Oncofusion

start / graph / gene-arid2

Proteinmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q47-swi-snf

ARID2 (BAF200) - the PBAF subunit whose loss ACCELERATES MLL-AF9 leukaemia, and which is nonetheless required to maintain it

also called BAF200 · KIAA1557 · Zipzap/p200 · ARID2_HUMAN

ARID2 marks out the PBAF flavour of the SWI/SNF machine. In mice, deleting it early makes MLL-AF9 leukaemia worse and shortens survival - but deleting it once the leukaemia exists slows it down. The direction of the effect depends entirely on when you remove it.

The same thing, for a scientist

ARID2 (Q68CP9, 1,835 aa) is the PBAF-defining ARID protein [E1][E2]. Two independent mouse studies in the same MLL-AF9 model reach a coherent but stage-dependent answer: Baf200 loss **accelerates tumour burden and shortens survival** [E3], and conditional deletion shows Arid2 **enhances leukaemogenesis when deleted at initiation yet is required during maintenance** [E4]. In normal haematopoiesis Baf200 is required: Tie2-Cre deletion causes perinatal death from defective erythropoiesis and impaired fetal-liver HSC expansion, and fetal-liver and bone-marrow HSCs show impaired long-term reconstitution [E3]. DepMap gives it nothing: rank 8,209/17,728, difference -0.0072, no lineage-split survival [E5]. **This entry is the clearest statement in the beat that disease stage, not subunit identity, can decide the sign of a SWI/SNF result - and that PBAF is, on present evidence, a suppressor arm in this disease rather than a target.**

What it looks like
loading AF-Q68CP9-F1…
AlphaFold AF-Q68CP9-F1 ↗
colourshape

This is a prediction, not a measurement. No one has seen this protein in this shape. The colours say how much the software trusts itself, and that is worth reading before anything else — the orange stretches are places the model is effectively saying this part has no fixed shape.

very high (90+)confident (70–90)low (50–70)very low (<50)
I have never read one of these pictures before

A protein is one long chain of amino acids that folds up. The ribbon traces that chain from one end to the other; it is not a wire, it is a simplification of where the backbone runs. Drag to rotate, scroll to zoom.

Domains are the stretches of chain that do a particular job, and they are what the coloured blocks above the picture mark out. Click one and the view flies to it. In this project the important question about any domain is usually whether the fusion still has it — the labels say lost where it does not.

Anything in yellow sticks is a bound small molecule, which is usually the drug.

Go deeper
Why 'required for maintenance' and 'loss accelerates leukaemia' are both true

Bluemn et al. deleted Arid2 at two times in one model. At initiation, loss enhances leukaemogenesis. At maintenance, Arid2 is required 4. Liu et al., using an MLL-AF9 model on a Baf200 conditional background, report acceleration and shortened survival, which is the initiation-side result 3.

The practical consequence is a warning about every other entry in this beat: a knockdown performed in an established line measures maintenance, and a knockout crossed into a transformation model measures initiation. Nearly all the SWI/SNF leukaemia literature is the former. A gene can be a maintenance requirement and a transformation suppressor simultaneously, and reporting only one is how a field ends up with SWI/SNF described as both oncogenic and tumour-suppressive without either side being wrong.

Facts and figures
uniprotQ68CP9
gene symbolARID2
protein nameAT-rich interactive domain-containing protein 2 (BAF200)
length AA1835
alphafold IDAF-Q68CP9-F1
accession resolutionUniProt REST gene_exact:ARID2 AND organism_id:9606 AND reviewed:true, run 2026-09-20; one reviewed row; gene_primary and length read back from the same record.
complexPBAF-defining
classificationstage-dependent: transformation suppressor at initiation, maintenance requirement in mouse, no human cell-line dependency signal
normal haematopoiesisTie2-Cre deletion perinatally lethal from defective erythropoiesis; fetal-liver and BM HSCs show impaired long-term reconstitution
depmap
{
  "rank": 8209,
  "of": 17728,
  "difference": -0.0072,
  "mean_kmt2a_r": -0.5631,
  "mean_other_leukaemia": -0.5559,
  "pan_essentiality_all_lines": -0.4023,
  "lineage_split_z_myeloid": -0.22,
  "lineage_split_z_b_lineage": 1.09,
  "verdict": "no dependency signal"
}
stage dependence
{
  "initiation": "deletion ENHANCES leukaemogenesis (Bluemn); loss accelerates burden and shortens survival (Liu)",
  "maintenance": "required (Bluemn, mouse); not detected (DepMap, human lines)"
}
depmap kmt2a selectivity
{
  "chronos_mean_kmt2a_rearranged": -0.5631,
  "chronos_mean_other_leukaemia": -0.5559,
  "difference": -0.0072,
  "pan_essentiality_all_lines": -0.4023,
  "selectivity_rank": 8209,
  "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
  • Any pan-SWI/SNF or PBAF-destabilising intervention risks the initiation-stage phenotype: in mice, removing Arid2 makes MLL-AF9 leukaemia worse.
  • Baf200 is required for fetal-liver HSC expansion and its loss is perinatally lethal in mice - the single most direct developmental-toxicity signal in this beat, and directly relevant to an infant population.
  • A maintenance-only assay cannot detect an intervention that accelerates relapse, which is exactly the failure mode this gene's biology predicts.
Still open
  1. Is the initiation-versus-maintenance sign flip a property of ARID2, of PBAF, or of SWI/SNF generally? Nobody has run the same two-timepoint design on SMARCA4 in this model, and every SMARCA4 result in this beat is a maintenance result.
  2. Does human KMT2A-rearranged disease ever acquire ARID2 loss at relapse, as the initiation result would predict for a suppressor?
  3. Why does mouse maintenance require Arid2 while human lines show a difference of -0.007? Species, assay kinetics and fusion partner are all uncontrolled between the two.
Every source behind this entry (6)
  1. E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    ARID2 is UniProt Q68CP9, 1,835 aa, primary gene ARID2, synonyms BAF200 KIAA1557.

