Hypothesis: KMT2A::AFF1 rewires splicing, because AFF1's own protein family are G-quadruplex RNA-binding splicing modulators and the fusion keeps AFF1
also called splicing landscape KMT2A-r · AFF family splicing · G-quadruplex RNA binding AFF1 · alternative splicing MLL-AF4
The second half of this fusion protein comes from a family of proteins whose day job includes deciding how RNA gets cut and stitched together. The fusion keeps that half, so it should change RNA stitching - but the RNA-stitching changes that have actually been found in this family of leukemias are explained by other proteins entirely, and nobody has tested whether the fusion's own half does anything.
The same thing, for a scientist
The AFF family (AFF1/AF4, AFF2/FMR2, AFF3/LAF4, AFF4/AF5q31) are nuclear RNA-binding proteins with high apparent affinity for G-quadruplex RNA structures; AFF2, AFF3 and AFF4 localise to **nuclear speckles** and modulate the splicing efficiency of a minigene containing a G-quadruplex in an alternatively spliced exon, and their overexpression disturbs nuclear speckle organisation [E1]. Three of the four family members (AFF1, AFF3, AFF4) form fusions with KMT2A [E1]. Since KMT2A::AFF1 retains essentially the whole AFF1 C-terminus, the fusion carries an intact RNA-binding, splicing-modulating module into a chromatin-tethered, hyper-elongating complex. **Correction and restructuring, cycle 35.** This node previously stated that no transcriptome-wide splicing analysis of KMT2A-rearranged leukaemia existed. That is wrong and is withdrawn. Three papers now bear directly on it, and together they move the general claim from open to substantially supported while moving the specific AFF1 mechanism from plausible to poorly supported: - A distinct transcriptome-wide splice-variant profile does separate KMT2A-r from KMT2A-wild-type paediatric T-ALL (n=171), but is attributed to the splicing regulators ESRP1 and MBNL3 and to CD44 promoter methylation, not to the fusion partner [E1 counter, PMID 39793727]. - The splicing dependency in KMT2A-r AML maps onto PTBP1 and RBM15, conventional RNA-binding proteins, whose loss dysregulates splicing genome-wide and collapses the KMT2A-r programme [E1 counter, PMID 42637878]. - Elongation-coupled splicing in MLL-r leukaemia runs through ECAS, an AFF4-ELL2-EAF1 module joined to a U2 snRNP-associated intron binding complex; AQR/XAB2 perturbation increases intron retention with a preference for long genes and cripples MLL-r leukaemia stem cell function in mouse cells [E2 counter, PMID 42748909]. So the surviving hypothesis is narrower and sharper than the one originally written: **splicing is rewired in KMT2A-rearranged leukaemia, and that is no longer in doubt; what is unsupported is that the rewiring runs through the retained AFF1 RNA-binding module rather than through AFF4, PTBP1/RBM15, trans-factor expression or a generic gene-length elongation effect.** No transcriptome-wide splicing analysis of KMT2A::AFF1 B-ALL specifically has been published that we can find, and AFF1 itself has never been tested for G-quadruplex RNA binding.
Why this is the most surprising gap in the sweep1 deeper
The field has spent twenty years on what the KMT2A half of the fusion does to chromatin. The AFF1 half is discussed almost entirely as an SEC-recruitment module — a docking site for AFF4, ELL, and P-TEFb. But the AFF family's independently characterised molecular function includes sequence-structure-specific RNA binding and splicing modulation 1. If you keep that module and tether it to thousands of actively elongating gene bodies, the default expectation is that co-transcriptional splicing changes.
The prediction is testable with data that already exists. Public RNA-seq of KMT2A::AFF1 primary infant ALL and of KMT2A::AFF1 lines with fusion knockdown (the Heidenreich group's siRNA systems) could be reanalysed for differential exon usage and intron retention without generating a single new sample.
The G-quadruplex angle sharpens it further: if the fusion's splicing effects are G-quadruplex-dependent, the affected exons should be enriched for predicted rG4 motifs, which is a computable, falsifiable prediction rather than a fishing expedition.
