KMT2A::MLLT1 versus KMT2A::MLLT3: paralogous partners whose chromosomes break on opposite sides of the same domain
also called MLL-ENL versus MLL-AF9 · ENL AF9 YEATS retention asymmetry · t(11;19) versus t(9;11)
ENL and AF9 are near-twin proteins, and both get joined to the same gene to cause leukaemia. But the chromosome almost always breaks in front of ENL's reading module and almost always behind AF9's, so the two fusions look alike on paper and carry opposite halves of the twin.
The same thing, for a scientist
MLLT1 (ENL, 559 aa, UniProt Q03111) and MLLT3 (AF9, 568 aa, UniProt P42568) are the two YEATS-family KMT2A partners, ~74% similar at gene level with YEATS domains sharing 88% alignment, each carrying an N-terminal YEATS reader and a C-terminal Anc1-homology domain (ENL 494-559, AF9 502-568) that binds AF4-family proteins and DOT1L. The decisive difference is not sequence but breakpoint geography. In 302 t(11;19) patients, 84.1% retain the full ENL YEATS domain; in 442 t(9;11) patients, 98.4% of AF9 breaks fall downstream of MLLT3 exon 4 and exclude the AF9 YEATS domain, with only two patients (0.5%) retaining it. The two fusions therefore differ by a whole functional module in almost every patient, despite being built from paralogues. A second asymmetry runs the other way at the wild-type level: ENL is required for AML maintenance whereas AF9 is dispensable in several AML lines while being required for haematopoietic stem-cell self-renewal. Their clinical profiles also diverge, in a lineage-dependent way already recorded elsewhere in this graph.
The same protein family, twice
ENL is the third most common KMT2A fusion partner and shares roughly 74% gene sequence homology with AF9; their YEATS domains share 88% alignment 1. Architecturally they are the same molecule twice: N-terminal YEATS acyl-lysine reader, long central disordered region, C-terminal AHD.
The AHD is functionally interchangeable in the assays that have been done. ENL 494-559 mediates AF5q31 and DOT1L binding and is required for MLL-ENL transformation; the portion of AF9 retained in MLL-AF9, which includes AHD residues 502-568, mediates the same associations and confers the same GAL4-dependent transactivation, Hoxa9 expression and myeloid transformation 2. Both fusions are also independent of wild-type ENL, unlike MLL-AF5q31 2.
On that account the two fusions ought to be near-equivalent. They are not.
The asymmetry that matters: breakpoints fall on opposite sides of the YEATS domain
| KMT2A::MLLT1, t(11;19) | KMT2A::MLLT3, t(9;11) | |
|---|---|---|
| patients analysed | 302 | 442 |
| partner YEATS domain retained | 84.1% (n=254) | 0.5% (n=2) |
| dominant break position | 5' of the partner gene or in intron 1 | downstream of MLLT3 exon 4 (98.4%) |
That is the finding 1. It is not a subtle enrichment; it is close to complete in both directions, and it means the two commonest YEATS-partner fusions in this disease are built from the opposite halves of the same protein family. KMT2A::MLLT1 carries reader plus AHD. KMT2A::MLLT3 carries AHD alone.
Anything attributed to 'the YEATS partners' as a class is therefore attributable to two different molecules. The graph's own subtraction entries for the two partners should be read that way: for MLLT3 the YEATS domain is a lost element, for MLLT1 it is a retained one, and experiments that re-supply YEATS in cis to KMT2A::MLLT3 are asking what KMT2A::MLLT1 already has.
The wild-type dependency runs the other way
At the level of the unrearranged proteins the asymmetry reverses. ENL, but not AF9, is required for disease maintenance in AML by CRISPR depletion, with ENL loss reducing RNA polymerase II occupancy and transcriptional initiation and elongation at ENL-enriched targets 3. AF9 is dispensable in several AML cell lines while being required for haematopoietic stem-cell self-renewal and expansion 1.
Putting the two asymmetries together produces an awkward result for drug discovery. The fusion that contains a YEATS domain (KMT2A::MLLT1) is the one whose cells do not need wild-type ENL 2; the fusion that lacks a partner YEATS domain (KMT2A::MLLT3) sits in a background where wild-type ENL is the dependency. A YEATS-directed drug therefore has a different mechanism in each, and neither mechanism is 'inhibit the fusion' in the simple sense.
