Oncofusion

start / graph / hyp-the-pbx-requirement-has-only-ever-been-tested-one-paralogue-at-a-time

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q74-interactome-gap

The PBX requirement in KMT2A-rearranged leukaemia is a family property, and every experiment that shaped this graph's view of it removed one paralogue at a time

also called PBX1 · PBX2 · PBX3 · paralogue redundancy · TALE cofactor

Three almost identical PBX proteins can each partner the HOX factors that drive this leukaemia, and every experiment so far has removed just one. If they substitute for each other, single-gene results would look weak even if the family as a whole were essential.

The same thing, for a scientist

Three observations that this graph holds separately do not fit together under the single-paralogue reading, and fit easily under a family-redundancy reading. **(1)** Li et al. 2013 report that PBX3, *not* PBX1 or PBX2, is consistently coexpressed with HOXA9 in MLL-rearranged AML, and that Pbx3 depletion inhibits MLL-fusion transformation - a transcript-level argument in a murine/myeloid setting [E1]. **(2)** Selleri et al. 2004 report by immunoblot that **Pbx2** is the predominant high-molecular-weight PBX protein in postnatal bone marrow and thymus, and that Pbx2-null mice have no haematopoietic phenotype, which the authors attribute to compensation by low-molecular-weight Pbx isoforms [E2]. **(3)** This graph's own DepMap join records PBX3 as a dependency in no human leukaemia line, mean Chronos effect **+0.042** across 57 leukaemia lines [E3]. **The hypothesis is that (1), (2) and (3) are all true and that the resolution is redundancy at the protein level within the PBX family.** Under it: the transcript coexpression that singles out PBX3 is a regulatory fact about which paralogue is induced, not about which is required; the absent knockout phenotype and the DepMap zero are what redundancy produces; and no published experiment could have distinguished redundancy from dispensability, because none has removed more than one paralogue. `gene-pbx3`'s own open questions ask for this test and note it has never been done in a human leukaemia line. **Prior probability 0.35** - the mechanism is generic and well precedented, the specific claim is untested, and the competing explanation (the PBX requirement is simply myeloid and murine) is at least as likely. **This hypothesis was written because PBX2 had no entry.** It cannot be stated, let alone tested, while one member of a three-member family is missing from the graph, which is a small concrete argument for closing interactome gaps even when the individual entry is thin.

Go deeper
What would kill it

Listed before the supporting argument, deliberately.

  • A triple PBX1/2/3 knockout in a KMT2A::AFF1 line with no growth or HOXA-programme phenotype. That kills it outright and simultaneously retires the PBX position in B-lineage disease.
  • A protein panel showing PBX1 and PBX2 are undetectable in KMT2A-rearranged cells while PBX3 is abundant. If the other paralogues are not expressed, they cannot compensate and the single-gene experiments were adequate.
  • A single PBX3 knockout producing a full phenotype in a human KMT2A-rearranged line. Redundancy predicts a blunted single-gene effect; a complete one falsifies the premise.
  • Evidence that the HMW/LMW isoform distinction, not the paralogue identity, is what Selleri's blot measures in a way that makes PBX2 abundance irrelevant to HOX dimer formation.

None of these requires a new technique or a new model. All four are blocked only by the fact that nobody has run the combinatorial arm.

The rival hypothesis, which this graph should prefer until the experiment is done

The PBX requirement is a myeloid and murine property and does not transfer to infant B-lineage KMT2A::AFF1 at all.

That explanation also accounts for all three observations, requires no redundancy, and is favoured by this project's own track record: gene-pbx3 already records that every positive PBX3 cell-state result in the corpus is murine and mostly myeloid, and the standing brief's rule 73 exists because generalising from murine MLL-AF9 myeloid disease to infant B-lineage KMT2A::AFF1 is this project's commonest documented synthesis failure.

The two hypotheses are distinguished by the same experiment, which is the practical reason to hold both: a triple knockout with a lineage control separates "redundant family requirement" from "no requirement in this lineage" in one design. Where they differ is in what they predict for the protein panel - redundancy predicts all three paralogues present, the lineage explanation makes no prediction about abundance at all.

Why the AFF4-PBX2 interaction row is not part of this argument

PBX2 arrived on this beat as an IntAct partner of AFF4 in the BioPlex atlas, and that row plays no role in this hypothesis.

