Oncofusion

start / graph / exp-yeats-cis-resupply-kmt2a-mllt3

Experimentmoderate confidence 3 of 5: One strong study, or several smaller consistent ones.Q11b-subtraction

Restore the AF9 YEATS reader in cis to MLL-AF9, and read out the ENL patients who already did the experiment for us

also called YEATS add-back MLL-AF9 · AF9 YEATS restoration · ENL exon-6/7 natural experiment

Nearly all MLL-AF9 leukemias have thrown away a reader module that the near-identical ENL fusion keeps. Put it back and see whether the leukemia gets weaker — and in parallel, find the rare patients whose ENL fusion threw it away and see whether their disease behaves like AF9.

The same thing, for a scientist

Two-arm programme. (a) In cis restoration: build MLL-AF9 variants restoring the AF9 YEATS domain (residues 1-138), with an aromatic-cage acetyl-lysine-binding-dead mutant as the discriminating control, and score immortalisation, Hoxa9/Meis1 output, H3K79me2 distribution and transplant latency. (b) Human natural experiment: assemble MLL-ENL cases with exon-6/7 breakpoints that exclude YEATS (~16% of patients) and compare lineage, age, outcome and, where material exists, SGC-iMLLT sensitivity against YEATS-retaining cases. Arm (b) costs almost nothing and tests the same hypothesis in humans.

Go deeper
The question

AF9 and ENL are close paralogues with homologous YEATS domains, fused to the same 5' partner, and their patient breakpoints sit at opposite ends of the protein: ~84% before ENL residue 5, ~92% at or after AF9 residue 376 1. For ENL the retained YEATS is load-bearing — deleting it reduces stem-cell frequency and lengthens latency 2. For AF9 nobody has asked the mirror question.

Two hypotheses are live. Functional selection: ENL YEATS helps and AF9 YEATS hinders, so patients are selected accordingly, and MLLT3 is a subtraction partner. Structural accident: the breakpoint distributions reflect the local recombination landscape at MLLT1 versus MLLT3, and the functional asymmetry is downstream noise.

Arm A — in cis restoration in MLL-AF9

Constructs (retroviral, murine myeloid progenitors, expression matched by immunoblot):

  1. MLL-AF9 patient-like (AF9 from ~residue 376, no YEATS). Positive control.
  2. MLL-YEATS-AF9 — AF9 YEATS (1-138) restored in cis.
  3. MLL-YEATS(cage-dead)-AF9 — the same, with aromatic-cage mutations abolishing acetyl-lysine binding while preserving the fold. The discriminating arm.
  4. MLL-linker-AF9 — inert spacer of matched length.

Readouts: serial replating; Hoxa9/Meis1; H3K79me2 and H3K9ac CUT&RUN with co-occupancy analysis (the specific prediction is that restored YEATS redistributes fusion occupancy and H3K79me2 towards H3K9ac-marked loci); transplant latency.

Interpretation: suppression by arm 2 but not arm 3 confirms an acetyl-reading constraint and makes MLLT3 a second confirmed subtraction partner. Suppression by both is steric. No suppression by either is the single most informative outcome for this project, because it would collapse the subtraction class back to MLLT10 and would mean the YEATS parallel — the reason this beat was commissioned at the size it was — is not a mechanism at all.

Arm B — the patients who already ran the experiment

Roughly 16% of MLL-ENL patients have breakpoints that do not fall at or before ENL residue 5 1, and this graph records an exon-6/7 architecture that excludes YEATS entirely (bp-kmt2a-intron-11-mllt1-e6-e7), producing a fusion architecturally equivalent to MLL-AF9.

Those patients are a human control group for the entire YEATS argument, and nobody has assembled them. The analysis requires no new wet-lab work: retrieve breakpoint-typed t(11;19) cases from the KMT2A recombinome resources and the collaborating diagnostic centres this graph already lists (collab-marschalek-meyer-frankfurt), stratify by YEATS retention, and compare lineage, age at diagnosis, and event-free survival.

Prediction if YEATS retention is load-bearing: the YEATS-excluding minority should resemble MLL-AF9 in lineage skew and should not respond to SGC-iMLLT. Prediction if the breakpoint distribution is a structural accident: no difference.

Arm B is cheap, fast and uses an existing cohort. It should be run first.