What differs clinically, and where to read it
This graph already holds the outcome comparison as separate entries rather than as assertions here: epi-partner-mllt1-prognostic-sign-flips-between-aml-and-all records MLLT1 as a high-risk partner in childhood AML and a favourable one in childhood ALL and T-ALL, and epi-partner-mllt3-t9-11-favourability-claim-and-its-non-replication records the contested favourability of t(9;11). The relevant point for this entry is structural, not prognostic: the two partners' prognostic signs are usually compared as if the fusions differed only by partner identity, when they in fact differ by the presence or absence of an entire reader domain in almost every patient.
One within-fusion observation sharpens that. Within t(11;19), significantly fewer AML than ALL patients retain the ENL YEATS domain (chi-square p<0.0001, df=1), while age group is not associated with retention (p=0.33, df=2) 1. So the lineage in which KMT2A::MLLT1 carries its worst prognosis is also the lineage in which the fusion is least likely to carry the reader. Whether that is causal, a consequence of shared breakpoint biology, or an artefact of a referral archive is untested, and the paper does not print the per-cell counts behind the figure.
The experiment this suggests
The cleanest comparison available is not a drug experiment. It is to build, in one cell background, KMT2A fused to (a) MLLT1 from exon 2, (b) MLLT3 from exon 5, (c) MLLT3 from exon 1, and (d) MLLT1 from exon 3, and ask whether the phenotype tracks the partner's identity or the presence of its YEATS domain. Two of those four constructs correspond to >84% and >98% of real patients respectively; the other two correspond to 0.5% and ~9% of real patients, so all four exist in humans and none is a synthetic curiosity.
If the phenotype tracks YEATS presence rather than partner identity, then 'MLLT1 versus MLLT3' is the wrong axis for the whole field and 'YEATS-in versus YEATS-out' is the right one - which would also reorganise how partner-stratified prognosis should be reported.
| partner a | MLLT1 |
|---|---|
| partner b | MLLT3 |
| uniprot a | Q03111 |
| uniprot b | P42568 |
| length a AA | 559 |
| length b AA | 568 |
| accession resolution | both accessions fetched from UniProt this session; primary gene names read back as MLLT1 (559 aa, 'Protein ENL') and MLLT3 (568 aa, 'Protein AF-9') |
| yeats alignment PCT | 88 |
| gene homology PCT | 74 |
| yeats retained PCT a | 84.1 |
| yeats retained PCT b | 0.5 |
| n patients a | 302 |
| n patients b | 442 |
| dominant break b | downstream of MLLT3 exon 4 (98.4%) |
| wild type enl required by fusion cells | false |
ahd a
{
"name": "AHD (ANC1 homology domain)",
"start": 494,
"end": 559
}ahd b
{
"name": "AHD (ANC1 homology domain)",
"start": 502,
"end": 568
}wild type required in aml
{
"ENL": true,
"AF9": false
}- A YEATS-domain antagonist has a different target in the two commonest YEATS-partner fusions: the fusion protein plus wild-type ENL in t(11;19), and wild-type ENL alone in t(9;11). A trial pooling 'YEATS-partner KMT2A leukaemia' would be pooling two mechanisms.
- Wild-type AF9 is dispensable in AML lines but wild-type ENL is not, so a pan-YEATS inhibitor loses the selectivity argument that an ENL-selective agent would have, while gaining the AF9-related haematopoietic stem-cell toxicity.
- Cross-partner extrapolation is the specific failure this entry exists to block: most mechanistic work on 'MLL-YEATS fusions' has been done in MLL-AF9 murine models, which lack the reader domain that 84% of t(11;19) patients have, so importing MLL-AF9 conclusions into t(11;19) imports conclusions about a protein missing a module.
- Is the breakpoint asymmetry a property of the genes' intron structure, of the chromatin state at the two loci, or of selection on the resulting protein? MLLT3's exon 4 boundary and MLLT1's intron 1 are both large introns; a breakpoint-cluster-region analysis of the two loci side by side would separate mechanism from selection.
- Does the AML-versus-ALL difference in ENL YEATS retention replicate in an independent archive? The finding is a single chi-square on a single referral series and would reorganise partner-stratified prognosis if real.
- Do the two MLL-AF9 patients who retain the AF9 YEATS domain differ clinically? n=2 in 442, and they are the natural experiment that the cis-resupply experiment is designed to imitate.