    UniProtKB Q68CP9 (ARID2_HUMAN) · UniProt Consortium · UniProtKB 2026
    What this evidence cannot show

    gene_exact:ARID2 AND organism_id:9606 AND reviewed:true, run 2026-09-20; one reviewed row; gene_primary and length read back from the fetched record.

    Evidence that argues against this (1)
    • ARID2 (1,835 aa) and ARID1A (2,285 aa) share the ARID-domain nomenclature and sit in different complexes; a reader who treats 'the ARID subunit' as one thing conflates cBAF with PBAF. UniProt Consortium 2026
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    ARID2 defines PBAF, distinguishing it from cBAF, which carries ARID1A or ARID1B.

    PMID 29374058DOI
    What this evidence cannot show

    Biochemical separation by density gradient and IP; the partition is treated as general.

    Evidence that argues against this (1)
    • PBAF also carries PBRM1, PHF10 and BRD7, so an ARID2 phenotype could reflect destabilisation of the whole PBAF assembly rather than an ARID2-specific function. Mashtalir N et al. 2018
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Baf200 (Arid2) is required for normal fetal and adult haematopoiesis - Tie2-Cre deletion causes perinatal death from defective erythropoiesis and impaired fetal-liver HSC expansion, and HSCs show impaired long-term reconstitution - while loss of Baf200 in a mouse MLL-AF9 leukaemogenesis model accelerates tumour burden and shortens host survival.

    PMID 29482581DOI
    What this evidence cannot show

    Mouse; three Cre drivers with transplantation; the leukaemia arm is one MLL-AF9 model with burden and survival as readouts. The paper frames this as a therapeutic opportunity to suppress leukaemia progression without interfering with normal haematopoiesis, which does not follow from its own data in the direction it implies - loss makes the leukaemia worse.

    Evidence that argues against this (1)
    • A conditional knockout crossed into a transformation model measures initiation; it cannot see a maintenance requirement, and Bluemn et al. report exactly that requirement in the same model. Bluemn T et al. 2022
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Arid2 has dual roles in MLL-AF9 leukaemogenesis: enhancing leukaemogenesis when deleted during initiation yet required during maintenance; Arid2 and Arid1b regulate largely distinct gene sets and converge oppositely on Etv4/Etv5.

    Differential roles of BAF and PBAF subunits, Arid1b and Arid2, in MLL-AF9 leukemogenesis. · Bluemn T, Schmitz J, Zheng Y, Burns R, et al. · Leukemia 2022
    PMID 35022500DOI
    What this evidence cannot show

    Mouse conditional knockouts with transcriptomic and genomic binding data; 'required during maintenance' is a growth phenotype, and the molecular mediator is not resolved beyond a set of dysregulated transcription factors.

    Evidence that argues against this (1)
    • DepMap, which is a pure maintenance assay in human lines, gives ARID2 a difference of -0.0072 and no lineage-split survival - so the maintenance requirement does not appear in human KMT2A-rearranged cells. Oncofusion knowledge graph et al. 2026
  5. E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    ARID2 ranks 8,209 of 17,728 with difference -0.0072 (KMT2A-r -0.5631, other leukaemia -0.5559, pan-essentiality -0.4023) and fails the lineage split (myeloid z -0.22, B-lineage +1.09).

    Genome-wide KMT2A-rearranged selectivity scan, 17,728 genes (DepMap Chronos via Open Targets) · Oncofusion knowledge graph, analyses/depmap-kmt2a-selective-scan-full/run.py · DepMap Chronos gene effect via Open Targets GraphQL 2026
    What this evidence cannot show

    11 index against 43 comparator lines. The B-lineage z of +1.09 - i.e. LESS dependent than the comparator arm - rests on five lines and should not be over-read as a protective effect.

    Evidence that argues against this (1)
    • A human cell-line maintenance assay cannot test the initiation-stage effect that both mouse papers report, so a null here does not contradict them. Bluemn T et al. 2022
  6. E6moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    `ARID2 AND (MLL OR KMT2A)` returns 9 PubMed records, of which the two fetched and read (PMIDs 35022500 and 29482581) are the two MLL-AF9 mouse studies cited here.

    What this evidence cannot show

    Bounded absence. Bounded to PubMed on 2026-09-20 over ARID2 and BAF200. Seven of the nine were judged on title only. The PBAF complex form was run and returns 14 records which were not individually title-checked, so a PBAF result in KMT2A-rearranged disease reachable only at complex level may have been missed.

    Evidence that argues against this (1)
    • 'PBAF' is a short token that also appears in unrelated contexts, so its 14-record count is an upper bound and the real on-beat number is likely lower - the opposite error to the one this caveat worries about. Michel BC et al. 2018
Connected to