Oncofusion

start / graph / concept-what-transfers-from-an-mll-af9-result-and-what-does-not

Explainerstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q29-mllt3-yeats-loss

The transfer rule: which MLL-AF9 results can be carried into t(11;19) and t(4;11), and which are statements about a protein missing a reading module

also called MLL-AF9 import rules · YEATS asymmetry consequence

Most of what the field knows about these leukaemias was learned with one fusion protein, and that protein is missing a part that the next-commonest fusion keeps. Some findings carry across anyway; the ones about how the protein finds its genes do not.

The same thing, for a scientist

This graph already records that murine MLL-AF9 is its largest single import class, with 46 entries citing one of five keystone papers from it. The YEATS asymmetry adds a constraint that has not been stated: **MLL-AF9 is a KMT2A targeting module fused to a 70-residue disordered docking peptide, while MLL-ENL is the same targeting module fused to an essentially intact ENL including a functional acyl-lysine reader** [E1][E2]. The two are not the same class of molecule, and the difference is precisely a chromatin-reading module. From that, a transfer rule with three tiers. **Transfers:** anything mapped to the KMT2A moiety - menin binding, LEDGF tethering, CXXC targeting, PAF1c - because that half is identical across all three fusions; and bottleneck claims about the downstream programme, which identify requirements rather than contacts. **Does not transfer to t(11;19):** any claim of the form *this fusion's chromatin targeting is entirely KMT2A-driven*, because deleting ENL's YEATS from MLL-ENL impairs fusion-protein binding, lowers Eya1 and Meis1, reduces stem-cell frequency and lengthens latency [E1]; and any negative result about YEATS inhibitors, since YEATS-containing MLL-ENL cells are more sensitive to SGC-iMLLT [E1]. **Does not transfer to t(4;11):** everything routed through the AHD, because AFF1 has no AHD and does not bind DOT1L directly - the single largest mis-transfer in this literature. And one reverse constraint that is easy to miss: **DOT1L binds the ENL and AF9 AHDs with similar affinity** [E3], so the partner difference is not a DOT1L-affinity difference and cannot be explained as one.

Go deeper
The three tiers, as a table
claim typesource fusiontransfers to t(11;19)?transfers to t(4;11)?why
Menin binding, MBM1/MBM2MLL-AF9yesyesKMT2A residues 6-25, identical in all three
LEDGF/PSIP1 tethering, IBM1/IBM2MLL-AF9yesyesKMT2A residues 123-152, identical
CXXC CpG targetingMLL-AF9yesyesKMT2A 1147-1195, identical
PAF1c contactMLL-AF9yesyesmapped to sequence flanking CXXC
HOXA9/MEIS1 requirementMLL-AF9yes, with careyes, with carea bottleneck claim, not a contact claim
DOT1L recruitment via the AHDMLL-AF9yesnoAFF1 has no AHD and does not bind DOT1L directly
Graded AHD occupancy sets H3K79me2/me3MLL-AF9probablynosame reason
Chromatin targeting is KMT2A-drivenMLL-AF9NOunknownMLL-ENL's binding depends on its own YEATS reader
The fusion cannot read acetylationMLL-AF9NOn/aMLL-ENL can and does
A YEATS inhibitor has no fusion targetMLL-AF9NOn/aSGC-iMLLT is more active against YEATS-containing MLL-ENL cells
The fusion cannot self-associate through the partnerMLL-AF9NOnoAF9's Ser run is discarded; ENL's is retained and ENL phase-separates
The fusion has no partner-encoded nuclear import signalMLL-AF9NOunknownAF9's predicted NLS at 295-300 is discarded; ENL's fusion carries nearly the whole protein
The three specific mis-transfers worth naming

1. Targeting. The most consequential sentence in this literature is some version of 'the MLL moiety provides targeting and the partner provides elongation machinery'. That is a fair description of MLL-AF9 and a false description of MLL-ENL in 84.1% of its patients, where deletion of the partner's reader impairs fusion-protein chromatin binding and reduces target-gene output 1. In MLL-ENL the partner contributes targeting too.

2. Multivalency. This graph already records that the fusion cannot oligomerise through AF9's Poly-Ser region because that region is discarded. The paralogue does not share that constraint: ENL's equivalent region rides along in the fusion, and ENL is the family member reported to undergo phase separation to enhance SEC recruitment. So every condensate and hub argument about 'MLL fusions' is a statement about a property that one of the two commonest YEATS fusions structurally lacks.

3. The negative controls. A YEATS-domain inhibitor tested against MLL-AF9 cells has no fusion target and is acting on wild-type AF9 and ENL only. A null there says nothing about t(11;19). Conversely the two complete remissions in the pinometostat trial were both t(11;19) and the report does not state how many t(9;11) patients were exposed 4 - so the clinical DOT1L evidence and the preclinical DOT1L evidence are built on different fusions in opposite directions.