Facts and figures
designTwo arms. (A) Four-construct in-cis restoration study in murine myeloid progenitors with transplant follow-on: MLL-AF9 patient-like, MLL-YEATS-AF9, MLL-YEATS(aromatic-cage-dead)-AF9, MLL-linker-AF9. (B) Retrospective stratification of breakpoint-typed MLL-ENL cases by YEATS retention (exon-2 versus exon-6/7 junctions), comparing lineage, age at diagnosis and event-free survival, plus ex vivo SGC-iMLLT sensitivity where viable material exists.
readoutArm A: serial replating, Hoxa9/Meis1 transcripts, H3K79me2 and H3K9ac CUT&RUN co-occupancy, transplant latency. Arm B: lineage, age, EFS by YEATS-retention stratum; SGC-iMLLT IC50 by stratum.
controlsArm A: aromatic-cage-dead YEATS point mutant, inert matched-length linker, empty vector, expression matched by immunoblot. Arm B: YEATS-retaining exon-2 MLL-ENL cases as the comparator stratum; MLL-AF9 cases as the architectural reference. NOT controlled for in arm A: change in fusion self-association caused by inserting a folded 138-residue module (So 2003, PMID 12957285) - both the wild-type and cage-dead arms carry the same fold, so oligomerisation effects are common to both; and vector-capacity effects of enlarging an already large MLL cassette (Wachter 2014, PMID 25016062).
cost tierlow for arm B, med for arm A
timeline3-6 months for arm B using existing breakpoint-typed cohorts; 9-15 months for arm A
sequencing noteRun arm B first. It is cheap, uses cohorts that already exist, and a null result there substantially lowers the prior for arm A.
falsifieshyp-subtraction-is-a-partner-class
arm b denominator riskThe ~92% / ~84% breakpoint figures come from one group's review of reported patient breakpoints. An independent series found 5 of 11 t(9;11) patients with AF9 breakpoints outside both published cluster regions (Langer 2003, PMID 12619163), so arm B must re-type breakpoints rather than reuse published architecture calls.
data sources available in graph
[
  "collab-marschalek-meyer-frankfurt",
  "breakpoint-kmt2a-partner-exon-preference",
  "strategy-data-and-sample-access-map"
]
How this could bite us back
  • Arm B compares patients treated across decades and protocols, so an outcome difference between breakpoint strata would be badly confounded. Lineage and age are the more robust endpoints; survival should be treated as hypothesis-generating only.
  • If arm A is null, the single published counterexample (MLLT10) stops looking like the first member of a class and starts looking like an exception, and the strategic conclusions of this beat should be revised downward accordingly.
  • Both arms of this experiment can produce a confident-looking null for reasons unrelated to YEATS: arm A through oligomerisation or packaging artefacts (So 2003; Wachter 2014), arm B through breakpoint mis-assignment (Langer 2003). A null here should not be allowed to collapse the subtraction class until those three controls have been run.
Still open
  1. Are there enough breakpoint-typed exon-6/7 MLL-ENL cases in existing cohorts to power arm B? If t(11;19) is ~5% of KMT2A-r disease and the YEATS-excluding architecture is ~16% of that, the absolute numbers may force an international pooling effort.
  2. Should arm A also be run in the reverse direction on MLL-ENL in a myeloid rather than lymphoid context, to control for the possibility that the ENL and AF9 difference is about lineage rather than about the reader?
  3. Before arm B is powered, how many t(11;19) cases in the available cohorts have breakpoints typed at nucleotide resolution rather than assigned to a published cluster region? An independent series put 5 of 11 t(9;11) AF9 breakpoints outside the described BCRs (Langer 2003), so the YEATS-retention strata may be mis-assigned at a rate that swamps the effect being looked for.
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.review

    Approximately 84% of MLL-ENL patients have breakpoints at or before ENL amino acid 5 (including the YEATS domain), whereas approximately 92% of MLL-AF9 patients have breakpoints at or after AF9 amino acid 376 (excluding YEATS while retaining the AHD); the differential retention is posed by the field as an unanswered question.

    PMID 37295550DOI
    What this evidence cannot show

    A review summarising reported patient breakpoints rather than a prospectively typed cohort; the exact denominator for the 16% YEATS-excluding ENL minority is not given, which is itself part of what arm B must establish. Verified against PubMed on 2026-09-18 and full text read the same day.

