Is the YEATS asymmetry just intron geometry? Computed from hg38: the direction is explained, the magnitude is not, and both genes fail the same prediction

https://api.genome.ucsc.edu/getData/track?genome=hg38;track=ncbiRefSeqSelect was queried for chr19:6,100,000-6,300,000 and chr9:20,300,000-20,700,000 on 2026-09-20. The returned records are NM_005934.4 (MLLT1, chr19:6,210,380-6,279,975, minus strand, 12 exons) and NM_004529.4 (MLLT3, chr9:20,341,668-20,622,499, minus strand, 11 exons) 4. Exon blocks were ordered 5-prime to 3-prime by strand, introns computed between consecutive blocks, and coding lengths accumulated across the CDS to map each exon to residues.

The mapping came out identical for the two genes over the reader:

This independently reproduces the primary paper's statement that the ENL YEATS domain is coded from exons 2-4 (amino acids 5-140) 1, and extends it: AF9's YEATS domain has exactly the same exon structure, which is what makes an intron-length comparison meaningful rather than apples-to-oranges.

Grouped by what a break there does to the fusion's YEATS content:

Under breakage uniform in intronic length: MLLT1 gives 13.9% / 61.1% / 25.0% and MLLT3 gives 0.51% / 62.9% / 36.6%.

Right, and it is not a small thing. The predicted YEATS-preserving fraction differs 26.9-fold between the two genes (13.9% against 0.51%) purely because of where the DNA is. In a cohort of 442 t(9;11) patients, 0.51% predicts 2.3 patients; the observed number of YEATS-retaining MLL-AF9 patients is 2 1. That agreement is close enough to be a coincidence and is still the single most economical explanation on offer for the rarity of a YEATS-positive MLL-AF9.

Wrong, and this is the informative part. Uniform breakage predicts that the partial class should dominate both genes, because introns 2-3 are the largest zone in each. It does not:

For MLLT1 the model over-predicts fourfold. For MLLT3 it over-predicts at least fiftyfold. In both genes the large 5-prime intron is cold, and the breaks pile up 3-prime of it.

And the model cannot see MLLT1's dominant break zone at all. 50.7% of t(11;19) patients break 5-prime of MLLT1 and 3-prime of ACER1 - outside the gene 1. No intronic model covers that, and it is the largest single class of MLL-ENL patients.

Three readings survive, and the entry's job is to say which the numbers discriminate.

1. Structural accident, in a refined form. Not 'breaks are uniform and MLLT3 has more DNA', which fails, but 'each locus has a breakage hotspot region, and the hotspot sits 3-prime of the reader in MLLT3 and 5-prime of it in MLLT1'. This is consistent with every number above, including the cold 5-prime introns, and it is a claim about topoisomerase-II cleavage and chromatin rather than about protein function. It is testable without a single patient: map TOP2 cleavage in etoposide-treated haematopoietic cells across both loci and ask whether the hotspots sit where the patients' breaks do.
2. Functional selection. AF9 YEATS suppresses transformation and ENL YEATS supports it, so the observed distributions are what survives. Nothing here refutes this, and the 2-patient agreement is weak evidence against it only because a selection model would predict zero, not two.
3. Both. Geography sets the supply of breakpoints and selection filters them. Under this reading the MLLT1 partial class (22 patients, 7.3%) is the population that should be followed clinically, because it is the one place where the supply is large and the filter is visible.

What the numbers do not support is the framing this graph has carried, in which the asymmetry is treated as a fact about the proteins. A 6.4-fold difference in intron-1 length between two genes whose exon structure is otherwise identical is a fact about the DNA.