It is worth saying explicitly, because the tempting version of this entry is "PBX2 contacts the SEC, therefore PBX2 matters", and that sentence is not supported: the row is a spoke-expanded AP-MS prey from two rounds of one pipeline in HEK293T, with no focused literature and a published transcript-level exclusion against it.

The hypothesis rests only on Li 2013, Selleri 2004 and the DepMap zero - three facts that were all available before this beat and that nobody had put side by side, because the graph had no PBX2 entry to put them in.

Facts and figures
statementThe HOXA9 cofactor requirement in KMT2A-rearranged leukaemia is satisfied redundantly by PBX1, PBX2 and PBX3, so every single-paralogue experiment underestimates it; PBX3's apparent primacy reflects which paralogue is transcriptionally co-induced, not which is required.
originliterature-gap
prior probability0.35
what this hypothesis does NOT rest onthe IntAct AFF4-PBX2 row. That is a spoke-expanded AP-MS prey from two rounds of one pipeline in HEK293T with no focused literature, and it plays no part in the argument.
shown for fusionNobody has looked in any fusion at the family level. Single-paralogue PBX3 results exist in murine MLL-AF9 and MLL-fusion transformation assays; no PBX experiment of any kind has been reported in human KMT2A::AFF1 B-lineage cells.
predictions
[
  "All three PBX proteins are detectable in KMT2A::AFF1 B-lineage cells, with PBX2 among the more abundant high-molecular-weight species as it is in murine bone marrow.",
  "Single PBX1, PBX2 or PBX3 knockout produces a blunted or absent growth phenotype in SEM and RS4;11, consistent with the DepMap zero.",
  "The triple knockout produces a growth and HOXA9/MEIS1-programme phenotype substantially larger than the sum of the singles.",
  "MEIS1 protein stability falls only when more than one PBX paralogue is removed.",
  "A pan-PBX interface peptide such as HXR9 produces a larger effect than any single-paralogue knockdown in the same cells - which, if the compound's selectivity is trusted, is a cheap partial pre-test."
]
experiments
[
  "exp-pbx-family-protein-panel-and-combinatorial-knockout-in-kmt2a-aff1"
]
analyses
[]
falsifiers
[
  "A triple PBX1/2/3 knockout in a human KMT2A::AFF1 line with no growth and no HOXA-programme phenotype kills the hypothesis outright.",
  "A validated protein panel showing PBX1 and PBX2 are undetectable in KMT2A-rearranged cells while PBX3 is abundant - non-expressed paralogues cannot compensate, and the single-gene experiments were then adequate.",
  "A single PBX3 knockout producing a full phenotype in a human KMT2A-rearranged line, since redundancy predicts a blunted single-gene effect.",
  "Demonstration that the PBX2 abundance Selleri measured is in an isoform class that cannot form HOX dimers, which would sever the only protein-level observation the hypothesis rests on."
]
rival hypothesis
{
  "statement": "The PBX requirement is a myeloid and murine property that does not transfer to infant B-lineage KMT2A::AFF1 at all.",
  "why_it_is_at_least_as_likely": "It explains all three observations with no redundancy, and this project's commonest documented synthesis failure is exactly this generalisation (standing brief rule 73). gene-pbx3 already flags that every positive PBX3 cell-state result in the corpus is murine and mostly myeloid.",
  "distinguished_by": "the same experiment - a combinatorial knockout with a non-KMT2A B-ALL lineage control"
}
How this could bite us back
  • If the hypothesis is right, the therapeutic implication is a pan-PBX agent, and a pan-PBX agent hits the dominant PBX protein of normal bone marrow. The redundancy that makes the family a target is the same redundancy that removes the therapeutic window.
  • If it is wrong in the direction of the rival hypothesis, this graph has been building a HOXA argument on a myeloid murine result for several cycles, and the correct response is to weaken gene-pbx3's claims rather than to add PBX2 alongside them.
  • A confirmed family redundancy would also mean that every DepMap-derived 'not a dependency' call on a paralogous transcription factor in this graph is uninterpretable rather than negative, which is a larger and less comfortable consequence than the PBX-specific one.
Still open
  1. Does PREP1/PKNOX1 belong in the redundant set? It is a MEINOX partner rather than a PBC-class protein, but this graph's mechanism-pbx-nuclear-retention-requires-a-meinox-partner makes the available-dimer pool depend on it, and a PBX-only knockout series may be measuring the wrong quantity.
  2. Is there any published compound knockout of two PBX paralogues in any haematopoietic system, murine or human? This beat did not search that specifically and the answer would change the prior substantially.
  3. If the family is redundant, is the shared HOX-PBX hexapeptide groove the only tractable target, and what does the on-target cost look like given that PBX2 is the dominant high-molecular-weight PBX in normal bone marrow?
Every source behind this entry (4)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In cytogenetically abnormal AML including MLL-rearranged AML, PBX3 but not PBX1 or PBX2 is consistently coexpressed with HOXA9; Pbx3 depletion inhibits MLL-fusion-mediated transformation and coexpressed PBX3 synergises with HOXA9 in vitro and in vivo.

