Oncofusion

start / graph / ana-q142-aff1-occupancy-does-not-mark-fusion-sites-in-primary-infant-kmt2a-aff1

Analysisstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q142-aff1-in-primary-infants

AFF1 occupancy does not distinguish fusion-occupied promoters from wild-type KMT2A promoters in primary infant KMT2A::AFF1 - reproduced at promoter resolution, and the corpus should stop reading a partner ChIP as a fusion readout

Finding the AFF1 protein sitting on a gene in a patient's leukaemia cells does not tell you the cancer-causing fusion protein is there, because normal AFF1 sits on many of the same genes anyway. A different protein, ENL, does tell you.

The same thing, for a scientist

**Question.** Sites where AFF1 and the retained KMT2A N-terminus are both present were expected to be the strongest available evidence of fusion occupancy in primary infant material. They are not. **Design.** 19,404 hg38 promoters (TSS +/- 1 kb, UCSC refGene, longest NM_ per symbol). Per patient state, using only that state's own KMT2A N- and C-terminal tracks: `ONCO` = top decile of N/C share ratio among above-median-N promoters (n = 971 in each of p279, p318, p454); `WT` = above-median N **and** above-median C with N/C within 1.25-fold of 1 (n = 3,531 / 3,038 / 3,699). There is no spike-in and no drug arm anywhere in this deposit, so every number is a WITHIN-SAMPLE share of promoter signal. A ratio between two site sets inside one sample is legitimate; a statement that any quantity rose or fell in absolute terms is not, and none is made. **Result - mean promoter share in ONCO divided by mean promoter share in WT** [E1]: | track | p279 (B-ALL) | p318 (AML, same patient) | p454 (B-ALL) | |---|---:|---:|---:| | IgG (the floor) | 0.75 | 0.86 | 1.08 | | KMT2A-N1 / N2 | 1.41 / 1.10 | 1.28 / 1.11 | 1.87 / 1.49 | | KMT2A-C1 / C2 | 0.27 / 0.36 | 0.18 / 0.35 | 0.36 / 0.49 | | **AFF1-1 (mouse mAb)** | **1.08** | **1.03** | **1.45** | | **AFF1-2 (rabbit pAb)** | **0.67** | **0.56** | **0.76** | | ENL R1 / R2 | 2.04 / 1.90 | 2.01 / 2.18 | 2.77 / 2.78 | | DOT1L R1 / R2 | 1.26 / 1.25 | 1.28 / 1.22 | 1.65 / 1.60 | | H3K4me3 R1 / R2 (negative control) | 0.98 / 0.97 | 1.05 / 1.06 | 1.04 / 1.06 | **The pre-registered branch that fires is the third one, 'neither'.** The two AFF1 antibodies disagree by 1.60x, 1.83x and 1.91x, which is more than either differs from 1 in p279 and p454. So the experiment does not resolve whether AFF1 is mildly enriched or mildly depleted at fusion sites, and the friendlier antibody was not picked. **But one branch is decisively excluded.** No AFF1 antibody comes near ENL's 1.9-2.8 in any state. Whatever AFF1's true value is, it is not a fusion-occupancy reporter in primary material. **The same result from the other direction.** Take each track's own top 971 promoters and ask how many are in `ONCO`. The null for a random 971 is 971 x 971 / 19,404 = **48.6** [E1]: | track | p279 | p318 | p454 | |---|---:|---:|---:| | KMT2A-C | 6 (0.1x null) | 2 (0.0x) | 7 (0.1x) | | IgG | 78 (1.6x) | 71 (1.5x) | 76 (1.6x) | | **AFF1-1** | 112 (2.3x) | 89 (1.8x) | 223 (4.6x) | | **AFF1-2** | 19 (0.4x) | 21 (0.4x) | 46 (0.9x) | | H3K4me3 (negative control) | 101 (2.1x) | 122 (2.5x) | 129 (2.6x) | | ENL | 169 (3.5x) | 171 (3.5x) | 265 (5.5x) | | KMT2A-N | 138 (2.8x) | 120 (2.5x) | 251 (5.2x) | **AFF1 with its better antibody beats the paper's own negative control in one state out of three.** In p318 H3K4me3 beats it outright. **This reproduces a published sentence and is reported as a reproduction.** The source paper states "AF4 does not effectively distinguish oncoprotein-target sites from wild-type sites in patient samples" [E2]. Knowing that before running is why this entry cannot be quoted as independent support for anything; what is new is the quantity, the per-antibody split, and the H3K4me3 benchmark placed beside it.

Go deeper
Why AFF1 is at wild-type KMT2A sites at all

AFF1 is a core component of the AF4-family/ENL-family/P-TEFb higher-order complex (AEP, the super elongation complex), which is recruited to wild-type KMT2A target chromatin in normal cells - the fusion's contribution is to make that recruitment constitutive, not to create it 3. So an AFF1 antibody reports the union of free SEC and fusion-bound SEC, and there is no reason it should mark fusion sites specifically. The result is the expected one once the complex biology is taken seriously; the surprise is only that the corpus has been treating partner occupancy as fusion occupancy.

