The taspase-1 N/C antibody control separates fusion-occupied from wild-type KMT2A promoters in every primary infant sample, and it is the C-terminal arm that does the work while being the less reproducible arm
Because the normal KMT2A protein is cut in half and the cancer fusion keeps only the front half, an antibody to the back half sees only the normal protein. That difference cleanly separates the genes the fusion sits on from the genes normal KMT2A sits on, in all three patient samples.
The same thing, for a scientist
**Why the control exists.** Wild-type KMT2A is cleaved by taspase 1 into N320 and C180 fragments that remain associated, so a C-terminal antibody reports **wild-type only** while an N-terminal antibody reports **wild-type plus fusion**. The source paper's entire site-calling method is this ratio, with a Bayesian N/C read-share model, N/C > 1.75 and p < 1e-5, giving 1,692 oncoprotein-target sites [E2]. This entry is an independent reproduction of the control's behaviour at promoter resolution, not its invention - the brief that dispatched this beat described the control as unused, which is true of this corpus and false of the source paper. **D1 - the control separates, in every state** [E1]. Among above-median-N promoters, the fraction exceeding the paper's own N/C cut of 1.75 is **16.3% (p279), 19.1% (p318), 13.1% (p454)**, against 6.3% / 8.3% / 6.3% below 1/1.75 in the same denominator - a clear asymmetric right tail, not a unimodal distribution around 1. The `WT` set's median N/C is 0.995-0.999, i.e. exactly 1 by construction and behaving. **The C-terminal arm is where the discrimination lives, and it is very clean** [E1]. Taking each track's own top 971 promoters and counting how many fall in the N/C-defined fusion set (random null = 48.6): | | p279 | p318 | p454 | |---|---:|---:|---:| | KMT2A-**C** top 971 in the fusion set | **6** (0.12x null) | **2** (0.04x) | **7** (0.14x) | | KMT2A-**N** top 971 in the fusion set | 138 (2.8x) | 120 (2.5x) | 251 (5.2x) | **The C-terminal antibody's best promoters essentially never coincide with fusion targets - 15 of 2,913 across three samples.** That is the sharpest single number in this beat and it is what makes the ratio informative: the denominator is genuinely fusion-free. **D3 - and the C arm is simultaneously the less reproducible one.** The two N antibodies agree at rho 0.905 / 0.846 / 0.903; the two C antibodies agree at **0.730 / 0.620 / 0.692**, in every patient [E1]. The pre-registered D3 boolean, *does C agree with C more than C agrees with N*, returned **False in all three** - but that boolean is confounded, because N and C are both expected to sit on wild-type KMT2A promoters and the wild-type sites outnumber the fusion sites 3-to-1 here. **The unconfounded statement is the one that survives: the wild-type-specific arm of the control is systematically the noisier arm.** The consequence is quantitative, not directional: the ONCO/WT depletion ratio for C1 is 0.27 / 0.18 / 0.36 and for C2 is 0.36 / 0.35 / 0.49, so the reagent choice changes the measured depth of depletion by up to 1.9-fold while never changing its sign. **A pre-registered test that failed and is reported as failed.** The D3 boolean was my instrument and it was the wrong instrument. Recorded here rather than quietly replaced, because a broken check is a claim about the checker.
Why this control is better than anything the corpus currently uses for the index fusion
Every fusion-occupancy claim in this project for KMT2A::AFF1 rests on one of: an epitope tag in an engineered line, a partner-protein antibody, or a transfer from KMT2A::MLLT3. The N/C control needs none of those - it works on uncultured patient material with two commercial antibody pairs, and it is internally controlled because both arms are measured in the same nuclei. It is also the only fusion-occupancy instrument in this corpus that does not require the cells to be modified.
What it cannot do
It cannot see fusion sites that wild-type KMT2A also occupies strongly - those land in the WT set by construction and are invisible to the ratio. It is therefore a lower bound on fusion occupancy, and a factor that is enriched at shared sites would be scored as non-discriminating even if it were fusion-bound there. This is the most likely way the AFF1 result in ana-q142-aff1-occupancy-does-not-mark-fusion-sites-in-primary-infant-kmt2a-aff1 could be understating AFF1.
