Oncofusion

start / graph / hyp-igf2bp-family-redundancy-kmt2a

Hypothesisspeculative confidence 1 of 5: Our own inference or analogy. Not shown to be true.Q6-interactome-rna-layer

Hypothesis: IGF2BP2 and IGF2BP3 buffer each other in KMT2A::AFF1 leukaemia, so single-gene CRISPR under-reports both

also called IGF2BP paralogue redundancy · IMP family buffering

If two of the three IGF2BP siblings can cover for each other in MLL/AF4 leukaemia, then deleting either one alone would look almost harmless even though the pair is essential. That would explain why mouse experiments say IGF2BP3 matters a lot and cell-line screens say it barely matters.

The same thing, for a scientist

The hypothesis: in KMT2A::AFF1 leukaemia IGF2BP2 and IGF2BP3 are functionally redundant on an overlapping set of target transcripts, so removing either alone produces a small phenotype while removing both produces a large one. It is motivated by a specific unexplained gap - Igf2bp3 deletion greatly attenuates MLL-Af4 disease in mice [E1] while IGF2BP3 has a DepMap Chronos mean of -0.137 with 1.6% of lines depleted [E4] - and by the observation that the two paralogues co-over-expressed in MLL/AF4-positive B-ALL are precisely IGF2BP2 and IGF2BP3 [E2], the same two with negative B-lymphoid index-arm differences in this project's scan [E4]. The alternative explanation for the same gap, which this hypothesis competes with rather than replaces, is that the mouse phenotype is a leukaemia-initiating-cell effect invisible to a two-week dropout assay [E1]. **The two are distinguishable**: redundancy predicts a large double-knockout effect on bulk proliferation, the initiation model predicts none.

Go deeper
Prior, stated honestly

Around 0.2. The motivating pattern is real but every component of it is weak, and the family's near-identical domain architecture is a reason to expect redundancy that is also a reason to be suspicious of it as an explanation - it is the explanation one reaches for whenever a paralogue family disappoints, and it is usually wrong.

Two things push it above a bare prior. First, the two datasets that nominate IGF2BP2 and IGF2BP3 together were generated by unrelated means 24. Second, the competing explanation - initiation versus proliferation - is not exclusive with this one, so a low prior on redundancy is not a high prior on anything else.

What would kill it

The falsifiers are listed in data.falsifiers and are all single experiments.

The cleanest is the double knockout. If IGF2BP2/IGF2BP3 double-knockout KMT2A::AFF1 cells proliferate like either single knockout, the hypothesis is dead in its strong form, and the initiation explanation is left holding the field.

The cheapest is compensation measurement: if IGF2BP2 protein does not rise when IGF2BP3 is deleted in MLL-Af4 cells, the buffering mechanism has no substrate. This is a western blot on material that already exists in at least two laboratories.

The most informative negative is the targetome overlap. Redundancy requires shared targets. If eCLIP in the same line shows IGF2BP2 and IGF2BP3 occupying largely non-overlapping transcript sets, the hypothesis fails on mechanism rather than on phenotype.

Why DepMap cannot settle this and what it did contribute

This project ran the co-dependency test and it returned nothing: pairwise Chronos correlations of 0.066-0.105 across 1,208 lines, negative inside the 54 leukaemia lines, and 4 lines depleted for two or more members against 2.52 expected under independence 4.

That is not evidence against the hypothesis, and the analysis says so in its own results file. Under redundancy every single knockout is near-neutral by construction, so the profiles contain no signal - which is indistinguishable from the family not mattering. A positive correlation would have indicated shared dependency context, not redundancy. The instrument cannot answer the question.

What the scan did contribute is the direction: among the three paralogues, only IGF2BP2 and IGF2BP3 have negative B-lymphoid index-arm differences (-0.123 and -0.135) while IGF2BP1 is +0.095 4. That narrows the hypothesis from 'the family' to a specific pair, which is what makes the double knockout a two-gene experiment rather than a three-gene one.

Facts and figures
statementIn KMT2A::AFF1 leukaemia IGF2BP2 and IGF2BP3 act redundantly on an overlapping target transcript set, so single-gene deletion of either under-reports the pair's combined requirement.
origindata-pattern
prior probability0.2
predictions
[
  "IGF2BP2 protein rises on IGF2BP3 deletion in MLL-Af4 cells, and the reverse.",
  "eCLIP of IGF2BP2 and IGF2BP3 in the same KMT2A::AFF1 line shows substantially overlapping target 3'UTRs.",
  "A double knockout produces a proliferation and self-renewal defect larger than the sum of the singles.",
  "A pan-IMP inhibitor outperforms a paralogue-selective IGF2BP3 inhibitor in a t(4;11) model."
]
experiments
[
  "exp-igf2bp-degron-self-renewal-xenograft"
]
analyses
[
  "ana-rna-layer-kmt2a-selectivity"
]
falsifiers
[
  "IGF2BP2/IGF2BP3 double-knockout KMT2A::AFF1 cells proliferate and self-renew like either single knockout - the strong form is dead.",
  "IGF2BP2 protein does not rise on IGF2BP3 loss in MLL-Af4 cells - no compensation mechanism.",
  "eCLIP shows IGF2BP2 and IGF2BP3 occupy largely non-overlapping transcripts in the same line - fails on mechanism.",
  "A demonstrably paralogue-selective IGF2BP3 inhibitor matches a pan-family one in a t(4;11) model - buffering is not operating."
]
How this could bite us back
  • If the hypothesis is right, a selective IGF2BP3 inhibitor is the wrong molecule and the programme that produced I3IN-002 is optimising for the wrong property.
  • If it is right, the resistance route to any single-paralogue inhibitor is already encoded in the genome - up-regulate the other one - and needs no mutation.
  • IGF2BP2 is high in healthy marrow and blood, so a pan-family inhibitor, which is what the hypothesis implies is needed, inherits a normal-tissue liability the selective one would not have.
  • This hypothesis is a convenient explanation for a disappointing dependency number and should be held to a higher bar for that reason. Paralogue redundancy is the standard rescue for a weak screen result and is usually wrong.
Still open
  1. Does IGF2BP2 protein rise on IGF2BP3 deletion in MLL-Af4 cells - a western blot on existing material?
  2. Do IGF2BP2 and IGF2BP3 occupy the same 3'UTRs in a KMT2A::AFF1 line?
  3. Is I3IN-002 paralogue-selective, which decides whether its MLL-Af4 activity supports or undercuts this hypothesis?
Every source behind this entry (5)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Igf2bp3 deletion significantly increases survival and greatly attenuates MLL-Af4-driven leukaemia in mice with minimal impact on baseline haematopoiesis, and MLL-Af4 leukaemia-initiating cells require Igf2bp3; the paper's own verb for the effect is that IGF2BP3 amplifies leukaemogenesis.