What this would mean for druggability
If confirmed, it opens a route that does not go through menin at all. An RNA-binding surface is a different pharmacology from a protein-protein interface: G-quadruplex-binding small molecules exist, and spliceosome-modulating compounds (E7107/H3B-8800 class) are clinical-stage. More importantly, it would predict a neoantigen source: fusion-specific mis-splicing generates novel junctions and therefore novel peptides in a disease whose defining problem is that it has almost no mutations and therefore almost no neoantigens (see comut-low-mutation-burden-infant-kmt2a-r). For an immunotherapy-oriented programme that is a potentially important consequence.
Honest grading
- Splicing is rewired in KMT2A-r leukaemia: established, cycle 35. A distinct splice-variant profile distinguishes KMT2A-r from KMT2A-wt paediatric T-ALL (n=171) [E1 counter]; PTBP1/RBM15 are KMT2A-r-biased dependencies whose loss dysregulates splicing [E1 counter]; AQR/XAB2 perturbation increases intron retention and destroys MLL-r leukaemia stem cell function [E2 counter].
- That it runs through AFF1: unsupported. Every mechanism so far identified is a different one - ESRP1/MBNL3 expression, CD44 promoter methylation, PTBP1, RBM15, and an AFF4-ELL2-EAF1 elongation module. None invokes AFF1.
- Direct evidence that the retained partner's family modulates splicing and binds G-quadruplex RNA: strong, but demonstrated for AFF2/AFF3/AFF4 rather than AFF1 itself 1. AFF1 was not the tested member. That was the precise weak link when this node was written and it is now the load-bearing one.
- KMT2A::AFF1 B-ALL splicing analysis: still none found. The published work is T-ALL and AML.
- The confound is severe. Perturbing almost anything in an MLL-r cell changes splicing: WDR5 WIN-site inhibitors alone produce an MDM4 splicing switch [E2 counter]. Attribution requires a fusion-knockdown arm, a partner-class comparison, and a control for gene length.
| statement | Splicing is rewired in KMT2A-rearranged leukaemia (now established). The open and specific claim is that in KMT2A::AFF1 the rewiring is driven in part by the retained AFF1 module's G-quadruplex RNA binding rather than solely by trans-factor expression (ESRP1/MBNL3, PTBP1, RBM15), by the AFF4-containing ECAS elongation-splicing module, or by a generic gene-length effect of altered elongation; and that if so it yields an rG4-enriched, partner-class-specific signature including novel junctions that constitute a neoantigen source in an otherwise mutation-poor cancer. |
|---|---|
| origin | literature-gap |
| prior probability | 0.2 |
| planned experiments note | Moved out of a reference field on 2026-09-19. It named a node that does not exist, which made the pointer read as a real cross-reference and, for hypotheses, silently satisfied the 'needs at least one experiment or analysis' rule. The intent is preserved here; the node still needs writing. |
predictions
[ "Recombinant AFF1 C-terminus binds G-quadruplex RNA in vitro with affinity comparable to AFF2/AFF3/AFF4 - the single foundational experiment", "Differential exon usage and intron retention in KMT2A::AFF1 B-ALL differ from those in KMT2A fusions with non-AFF partners (MLLT1, MLLT3, MLLT10) after matching for expression and gene length", "Exons whose usage changes in KMT2A::AFF1 are enriched for predicted RNA G-quadruplex motifs relative to matched unaffected exons, and this enrichment survives correction for gene length and GC content", "The KMT2A::AFF1 signature is not explained by ESRP1, MBNL3, PTBP1 or RBM15 expression differences, which must be measured in the same samples", "Fusion knockdown reverses the signature, while DOT1L inhibition and AQR perturbation produce a different, length-biased signature (separating an RNA-binding effect from elongation and chromatin-mark effects)", "Novel junctions predicted to be translated yield HLA-presentable peptides detectable by immunopeptidomics" ]
