Oncofusion

start / graph / hyp-ncbaf-not-canonical-baf-is-the-kmt2a-dependency

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q47-swi-snf

The SWI/SNF dependency in KMT2A-rearranged leukaemia belongs to ncBAF, not to canonical BAF - and the drug in the clinic is aimed at the wrong module

also called ncBAF hypothesis · BRD9 over BRG1

All the classic experiments on SWI/SNF in MLL-rearranged leukaemia were done before anyone knew the machine comes in three flavours. When you sort the modern dependency data by flavour, only the third flavour - the one containing BRD9 - shows a signal specific to KMT2A-rearranged cells. If that is right, the drug now in trials targets the wrong part.

The same thing, for a scientist

**Statement.** The KMT2A-rearrangement-selective component of the mSWI/SNF dependency resides in **ncBAF** (BRD9, BICRA, SS18, with SMARCD1 as the shared BAF60), while the requirement for canonical BAF and PBAF subunits is a general leukaemia-maintenance requirement shared with BCR-ABL, RUNX1::RUNX1T1 and T-lineage disease. **Origin: data-pattern plus literature-gap.** Four of 28 SWI/SNF panel genes survive a lineage split; three are ncBAF [E1]. Genome-wide, BRD9 39, SMARCD1 59, SS18 224, BICRA 517 against SMARCA4 16,333, ARID1A 12,593, SMARCB1 14,024, ACTL6A 17,410 [E2]. Every functional MLL-rearranged SWI/SNF paper predates the 2018 definition of ncBAF and perturbed shared or cBAF subunits, so no published experiment could have made this distinction [E3][E4]. Shi et al. explicitly report the canonical-subunit requirement extending to non-MLL leukaemias [E5], and Cruickshank et al. report the same for T-ALL lines [E6]. **Prior probability: 0.3.** It is a re-reading of existing data, not a new observation, and the largest single alternative explanation - paralogue buffering hiding a canonical-BAF dependency from single-gene CRISPR - is unexcluded and applies to every cBAF slot. **Why it matters.** FHD-286 inhibits the shared ATPase, which is in all three complexes; a BRD9 degrader hits ncBAF alone. If the hypothesis is right, the selective agent exists (BRD9 degraders are published and one is in the clinic in synovial sarcoma) and has never been tested in KMT2A-rearranged leukaemia with the rearrangement as the selection criterion.

Go deeper
Predictions
  1. A BRD9 degrader will show a wider therapeutic index than FHD-286 in KMT2A-rearranged models, and specifically will retain activity in KMT2A-rearranged B-lineage lines where FHD-286 is reported inactive 7.
  2. A SMARCA4 + SMARCA2 double knockout will produce a large dependency in KMT2A-rearranged lines with no selectivity against other leukaemias - i.e. it will confirm the ATPase module matters and confirm it is not selective.
  3. BRD9 or BICRA CUT&RUN in a KMT2A-rearranged line will show ncBAF occupancy at a set of loci distinct from the fusion's own targets, consistent with the E1-E5 MYC-enhancer geometry rather than with HOXA cluster occupancy 8.
  4. In primary KMT2A-rearranged blasts, ncBAF subunit expression will not be elevated - the dependency is a wiring property, not an abundance one - so a transcriptomic biomarker will fail.

Falsifiers.

