AKAP95's requirement in KMT2A-rearranged leukaemia runs entirely through IDR co-condensation with the retained fragment, and not at all through DPY30 - so a DPY30-binding-dead AKAP95 will fully rescue
also called AKAP95 route hypothesis · condensate-not-WRAD
AKAP95 touches the KMT2A machinery in two ways. We propose that only one of them - the droplet-forming one - matters for the leukaemia, and that an AKAP95 which cannot grip DPY30 will work perfectly well in a leukaemia cell.
The same thing, for a scientist
**Statement.** In a KMT2A-rearranged cell, AKAP95's contribution is mediated by IDR1 (101-390) co-condensation with the retained KMT2A N-terminal fragment, and the AKAP95-DPY30 contact contributes nothing measurable. **Prediction:** an AKAP95 allele that cannot bind DPY30 but retains IDR1 will rescue an AKAP95 knockdown in THP-1, MOLM-13 and MV4-11 to wild-type levels, while a condensation-dead allele that retains DPY30 binding will not. **Prior probability 0.6**, on three grounds: (i) the DPY30 route requires a SET domain the fusion does not have, and the only enzymatic demonstration is on KMT2B, not KMT2A [E1]; (ii) the 2026 paper's rescue experiment already shows the ZF- and PS-mutants failing while making no DPY30 argument at all [E2]; (iii) `hyp-wrad-reaches-kmt2a-fusions-only-as-a-second-shell-passenger` and `concept-dpy30-what-is-established-for-wild-type-kmt2a-and-what-for-the-fusion` independently place DPY30 outside the fusion's first shell. **The reason it is not 0.9** is that the two AKAP95 regions overlap in 101-210, no published construct separates them, and the 2013 negative control AKAP95(101-692) may have been condensation-competent all along - which would mean the existing data are consistent with either answer.
What would kill it, and what a partial result would mean
Cleanly falsified if a DPY30-binding-dead, condensation-competent AKAP95 fails to rescue while wild-type does, in at least two MLLr lines of different fusion class (one AF9-type, one AFF1-type), with equal expression by immunoblot.
Cleanly confirmed if that allele rescues to within 15% of wild-type in both lines, and a condensation-dead allele that retains DPY30 binding does not.
The awkward middle is the likely outcome and needs pre-registering. If both alleles partially rescue, the honest reading is that the two routes are additive, not that one is dominant - and the hypothesis as stated ('contributes nothing measurable') is refuted even though the condensate route is the larger term. Pre-register that reading now, because the temptation afterwards will be to report 'the condensate route dominates' as confirmation.
A third possibility that neither arm expresses: the DPY30 contact might matter in the normal cell and not in the leukaemic one, which is the configuration that would make it therapeutically interesting in exactly the opposite direction - a route to spare. Nothing in either arm of the experiment detects that; it needs a normal CD34+ arm.
The cheap prior experiment that should run first
Before building any mutant, assay AKAP95(101-692) for MLL1T co-condensation. It is an existing 2013 construct. The 2026 fragment mapping (IDR1 = 101-390) predicts it condenses normally.
- If it condenses, then the 2013 paper's central negative control was condensation-competent, its coactivation defect is attributable to the DPY30 route alone, and the two routes are already partly separated by a reagent that has existed for thirteen years - which would raise this hypothesis's prior and make the full mutant series unnecessary.
- If it does not condense, IDR1's boundary is wrong or context-dependent, the PS-mutant is uninterpretable as a route-specific reagent, and the whole separation-of-function programme needs a different design.
Either answer is worth having and the experiment is one transfection.
| statement | In a KMT2A-rearranged cell, AKAP95's requirement is mediated by IDR1 (101-390) co-condensation with the retained KMT2A N-terminal fragment, and the AKAP95-DPY30 contact contributes nothing measurable to the phenotype. |
|---|---|
| origin | literature-gap |
| prior probability | 0.6 |
predictions
[ "A DPY30-binding-dead, condensation-competent AKAP95 rescues an AKAP95 knockdown in THP-1, MOLM-13 and MV4-11 to within 15% of wild-type rescue, at equal expression by immunoblot.", "A condensation-dead, DPY30-binding-competent AKAP95 does not rescue in the same lines.", "AKAP95(101-692), the 2013 negative control, co-condenses with MLL1T normally, because IDR1 is 101-390 and is intact in it.", "DPY30 L69D, which abolishes AKAP8 binding, does NOT phenocopy AKAP95 loss in an MLLr line - although it will have its own WRAD-assembly phenotype, so the comparison needs an AKAP95-binding-selective DPY30 allele if one can be found." ]
experiments
[ "exp-separate-the-dpy30-route-from-the-mll1t-route" ]
analyses
[ "ana-q166-per-claim-verification-of-pmc3813012" ]
falsifiers
[ "A DPY30-binding-dead, condensation-competent AKAP95 allele fails to rescue AKAP95 knockdown in at least two MLLr lines of different fusion class (one KMT2A::MLLT3 or MLL-AF9-type, one KMT2A::AFF1-type such as MV4-11 or SEM), at matched expression, with the rescue deficit exceeding 30% of the wild-type rescue in both. PRE-REGISTERED READING OF THE MIDDLE OUTCOME: if both single-route-dead alleles partially rescue, the hypothesis as stated is REFUTED - the routes are additive - and must not be rewritten as 'the condensate route dominates'.", "AKAP95(101-692) is found NOT to co-condense with MLL1T. That would mean the IDR1 boundary is wrong or context-dependent, the PS-mutant is not a route-specific reagent, and the mapping this hypothesis rests on is unsafe.", "An AKAP95 allele that cannot condense but retains DPY30 binding rescues fully. That inverts the hypothesis." ]
- If this hypothesis is right, every AKAP95 result obtained through the DPY30/WRAD frame - which is how this protein entered the field and how this graph first recorded it - is about the wrong route, and a programme built on H3K4 methylation readouts will measure a real signal that is not the one carrying the phenotype.
