The YEATS asymmetry is breakage geography, not selection: the reader's fate is decided by where topoisomerase II cuts, before any leukaemia exists
also called geography hypothesis · YEATS accident hypothesis
The two near-identical partner genes lose or keep their reader module because of where in the DNA the chromosome happens to break, not because one version makes better cancer. If that is right, the difference is a fact about chromosome breakage and not about what the leukaemia needs.
The same thing, for a scientist
This graph's `subtraction-mllt3-yeats` names two competing explanations for the YEATS asymmetry - functional selection or structural accident - and leaves the verdict open. This hypothesis takes the accident side and makes it specific enough to kill. **Claim: the fraction of patients retaining the partner's YEATS domain is set by where the breakage hotspot of each locus sits relative to the YEATS-coding exons, and requires no selection on the resulting protein.** Three observations make it more than a null. The two genes have **identical exon-to-residue mapping over the reader** (exon 1 aa 1-4, exon 2 aa 5-64, exon 3 aa 65-92, exon 4 aa 93-140) but a 6.4-fold difference in intron 1 length (9,013 bp against 1,410 bp) [E1]. Uniform intronic breakage predicts 13.9% YEATS retention for MLLT1 against 0.51% for MLLT3 - and 0.51% of 442 is 2.3 patients against **2 observed** [E1][E2]. And MLLT3 is the **commonest partner in therapy-induced KMT2A leukaemia** (49 of 142), with its adult KMT2A-side breakpoint shift explicitly attributed to therapy induction [E3] - i.e. this locus's breakpoint distribution demonstrably responds to a DNA-damaging exposure rather than to a protein-level advantage. The hypothesis is prior-probability moderate rather than high, because the same computation **fails** on the partial-YEATS class in both genes, which means breakage is not uniform and the geography must be hotspot-shaped rather than length-shaped.
What the hypothesis predicts
- A topoisomerase-II cleavage map over MLLT1 and MLLT3 in etoposide-treated human haematopoietic progenitors will show hotspots that coincide with the observed patient breakpoint zones - concentrated 5-prime of the MLLT1 reader (intergenic between ACER1 and MLLT1, and intron 1) and 3-prime of the MLLT3 reader (introns 4 onward) - without reference to any protein phenotype.
- The cold 5-prime introns will be cold in the cleavage map too. MLLT1 intron 3 (31,514 bp) and MLLT3 intron 2 (163,867 bp) are the largest zones in each gene and produce far fewer patients than length predicts; under this hypothesis they should be cleavage-poor.
- A partner with a YEATS-coding block positioned like MLLT1's should retain its reader, and one positioned like MLLT3's should lose it, independent of what the reader does. The YEATS family has other members and the prediction is testable across them wherever breakpoint data exist.
- The 2x2 backbone-swap matrix will show that the reader contributes similarly on both backbones. Under selection, AF9's reader should be neutral or harmful in an AF9 fusion; under geography there is no reason for that.
- Therapy-induced t(9;11) breakpoints will differ in distribution from de novo t(9;11) breakpoints on the partner side, not only on the KMT2A side. The KMT2A-side difference is already reported 3; the partner side has never been examined.
What would kill it
- A cleavage map that does not match. If TOP2 hotspots in both loci sit in the large 5-prime introns, where patients' breaks do not, then the patient distribution is being shaped after breakage, and selection is back.
- A backbone-dependent reader effect. If restoring YEATS to MLL-AF9 reduces transformation while deleting it from MLL-ENL also reduces transformation, the reader is doing opposite things on the two backbones and only selection explains the patient distribution.
- A YEATS-positive MLL-AF9 patient with an ordinary intragenic breakpoint. The two known retainers broke upstream of MLLT3 exon 1, which is the spliced architecture; the hypothesis predicts intron-1 retainers should be about as rare (1,410 bp of target), and a cluster of them would need another explanation.
- Equal cleavage susceptibility. If the two loci have similar TOP2 cleavage density per base in the relevant zones and the patient distributions still differ 26.9-fold, geography is insufficient.
The hypothesis is not killed by the failure of the uniform-length model on the partial class; that failure is what forces the hotspot form of the hypothesis rather than the length form, and it is stated in the analysis that produced it 1.
Why it matters to this project rather than to genome biology
If the asymmetry is geography, three things this graph carries change.