    Evidence that argues against this (1)
    • The breakpoint denominators that arm B depends on are softer than the percentages suggest, and an independent cohort found substantial breakpoint heterogeneity in the AF9/MLLT3 partner itself. Sequencing t(9;11) breakpoints in 11 children and two cell lines, Langer et al. placed AF9 breakpoints within BCR1 in four patients and two cell lines and within BCR2 in two, but found five patients whose AF9 breakpoints lay OUTSIDE both previously described cluster regions - in the centromeric region of intron 4, and in one case in intron 3. Nearly half of a small independent series therefore falls outside the architecture from which the '~92% break at or after AF9 residue 376' figure is built. Arm B's stratification of MLL-ENL cases by YEATS retention will inherit the same problem: it requires breakpoint typing at a resolution that the historical literature has not uniformly applied. Langer T et al. 2003
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Deleting the YEATS domain from MLL-ENL impairs fusion-protein binding and reduces Eya1 and Meis1 expression; point mutations disrupting YEATS binding to acetylated histones decrease stem-cell frequency and increase MLL-ENL leukemia latency; YEATS-containing MLL-ENL cells are more sensitive to the YEATS inhibitor SGC-iMLLT than control AML cells.

    The ENL YEATS epigenetic reader domain critically links MLL-ENL to leukemic stem cell frequency in t(11;19) Leukemia · Hu H, Saha N, Yang Y, Ahmad E, Lachowski L, Shrestha U, Premkumar V, Ropa J, Chen L, Teahan B, Grigsby S, Marschalek R, Nikolovska-Coleska Z, Muntean AG · Leukemia 2022
    PMID 36435883DOI
    What this evidence cannot show

    Supplies the experimental template, the point-mutation strategy and the SGC-iMLLT readout, in the deletion direction and in the ENL context. Verified against PubMed on 2026-09-18.

    Evidence that argues against this (2)
    • Restoring a folded reader domain in cis changes more than reading. So et al. showed that coiled-coil oligomerisation domains of cytoplasmic MLL partners are necessary and sufficient for transformation, and that homodimerising MLL with a synthetic module that has no chromatin function reproduces bona fide MLL fusion activity including Hox activation and enhanced self-renewal. Inserting the 138-residue AF9 YEATS domain between MLL and the AF9 C-terminus alters spacing and self-association geometry in a chimera whose activation is known to be oligomerisation-sensitive. The aromatic-cage-dead arm controls for acetyl-lysine reading; it does not control for a change in oligomerisation, since both arms carry the same fold in the same position. If arms 2 and 3 suppress equally the study concludes 'steric' and stops, which is a null rather than a result. So CW et al. 2003
    • Arm A's readout is vulnerable to construct and vector effects independent of the biology. Wachter et al. built a size-unconstrained Sleeping Beauty system precisely because MLL fusion cDNA cassettes are large and hard to package into retro- or lentiviral particles, and found that among eight MLL alleles tested only 1 of 4 direct MLL fusions was oncogenic while 3 of 4 reciprocal fusions were. Adding ~414 bp of YEATS coding sequence to an MLL-AF9 cassette moves it further against the packaging ceiling, so a failure to transform in arm 2 is compatible with reduced titre or product truncation. Expression matching by immunoblot is necessary but does not exclude integration-site or product-integrity effects. Wachter K et al. 2014
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    The AF9 YEATS domain binds H3K9ac via a serine-lined aromatic sandwiching cage, colocalises with H3K9ac genome-wide, and that colocalisation is important for chromatin recruitment of DOT1L — supplying both the mutations for the cage-dead arm and the specific genomic prediction for the CUT&RUN readout.

    AF9 YEATS domain links histone acetylation to DOT1L-mediated H3K79 methylation · Li Y, Wen H, Xi Y, Tanaka K, Wang H, Peng D, Ren Y, Jin Q, Dent SYR, Li W, Li H, Shi X · Cell 2014
    PMID 25417107DOI
    What this evidence cannot show

    Intact AF9, not the fusion. Verified against PubMed on 2026-09-18.

    Evidence that argues against this (1)
    • The 'acetyl-reading' framing that the discriminating arm depends on is too narrow, and the cage-dead mutant is less discriminating than it looks. Li et al. showed the AF9 YEATS domain binds crotonyllysine with SELECTIVELY HIGHER affinity than acetyllysine, through an extended aromatic sandwiching cage whose crotonyl preference comes from pi-aromatic and hydrophobic contacts conserved across YEATS but not bromodomains, and that AF9 co-localises with crotonylated histone H3 in cells and positively regulates gene expression in a YEATS-dependent manner. Mutating the cage removes crotonyl reading as well as acetyl reading, so a null in the cage-dead arm cannot be attributed to H3K9ac; and the CUT&RUN prediction should be scored against H3K9cr as well as H3K9ac or it will be scored against the wrong mark. Li Y et al. 2016
Connected to