- If it is wrong and both routes contribute, then an AKAP95-directed agent is a WRAD-pathway agent as well as a condensate agent, and the clean therapeutic window from the Akap95-/- mouse becomes much harder to believe, because WRAD is required in normal haematopoiesis.
- Is there an AKAP95 point mutant that kills DPY30 binding without touching IDR1? The two published mappings disagree about where the DPY30 determinant is, so the reagent cannot be designed until that is settled - which makes the head-to-head mapping experiment a prerequisite, not a parallel task.
- Does the answer differ by fusion class? Every mechanistic experiment is in AF9-class fusions. KMT2A::AFF1 joins the super elongation complex and has a different condensate environment, so the two routes may not be weighted the same way in the index fusion.
- What is the AKAP95 requirement in a KMT2A-rearranged cell under menin inhibition? If menin inhibition displaces the fusion from chromatin, the co-condensation route should lose its anchor and AKAP95 should become dispensable - a strong, cheap, and completely untested prediction of this hypothesis.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The AKAP95-DPY30 route's only functional demonstration is enhancement of H3K4 methylation by the MLL2/KMT2B complex on recombinant chromatin; the paper's Discussion states that whether AKAP95 enhances methylation by other MLL complexes remains to be examined, and no MLL1/KMT2A methyltransferase assay with AKAP95 exists. KMT2A::AFF1 loses the SET domain.
Regulation of transcription by the MLL2 complex and MLL complex-associated AKAP95. · Jiang H, Lu X, Shimada M, Dou Y, Tang Z, Roeder RG · Nat Struct Mol Biol 2013What this evidence cannot show
Full text read; supplementary unavailable. The SET-domain loss is this graph's standing position from fusion-kmt2a-aff1 and is not in dispute in the field.
Evidence that argues against this (1)
- This item establishes that Route A cannot act on the fusion PROTEIN, and the hypothesis silently upgrades that to 'Route A contributes nothing to the disease'. The upgrade does not follow. A KMT2A-rearranged blast retains a wild-type KMT2A allele and normal KMT2B, so AKAP95 could support the leukaemic transcriptional programme through Route A acting on those enzymes, and the hypothesis would be false with this evidence item entirely intact. The rescue experiment is designed to detect exactly that, which is why a partial rescue by both alleles is pre-registered as a refutation rather than a nuance. Jiang H et al. 2013
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
In AKAP95-knockdown THP-1 and MOLM-13 cells, AKAP95 WT but not the zinc-finger mutant or the phase-separation mutant rescues the growth defect, and only AKAP95 WT partially co-localises with endogenous MLL1 N-terminal fragments. Neither mutant was tested for DPY30 binding, and no DPY30-binding mutant appears anywhere in the study.
AKAP95 condensates regulate transcription and can be targeted in MLL-fusion driven oncogenesis. · Yan X, Luan Y, Zhu Y, Cai Y, Ye F, Yang L, Zhu K, Hu J, Luo S, Lin J, Wang Y, Huang Y, Sun P, Yang Z, Shi B, Jiang H, et al. · Nat Commun 2026What this evidence cannot show
Full text read. The expected_absent_terms entry is the point: the 2026 paper never mentions DPY30 in its rescue design, so its mutant series cannot distinguish the two routes. Rescue was scored as proliferation; no transcriptional or chromatin rescue readout is reported for the mutants.
Evidence that argues against this (1)
- The PS-mutant's six tyrosines lie in aa101-210, inside the region that Route A also requires, so its failure to rescue is equally consistent with a DPY30-route defect. The experiment that is claimed to support the condensate route does not exclude the WRAD route. Yan X et al. 2026
the hypothesis is the claim that Route B carries everything
if the DPY30 route contributes nothing, AKAP95 is another case of WRAD not reaching the fusion - but by a route that does
three gated stages; Stage 0 is one transfection that re-reads the 2013 negative control