analyses
[ "ana-noncoding-gap-ledger-kmt2a-r" ]
falsifiers
[ "The AFF1 C-terminus does not bind G-quadruplex RNA, breaking the family-inference that the hypothesis rests on", "The KMT2A::AFF1 splicing signature is equally present in KMT2A fusions that do NOT retain an AFF-family partner, showing it is a generic consequence of the KMT2A half or of altered elongation", "The signature is fully accounted for by differential expression of ESRP1, MBNL3, PTBP1 or RBM15", "Affected exons show no rG4 enrichment once gene length and GC content are controlled, i.e. the changes are the generic long-gene intron-retention pattern seen on AQR perturbation", "Fusion knockdown does not reverse the splicing signature" ]
evidence grade by class
{
"splicing_rewired_in_kmt2a_r": "established (T-ALL cohort, AML CRISPR screen, ECAS/AQR)",
"direct_kmt2a_aff1_b_all_splicing_analysis": "none found",
"aff1_mechanism": "none; AFF1 itself untested for G-quadruplex RNA binding",
"competing_explanations": "ESRP1/MBNL3 expression; CD44 promoter methylation; PTBP1/RBM15; AFF4-ELL2-EAF1/AQR elongation module with long-gene intron-retention bias",
"retained_partner_family_function": "strong but tested on AFF2/3/4, not AFF1",
"splicing_factor_mutations_in_disease": "absent"
}prior probability history
[
{
"cycle": 1,
"value": 0.45,
"basis": "AFF-family G-quadruplex RNA binding plus disrupted H3K36me3 at fusion targets"
},
{
"cycle": 35,
"value": 0.2,
"basis": "downgraded: the KMT2A-r splicing signatures now published are attributed to ESRP1/MBNL3 (PMID 39793727), PTBP1/RBM15 (PMID 42637878) and an AFF4-ELL2-EAF1/AQR elongation module with a gene-length bias (PMID 42748909). The general 'splicing is rewired' claim was promoted to established and removed from the hypothesis; what remains is the AFF1-specific mechanism, and it has acquired three competing explanations and no supporting data."
}
]corrections
[
{
"cycle": 35,
"date": "2026-09-19",
"withdrawn": "'we could find no transcriptome-wide splicing analysis of KMT2A::AFF1 leukemia' as a statement about KMT2A-rearranged leukaemia generally, and the framing of 'splicing is rewired' as the open question",
"replaced_with": "splicing is rewired in KMT2A-r leukaemia (established); the open claim is the AFF1-specific mechanism, and no KMT2A::AFF1 B-ALL splicing analysis specifically has been found",
"source": "PMID 39793727 (T-ALL, n=171), PMID 42637878 (PTBP1/RBM15 in KMT2A-r AML), PMID 42748909 (ECAS/AQR in MLL-r)"
}
]planned experiments
[ "exp-noncoding-dependency-crispr-screen-kmt2a-aff1" ]
- Spliceosome-modulating drugs are broadly toxic and have narrow therapeutic windows in adults; an infant is the least forgiving setting for that class.
- If the splicing signature turns out to be a generic consequence of transcriptional hyper-elongation, it is not fusion-selective and provides no therapeutic handle.
- G-quadruplex-binding small molecules are notoriously promiscuous, binding DNA and RNA G4s genome-wide, so selectivity claims would need unusually careful validation.
- Neoantigens from mis-splicing may be subclonal or lost under therapy, so a vaccine strategy built on them risks immune escape by the same route as any single-antigen approach.
- Splicing changes are trivially easy to induce in an MLL-r cell - WDR5 inhibition alone produces an MDM4 splicing switch - so a splicing signature found without a fusion-knockdown arm will be uninterpretable and may be a drug or elongation artefact rather than fusion biology.
- Does AFF1 itself bind G-quadruplex RNA? A single in-vitro binding experiment on the retained C-terminus decides whether this hypothesis has a foundation.
- Is there a differential-splicing analysis of KMT2A::AFF1 **B-ALL** specifically? The published KMT2A-r splicing work is T-ALL and AML; the B-lineage infant case the lab cares about is still unexamined.