  • A BRD9 degrader with no effect in KMT2A-rearranged lines at concentrations active in synovial sarcoma kills the hypothesis outright.
  • A combinatorial CRISPR screen showing a strong KMT2A-selective dependency on the SMARCA4+SMARCA2 pair, or on ARID1A+ARID1B, relocates the dependency to canonical BAF and kills it.
  • Demonstration that BRD9, BICRA, SS18 and SMARCD1 are all depleted in KMT2A-rearranged lines because of a shared upstream cause unrelated to complex identity - for example all four being MYC-dependent - would make the module reading an artefact.
  • An expression control showing SMARCD1 and BICRA simply are the paralogues expressed in the index lines would remove two of the four survivors.
Facts and figures
statementThe KMT2A-rearrangement-selective part of the mSWI/SNF dependency is in ncBAF (BRD9, BICRA, SS18, SMARCD1); the canonical BAF and PBAF requirement is a general leukaemia maintenance requirement.
origindata-pattern
prior probability0.3
largest unexcluded alternativeparalogue buffering hides a canonical-BAF dependency from every single-gene CRISPR knockout in this panel
predictions
[
  "a BRD9 degrader retains activity in KMT2A-rearranged B-lineage lines where FHD-286 is reported inactive",
  "SMARCA4+SMARCA2 double knockout gives a large but unselective dependency",
  "BRD9/BICRA CUT&RUN occupies loci distinct from the fusion's own targets",
  "ncBAF subunit expression is not elevated in KMT2A-rearranged blasts, so no transcriptomic biomarker will work"
]
experiments
[
  "exp-ncbaf-versus-cbaf-in-kmt2a-rearranged-lines"
]
analyses
[
  "ana-swi-snf-lineage-split-dependency-scan",
  "ana-swi-snf-enumeration-and-bounded-absences"
]
falsifiers
[
  "a BRD9 degrader inactive in KMT2A-rearranged lines at concentrations active in synovial sarcoma",
  "a combinatorial CRISPR screen showing strong KMT2A-selective dependency on SMARCA4+SMARCA2 or ARID1A+ARID1B",
  "evidence that all four survivors are depleted through a shared upstream cause unrelated to complex identity",
  "an expression control showing SMARCD1 and BICRA are simply the paralogues expressed in the index lines"
]
How this could bite us back
  • If this is right, a programme that follows the clinical-stage agent (FHD-286) is targeting the unselective module, and its toxicity profile is that of a gene required for normal development while its selectivity is not.
  • If this is wrong because of paralogue buffering, then acting on it redirects effort away from the module the existing drug hits - the more expensive error of the two.
  • BRD9 degraders inherit the differentiation-induction liability and, in a disease with an ncBAF-independent menin axis, would still need combining.
Still open
  1. Which complex is at the E1-E5 MYC enhancers? Brg1 ChIP cannot say, and a BICRA or ARID1A complex-resolved experiment would decide the hypothesis in one figure.
  2. Is there any KMT2A-rearranged patient sample with ncBAF subunit loss or amplification?
  3. Does menin inhibition change ncBAF occupancy? If the ncBAF arm is fusion-independent it should not, which would make it a ready-made escape route and a combination rationale at once.
  4. BRD9::NUTM1 is a comparator this graph does not have. Both records that contaminate the ncBAF query describe it in INFANT B-ALL LACKING KMT2A REARRANGEMENT - the KMT2A-wt group of the same cohorts this project studies - and it carries a FAVOURABLE prognosis: 2-year EFS 80% (95% CI 52-100%) against 36-37% for KMT2A-r in the same series (PMID 40000924). NUTM1 has no entry here and BRD9::NUTM1 is named in exactly one entry, a collaboration record. Same disease, same age, opposite outcome, and it is PI dimension 5.
Every source behind this entry (8)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    Of 28 SWI/SNF panel genes scored with the KMT2A-rearranged arm split by lineage, only SMARCD1, BRD9, BICRA and SS18 are depleted in both halves; three of the four are ncBAF-defining or ncBAF-shared and none is cBAF-defining or PBAF-defining.

    SWI/SNF panel lineage-split DepMap dependency scan, 28 genes · Oncofusion knowledge graph, analyses/swi-snf-lineage-split/run.py · DepMap Chronos gene effect via Open Targets GraphQL 2026
    What this evidence cannot show

    Eight myeloid and five B-lineage index lines, none infant; thresholds inherited rather than derived; single-gene CRISPR is blind to paralogue-buffered dependencies, which biases against every canonical-BAF slot in the panel.

    Evidence that argues against this (1)
    • The bias against canonical BAF is systematic and large enough to produce this entire result on its own; the hypothesis is stated at prior 0.3 for exactly this reason. Mashtalir N et al. 2018
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    Genome-wide over the same lines, ncBAF genes rank BRD9 39, SMARCD1 59, SS18 224 and BICRA 517 of 17,728, while canonical subunits rank SMARCA4 16,333, ARID1A 12,593, SMARCB1 14,024 and ACTL6A 17,410, the last three with positive differences.

    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

    The two scans share underlying data, so their agreement is not independent. ncBAF genes have pan-essentiality at or above zero, which inflates a z-derived rank relative to raw effect.

    Evidence that argues against this (1)
    • SMARCA4's KMT2A-r mean of -0.4469 is a genuine dependency; a positive difference means unselective, not unnecessary, and an unselective dependency can still be the best drug target. Fiskus W et al. 2024
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    ncBAF was defined in 2018; before that no experiment could attribute a leukaemia phenotype to one of the three complexes.

    PMID 29374058DOI
    What this evidence cannot show

    A dating argument, not a search; a post-2018 complex-resolved leukaemia study could exist and was not found by `ncBAF AND leukemia` title-checking, which this beat did not complete.

    Evidence that argues against this (1)
    • BRD9's AML dependency was published in 2016 and was already an ncBAF result in substance, so the field had the data before the vocabulary and the dating argument is softer than it looks. Hohmann AF et al. 2016
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    ncBAF localises uniquely to CTCF sites and promoters and is a synthetic-lethal target where cBAF is perturbed, functioning distinctly from fusion-oncoprotein-bound complexes.

    A non-canonical SWI/SNF complex is a synthetic lethal target in cancers driven by BAF complex perturbation. · Michel BC, D'Avino AR, Cassel SH, Mashtalir N, et al. · Nat Cell Biol 2018
    PMID 30397315DOI
    What this evidence cannot show

    Synovial sarcoma and rhabdoid tumour; the conditionality on cBAF perturbation is a problem for transferring the logic to leukaemia, which has no known cBAF lesion.