- Is the phenotype driven by partner identity or by YEATS presence? The four-construct isogenic panel in this entry has never been built.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In 302 t(11;19) patients 84.1% retain the full ENL YEATS domain, whereas in 442 t(9;11) patients AF9 breaks fall downstream of MLLT3 exon 4 in 98.4% of cases and only two patients (0.5%) retain the AF9 YEATS domain; ENL and AF9 YEATS domains share 88% alignment and the genes ~74% homology; within t(11;19), significantly fewer AML than ALL patients retain YEATS (chi-square p<0.0001) while age group does not associate with retention (p=0.33).
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
Both breakpoint sets come from the same Frankfurt diagnostic archive, a referral series, so the two arms share whatever ascertainment bias it carries - which is a strength for the comparison and a weakness for the absolute percentages. Domain retention is inferred from genomic break position, not from sequenced protein. The AML-versus-ALL retention difference is presented as a chi-square p-value on a summary figure without printed cell counts, so the effect size cannot be recovered from the paper. The 88% alignment figure is cited by the authors to another source rather than computed here.
Evidence that argues against this (1)
- Two smaller earlier series reported ENL YEATS retention in 19 of 23 and in 10 of 15 t(11;19) patients - 83% and 67%. The 67% figure, from the series cited here, is well below 84.1%, so the retention rate is less settled than a single decimal place suggests. Fu JF et al. 2007
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
ENL residues 494-559 and the equivalent AF9 AHD residues 502-568 both mediate AF5q31 and DOT1L association and both confer transactivation, Hoxa9 expression and myeloid transformation in the corresponding KMT2A fusions; both MLL-ENL- and MLL-AF9-transformed cells are unaffected by wild-type ENL knockdown, unlike MLL-AF5q31-transformed cells.
A higher-order complex containing AF4 and ENL family proteins with P-TEFb facilitates oncogenic and physiologic MLL-dependent transcription. · Yokoyama A, Lin M, Naresh A, et al. · Cancer Cell 2010What this evidence cannot show
Retroviral myeloid transformation assays in murine progenitors; functional equivalence in a colony assay is a low bar for concluding the two AHDs are interchangeable in patients. The MLL-ENL constructs used here are of the YEATS-containing form only by virtue of the vector, not by design, so the comparison is not explicitly YEATS-controlled.
Evidence that argues against this (1)
- The breakpoint data show that the real MLL-AF9 protein in 98.4% of patients lacks the AF9 YEATS domain while the real MLL-ENL protein in 84.1% of patients retains ENL's, so an assay that finds the two AHDs equivalent is comparing the halves the two fusions do not differ in. Hu H et al. 2023
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
ENL, but not its paralogue AF9, is required for disease maintenance in acute myeloid leukaemia; CRISPR-Cas9 depletion of ENL increases terminal myeloid differentiation and suppresses leukaemia growth in vitro and in vivo.
ENL links histone acetylation to oncogenic gene expression in acute myeloid leukaemia. · Wan L, Wen H, Li Y, et al. · Nature 2017What this evidence cannot show
The ENL-versus-AF9 contrast is drawn across AML cell lines that are mostly not t(11;19) and not t(9;11), so it is a statement about the wild-type paralogues in a leukaemic background rather than about either fusion. Paralogue redundancy in CRISPR screens is assay-dependent: a gene can score non-essential because its paralogue compensates in that line.
Evidence that argues against this (1)
- In a t(11;19) line the fusion neither incorporates nor requires wild-type ENL, so a general 'ENL is required in AML' claim does not extend to the fusion that contains ENL. Yokoyama A et al. 2010
- related to →KMT2A::MLLT1 (MLL-ENL), t(11;19)(q23;p13.3)
84.1% of 302 patients retain the partner YEATS domain
- related to →KMT2A::MLLT3 (MLL-AF9), t(9;11)(p21.3;q23.3)
0.5% of 442 patients retain the partner YEATS domain; 98.4% break downstream of MLLT3 exon 4
AF9, UniProt P42568, 568 aa, AHD 502-568
quantifies how nearly universal the AF9 YEATS loss is: 98.4% of 442 patients
the cis-resupply experiment reconstructs what KMT2A::MLLT1 already has, and two real MLL-AF9 patients are the natural version of it
the lineage in which MLLT1 is a high-risk partner is also the one least likely to retain the reader domain
the functional content of the module whose presence separates the two fusions
structural counterpart to the contested t(9;11) favourability claim
verified the 98.4/442/exon-4 figures from the primary text
identical exon structure, 6.4-fold different intron 1