- Do KMT2A fusions with AFF-family partners (AFF1, AFF3, AFF4) differ from non-AFF partners (MLLT1, MLLT3, MLLT10, ELL, AFDN) in splicing phenotype? If so, partner class predicts a whole biological axis nobody is stratifying on.
- Could fusion-specific mis-splicing supply neoantigens in a cancer with almost no somatic mutations, and would they be shared across patients and therefore vaccine-tractable?
- Is the intron-retention phenotype in MLL-r leukaemia (AQR/ECAS) simply the generic long-gene signature of impaired elongation? If so it is not partner-specific and this hypothesis has no independent content; gene length must be a covariate in every analysis proposed here.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
The AFF family (AFF1/AF4, AFF2/FMR2, AFF3/LAF4, AFF4/AF5q31) are nuclear RNA-binding proteins; AFF3 and AFF4, like AFF2, localise to nuclear speckles, bind RNA with high apparent affinity for G-quadruplex structures, modulate in vivo the splicing efficiency of a minigene containing a G-quadruplex in an alternatively spliced exon, and when overexpressed interfere with nuclear speckle organisation or biogenesis; three family members form fusions with MLL/KMT2A in acute lymphoblastic leukemia.
Functional characterization of the AFF (AF4/FMR2) family of RNA-binding proteins: insights into the molecular pathology of FRAXE intellectual disability · Melko M, Douguet D, Bensaid M, Zongaro S, Verheggen C, Gecz J, Bardoni B · Human Molecular Genetics 2011What this evidence cannot show
The speckle-localisation and splicing-modulation experiments were performed on AFF2, AFF3 and AFF4, not on AFF1 itself; AFF1 relevance is inferred from family membership and is therefore our extrapolation. Work is in HeLa and fibroblasts with a reporter minigene, not in haematopoietic cells, and the disease framing is FRAXE intellectual disability rather than leukemia. Overexpression artefacts are a real risk for speckle-morphology claims.
Evidence that argues against this (2)
- Somebody did look, and the answer they got does not need AFF1. In 171 paediatric T-ALL samples, KMT2A-rearranged cases show a distinct transcriptome-wide gene- and splice-variant expression pattern versus KMT2A-wild-type, including the oncogenic CD44 variant 3. The authors attribute it to expression of the splicing regulators ESRP1 and MBNL3 and to activating methylation of the CD44 promoter - a trans-factor and chromatin account, with no role for the fusion partner's RNA-binding module and no G-quadruplex involvement reported. Two things follow. (i) This node's claim that no transcriptome-wide splicing analysis of KMT2A-rearranged leukaemia exists is withdrawn; the surviving narrower claim is that none exists for KMT2A::AFF1 B-ALL. (ii) A KMT2A-r splicing signature can be produced without invoking the AFF1 half at all, so observing one in KMT2A::AFF1 would not by itself support this hypothesis. The cohort is T-ALL, and the abstract does not give the partner-gene composition, so this is a caution rather than a refutation of the AFF1 mechanism. Ramilo Amor A et al. 2025
- A domain-focused CRISPR screen across cancers found the splicing factor PTBP1 and the m6A reader RBM15 as dependencies in AML with a bias toward the KMT2A-rearranged subtype; all four PTBP1 RNA-binding domains are required for KMT2A-r AML proliferation, and PTBP1 loss causes widespread splicing dysregulation and disrupts the KMT2A-r-essential transcriptional programme. The splicing dependency that actually exists in KMT2A-r leukaemia has so far mapped onto conventional RNA-binding proteins, not onto the retained fusion partner - and it was found in AML, where the partner is generally not AFF1. Any AFF1-specific splicing effect must be demonstrated on top of this, not instead of it. Cao Z et al. 2026
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper
KMT2A fusion binding spreads into gene bodies and disrupts normal H3K36me3 patterning, the chromatin mark most tightly coupled to co-transcriptional exon definition, giving a chromatin-level reason to expect altered splicing at fusion targets.