    Evidence that argues against this (1)
    • If the ncBAF vulnerability is strictly conditional on cBAF perturbation, the KMT2A-rearranged BRD9 dependency must have a different cause and this hypothesis's mechanistic frame is wrong even if its dependency observation is right. Chatterjee SS et al. 2018
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The canonical-subunit requirement in MLL-AF9 cells extends to BCR-ABL B-ALL, KASUMI-1, JURKAT, REH and K-562, and the authors state it is not restricted to myeloid leukaemia or MLL fusion subtypes.

    Role of SWI/SNF in acute leukemia maintenance and enhancer-mediated Myc regulation. · Shi J, Whyte WA, Zepeda-Mendoza CJ, Milazzo JP, et al. · Genes Dev 2013
    PMID 24285714DOI
    What this evidence cannot show

    Supplementary sensitivity panels, qualitative. This claim comes from the PMC full text (PMC3877755, read 2026-09-20), not from the abstract - the quoted sentence does not appear in the abstract at all - so it scores low against the abstract in scripts/verify-claims.py by construction. That low score is expected and is not drift.

    Evidence that argues against this (1)
    • HEL and CMK were much less affected, so the canonical requirement is not universal either, and 'general leukaemia dependency' is itself an approximation. Shi J et al. 2013
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Jurkat and T-ALL1 T-lineage leukaemia lines also require full SMARCA4 expression for unperturbed proliferation.

    SWI/SNF Subunits SMARCA4, SMARCD2 and DPF2 Collaborate in MLL-Rearranged Leukaemia Maintenance. · Cruickshank VA, Sroczynska P, Sankar A, Miyagi S, et al. · PLoS One 2015
    PMID 26571505DOI
    What this evidence cannot show

    shRNA proliferation assays in two T-ALL lines within a paper whose title claims MLL-rearranged leukaemia maintenance.

    Evidence that argues against this (1)
    • Two T-ALL lines is a small sample on which to remove fusion specificity from a headline result. Shi J et al. 2013
  7. E7weak confidence 2 of 5: Preliminary, small, indirect, or not yet peer reviewed.preprint

    FHD-286 and BRM014, which inhibit the shared ATPase, are reported inactive in KMT2A-R B-ALL lines while active in Ph-like B-ALL; the ncBAF-selective route has never been tested in KMT2A-rearranged leukaemia.

    Selective Sensitivity of Ph-like B-ALL to BRG1 Inhibition Reveals a Novel Targeted Therapy Strategy. · Abhinav Ayyadevara VSS, Gaur S, Paik A, Perencsik R, et al. · bioRxiv 2025
    PMID 40672274DOI
    What this evidence cannot show

    bioRxiv preprint, abstract only. The second half of the claim - that ncBAF-selective agents are untested in this disease - is now a TITLE-CHECKED bounded absence, and it SURVIVES. Re-run 2026-09-22: `(ncBAF[tiab] OR BRD9[tiab]) AND (KMT2A[tiab] OR MLL[tiab]) AND (leukemia[tiab] OR leukaemia[tiab])` returns 2 records and BOTH WERE FETCHED. Neither is ncBAF pharmacology: PMID 41460961 is 'Comprehensive molecular and functional analysis of NUTM1-rearranged leukemia' (Blood 2026) and PMID 40000924 is an infant BCP-ALL prognostic study (Int J Lab Hematol 2025) -- BRD9 appears in both as the 5-PRIME FUSION PARTNER of BRD9::NUTM1, not as an ncBAF subunit being inhibited. Breadth control through the identical code path: `BRD9[tiab] AND leukemia[tiab]` = 28, so the query is alive and the 2 is real. The previous version of this caveat recorded 3 records 'none of which was fetched'; fetching them is what turned a count into a bound.

    Evidence that argues against this (1)
    • BRD9 inhibitors were shown active in AML lines in 2016 without KMT2A selection, so 'never tested' means never tested with the rearrangement as the selection criterion, not never tested in leukaemia. Hohmann AF et al. 2016
  8. E8strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    SWI/SNF in MLL-AF9 cells occupies a distal MYC enhancer cluster 1.7 Mb away and most direct MLL-AF9 target genes do not respond to Brg1 loss, so the complex's address is distinct from the fusion's.

    Role of SWI/SNF in acute leukemia maintenance and enhancer-mediated Myc regulation. · Shi J, Whyte WA, Zepeda-Mendoza CJ, Milazzo JP, et al. · Genes Dev 2013
    PMID 24285714DOI
    What this evidence cannot show

    Mouse ChIP-seq plus human 3C; the 'majority of direct targets' statement is data not shown.

    Evidence that argues against this (1)
    • That occupancy was mapped for Brg1, a subunit shared by all three complexes, so it does not tell you which complex sits at E1-E5 - which is precisely what this hypothesis needs and what nobody has measured. Michel BC et al. 2018
Connected to