IgG is not the confound
The IgG track's own ONCO/WT ratio is 0.75 / 0.86 / 1.08 and its correlation with KMT2A-N is 0.363 / 0.239 / 0.326, so the site sets are not an accessibility artefact. The pre-registered U1 guard did not fire. IgG is however very sparse - 80% / 76% / 72% of promoters are exactly zero - so IgG-derived ratios are the noisiest numbers in this beat and are used only as a floor, never as a denominator.
| script | analyses/q142-aff1-in-primary-infants/analyze.py |
|---|---|
| status | run |
| result summary | N/C control separates in all three states; C-arm top promoters hit the fusion set at 0.04-0.14x null; C1~C2 rho 0.62-0.73 vs N1~N2 0.85-0.91. |
| pre registered outcome | D1 holds for direction; D3 fires for magnitude and for the reagent |
- If N/C ratio is proposed as a clinical fusion-occupancy assay, its denominator is the less reproducible half, so a patient sample with a weak C library will look like it has more fusion everywhere - the failure mode inflates the thing the assay is for.
- Would a third C-terminal antibody close the C1/C2 gap, or is the C180 fragment genuinely harder to profile in primary material?
- Does the reciprocal AFF1::KMT2A product, where it is expressed, carry any epitope either antibody pair can see?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset
In all three primary KMT2A::AFF1 samples in GSE252378, 13-19% of above-median-N promoters exceed N/C 1.75; the C-terminal antibodies' own top 971 promoters include only 2-7 of the N/C-defined fusion set against a null of 48.6; and the two C-terminal antibodies agree with each other (rho 0.62-0.73) less well than the two N-terminal antibodies do (0.85-0.91).
GSE252378 - AutoCUT&RUN and AutoCUT&Tag profiles of 36 MLL-rearranged leukemia samples (171 GSMs, hg38) · Janssens DH, Henikoff S (Fred Hutchinson Cancer Center / Van Andel Institute) · NCBI Gene Expression Omnibus 2024What this evidence cannot show
Promoter-resolution reproduction over 19,404 refGene windows; the paper's intervals are peaks. Within-sample shares only; no spike-in exists. The C1/C2 reproducibility gap could reflect the reagents, the lower abundance of the C180 fragment, or both, and this design cannot separate them.
Evidence that argues against this (1)
- A C-terminal antibody that simply worked poorly would also produce few top promoters in the fusion set and poor replicate agreement - i.e. the cleanest-looking number here, 2-7 of 971, is equally consistent with excellent specificity and with a weak reagent. The N/C ratio inherits that ambiguity, and no orthogonal fusion-occupancy measurement exists in these samples to break it. Janssens DH et al. 2024
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The source paper calls oncoprotein target sites from the ratio of KMT2A N-terminal to C-terminal CUT&RUN signal, with a Bayesian model, N/C > 1.75 and p < 1e-5, yielding a master list of 1,692 sites across 36 samples.
MLL oncoprotein levels influence leukemia lineage identities · Janssens DH, Duran M, Otto DJ, Wu W, Xu Y, Kirkey D, Mullighan CG, Yi JS, Meshinchi S, Sarthy JF, Ahmad K, Henikoff S · Nature Communications 15:9341 2024What this evidence cannot show
Read from the paper's methods in PMC11522475. The 1,692-site list is a supplementary table and was not obtained or used here.
Evidence that argues against this (1)
- The paper itself notes the C-terminal antibody is specific to the wild-type protein, an assumption from taspase-1 biology rather than a measurement made in these samples; if any fusion retained C-terminal epitope reactivity - for instance through a reciprocal AFF1::KMT2A product carrying the C terminus - the control would be partially blinded. Janssens DH et al. 2024
- related to →KMT2A / MLL1 — the gene that breaks
taspase-1 N320/C180 split used as a built-in wild-type control
- related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
defines fusion-occupied promoters in primary infant material
source deposit
the site sets come from the N/C control
site sets from the N/C control