The in-cis add-back experiment changes meaning. Restoring YEATS to MLL-AF9 would no longer be 'undoing what selection did'; it would be building a molecule nature never tested. Still worth doing, but it answers a different question and should not be described as reconstructing a selected state.
The transfer rule survives intact and gets stronger. Whatever causes the asymmetry, MLL-AF9 and MLL-ENL differ by a functional reader in almost every patient, and results do not transfer. Geography removes any temptation to argue that the difference must be functionally unimportant because it is 'just' an accident - the patients' proteins differ either way.
The therapy-induced arm becomes the hypothesis's best natural experiment. 49 of 142 therapy-induced KMT2A leukaemias are MLLT3, and the recombinome already links an adult MLLT3 breakpoint shift to therapy induction 3. Partner-side breakpoint distribution stratified by prior topoisomerase-II exposure is a query against an existing archive, needs no new patients, and directly tests whether the drug moves the breakpoints.
| statement | The proportion of patients whose KMT2A fusion retains the partner's YEATS domain is set by the position of each partner locus's breakage hotspot relative to its YEATS-coding exons, and requires no selection on the resulting fusion protein. |
|---|---|
| origin | data-pattern |
| prior probability | 0.45 |
| known weakness | The uniform-length form of the model over-predicts partial-YEATS inclusion fourfold in MLLT1 and at least fiftyfold in MLLT3, so the hypothesis must be stated in hotspot terms and the length calculation is an indicator rather than a fit. |
predictions
[ "A TOP2 cleavage map in etoposide-treated human haematopoietic progenitors will show hotspots coinciding with the observed patient breakpoint zones in both loci", "MLLT1 intron 3 and MLLT3 intron 2, the largest and coldest zones, will be cleavage-poor", "The 2x2 backbone-swap matrix will show a similar reader contribution on both backbones", "Partner-side breakpoint distribution in t(9;11) will differ between therapy-induced and de novo cases", "Other YEATS-family or non-YEATS partners with a 5-prime-positioned hotspot will retain their N-terminal modules" ]
experiments
[ "exp-yeats-backbone-swap-matrix-mllt1-mllt3", "exp-yeats-cis-resupply-kmt2a-mllt3" ]
analyses
[ "ana-yeats-asymmetry-tested-against-intron-geometry" ]
falsifiers
[ "A TOP2 cleavage map whose hotspots sit in the large 5-prime introns, where patients' breaks do not", "A backbone-dependent reader effect: restoring YEATS reduces MLL-AF9 transformation while deleting it reduces MLL-ENL transformation", "A cluster of YEATS-positive MLL-AF9 patients with ordinary intragenic MLLT3 intron-1 breakpoints, exceeding what 1,410 bp of target predicts", "Similar TOP2 cleavage density per base in the relevant zones of both loci despite a 26.9-fold difference in patient retention" ]
computed support
{
"MLLT1_intron1_bp": 9013,
"MLLT3_intron1_bp": 1410,
"predicted_retention_pct": {
"MLLT1": 13.9,
"MLLT3": 0.51
},
"observed_MLLT3_retainers": 2,
"predicted_MLLT3_retainers_of_442": 2.3
}- If the asymmetry is geography, then the therapy that causes therapy-induced KMT2A leukaemia is also what decides which fusion protein the patient gets - a dose-and-agent-dependent property of treatment rather than of the disease, and one that would apply to every new DNA-damaging agent.
- A geography result would weaken the rationale for partner-specific drug development built on the premise that the retained modules were selected for, without weakening the transfer rule that makes partner-blind development risky.
- Does a published topoisomerase-II cleavage map covering MLLT1 and MLLT3 already exist in a haematopoietic system? This beat did not find one and did not search exhaustively for it.
- Can the recombinome archive be queried for partner-side breakpoint position stratified by therapy-induced status? The data exist; the cross-tabulation has never been published.
- If geography is right, why does MLLT1 have a large intergenic hotspot 5-prime of the gene and MLLT3 apparently not? That asymmetry is not explained by anything in this hypothesis.
- E1moderate confidence 3 of 5: One strong study, or several smaller consistent ones.database
MLLT1 and MLLT3 have identical exon-to-residue mapping over the YEATS domain in hg38 but MLLT1 intron 1 is 9,013 bp against MLLT3's 1,410 bp, so uniform intronic breakage predicts 13.9% against 0.51% YEATS retention - 2.3 of 442 t(9;11) patients against 2 observed - while the same model over-predicts the partial-YEATS class fourfold in MLLT1 and at least fiftyfold in MLLT3.