What would test the transfer directly, and why the obvious experiment is not enough

The obvious design is to put the AF9 YEATS domain back into MLL-AF9 in cis, which this graph already proposes as exp-yeats-cis-resupply-kmt2a-mllt3. That measures whether the reader is sufficient to change MLL-AF9's behaviour, and it confounds two variables: adding a folded 138-residue module changes the molecule's size, valency and expression as well as adding a reader.

The design that isolates the variable is a 2x2 paralogue swap, written up as exp-yeats-backbone-swap-matrix-mllt1-mllt3: AF9 backbone with and without its YEATS, ENL backbone with and without its YEATS, one vector, one progenitor pool, expression matched by immunoblot, with aromatic-cage-dead readers as the discriminating control in both YEATS-containing arms. That separates reader present or absent from which backbone it sits on, which no published experiment does.

One caution that should be stated before anyone runs it. A companion analysis in this beat finds that the intron geometry of the two loci predicts the direction of the YEATS asymmetry without invoking any selection at all - MLLT1 intron 1 is 9,013 bp and MLLT3 intron 1 is 1,410 bp, and uniform intronic breakage predicts 13.9% against 0.51% YEATS retention, against 2 observed retainers in 442 patients 5. If the asymmetry is breakage geography, the add-back experiment is measuring what a never-selected molecule does, which is interesting but is not a statement about what the leukaemia needs.

Facts and figures
source fusionKMT2A::MLLT3 (MLL-AF9)
import class entryimport-class-murine-mll-af9-myeloid
reverse constraintDOT1L binds the ENL and AF9 AHDs with similar affinity, so the partner difference is not a DOT1L-affinity difference
direct testexp-yeats-backbone-swap-matrix-mllt1-mllt3
prior partial testexp-yeats-cis-resupply-kmt2a-mllt3
confound on the whole programmeana-yeats-asymmetry-tested-against-intron-geometry
transfers
[
  "menin binding (KMT2A 6-25)",
  "LEDGF/PSIP1 tethering (KMT2A 123-152)",
  "CXXC CpG targeting (KMT2A 1147-1195)",
  "PAF1c contact",
  "HOXA9/MEIS1 bottleneck claims"
]
does not transfer to t 11 19
[
  "chromatin targeting is entirely KMT2A-driven",
  "the fusion cannot read histone acylation",
  "a YEATS inhibitor has no fusion target",
  "the fusion cannot self-associate through the partner",
  "the fusion has no partner-encoded nuclear import signal"
]
does not transfer to t 4 11
[
  "DOT1L recruitment through the AHD",
  "graded AHD occupancy sets H3K79 methylation",
  "any AHD-competition argument"
]
How this could bite us back
  • A drug programme justified by MLL-AF9 preclinical data and enrolled partner-agnostically is exposing t(11;19) and t(4;11) patients on the strength of a molecule that differs from theirs by a functional module or by an entire domain class.
  • The reverse is worse for the partner-specific strategies: a YEATS inhibitor developed on t(11;19) rationale has no fusion target in 99.5% of t(9;11) patients, who are the largest named partner group in the menin-inhibitor registration trial.
  • If this transfer rule is right, the field's habit of writing 'MLL-rearranged leukaemia' as a single entity in preclinical papers is the mechanism by which the mis-transfers propagate, and no amount of care at the point of reading fixes it.
Still open
  1. How many entries in this graph make a targeting claim sourced from MLL-AF9 and applied to KMT2A::MLLT1 or KMT2A::AFF1 without qualification? The import-class entry counts citations, not claim transfers.
  2. Does MLL-ENL's YEATS-dependent chromatin binding survive in a human t(11;19) line, or is it a property of the murine retroviral model?
  3. Is there any assay in which MLL-AF9 and MLL-ENL have been run side by side in one background with matched expression? This beat found none.
Every source behind this entry (6)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The ENL YEATS domain is retained in 84.1% of MLL-ENL patients and is crucial for MLL-ENL-mediated leukemogenesis: its deletion impairs fusion-protein binding and decreases expression of pro-leukemic genes such as Eya1 and Meis1, point mutations disrupting YEATS binding to acetylated histones decrease stem-cell frequency and increase leukemia latency, and YEATS-containing MLL-ENL leukemic cells show increased sensitivity to the YEATS inhibitor SGC-iMLLT; by contrast 98.4% of 442 t(9;11) patients break downstream of MLLT3 exon 4 and exclude the AF9 YEATS domain.

    PMID 36435883DOI
    What this evidence cannot show

    Read verbatim from the PMC HTML body (PMC11246743) on 2026-09-20 after `fullTextXML` returned HTTP 500. The functional work is retroviral MLL-ENL in murine progenitors with transplantation - a species and lineage import in its own right - and the SGC-iMLLT comparison is against control AML cells rather than against an isogenic YEATS-deleted line.

