Does the fusion 'lose restraint'? The pattern of KMT2A-side loss, and why the tidy reading does not survive contact with the ledger

The strong form: KMT2A fusions transform because the break separates a targeting module from a regulatory module, and it is the loss of regulation that drives the disease.

What is genuinely in its favour:

• The break position is remarkably consistent, and it falls between the targeting apparatus and everything downstream. That is not obviously a coincidence.
• At least two independent brakes sit in the lost region and nowhere else: PHD2's CDC34-dependent E3 ligase activity, which limits MLL1 abundance and target occupancy [see subtraction-kmt2a-phd2-e3-ligase], and the PHD3/CyP33 activator-to-repressor switch. Two independent restraints lost in one cut is a real pattern, not a rhetorical one.
• Re-inserting PHD3 into MLL-ENL blocks immortalisation - the only in-cis add-back result that exists, and it points the hypothesis's way.
• The 1994 founding observation was exactly this shape: activation domain lost, repression domain and DNA-binding retained [see subtraction-kmt2a-tad-cbp-contact].

1. The best-evidenced loss is an activating output, not a brake. Mll1 deletion selectively depletes H4K16 acetylation at MLL1 target genes with reduced transcription, and that H4K16ac depends on recruited MOF - which binds the MLL1 C-terminus the fusion discards. The single loss with the most functional evidence behind it therefore makes the protein less able to activate, which is the opposite of the restraint story's direction.

2. The restraint may be outsourced rather than removed. MLL1 fusion proteins lack the PHD fingers and require prebinding of a wild-type MLL1 complex for stable HoxA9 association 1. If a PHD-containing wild-type complex is sitting at the locus the fusion occupies, the brake has not left the building. What the fusion has acquired is exemption, not absence.

3. Loss of restraint is not sufficient. Induced dimerisation of a truncated MLL immortalises bone marrow and imposes a reversible myeloid differentiation block with Hoxa7, Hoxa9 and Meis1 upregulation; dimerised MLL binds a CpG island at Hoxa9 with higher affinity than undimerised MLL; and the authors' model is that either dimerisation/exon duplication or fusion to a transcriptional activator produces Hox upregulation and transformation 2. The truncated MLL in that system is a protein that has lost every module in this beat's ledger. It needed something added before it transformed anything.

That third point is the decisive one. If losing restraint were the mechanism, a truncated MLL should be oncogenic on its own. The experiment closest to that test says it is not - something must be added, whether a partner's elongation machinery or an artificial dimerisation.

Losses with a demonstrated functional consequence (2 of 15):

• PHD3 - in-cis add-back blocks replating immortalisation in MLL-ENL. One study, one fusion, in vitro, unreplicated (ana-phd3-addback-replication-audit). And the consequence points at restoration breaking the fusion, not at loss building it.
• MOF contact - indirect but strong: Mll1 deletion depletes H4K16ac at target genes with reduced transcription. Measured on wild-type protein, not on the fusion.

Losses demonstrated NOT to matter (1 of 15):

• SET domain - dispensable for HSC maintenance and for MLL-AF9 leukemogenesis. Contradicted pharmacologically by MM-401, unresolved.

Losses that are merely absent, with no demonstrated consequence (11-12 of 15):

• PHD1, PHD4, the pre-PHD C2HC finger, the bromodomain (as a separable unit), the HCF-binding motif, FYRN, FYRC, both taspase1 cleavage sites, and - functionally - the TAD, whose job the partner does differently.

The ratio is the finding. Eleven or twelve of fifteen lost modules have never been asked whether their loss matters. Writing a mechanistic narrative over that base is the mirror image of the error this project made before, when it read 'retained' as 'required'. Here the temptation is to read 'lost' as 'load-bearing', and for roughly four fifths of the ledger there is no evidence either way.

What was withdrawn. This entry's first argument against the restraint reading was that the best-evidenced KMT2A-side loss is an activating output - the MOF/KAT8 contact - rather than a brake. That argument rested on this graph's own ranking in subtraction-kmt2a-mof-contact, which has since been corrected. MOF was finally tested inside a fusion-driven leukemia: it is required, but not through the fusion's target programme (no clear Meis1 or Hoxa downregulation), the mechanism is global H4K16ac loss and DNA-damage-repair failure, and it is equally required in NUP98-fusion AML 5. MOF is a dependency of the cell. It is not the KMT2A half's best-evidenced loss, so it cannot be the counterexample that sinks the restraint reading.

What replaces it. The losses with measured consequences turn out to be brakes after all - but not the brakes the story names. Three independent determinants of KMT2A turnover sit C-terminal to the breakpoint and are gone from every der(11) fusion: the PHD2 E3 ligase, the taspase1 cleavage sites, and the FYRN degron that marks an unpartnered MLL N-terminal fragment for destruction. Disabling each of them in wild-type MLL1 produces the same phenotype: more protein, longer-lived, more chromatin-bound 6.

So the corrected verdict on the 'loss of restraint' reading is not that it is over-stated, but that it is aimed at the wrong faculty. The fusion is not conspicuously freed from brakes on what it does. It is freed from brakes on how much of it there is and how long it stays. That is written up as hyp-kmt2a-half-loses-abundance-control-not-activity-control.

What still stands against the strong form. Points (2) and (3) are unaffected and one of them is stronger: the PHD function is outsourced rather than removed, and the wild-type Mll allele is now known to be genetically required for MLL-AF9 leukemogenesis and maintenance 7, which is a stronger statement than the prebinding model it previously rested on. And loss of restraint remains demonstrably insufficient on its own.

The honest caveat. All three degradation determinants were measured on wild-type MLL1, never on a fusion, and one of them (taspase1) has a well-executed study reading it in the opposite direction. Three wild-type results pointing one way is a pattern worth testing, not a conclusion.