Oncofusion

start / graph / hyp-kmt2a-fusion-creates-an-fa-pathway-dependency

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q-fa-replication

KMT2A rearrangement creates a requirement for the FA ubiquitin-transfer module — the claim, its prior, and the four results that would kill it

also called Fanconi dependency hypothesis · FA module dependency in KMT2A-r leukaemia

The claim in one sentence: leukaemia cells driven by a KMT2A fusion need a specific part of the Fanconi DNA-repair machinery more than other leukaemia cells do. The evidence is a computer analysis of cell-line data that survives every statistical check this project could run — and no experiment at all.

The same thing, for a scientist

**Statement.** In leukaemia cells carrying a KMT2A rearrangement, loss of the FA ubiquitin-transfer module — UBE2T (E2), FANCL (E3), FAAP100 (catalytic scaffold), and the FANCA–FANCG–FANCF assembly arm — imposes a fitness cost that it does not impose on leukaemia cells without the rearrangement, and this holds within each lineage. **Prior probability 0.35**, and the number is that high only because of how much survived testing and how little of it is experimental. Supporting: a 26-gene set fixed before any result was read beats its null at p = 5e-05; eight of the set land in the top 40 of 8,269 by raw effect; the enrichment is present in both lineage halves independently, is stronger in the B-lineage half, and holds at -0.334 and -0.308 when lineage is held exactly constant; a correlation-preserving line-label permutation gives p = 0.0002 for the eight; none of the eight is pan-essential; and an independent published CRISPR screen already found UBE2T and FANCL to be AML dependencies. Against: no experiment has ever tested an FA gene in a KMT2A-rearranged cell; the obligate core subunits do not move together, so the canonical pathway is probably not the unit; the substrate FANCD2 is weak; every number comes from eleven immortalised lines, none infant-derived; a complete rival mechanism with actual experimental support exists (ALDH2 silencing and endogenous aldehydes); and humans with no FA pathway at all still develop KMT2A-rearranged leukaemia.

Go deeper
What is being claimed, and what is not

Claimed. A quantitative fitness requirement, in KMT2A-rearranged leukaemia cells, for a specific module of FA proteins, over and above whatever the lineage requires.

Not claimed.

  • Not that the pathway is required. The obligate core subunits span a factor of five and the canonical substrate FANCD2 is weak, so the dependency is probably not the pathway doing its textbook job 4.
  • Not that it is absolute. Fanconi anaemia patients develop KMT2A-rearranged leukaemia 6.
  • Not that transcription–replication conflict is the reason. That is a separate, mostly unmeasured hypothesis kept in mech-transcription-replication-conflict-kmt2a-r, and the dependency claim stands or falls independently of it.
  • Not that it is a drug target. The therapeutic window is roughly a quarter to a half a Chronos unit, against a gene family whose human loss-of-function phenotype is bone-marrow failure.
Why the prior is 0.35 and not higher or lower

Pushing it up. Four independent objections were tested and none survived: lineage (both halves, lymphoid stronger, holds within lineage), pan-essentiality (none of the eight is common-essential), subunit correlation (line-label permutation), and global line sensitivity (per-line offset correction, with ACTB behaving as the contaminant control should). An externally published CRISPR screen independently nominated two of the eight in acute leukaemia 3. The gene set was fixed before any result was read, which removes the commonest way a pathway result is manufactured.

Holding it down. The evidence is entirely one dataset analysed several ways, which is not the same as several datasets. Eleven cell lines. Zero experiments. A rival mechanism with an experiment behind it. And the strongest single argument against — that the complex does not behave like a complex — is an argument this beat generated itself and could not resolve, because it might equally be CRISPR reagent quality 4.

0.35 is 'more likely than not to be a real property of these cells, and considerably less likely than not to be a real property of the disease'.

Predictions
  1. Knockdown of FANCL or UBE2T reduces engraftment of a KMT2A-rearranged patient-derived xenograft more than of a lineage-matched non-rearranged one.
  2. FA-gene dependency is stronger in KMT2A-rearranged than in non-rearranged leukaemia cells after conditioning on ALDH2 expression.
  3. Inducible knockdown of the fusion in an isogenic line reduces the FA-gene requirement.
  4. Within the FA annotation, the dependency concentrates on the ubiquitin-transfer module (UBE2T, FANCL, FAAP100, FANCA, FANCF, FANCG) and not on FANCB, FANCC, FANCE, FANCM or FANCD2 — a prediction already partly observed, and the observation that generated it, so it counts only if it holds in a new dataset.
  5. FANCD2 K561R knock-in is better tolerated than FANCL knockout in a KMT2A-rearranged line.
  6. The dependency is weaker in quiescent leukaemia-initiating cells than in bulk blasts.
Falsifiers

Stated so that a single result can end this.

