The B-ALL and the AML of one infant share no more fusion-occupied promoters than two unrelated infants do - lineage, not patient identity, sets where KMT2A::AFF1 sits
One child's leukaemia was profiled twice, before and after it changed from a lymphoid to a myeloid cancer. The genes the fusion protein sat on in the two states overlapped no more than the genes it sat on in two completely different children.
The same thing, for a scientist
🔵 **REINSTATED IN PART 2026-09-24 (cycle 131). YOUR ORDERING SURVIVES A PROPER TEST; YOUR MAGNITUDE DOES NOT.** The cycle-128 banner below superseded this entry on the strength of a **92.6% retention figure that does not exist** — two independent per-sample counts read as one statistic — and of a **null belonging to a different pair**. **q146 recomputed on one universe with a structure-respecting null:** in **q144's own universe with q144's own rule**, same-patient gives **0.397** against **0.434** for `p279`-vs-`p454`, and `J(p279,p318) < J(p279,p454)` in **6 of 6** factorial cells. **That is this entry's ordering.** 🟠 **But the effect is far smaller than either beat implied and may not be about the patient at all.** On the pre-registered one-universe test the pair sits at the **37th percentile of 27 unrelated AFF1 pairs** — the pre-registered **NEITHER** fired. Every unrelated AFF1 pair shows the same excess, so **the excess is a fusion-level property**. And **GSE252378 carries an institution batch effect** (p = 0.0016) which confounds every same-patient comparison in the flattering direction, because a same-patient pair is same-institution by construction. **So: the direction you reported is supported; 'lineage beats patient identity' as a mechanism is not.** --- 🔴 **SUPERSEDED 2026-09-24 (cycle 128). THE ORDERING THIS ENTRY REPORTS IS NOT SUPPORTED.** Q144 measured the same `p279`/`p318` pair across a 27-sample cohort with an independence expectation and found **Jaccard 0.400 — the switch keeps 1,894 of 2,045 calls (92.6%)**, ranking **12th of 351 pairs, top 3.4%.** This entry's 0.117-versus-0.149 comparison rests on **three pairs with no chance floor**; against q144's null of **0.069** both values sit near chance and a 0.032 gap between them carries no information. 🔴 **It also merged two phenomena that behave oppositely.** The **untreated** switch (`p279→p318`) keeps 92.6% of calls; the **revumenib-treated** switch (`148752→152985`) falls **345 → 0**. *A lineage switch does not replace the fusion's addresses; a menin inhibitor erases them.* 🟠 **This beat was not careless — it was under-powered by my brief**, which told it the deposit held three primary samples. It holds far more. See `ana-two-beats-measured-the-same-lineage-switch-and-got-opposite-answers-and-the-better-powered-one-wins`. --- p279 (infant B-ALL at presentation) and p318 (the AML that the same patient's disease became under ALL-directed chemotherapy) are one patient in two lineage states; p454 is an unrelated infant B-ALL [E2]. **Overlap of the N/C-defined fusion-target promoter sets (971 per sample, Jaccard)** [E1]: | pair | relationship | Jaccard | |---|---|---:| | p279 ^ p318 | **same patient**, B-ALL vs AML | **0.117** | | p279 ^ p454 | different patients, **both B-ALL** | **0.149** | | p318 ^ p454 | different patients, AML vs B-ALL | 0.124 | **The within-patient pair is the LOWEST of the three.** The same ordering appears in every genome-wide profile, for every antibody class [E1]: | track | p279 vs p318 (same patient) | p279 vs p454 (both B-ALL) | |---|---:|---:| | KMT2A-N | 0.851 | **0.920** | | AFF1 | 0.831 | **0.893** | | H3K4me3 | 0.937 | **0.970** | | KMT2A-C | 0.729 | - | | IgG (floor) | 0.144 | 0.175 | **Three independent factor classes, three times the same direction: two B-ALLs from different children resemble each other more than one child's B-ALL resembles its own AML.** **A worked locus.** EBF1, the B-lineage transcription factor, is a fusion target in both B-ALLs (KMT2A-N 7.84x and 8.95x mean promoter, N/C 3.09 and 3.68, called ONCO in both) and is **not** one in the AML from the same patient as the first of them (N 2.54x, N/C 1.02, not called) [E1]. PAX5, CDK6, MEIS1, JMJD1C/KDM3C, FLT3 and PROM1 remain fusion targets in all three states; HOXA9/HOXA10 is a target in none of them by the N/C criterion. **What this is evidence for.** It is a within-deposit, within-protocol, within-build reproduction of the source paper's central claim that MLL oncoproteins relocate with lineage [E3]. It is one patient, so it is an existence proof rather than a rate.
