Attribution audit 5 of 5 - where KMT2A::AFF1 sits on chromatin: the field's standard attribution heuristic is invalidated by the paper that founded it, but two real survivors exist

This beat looked for that case in the KMT2A families and did not find one, and the reason is structural rather than accidental. The recurrent KMT2A partners - AFF1, MLLT1, MLLT3, MLLT10, ELL, AFDN - are all broadly expressed housekeeping-range transcription and elongation proteins, so the wild-type allele is present and expressed in every cell that carries the fusion. There is no KMT2A partner analogous to a lineage-restricted gene that is silent in the tumour's cell of origin.

What exists instead is the reciprocal-negative arm, which removes one confounding species but not the wild-type one: of 460 KMT2A::MLLT1 fusions about 241 are spliced fusions expressing only the direct product and no reciprocal transcript, with smaller equivalents for ELL (10 of 119) 4. That arm makes an anti-MLL-C reagent interpretable, because it removes the reciprocal source; it does nothing for an anti-partner reagent, because wild-type MLLT1 is still there.

So the honest answer to that part of the brief is that the class is empty in this disease, and the surviving cases are the tagged and degron-subtracted ones instead.

KMT2A* is the N-terminally truncated protein made from a gene-internal promoter that coincides with breakpoint cluster region hot spot II; its mRNA encodes a truncated MLL retaining the C-terminal functional regions including the SET domain 5. In an MLL-N-minus-MLL-C subtraction it behaves as a C-only species: it adds MLL-C signal without adding MLL-N signal. Its effect is therefore to make loci look less fusion-bound than they are - a conservative bias, but a real one, and it is unmeasured because no ChIP in this disease has ever distinguished KMT2A* from wild-type MLL-C.

The reciprocal product does the opposite of nothing here: it carries MLL-C and not MLL-N, so it too inflates the subtracted-away term. Both biases run in the same direction, which is why Heuristic 1 is usable at all.

Knock a degron plus an epitope tag into the endogenous KMT2A::AFF1 allele in SEM, C-terminally on the AFF1 side. One construct gives (a) anti-tag CUT&RUN, which is unambiguous attribution at endogenous abundance, and (b) degron subtraction against the same cells' MLL-N profile, which cross-validates every existing MLL-N-based map in the literature. Nothing else in this audit would settle as much per experiment (exp-endogenous-tag-and-degron-knock-in-kmt2a-aff1).

An interim check costing one CUT&RUN run: profile MLL-N and MLL-C in the same cells with a spike-in normalisation, and publish the subtracted track rather than the MLL-N track. The field currently publishes the unsubtracted one and calls it MLL-AF4.

Alongside the epitope tag and the degron there is an older fusion-specific perturbation: siRNA directed at MLL-AF4, which impairs clonogenicity of SEM and RS4;11 and reduces xenograft morbidity 6. If the oligonucleotides span the junction, they are as fusion-specific as a tag, and a ChIP performed after such a knockdown attributes by perturbation in exactly the way degron subtraction does - in a patient-derived line rather than an engineered one, which would make it stronger than the degron evidence on that axis.

This matters directly for the occupancy verdict, because Kerry et al. use MLL-AF4 siRNA knockdown followed by MLL-N, AF4-C, Menin and PAF1 ChIP 1, and the laboratory that developed the fusion-directed siRNAs is on their author list. If those are junction-spanning siRNAs, the SEM occupancy attribution is considerably stronger than this audit's headline verdict allows. If they target retained parent sequence, the perturbation depletes a wild-type parent too and adds nothing.

This beat could not settle it: neither paper's oligonucleotide sequences were reached (PMID 16046533 has no PMCID; the supplemental experimental procedures of PMID 28076791 were not retrieved). It is a ten-minute check for anyone with the supplementary files, and it is the single highest-value unresolved item in this audit.

This entry's top open question asked whether the MLL-AF4 siRNA used for occupancy subtraction in SEM spans the fusion junction. If it does, the subtraction is fusion-specific in a patient-derived line, which is stronger than anything else in this audit.

It does. Retrieved 2026-09-20 from the supplemental experimental procedures of PMID 28076791 (Europe PMC supplementaryFiles for PMC5263239, mmc1.pdf), verbatim:

siMA6 (sense, AAGAAAAGCAGACCUACUCCA; antisense, UGGAGUAGGUCUGCUUUUCUUUU), targeting the MLL exon 9 and AF4 exon 4 MLL-AF4 fusion site present in SEM cells. As control siRNAs we used the mismatch control siMM.

Three things follow.

1. The reagent is fusion-specific by sequence, not by inference. It is named to the junction, the sequences are published, and it carries a mismatch control rather than a scrambled one - the stricter comparator.

2. It is endogenous and patient-derived. SEM is a t(4;11) patient line, not a transduced model. That makes MLL-AF4 siRNA subtraction the only fusion-attributable perturbation of an endogenous KMT2A::AFF1 protein in this literature - the tag and degron systems that do this properly are all MLL::AF9 or overexpression.

3. The verdict on occupancy moves up. The qualitative map was already accepted; what was in doubt was whether any attribution in a patient line existed at all. It does. What still does not survive is quantitative occupancy - the siRNA reduces the fusion rather than removing it, and an MLL-N peak in a t(4;11) cell still sums wild-type KMT2A and fusion, so what fraction of the peak is fusion remains unmeasured. The knock-in proposed in exp-endogenous-tag-and-degron-knock-in-kmt2a-aff1 is still the experiment that settles that.

The sequences come from PMID 16046533, which this graph had separately been citing for a minute-scale elongation claim it never made. The same paper is the source of the field's only junction-specific reagent, which is the more useful thing about it.