Across one infant's B-ALL to AML lineage switch, HOXA occupancy collapses ~28-fold and IRX1 falls only ~2.2-fold
One baby's leukaemia changed from a B-cell type to a myeloid type under treatment. The HOXA genes lost the driver protein almost entirely; IRX1 kept most of it.
The same thing, for a scientist
p279 and p318 are **one patient**: the infant KMT2A::AFF1 B-ALL at diagnosis and the AML it lineage-switched into under ALL-directed chemotherapy, stated reciprocally in both GEO `!Sample_description` fields and in the deposit paper [E1]. Across that switch, KMT2A-N occupancy at HOXA9/HOXA10 falls from **2.92x to 0.23x (~12.6-fold at 20 kb; ~28-fold at the HOXA9 promoter, 3.43x to 0.12x)** while IRX1 gene-body occupancy falls only from **6.08x to 2.68x (~2.2-fold)** [E2]. The **ratio** IRX1:HOXA9 therefore moves from ~1.0 to ~17 — strongly reciprocal — while the **absolute** IRX1 signal goes down, not up.
A registered prediction that failed, recorded as failed
The pre-registration stated, blind: "if IRX1 is the reciprocal of HOXA, IRX1 should be HIGHER in p318 than in p279". It is lower. The reciprocity that does appear is in the ratio, not the level.
That distinction is not a rescue — it is the reason the ratio was the registered primary readout in the first place. GSE252378 has no spike-in, so between-sample absolute comparisons are not licensed at all and the failed prediction was, strictly, not a scoreable one. It is recorded because a prediction that could not have been scored is a defect in the pre-registration, and because the direction it assumed — that the two programs trade places one-for-one — is the intuition this beat is correcting.
Three confounds, all present at once
p318 differs from p279 in lineage (B to myeloid), in time, and in treatment (p318 is a post-chemotherapy relapse AML). n = 1 pair. Nothing here separates those three, and no statement of the form "the lineage switch causes X" is licensed by this comparison.
What the pair does give, uniquely in the deposit, is a within-patient comparison where germline background and clonal origin are held constant — which is worth more than it looks for a locus-ratio readout, because the ratio cancels library scale and the patient cancels genetic background.
Concordance with the deposit paper's own model
Janssens et al. report that in lineage-switching samples oncoprotein levels are reduced and the oncoprotein preferentially activates granulocyte-monocyte progenitor genes 1. A reduced total oncoprotein is exactly the regime where a within-sample share is the right readout and an absolute comparison is the wrong one. The observation here — that HOXA loses it far faster than IRX1 does — is a statement about redistribution of a shrinking pool, and that is the only statement the data supports.
| patient | p279 (B-ALL, dx) -> p318 (AML, post-chemo relapse) |
|---|---|
| n pairs | 1 |
| registered prediction | IRX1 higher in p318 than p279 - REFUTED, IRX1 is lower |
hoxa9 hoxa10 20kb
{
"p279": 2.924,
"p318": 0.226
}hoxa9 promoter 2kb
{
"p279": 3.428,
"p318": 0.022
}irx1 gene
{
"p279": 6.076,
"p318": 2.684
}ratio irx1 over hoxa9 2kb
{
"p279": 1.03,
"p318": 73.4
}confounds
[ "lineage", "time", "chemotherapy exposure" ]
- Lineage switch is a recognised escape from lineage-directed immunotherapy. If the IRX1 program is what survives the switch, targeting it would be targeting the escape route rather than the bulk disease - but that rests on one patient.
- Are there other paired pre- and post-switch KMT2A-r samples anywhere with oncoprotein profiling? The deposit states three lineage-switching samples; only this pair is KMT2A::AFF1 with a full panel.
- Does IRX1 occupancy predict which patients lineage-switch? p279 is the only switcher here and it is the one double-positive sample, which is suggestive and completely unpowered.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
p279 and p318 are the same patient: an infant KMT2A::AFF1 B-ALL and the AML resulting from its lineage switch; in lineage-switching samples oncoprotein levels are reduced and the oncoprotein preferentially activates GMP genes.
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 · Nat Commun 2024What this evidence cannot show
The pairing is stated reciprocally in both samples' GEO !Sample_description fields and in the paper. The GMP-program claim is the paper's conclusion across its lineage-switching samples, not a measurement made in this beat.
Evidence that argues against this (1)
- The paper's reduced-oncoprotein claim is supported by its own spike-in-free CUT&RUN plus orthogonal measurements not reproduced here; a within-sample share analysis like this one cannot independently confirm that levels fell. Janssens DH et al. 2024
- E2moderate confidence 3 of 5: One strong study, or several smaller consistent ones.dataset
Across the p279-to-p318 switch, KMT2A-N at HOXA9/HOXA10 falls from 2.92x to 0.23x while IRX1 gene-body occupancy falls from 6.08x to 2.68x, moving the IRX1:HOXA9 ratio from ~1.0 to ~17.
q145 re-analysis of GSE252378: IRX1 and HOXA oncoprotein occupancy across seven KMT2A::AFF1 patients (analyses/q145-irx1-in-hoxa-negative-infants) · Iqbal Lab oncofusion knowledge graph, beat q145 · internal analysis over deposited GSE252378 bigWigs 2026What this evidence cannot show
n = 1 patient pair, confounding lineage, time and chemotherapy exposure. No spike-in, so only the ratio is interpretable and the absolute fold-changes quoted are within-sample shares, not amounts. Every number is a WITHIN-SAMPLE share: an interval's mean coverage divided by that same library's mean promoter coverage, so 1.0 = an average promoter in that library. **GSE252378 has no spike-in and no drug arm**, so no between-sample absolute statement is licensed; the only cross-sample quantity used is a ratio of two loci measured inside one library, where the unknown per-library scale cancels.
Evidence that argues against this (1)
- One deposit, one lab, one assay; seven patients. A second cohort profiled by a different group could reverse any of this. Janssens DH et al. 2024
- supports →IRX1 — Iroquois homeobox 1, the gene that names the HOXA-negative half of infant t(4;11) ALL
IRX1 falls 2.2-fold while HOXA falls ~28-fold across one patient's lineage switch
the switched AML retains the IRX1 address
p318 is the AML arising from an infant B-ALL lineage switch