  • ALDH2 expression predicts FA-gene dependency across the DepMap leukaemia panel and KMT2A status adds nothing after conditioning on it. This is a desk analysis and it should be run first.
  • FA genes score neutrally in a published in vivo or PDX CRISPR screen in KMT2A-rearranged leukaemia. This may already be sitting in a supplementary table.
  • Per-guide DepMap data show the core-complex spread is reagent quality, which would mean the coherent module is an illusion and the whole set is behaving as one poorly measured complex.
  • FANCL or UBE2T knockdown does not impair engraftment of a KMT2A-rearranged PDX.
  • Primary KMT2A-rearranged blasts show no FA dependency where the lines do.
  • The eight genes' selectivity disappears in a newer DepMap release, which for an effect of this size is a live possibility.
Facts and figures
statementKMT2A rearrangement imposes a quantitative requirement for the FA ubiquitin-transfer module (UBE2T, FANCL, FAAP100, FANCA, FANCF, FANCG) plus FANCI and BRIP1, over and above the lineage baseline, in leukaemia cells.
origindata-pattern
prior probability0.35
predictions
[
  "FANCL or UBE2T knockdown reduces engraftment of a KMT2A-rearranged PDX more than of a lineage-matched non-rearranged one",
  "the dependency survives conditioning on ALDH2 expression across the DepMap leukaemia panel",
  "inducible fusion knockdown in an isogenic line reduces the FA-gene requirement",
  "the dependency concentrates on the ubiquitin-transfer module and not on FANCB, FANCC, FANCE, FANCM or FANCD2, in a dataset other than this one",
  "FANCD2 K561R knock-in is better tolerated than FANCL knockout in a KMT2A-rearranged line",
  "the dependency is weaker in quiescent leukaemia-initiating cells than in bulk blasts"
]
experiments
[
  "exp-replication-stress-readout-kmt2a-r"
]
analyses
[
  "ana-fa-pathway-lineage-and-kmt2a-specificity",
  "ana-effect-size-vs-z-reranking"
]
falsifiers
[
  "ALDH2 expression predicts FA-gene dependency across the DepMap leukaemia panel and KMT2A status adds nothing after conditioning on it",
  "FA genes score neutrally in a published in vivo or PDX CRISPR screen in KMT2A-rearranged leukaemia",
  "per-guide DepMap data show the core-complex spread is reagent quality rather than biology",
  "FANCL or UBE2T knockdown does not impair engraftment of a KMT2A-rearranged patient-derived xenograft",
  "primary KMT2A-rearranged blasts show no FA dependency where the cell lines do",
  "the eight genes' selectivity does not reproduce in a newer DepMap release"
]
How this could bite us back
  • If true and drugged, the on-target toxicity is Fanconi anaemia: progressive bone-marrow failure and a raised secondary-malignancy risk, in children who would otherwise live for decades after cure.
  • Disabling fork protection raises the mutation rate in surviving cells, which is the standard route to resistance and, in a cured child, to a second cancer.
  • If the dependency is downstream of the fusion's transcriptional output, menin inhibition would relieve it, and the combination of a menin inhibitor with an FA-directed agent would be antagonistic rather than additive.
  • None of the eight has a small-molecule ligand, and only UBE2T has a catalytic site of the kind that usually yields one.
Still open
  1. Does the dependency survive conditioning on ALDH2 expression? Everything else is secondary to this.
  2. Is the unit of dependency the ubiquitin-transfer module, and what does it ubiquitinate if not FANCD2?
  3. Does fusion partner matter — the eleven lines carry KMT2A::AFF1, KMT2A::MLLT3, KMT2A::MLLT1 and KMT2A::AFDN, which is enough to look?
  4. Would the same hypothesis survive in a DepMap release with more KMT2A-rearranged lines?
Every source behind this entry (6)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    The 26-gene FA/BRCA set, fixed before any result was read, has mean rank 1852 of 8269 by z and 2075 by raw effect against a null median of 4135, p = 5e-05 on 20,000 permutations for both statistics, with FANCF, FANCI, FANCA, BRIP1, FANCL, FANCG, UBE2T and FAAP100 at ranks 11, 15, 20, 22, 26, 27, 33 and 36 by raw effect.

    Effect-size versus z re-ranking of the genome-scale lineage-consistency sweep, with a permutation test of the Fanconi/FA-BRCA pathway · Oncofusion knowledge graph, analyses/effect-size-vs-z-reranking; DepMap Chronos via Open Targets 2026
    What this evidence cannot show

    Eleven immortalised KMT2A-rearranged lines against 43 other leukaemia lines. The FA/BRCA membership list is one of several defensible annotations.

    Evidence that argues against this (1)
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    The enrichment survives every test this beat could run: both lineage halves independently at p = 5e-05 with the lymphoid half stronger; within-lineage contrasts at -0.334 (AML) and -0.308 (B-ALL) on the eight headline genes; a correlation-preserving line-label permutation at p = 0.0002 for those eight; and none of the eight pan-essential.