UCSC Genome Browser REST API, hg38 ncbiRefSeqSelect track, NM_005934.4 (MLLT1) and NM_004529.4 (MLLT3) · UCSC Genome Browser 2026What this evidence cannot show
Computed in this beat from one canonical RefSeq transcript per gene, using intron length as a proxy for breakage probability. The proxy is demonstrably wrong at the level of the partial class, which is why the hypothesis is stated in hotspot rather than length terms. Full method and failure modes in `ana-yeats-asymmetry-tested-against-intron-geometry`.
Evidence that argues against this (1)
- The agreement between 2.3 predicted and 2 observed retainers is a single-digit comparison in which a factor-of-two error would be invisible, and the model that produces it fails by an order of magnitude elsewhere in the same gene; it should be read as suggestive, not as a fit. Hu H et al. 2023
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
AF9 genomic breaks occur almost exclusively downstream of MLLT3 exon 4 (98.4%, n=442) with only two patients (0.5%) breaking upstream of MLLT3 exon 1, whereas 50.7% of 302 t(11;19) patients break 5-prime of MLLT1 and 3-prime of ACER1 and a further 33.4% within MLLT1 intron 1, giving 84.1% full ENL YEATS retention.
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, et al. · Leukemia 2023What this evidence cannot show
Read verbatim from the PMC HTML body (PMC11246743) on 2026-09-20; both distributions are re-analyses of one referral archive, so they are not independent cohorts. Genomic breakpoints, not protein.
Evidence that argues against this (1)
- That the MLLT1 majority breaks outside the gene entirely, in the ACER1-MLLT1 intergenic region, is not explained by any property of MLLT1's introns, so the geography hypothesis must also account for an intergenic hotspot that has no MLLT3 counterpart in the data. Meyer C et al. 2023
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
MLLT3 is the commonest partner in therapy-induced KMT2A leukaemia (49 of 142 cases; 47 of 101 t-AML cases), and the shift of KMT2A breakpoints towards intron 11 in adult MLLT3 patients was clearly linked to therapy-induced leukaemia while the same shift was not seen for ELL, EPS15 or SEPTIN6.
The KMT2A recombinome of acute leukemias in 2023. · Meyer C, Larghero P, Almeida Lopes B, et al. · Leukemia 2023What this evidence cannot show
Read in full text from the PMC HTML page (PMC10169636) on 2026-09-20. This is a KMT2A-side observation; no partner-side breakpoint distribution stratified by aetiology has been published for MLLT3 or for any partner. Therapy-induced status is as submitted, not adjudicated.
Evidence that argues against this (1)
- MLLT3's dominance among therapy-induced cases (34.5%) is close to its baseline share of all AML (30.4%), so the count alone is not evidence of a special relationship with topoisomerase-II-induced breakage; only the intron-11 shift is. Meyer C et al. 2023
- E4established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Breakpoint distributions within KMT2A are demonstrably non-uniform and shift with age, lineage and partner, with intron 11 breaks associated with ALL and younger patients and intron 9 breaks with AML and older patients, which the authors read as evidence of different aetiologies across age groups.
The MLL recombinome of acute leukemias in 2017. · Meyer C, Burmeister T, Gröger D, et al. · Leukemia 2018What this evidence cannot show
A KMT2A-side result, cited here as the existence proof that breakage in this disease is locus-position-dependent and exposure-dependent rather than uniform. It says nothing directly about MLLT1 or MLLT3.
Evidence that argues against this (1)
- An age- and lineage-dependent breakpoint distribution is equally compatible with selection operating differently in different lineages, so non-uniformity on the KMT2A side supports the plausibility of the geography hypothesis without discriminating it from selection. Wan L et al. 2017
prediction 3: is the reader contribution backbone-dependent
takes the structural-accident side and makes it falsifiable
the computed intron geometry
- depends on →MLLT3 is the commonest partner in therapy-induced KMT2A leukaemia: 49 of 142, and 47 of those are t-AML
the exposure-dependent breakpoint shift
if geography decides, subtraction is not a designed partner class