    PBX3 is an important cofactor of HOXA9 in leukemogenesis. · Li Z, Zhang Z, Li Y, et al., Chen J · Blood 2013
    PMID 23264595DOI
    What this evidence cannot show

    Read as abstract, not full text. Cytogenetically abnormal AML - myeloid, not B-lineage - and the transformation assays are murine. The exclusion of PBX1 and PBX2 is transcript coexpression, not a functional test and not a protein measurement. Cited in 5 other entries of this graph.

    Evidence that argues against this (1)
    • A coexpression argument identifies which paralogue is co-regulated, which is a different question from which is required. Under redundancy, the co-regulated one is the one whose loss is compensated, so this result is compatible with PBX3 being both the induced paralogue and a non-dependency. Selleri L et al. 2004
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Pbx2-null mice have no developmental or haematopoietic phenotype despite Pbx2 being the predominant high-molecular-weight-isoform Pbx protein in postnatal bone marrow and thymus, which the authors attribute to compensation by low-molecular-weight Pbx isoforms.

    The TALE homeodomain protein Pbx2 is not essential for development and long-term survival. · Selleri L, DiMartino J, van Deursen J, et al., Cleary ML · Mol Cell Biol 2004
    PMID 15169896DOI
    What this evidence cannot show

    Murine, germline, steady-state, normal haematopoiesis, read as abstract. The authors' compensation explanation is their inference from protein levels, not a demonstrated redundancy - they did not make a compound knockout. No leukaemia challenge is reported in the abstract. This is the only protein-level PBX measurement in haematopoietic tissue that this beat located, in either species.

    Evidence that argues against this (1)
    • A germline knockout gives lifelong opportunity for developmental compensation that an acute deletion in a transformed cell does not, so the absent phenotype may be a property of the knockout design rather than evidence of standing redundancy in a leukaemic cell. Li Z et al. 2013
  3. E3moderate confidence 3 of 5: One strong study, or several smaller consistent ones.analysis

    PBX3 is not a dependency in any human leukaemia line in DepMap - mean Chronos effect +0.0418 across 57 leukaemia lines and +0.0297 across 1,258 lines, with 0.08% of lines below -0.5 - and the PBX family has never been tested combinatorially in a human leukaemia line.

    What this evidence cannot show

    An internal entry of this project, quoting a DepMap join it retrieved on 2026-09-20; the figure was not re-fetched this cycle and carries no DepMap release identifier, which brief rule 83 requires. A near-zero Chronos effect in a paralogue-redundant family is uninterpretable rather than negative - which is the hypothesis, so citing the zero in support of it is close to circular and is flagged as such.

    Evidence that argues against this (1)
    • DepMap's leukaemia panel is dominated by adult myeloid lines and contains few KMT2A::AFF1 B-lineage lines, so a null there is weak evidence about the index disease independently of any redundancy argument. This graph has already had to correct itself twice about reading DepMap numbers without a lineage split. Li Z et al. 2013
  4. E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    PBX2 is UniProt P40425, 430 aa, exactly the length of PBX1 (P40424) and four residues from PBX3 (P40426), and it had no entry in this graph before this cycle - gene-pbx3's paralogue_discrimination field recorded it as 'not resolved this cycle'.

    What this evidence cannot show

    Retrieved 2026-09-21. This item is about the graph's own completeness rather than about biology; it is included because the hypothesis is not expressible without a PBX2 node and that is the point being made.

    Evidence that argues against this (1)
    • The absence of a node is not evidence for a hypothesis. That a gap in the graph prevented an argument being stated says something about the graph's coverage and nothing about whether the argument is right. Selleri L et al. 2004
Connected to