    What this evidence cannot show

    All of it is one dataset analysed several ways. Four B-lineage index lines. No fusion-specificity control was possible because the DepMap leukaemia slice holds one BCR::ABL1 line.

    Evidence that argues against this (1)
    • Robustness to statistical objections is not robustness to the dataset. Every test here shares the same eleven cell lines, the same Chronos values and the same Cellosaurus annotations, so they are not independent confirmations of each other. Ben-David U et al. 2018
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    An independent domain-focused CRISPR screen identified UBE2T and FANCL — two of the eight — as selective dependencies in acute myeloid leukaemia, which is external experimental support that an FA dependency exists in acute leukaemia.

    PMID 33893150DOI
    What this evidence cannot show

    AML against other cancer lineages, not KMT2A-rearranged against other leukaemia. It attributes the dependency to ALDH2 silencing, not to any fusion.

    Evidence that argues against this (1)
    • This is simultaneously the best external support and the leading rival: if ALDH2 status explains the dependency, KMT2A rearrangement is incidental and this hypothesis is wrong even though the dependency is right. Yang Z et al. 2021
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    The obligate FA core subunits do not behave alike — FANCF -0.437 to FANCB -0.079 — and the canonical substrate FANCD2 is weak (-0.150, line-permutation p = 0.071) while its heterodimer partner FANCI is strong (-0.361), so the dependency is probably not the canonical pathway performing its known reaction.

    Is the Fanconi signal lineage-wide, fusion-specific, and separable from pan-essentiality? · Oncofusion knowledge graph, analyses/fa-pathway-lineage-and-fusion-specificity; DepMap Chronos via Open Targets 2026
    What this evidence cannot show

    Per-guide CRISPR reagent quality was not checked and could produce the same pattern; FANCB is X-linked and FANCM is often reported to act independently of the core complex.

    Evidence that argues against this (1)
    • The subset that does carry the signal is not random — it is the ubiquitin-transfer machinery plus its assembly arm — and the published AML screen independently nominated the catalytic pair from that subset, which is what a module-level dependency would look like. Yang Z et al. 2021
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.database

    No published study has tested any FA gene in a KMT2A-rearranged cell: symbol-and-alias searches for FANCI, FANCL, UBE2T, FAAP100, BRIP1 and FANCJ against KMT2A or MLL return zero records each, and the joint FANCA/FANCF/FANCG query returns five of which four are incidental mutation-frequency lists.

    What this evidence cannot show

    Counted absences are only as good as the symbol set, and this beat found one case — MED23 in a Mediator-complex paper — where a gene studied in a screen was invisible to every symbol query. The same could be true here for an FA gene buried in a complex-level or pathway-level paper.

    Evidence that argues against this (1)
    • The pathway is not unstudied even where the genes are: a Hoxa-dependency transcriptome in MLL-rearranged mouse leukaemia returns an FA-pathway-enriched signature, which no gene-symbol query would reach. Kettyle LM et al. 2019
  6. E6moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Humans with biallelic FA-gene loss develop KMT2A-rearranged leukaemia, so the requirement cannot be absolute.

    PMID 10679915
    What this evidence cannot show

    One case report; complementation group not given in the abstract; therapy-related context. **The KMT2A rearrangement in this case is THERAPY-RELATED.** A boy with Fanconi anaemia developed early pre-B ALL with a normal karyotype, was treated with an ALL regimen including 140 mg/m2 of doxorubicin — a topoisomerase II inhibitor — and relapsed 38 weeks later as AML carrying t(11;16)(q23;p13) MLL-CBP. So the KMT2A lesion arose after topoisomerase II exposure, which is a well-established cause of exactly this rearrangement class, and the Fanconi background may be permissive rather than causal. Surfaced by scripts/verify-claims.py, which ranked this claim near the bottom on claim-to-abstract overlap.

    Evidence that argues against this (2)
    • Only the absolute claim is excluded. A fitness cost of a quarter to a half a Chronos unit is entirely compatible with a clone arising and expanding in an FA-null host, and KMT2A fusions are notably NOT part of the recurrent lesion spectrum catalogued in FA-associated myelodysplasia and leukaemia, which points weakly the other way. Quentin S et al. 2011
    • This case bounds the FA dependency less than the claim implies. It is a single therapy-related leukaemia in a DNA-repair-deficient child after topoisomerase II exposure, which is the standard route to a KMT2A rearrangement and is if anything expected to be easier in an FA background. It shows a person lacking FA function can acquire a KMT2A rearrangement; it does not show that FA function is dispensable for a KMT2A-rearranged leukaemia to be maintained, which is what the dependency claim is about. Genes, Chromosomes and Cancer 2000
Connected to