Redesign: the proximity-labelling map of KMT2A::AFF1 cannot be controlled with the wild-type allele - it has to be controlled with time

The original design's three-way partition - shared with wild-type (retained-fragment partners), unique to wild-type (discarded-fragment partners), unique to the fusion (partner-supplied contacts) - assumes the two tagged proteins occupy different neighbourhoods except where they genuinely share a partner. They do not.

• Wild-type MLL1 is at the fusion's loci, by requirement. Fusion proteins lack the PHD fingers and require prebinding of a wild-type MLL1 complex for stable HoxA9 association 1. So at exactly the loci the experiment cares about, the control tag is sitting a few nanometres from the experimental tag.
• And it is not incidental. The wild-type Mll allele is genetically required for MLL-AF9 leukemogenesis and maintenance, with menin recruiting both molecules to HOX loci 2. This is not a bystander allele that happens to be nearby.

TurboID's labelling radius is roughly 10 nm. Two molecules recruited to the same promoter by the same adaptor are inside it. The consequence is that the 'shared' bin will be enormous and uninterpretable, the 'fusion-unique' bin will be depleted of anything the wild-type complex also touches - menin, LEDGF, PAF1C, the very partners that matter - and the 'wild-type-unique' bin will contain the C-terminal partners the graph already knows about from 2004 biochemistry. The design returns its least informative bin at full strength and its most informative bin attenuated.

It is worth being clear that this is not a flaw in the idea of proximity labelling here. It is a flaw in using a second protein at the same locus as the control.

The original design places TurboID at the 3' end of the fusion ORF, reasoning that this keeps it away from the N-terminal menin-binding motifs. That reasoning is right about the N-terminus and wrong about the C-terminus. The 3' end of KMT2A::AFF1 is the AFF1 C-terminal homology domain (CHD), and the CHD is not inert: the AFF4 CHD structure shows an eight-helix fold that mediates AFF4 homodimerisation and AFF1-AFF4 heterodimerisation, binds RNA and DNA, and contains a surface loop that is a substrate for the P-TEFb kinase CDK9 3.

So a C-terminal cassette is fused directly onto the fusion's dimerisation surface, its nucleic-acid-binding surface and a CDK9 phosphosite. The design's own risk mitigation - checking HOXA9/MEIS1 expression and proliferation against parental - would not necessarily catch a partial impairment of AFF dimerisation, and a partial impairment is exactly what would distort a proximity map.

The graph already has exp-aff1-chd-dimerisation-separation-of-function and hyp-aff1-chd-dimerisation-in-kmt2a-fusion, so this is a surface the project has independently decided matters.

1. Move the tag off the CHD. Two options, in order of preference:

• Internal knock-in at the fusion junction. The KMT2A-AFF1 junction sequence exists only in the fusion allele, so a cassette placed there is fusion-specific by sequence without needing an allele-matched control at all, and it avoids both the menin-binding N-terminus and the CHD. The junction sits in AFF1's intrinsically disordered N-terminal region, which is the most insertion-tolerant part of the protein - though it is also where the SEC-binding ALF/pSER elements lie, so SEC co-IP must be checked explicitly rather than inferred from proliferation.
• miniTurbo rather than TurboID, at the C-terminus, with a long flexible linker, if the junction proves intolerant. Smaller cassette, shorter labelling window, less steric load on the CHD. This is the fallback, not the plan.

2. Replace the allele control with a degron control. Knock a dTAG or mAID degron into the same fusion allele alongside the biotin ligase. Then run, in one line and one chromatin state:

• Arm A. Biotin pulse, 20-30 min (miniTurbo), vehicle.
• Arm B. Degrader for the minimum time that clears the fusion, then the identical biotin pulse.
• Arm C. Nuclear free miniTurbo, for abundance and compartment bias.
• Arm D. Untagged parental.

The readout is A minus B: proteins whose labelling collapses when the fusion is gone, in the same cell, at the same loci, with the wild-type allele untouched throughout. That is a fusion-dependent proximity set. Everything that labels equally in A and B is a neighbourhood resident, which includes wild-type MLL1's own partners - and now it is a scored category rather than a contaminant.

3. Add a menin-inhibitor eviction arm as an orthogonal displacement, since revumenib-class eviction is well characterised in this graph. Labelling lost on eviction is chromatin-dependent; labelling retained is chromatin-independent. That splits the fusion-dependent set into on-chromatin and off-chromatin contacts, which no affinity purification can do.

4. Keep, but reinterpret, the wild-type-allele line. It is still worth building - not as the discriminating control, but as a map of the wild-type neighbourhood in a leukemic cell, which nobody has either. It answers a different question and should be written up as such.

The redesign does not rescue everything the original claimed, and it is better to say so than to propose an experiment that cannot work.

Allele-resolved static proximity labelling at shared loci is not achievable with current methods. If two molecules are recruited to the same promoter by the same adaptor, no tag on either one can report which of them a labelled protein was 'really' near. Split-TurboID reconstituted across the fusion junction would give molecular specificity for the ligase, but not spatial specificity for the labelling - the biotinoyl-AMP intermediate still diffuses over the same ~10 nm, and it would label the wild-type complex's partners just as readily. Shrinking the radius with APEX2 helps marginally and costs a peroxide treatment that perturbs chromatin.

So the honest reformulation is that the question 'which proteins are near the fusion rather than near wild-type KMT2A' is not answerable, and it is also not the question worth asking. The answerable and more useful question is 'which proteins are near the fusion because of the fusion', and that is a time-resolved question, not an allele-resolved one. The degron design answers it.

Two residual limitations that no design removes: proximity is not binding, and SEM and RS4;11 are cell lines rather than infant primary blasts.