Oncofusion

start / graph / ana-depmap-usp2-maml2-lineage-split

Analysismoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q-usp2-maml2-second-bcr

Lineage-split DepMap: neither USP2 nor MAML2 is a dependency in KMT2A-rearranged lines, and USP22 is

also called USP2/MAML2 dependency lineage split · deferred-partner DepMap round for USP2 and MAML2

Before saying a gene is needed by KMT2A-rearranged leukaemia cells, you have to check both kinds of those cells - the myeloid ones and the B-cell ones - because the available cell panel is mostly myeloid and a myeloid gene looks important for free. Doing that check, neither USP2 nor MAML2 is needed by these cells, while USP22 clearly is.

The same thing, for a scientist

The standing confound in this repo is that the KMT2A-rearranged DepMap index arm is myeloid-enriched (eight AML lines against five B-ALL), so a myeloid-identity gene scores as KMT2A-selective with no KMT2A biology involved - established by ana-lineage-confound-myeloid-tf-dependencies and ana-lineage-consistent-dependency-null. This run applies the same split to this beat's genes. Chronos gene-effect scores were pulled through the Open Targets Platform, restricted to leukaemia lines, the KMT2A-rearranged arm split into myeloid and B-lineage halves, and each half expressed in standard deviations of the other-leukaemia arm. **USP2 scores +0.58 (myeloid) and +0.31 (B-lineage); MAML2 scores +0.33 and +0.29; USP8 scores +0.39 and +0.24.** All three are slightly *less* depleted in KMT2A-rearranged lines than in other leukaemias, i.e. no dependency in either half and no lineage artefact to disentangle. In the same run MEN1 scores -2.29 and -2.28, KMT2A -1.48 and -1.22, and USP22 -1.79 and -0.76 [E1]. **The result's meaning is bounded hard by one fact: the DepMap panel contains no KMT2A::USP2 and no KMT2A::MAML2 line.** This measures the wild-type partner gene in cells carrying other KMT2A fusions, which is a different question from whether the fusion protein's partner-derived domain is required.

Go deeper
Results2 deeper

Standard deviations of the other-leukaemia arm; negative means more depleted, i.e. more essential in KMT2A-rearranged lines.

geneKMT2A-r myeloid (n=8)KMT2A-r B-lineage (n=5)gapOCI-AML3U-937verdict
USP2+0.58+0.310.28-0.76+0.19no dependency in either half
MAML2+0.33+0.290.03-0.11+0.59no dependency in either half
USP8+0.39+0.240.14+1.14+0.95no dependency in either half
USP22-1.79-0.76-1.03-0.95-2.31lineage-independent dependency
MEN1-2.29-2.28-0.01-2.80+0.36lineage-independent dependency
KMT2A-1.48-1.22-0.26-2.05-0.32lineage-independent dependency

No gene in the panel was classified as a myeloid-identity artefact, because none showed the artefact signature (depleted in the myeloid half, flat in the B-lineage half) 1.

USP22 as the built-in positive control, and as the disambiguation test

USP22 was included deliberately for two reasons. First, it is an established lineage-independent dependency in this repo and therefore a positive control that the pipeline still works. Second, and specific to this beat, it is the gene most likely to be confused with USP2, and putting both in the same run produces the disambiguation in the hardest possible form: run side by side, in the same script, on the same lines, USP2 scores +0.58/+0.31 and USP22 scores -1.79/-0.76. Substituting one symbol for the other would not blur a conclusion, it would invert it. That result is carried in gene-usp2 as the discriminating fact.

Why the gap column is the wrong thing to look at here

The gap between halves is the diagnostic for a lineage artefact, and it only means something when at least one half is depleted. USP2's gap of 0.28 is a difference between two non-dependencies and carries no information. Reporting 'USP2 gap 0.28, no artefact' as though it were a finding would be a category error; the finding is simply that both halves are flat.

The limitation that decides how much this is worth1 deeper

DepMap has no KMT2A::USP2 line, no KMT2A::MAML2 line and no KMT2A::USP8 line. There is no such cell line anywhere - these fusions have never been modelled. So this analysis answers: do KMT2A-rearranged leukaemia cells, of the fusions we can culture, require wild-type USP2 or MAML2? Answer: no. It does not and cannot answer: does a KMT2A::USP2 cell require the UCH activity of its own fusion protein?

Those two questions have different answers in principle. A fusion can depend on a domain whose wild-type parent gene is entirely dispensable, and the KMT2A-partner literature is full of examples - the partner's contribution is usually a module ripped out of context, not the gene's normal function. This is exactly the kind of null result that becomes a wrong conclusion if the caveat is dropped in transmission, so it is stated here, in gene-usp2, in gene-maml2 and in gene-usp8.

One line worth noting despite the above

OCI-AML3, a pathway-equivalent NPM1c line carried as a myeloid reference, scores -0.76 for USP2 against a flat KMT2A-rearranged arm 1. A single line is not a result and this is reported rather than interpreted. It is mentioned only so that a reader who opens the results file does not think it was hidden.