What would overturn this

An AFF1 reagent with a fusion-junction-specific epitope, or an epitope-tagged fusion allele knocked into a primary sample, would separate fusion-bound from free AFF1 and could show strong fusion-site specificity that a total-AFF1 antibody cannot see. Nothing in this analysis argues that AFF1 is absent from fusion sites - only that its presence carries no information about the fusion.

Interval definition, stated because it is not the paper's

The paper calls 1,692 oncoprotein-target sites with a Bayesian N/C read-share model at a threshold of N/C > 1.75 and p < 1e-5 over its own peak intervals 2. This analysis uses fixed 2 kb promoter windows, which is coarser and differently anchored, so numbers are not expected to match the paper's. Sensitivity was checked: halving and doubling the pseudocount moves the median ONCO N/C ratio from 3.07 to 1.90 in p279 while leaving the ordering of the tracks unchanged.

Facts and figures
scriptanalyses/q142-aff1-in-primary-infants/analyze.py
statusrun
result summaryAFF1 ONCO/WT 1.03-1.45 (mouse mAb) and 0.56-0.76 (rabbit pAb) vs ENL 1.90-2.78 and H3K4me3 0.97-1.06; AFF1 does not discriminate.
n onco971
n promoters19404
inputs
[
  "GSE252378 bigWigs",
  "UCSC refGene hg38"
]
n wt
[
  3531,
  3038,
  3699
]
How this could bite us back
  • If AFF1 occupancy is used as a pharmacodynamic readout for a drug aimed at the fusion, a compound that removes the fusion while leaving free SEC in place will look inactive; conversely a compound that disrupts the SEC globally will look like it hit the fusion.
  • ENL is the discriminating track here, and ENL is itself a drug target - so an ENL-based occupancy readout would be confounded the moment an ENL inhibitor enters the same experiment.
Still open
  1. Is there any antibody or tag that reports fusion-bound AFF1 separately from free SEC AFF1 in primary cells?
  2. Do the entries in this corpus that cite AFF1 or AFF4 occupancy as evidence of fusion presence need to be re-read? A census has not been run.
Every source behind this entry (3)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.dataset

    Across the three primary KMT2A::AFF1 states in GSE252378, ENL is enriched 1.9-2.8-fold at N/C-defined fusion promoters over wild-type KMT2A promoters while the two AFF1 antibodies give 1.03-1.45 and 0.56-0.76 respectively, bracketing the IgG floor of 0.75-1.08 and the H3K4me3 control of 0.97-1.06.

    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 2024
    What this evidence cannot show

    Computed in this beat from the deposited hg38 bigWigs over 19,404 refGene promoters; scripts and the pre-registration are in analyses/q142-aff1-in-primary-infants. Promoter windows are not the paper's peak intervals. Two patients only, one of them sampled twice. No spike-in, so all values are within-sample shares. IgG is very sparse (72-80% of promoters zero), so its ratio is the noisiest row in the table.

    Evidence that argues against this (1)
    • AFF1-1 does reach 4.6x the null in p454 - higher than H3K4me3's 2.6x - so a reader who trusts the mouse monoclonal over the rabbit polyclonal can argue AFF1 IS a fusion reporter in a clean, non-lineage-switched infant B-ALL and that p279/p318 are degraded by low oncoprotein levels. This analysis cannot exclude that reading; it can only say the two reagents disagree and the deposit contains no third one. Janssens DH et al. 2024
  2. E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    The source paper concludes that AF4 co-occupancy cannot validate oncoprotein-target calls whereas DOT1L and ENL can, and defines target sites by an N/C ratio above 1.75.

    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 2024
    PMID 39472576DOI
    What this evidence cannot show

    Full text read from PMC11522475. The paper's conclusion rests on its own peak intervals and on the two AF4 antibodies POOLED, which is a different measurement from either antibody alone.

    Evidence that argues against this (1)
    • The paper's sentence is about its own interval set; a claim about promoters is a different claim, and a factor could in principle discriminate at enhancers while failing at promoters. Nothing here tests non-promoter intervals. Janssens DH et al. 2024
  3. E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper

    AF4-family and ENL-family proteins with P-TEFb form a higher-order complex that is normally recruited to wild-type MLL-target chromatin, and MLL fusions cause that recruitment to become constitutive.

    PMID 20153263DOI
    What this evidence cannot show

    Biochemistry and mouse transformation assays, not primary human infant leukaemia; establishes that free AFF1 reaches KMT2A targets, which is the mechanism this entry invokes, but does not itself measure primary patient chromatin.

    Evidence that argues against this (1)
    • If in a KMT2A::AFF1 cell essentially all AFF1 were sequestered into the fusion complex, the wild-type-site AFF1 signal would have to come from somewhere else and the mechanism offered here would be wrong; no AFF1 stoichiometry measurement in primary infant material exists to check it. Janssens DH et al. 2024
Connected to