907498, 33454011). ARID4B[tiab] AND NuRD[tiab] -> 0. SAP180[tiab] -> 2 (26609676, 12724404). BRCAA1[tiab] AND leukemia[tiab] -> 0. Positive control MEN1[tiab] AND MLL[tiab] -> 55. · NCBI PubMed E-utilities 2026
   ▸What this evidence cannot show

   NAMES TRIED: ARID4B, SAP180, BRCAA1. NOT tried as bare tokens: RBBP1L1 and RBP1L1 (low-yield synonyms). One tag per term throughout, querytranslation recorded, run through pubmed_safe so the zeros are real zeros rather than empty-translation failures. THE TWO ARID4B-plus-SIN3A AND THE TWO SAP180 RECORDS WERE NOT READ - they are counted, not assessed, and no claim in this entry rests on them. ARID4B[tiab] AND NuRD[tiab] returning 0 is the useful negative here: it supports treating ARID4B as Sin3-specific rather than shared.

   ▸Evidence that argues against this (1)

   • A near-empty literature is not evidence of unimportance, and in this corpus it has repeatedly been the opposite - several of this graph's best-shaped dependencies have zero or one record. But the converse risk applies to this entry specifically: with one KMT2A record, one marginal AP-MS bait and a null DepMap rank, almost everything stated here rests on a single 2023 paper in a lower-scrutiny venue whose ARID4B-specific data I did not read. This repository 2026

• part of →SIN3A (Q96ST3) - the scaffold of the OTHER HDAC1/2 corepressor complex, a 5' fusion partner in infant B-ALL and in myelodysplasia, and present in three MLL-fusion AP-MS columns but not the index fusion's

  SAP180: subunit of the SIN3A corepressor complex; annotated to SIN3A 687-829 alongside HDAC1

• depends on →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)

  validated CRISPR vulnerability in a KMT2A-r ALL panel containing two KMT2A::AFF1 lines

• co-occurs with →MBD3 (O95983) - the methyl-CpG-binding subunit that does NOT bind methylated DNA, and a validated KMT2A::AFF1-selective CRISPR vulnerability that is absent from every column of the pan-fusion interactome

  the two epigenetic hits of the same epigenome-wide CRISPR screen: one Sin3 subunit, one NuRD subunit

• binds →KMT2A::MLLT3 (MLL-AF9), t(9;11)(p21.3;q23.3)

  present in the MLL-AF9 column, p = 0.0138, 10|7 spectra, neg ctrl 0|0

• ← supportsThe MLL-fusion interactome recovers corepressor SCAFFOLDS and not their HDAC1/2 catalytic core: NuRD splits 5/7 remodelling arm versus 0/7 deacetylase arm (Fisher p = 0.021), and HDAC1, HDAC2, RBBP4 and RBBP7 are absent from all seven baits

  one of four Sin3-specific subunits recovered; the shared HDAC1/2 core is 0 of 4