    Evidence that argues against this (1)
    • A YEATS-deletion phenotype in MLL-ENL could reflect loss of a 138-residue folded module rather than loss of acyl-lysine reading; the authors address this with reader-dead point mutants, which is the right control, but those mutants were scored on stem-cell frequency and latency rather than on the full chromatin-binding readout. Hu H et al. 2023
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    The AF9 ANC1 homology domain is intrinsically disordered and recruits multiple transcription factors through coupled folding and binding, functioning as a dynamic signalling hub rather than a structured module.

    PMID 23260655DOI
    What this evidence cannot show

    Isolated peptides in vitro, no fusion protein and no chromatin. Cited here to characterise what the AF9 moiety of the fusion is, not to support any claim about its activity in a cell.

    Evidence that argues against this (1)
    • Describing the retained AF9 fragment as 'a 70-residue disordered docking peptide' understates it if the retained region in patients extends from residue ~376 rather than ~499, which would add roughly 120 disordered residues of unknown function between the breakpoint and the AHD. Hu H et al. 2023
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    DOT1L peptides bind ENL with affinities similar to those measured for AF9, so the two paralogous AHDs do not differ appreciably in DOT1L affinity; and structure-guided mutation of the AF9 AHD gives graded rather than binary loss of DOT1L recruitment, with graded consequences for H3K79 methylation and transformation.

    PMID 25921540DOI
    What this evidence cannot show

    Read in full text from the PMC HTML page (PMC4426023) on 2026-09-20. Fluorescence polarization on MBP-tagged isolated domains against synthetic peptides; the ENL comparison is in a supplementary figure. The transformation readout is serial replating in murine progenitors.

    Evidence that argues against this (1)
    • Equal affinity in a peptide assay does not entail equal recruitment in a cell, where local concentration, competing partners and chromatin residence differ; this graph records that ENL's DOT1L and AF4-family contacts are mutually exclusive, which is a difference in competition rather than in affinity. Wan L et al. 2017
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In the 51-patient phase 1 of the DOT1L inhibitor pinometostat the only two complete remissions were both in t(11;19) patients, and the report does not state how many t(9;11) patients were enrolled.

    PMID 29724899DOI
    What this evidence cannot show

    Read in full text from the PMC HTML page (PMC6265654) on 2026-09-20. A dose-escalation study with no efficacy power; the absence of t(9;11) reporting is an absence of reporting, not evidence about t(9;11) responses.

    Evidence that argues against this (1)
    • The authors themselves offer schedule inadequacy rather than fusion identity as the first explanation for modest activity, and the maximum tolerated dose was not reached, so the partner pattern in two responders is a very weak signal. Stein EM et al. 2018
  5. E5moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database

    In hg38, MLLT1 intron 1 is 9,013 bp and MLLT3 intron 1 is 1,410 bp while the two genes have identical exon-to-residue mapping over the YEATS domain (exon 1 aa 1-4, exon 2 aa 5-64, exon 3 aa 65-92, exon 4 aa 93-140), so uniform intronic breakage predicts 13.9% YEATS retention for MLLT1 against 0.51% for MLLT3 - 2.3 patients in a cohort of 442, against 2 observed.

    What this evidence cannot show

    Computed in this beat from one canonical RefSeq transcript per gene; intron length is a poor proxy for breakage probability, and the same analysis shows the model fails badly on the partial-YEATS class in both genes. Full method and failure modes are in `ana-yeats-asymmetry-tested-against-intron-geometry`.

    Evidence that argues against this (1)
    • The same computation over-predicts partial YEATS inclusion fourfold in MLLT1 and at least fiftyfold in MLLT3, so uniform breakage is demonstrably wrong as a model and the agreement on the retention class may be coincidence. Hu H et al. 2023
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    MLLT3 is required for human haematopoietic stem-cell self-renewal and engraftment, whereas ENL but not AF9 is required for disease maintenance in acute myeloid leukaemia - so the two wild-type proteins differ in requirement in opposite directions from their fusion architectures.

    MLLT3 governs human haematopoietic stem-cell self-renewal and engraftment. · Calvanese V, Nguyen AT, Bolan TJ, et al. · Nature 2019
    PMID 31776511DOI
    What this evidence cannot show

    Two different systems: human cord-blood HSPC transplantation for MLLT3 and CRISPR depletion in AML lines and mouse models for the ENL/AF9 comparison. The juxtaposition is this entry's, not either paper's.

    Evidence that argues against this (1)
    • Several of the AML lines in which AF9 was found dispensable are themselves KMT2A::MLLT3, where an AF9 knockout removes the wild-type allele while leaving the fusion's AHD intact - so 'AF9 is dispensable' in those lines is partly a statement about redundancy with the fusion. Wan L et al. 2017
Connected to