The obvious alternative explanation, and why it does not dispose of the result
p318 is a post-chemotherapy relapse: therapy exposure, clonal selection and lineage change are completely confounded and this deposit cannot separate them. But the confound explains why p279 and p318 differ; it does not explain why they differ MORE than two unrelated patients do. For that, the shared thing between p279 and p454 has to be lineage, because it is not the patient, not the hospital (Texas Children's for all three) and not the batch (p279 and p318 were processed together, p454 separately - which if anything should have pushed the same-patient pair together).
A caution about the batch structure, stated because it cuts the other way
p279 and p318 KMT2A libraries share a run date (230626) and p454's share a different one (also 230626 for the AutoCUT&RUN, per the GEO title strings); the AFF1 libraries for all three are 240410. So the batch structure does not cleanly separate the patient pair from p454 and cannot be used to rescue the result either. The honest statement is that the comparison is not batch-controlled in either direction.
| script | analyses/q142-aff1-in-primary-infants/analyze.py |
|---|---|
| status | run |
| result summary | Same-patient cross-lineage Jaccard 0.117 < different-patient same-lineage 0.149; the ordering holds for KMT2A-N, AFF1 and H3K4me3 correlations. |
- If fusion-occupancy is lineage-contingent, a therapy validated against the B-ALL occupancy map may miss its targets in the myeloid relapse it selects for - which is exactly the clinical sequence this patient went through.
- CD19-directed therapy and lineage switch are already linked in this corpus; an occupancy-based target list built at diagnosis is a diagnosis-state artefact.
- Do the other two lineage-switching pairs in GSE252378 (KMT2A::ENL 148752/152985 and KMT2A::EPS15 R3595/R3652) show the same ordering? Both are in the deposit and neither was analysed here.
- Is EBF1's loss of fusion occupancy upstream or downstream of the B-to-myeloid switch?
its 0.117-versus-0.149 ordering rests on three pairs with no chance floor; q144 measures the same pair at Jaccard 0.400, keeping 1,894 of 2,045 calls, 12th of 351 pairs, against an independence expectation of 0.069
- supports →Lineage switch in KMT2A-rearranged leukemia — the same clone changes what kind of blood cell it is
occupancy-level evidence in one KMT2A::AFF1 patient
- related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
fusion target set is lineage-contingent
the AML state of the same patient
source deposit
- related to →Lineage switch as antigen escape from CD19-directed therapy in KMT2A-rearranged leukemia
occupancy changes with the lineage the therapy selects
global rho 0.83 and fusion-target Jaccard 0.12 in the same pair
a worked locus inside the same comparison
the occupancy comparison that defines the state
the Jaccard comparison
supplies the independence null that comparison was missing; direction unaffected, scale corrected
on this beat's universe and calling rule the paired patient is the MOST similar pair in the cohort, not the least
q142's direction survives against the closest comparators but not against the whole AFF1 arm
q142's quantile rule reproduced in three universes
q142's direction, unanimous across all six factorial cells against matched comparators
q142's p279-anchored ordering replicates in 6 of 6 cells including q144's own universe and rule