MLL-AF4 Spreading Identifies Binding Sites that Are Distinct from Super-Enhancers and that Govern Sensitivity to DOT1L Inhibition in Leukemia · Kerry J, Godfrey L, Repapi E, Tapia M, Blackledge NP, Ma H, Ballabio E, O'Byrne S, Ponthan F, Heidenreich O, Roy A, Roberts I, Konopleva M, Klose RJ, Geng H, Milne TA · Cell Reports 2017What this evidence cannot show
The paper reports disrupted H3K36me3 patterns but performs no splicing analysis; the link from H3K36me3 disruption to altered splicing is established in the general literature but is an inference here.
Evidence that argues against this (2)
- Elongation-coupled splicing in MLL-rearranged leukaemia has now been given a mechanism that is not AFF1's RNA binding. Xiang et al. identify ECAS, a bipartite chromatin assembly linking an elongation module (AFF4-ELL2-EAF1) to a U2 snRNP-associated intron binding complex and explicitly distinct from the canonical P-TEFb-containing SEC. Acute perturbation of AQR/XAB2 impairs productive gene-body elongation without altering pause release, increases intron retention and preferentially affects long genes; in mouse haematopoietic cells Aqr perturbation profoundly impairs MLL-rearranged leukaemia initiation, maintenance and leukaemia stem cell function. This both strengthens and complicates the node. It strengthens the premise that elongation and splicing are coupled in this disease and that the coupling is functionally required. It complicates the mechanism badly: the module doing the coupling is AFF4, not AFF1; the effect is a gene-length effect, which is the classic non-specific signature of impaired elongation and is exactly what an rG4-driven mechanism would need to be distinguished from; and the leukaemia data are in mouse cells with no fusion partner named in the abstract. Xiang Y et al. 2026
- A worked example of the confound this hypothesis is most exposed to. In MLL-rearranged cells, WDR5 WIN-site inhibitors produce changes in alternative mRNA splicing that inactivate the p53 antagonist MDM4 - a real, functionally consequential splicing change in exactly this disease that is entirely drug-imposed and downstream of nucleolar stress. Splicing changes observed in a KMT2A-r cell are therefore easy to generate by perturbation and hard to attribute to the fusion; a fusion-knockdown arm is not optional. Howard GC et al. 2024
Correction and counter (Ramilo Amor 2025, PMID 39793727): a transcriptome-wide splice-variant analysis of KMT2A-r leukaemia does exist (171 paediatric T-ALL) and attributes the KMT2A-r signature to the splicing regulators ESRP1 and MBNL3 and to CD44 promoter methylation, not to the fusion partner - so a KMT2A-r splicing signature can arise with no AFF1 involvement
Counter-evidence (Cao 2026, PMID 42637878): the splicing dependency in KMT2A-r leukaemia maps onto conventional RNA-binding proteins - PTBP1 and RBM15 are KMT2A-r-biased AML dependencies and PTBP1 loss dysregulates splicing genome-wide - rather than onto the retained fusion partner
Counter-evidence (Xiang 2026, PMID 42748909): elongation-coupled splicing in MLL-r leukaemia runs through ECAS, an AFF4-ELL2-EAF1 module joined to a U2 snRNP-associated intron binding complex; AQR/XAB2 perturbation increases intron retention with a long-gene preference and cripples MLL-r leukaemia stem cell function in mouse cells - AFF4 and gene length, not AFF1 G-quadruplex binding
- ← related toWDR5 — the partner that is not a partner: it binds the half of KMT2A the fusion throws away
Confound counter (Howard 2024, PMID 38682900): WDR5 WIN-site inhibition alone produces alternative-splicing changes that inactivate MDM4 in MLL-rearranged cells - a drug-imposed splicing change in exactly this disease, showing that a splicing signature without a fusion-knockdown arm cannot be attributed to the fusion
AFF1 has its own splicing connections, so the fusion may rewire splicing
Named by this hypothesis as the analysis that tests it; the reference was dangling until now.
SRRM4-dependent alternative splicing inactivates REST by producing isoform 3
named in that entry's planned_experiments