Facts and figures
scriptanalyses/depmap-usp2-maml2-lineage-split/run.py
statusrun
result summaryUSP2 +0.58/+0.31, MAML2 +0.33/+0.29, USP8 +0.39/+0.24 (myeloid/B-lineage z of the other-leukaemia arm): no dependency in either half. USP22 -1.79/-0.76, MEN1 -2.29/-2.28, KMT2A -1.48/-1.22: lineage-independent dependencies. No myeloid-identity artefacts in this panel. Hard limit: no KMT2A::USP2, ::MAML2 or ::USP8 line exists in DepMap or anywhere else.
identifier failuresNone. The query asserts approvedSymbol for every Ensembl id and aborts on a mismatch; all six resolved to the expected symbol.
both halves reportedtrue
inputs
[
  "DepMap Chronos gene effect via Open Targets Platform GraphQL",
  "KMT2A-rearranged myeloid lines: THP-1, MOLM-13, MOLM-14, NOMO-1, SHI-1, OCI-AML-2, ML-2",
  "KMT2A-rearranged B-lineage lines: SEM, SEMK2, RS4;11, HB11;19, KOPN-8",
  "pathway-equivalent references: OCI-AML3 (NPM1c), U-937 (PICALM::MLLT10)",
  "genes: USP2, MAML2, USP22, USP8, MEN1, KMT2A"
]
ensembl IDS asserted
{
  "USP2": "ENSG00000036672",
  "MAML2": "ENSG00000184384",
  "USP22": "ENSG00000124422",
  "USP8": "ENSG00000138592",
  "MEN1": "ENSG00000133895",
  "KMT2A": "ENSG00000118058"
}
How this could bite us back
  • A clean null is the most transmissible kind of wrong conclusion, because it is short and quotable. 'USP2 is not a dependency' will survive into a summary long after 'in a panel containing no KMT2A::USP2 line' has been dropped.
  • The absence of any minor-BCR cell line means that every functional-genomics claim this project makes about KMT2A-rearranged leukaemia is, strictly, a claim about major-BCR fusions. That limitation applies well beyond this beat.
  • USP22's dependency in both halves is real and is already carried elsewhere in this graph, but it is a dependency of KMT2A-rearranged cells in general and is not evidence about USP2, despite the symbols.
Still open
  1. Would a CRISPR screen in an engineered KMT2A::USP2 cell give a different answer? Since no such cell exists, this is the gating question for every dependency statement about this genotype.
  2. Is there any KMT2A-rearranged line with a minor-BCR breakpoint at all? Every line in this panel carries a major-BCR fusion, so the whole BCR2 class - PHD-retaining fusions - is unrepresented in functional genomics.
  3. Does splitting by an expression-derived lineage score rather than a lineage label change any verdict in this panel? The label-based split is the repo's standard and has never been benchmarked against the alternative.
Every source behind this entry (2)
  1. E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset

    In DepMap Chronos gene-effect data accessed through the Open Targets Platform and restricted to leukaemia lines, USP2 (+0.58/+0.31), MAML2 (+0.33/+0.29) and USP8 (+0.39/+0.24) show no depletion in either the myeloid or the B-lineage half of the KMT2A-rearranged arm, while USP22 (-1.79/-0.76), MEN1 (-2.29/-2.28) and KMT2A (-1.48/-1.22) are depleted in both halves.

    What this evidence cannot show

    Eight myeloid and five B-lineage KMT2A-rearranged lines, so a single line moves a mean and the B-lineage half in particular is fragile. The split is by lineage LABEL rather than by an expression signature, and KMT2A-rearranged disease includes genuinely ambiguous-lineage cases that a label forces into one bin. Decisively, the panel contains no line carrying any of the three fusions under study, so this bounds the requirement for the wild-type partner gene and not the requirement for the fusion's partner-derived domain. Every Ensembl id was asserted to resolve to its expected approvedSymbol by the query itself, which aborts on mismatch; none aborted.

    Evidence that argues against this (1)
    • A null dependency in a panel lacking the genotype is close to uninformative about that genotype, and the KMT2A-partner literature is largely a catalogue of partners whose contribution to the fusion is unrelated to the parent gene's normal essentiality. Reading 'USP2 is not a dependency' as 'USP2 does not matter in KMT2A::USP2 leukaemia' would be the error this caveat exists to prevent. Meyer C et al. 2019
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    The lineage split is necessary rather than optional, because the KMT2A-rearranged DepMap index arm is myeloid-enriched (roughly eight AML lines to five B-ALL), so a myeloid-identity gene scores as KMT2A-selective with no KMT2A biology involved.

    What this evidence cannot show

    An internal methods reference, not an external publication, and it inherits every limitation of the DepMap panel it describes. The panel composition (THP-1, MOLM-13, MOLM-14, NOMO-1, SHI-1, OCI-AML-2, ML-2 myeloid; SEM, SEMK2, RS4;11, HB11;19, KOPN-8 B-lineage) is fixed by what DepMap screened and cannot be rebalanced.

    Evidence that argues against this (1)
    • Lineage labels are a crude proxy: KMT2A-rearranged disease is characteristically lineage-ambiguous and includes MPAL, so binning each line as myeloid or B-lineage imposes a distinction the biology does not respect. A gene that matters in the ambiguous state could score flat in both halves. Meyer C et al. 2019
Connected to