    PMID 34321607DOI
    What this evidence cannot show

    Abstract read; full text not read. Murine, chimeric MLL-Af4 construct, deletion applied before and during leukaemogenesis. A single-gene deletion, which is the design that cannot detect the buffering this hypothesis proposes.

    Evidence that argues against this (1)
    • If a single Igf2bp3 deletion already greatly attenuates disease in mouse, the strong form of the redundancy hypothesis is in trouble: a buffered pair should show little phenotype on single deletion. The hypothesis has to be the weaker claim that buffering operates in human cell lines in culture and not in the murine in vivo setting, which is a less attractive position. Oncofusion knowledge graph (internal analysis) 2026
  2. E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    In B-ALL, MLL/AF4-positive disease is characterised by over-expression of both IGF2BP2 and IGF2BP3, while IGF2BP1 over-expression instead characterises ETV6/RUNX1-positive disease.

    PMID 21414819DOI
    What this evidence cannot show

    RT-qPCR transcript levels, 2011, single centre, no protein and no function. Co-expression is a necessary but nowhere near sufficient condition for redundancy.

    Evidence that argues against this (1)
    • Co-expression is equally predicted by a shared upstream regulator, and one is known: MLL-AF4 induces genes encoding RNA-binding proteins as a counter-regulatory response. Under that model the two paralogues rise together because the fusion drives both, with no functional relationship between them at all. Hanewald T et al. 2026
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database

    The three IMP-family paralogues share domain architecture and near-identical length - IGF2BP1 577 aa, IGF2BP2 599 aa, IGF2BP3 579 aa - which is the structural basis for expecting overlapping substrate preference.

    What this evidence cannot show

    Domain architecture and length similarity are a reason to hypothesise shared substrates, not a measurement of them. The accessions establish identity and size only; the substrate overlap has not been measured in a KMT2A::AFF1 line.

    Evidence that argues against this (1)
    • Structural similarity within an RNA-binding protein family routinely fails to predict target overlap, because targeting is set by cooperating factors and by which transcripts are present, not by the RNA-binding domain alone. Kabbani LES et al. 2025
  4. E4moderate confidence 3 of 5: One strong study, or several smaller consistent ones.analysis

    DepMap gives IGF2BP1/2/3 pairwise Chronos correlations of 0.066, 0.105 and 0.078 across 1,208 lines, negative (-0.275, -0.063, -0.123) within 54 leukaemia lines, with 4 lines depleted for two or more members against 2.52 expected under independence; IGF2BP2 and IGF2BP3 have negative B-lymphoid index-arm differences (-0.123, -0.135) while IGF2BP1 is +0.095.

    What this evidence cannot show

    Single-gene CRISPR cannot detect paralogue buffering: under redundancy each single knockout is near-neutral and the profiles carry no signal, which is indistinguishable from irrelevance. The scan is cited for the paralogue split, not for the absence of co-dependency.

    Evidence that argues against this (1)
    • The absence of any co-dependency signal is at least as consistent with the family simply not mattering in these lines, and that reading requires no additional assumption. The hypothesis survives this data only because the data cannot test it, which is a weak position to be in. Oncofusion knowledge graph (internal analysis) 2026
  5. E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    A small-molecule IGF2BP3 inhibitor, I3IN-002, disrupts in situ IGF2BP3-target mRNA binding, alters IGF2BP3-dependent gene expression, and shows anti-leukaemic activity in a syngeneic MLL-Af4 transplantation model, with on-target engagement supported by thermal shift and DARTS assays.

    PMID 41307127DOI
    What this evidence cannot show

    Abstract read; full text not read. A tool compound, described by its own authors as paving the way for potent and selective inhibitors rather than being one. Paralogue selectivity is not stated in the abstract.

    Evidence that argues against this (1)
    • If the compound is not paralogue-selective, its efficacy is evidence for the pan-family hypothesis rather than for IGF2BP3 specifically - and if it is selective, its efficacy in the MLL-Af4 model argues against the redundancy hypothesis, because a buffered target should not be druggable alone. The abstract does not say which, so the compound currently cuts both ways. Oncofusion knowledge graph (internal analysis) 2026
Connected to