A pre-registered, institution-matched serial-pair replication: does a patient's own target set predict itself better than another child's does?
To settle whether one child's leukaemia keeps its own set of target genes, you need several children profiled twice at the same hospital with the same method - including children whose disease did not change type. The public record has one such pair, which is not enough.
The same thing, for a scientist
The question 'does a patient's oncoprotein-target set predict itself across time better than another patient's set does' is currently answerable at n = 1 in the entire public record, and that single pair is confounded with a lineage switch and with chemotherapy. The design needed is serial AutoCUT&RUN with paired KMT2A N- and C-terminal antibodies in >= 8 KMT2A-rearranged children, all processed at one site in interleaved batches, sampled at diagnosis and at relapse, with at least half the relapses being SAME-lineage. The same-lineage serial pairs are the arm that isolates patient identity from lineage change; the switching pairs are the contrast.
Design
- Arm A (the missing control): >= 4 same-lineage serial pairs. Diagnosis and same-lineage relapse, one patient. This arm does not exist anywhere in the public record for KMT2A oncoprotein profiling and is the only way to separate 'patient identity does not persist' from 'lineage switch replaces the target set'.
- Arm B: >= 4 lineage-switching serial pairs, at least two of them chemotherapy-only (no menin inhibitor), because the two existing switch pairs are one chemotherapy switch and one menin-inhibitor switch and the drug arm calls zero sites.
- Batch design: interleave. Every batch contains both timepoints of some patients and single timepoints of others, so institution and batch are not collinear with patient. The measured institution effect here (within-site excess 1.456 versus 1.372 across, p = 0.0016) is small enough to be designed out and large enough to matter if it is not.
- Include a fusion-negative calibrator in every batch (CD34+ or a KMT2A-wild-type leukaemia). Its absence from GSE252378 is why the published 1.75 / 1e-5 threshold cannot be reproduced at its stated specificity and why a same-epitope antibody contrast calls ~60% as many promoters as the real N-versus-C contrast.
- Match depth by design, not by down-sampling afterwards: target equal per-arm fragment totals, because an absolute calling rule's set size tracks depth.
Readout and the pre-registered analysis
Primary: excess over an activity-stratified permutation null of the Jaccard between the two timepoints of one patient, against the distribution of the same statistic over institution-matched, depth-matched, anchored unrelated pairs. Report the coverage (universe size and comparable fraction) above every rate. Pre-register the calling rule AND report the alternative rule, because the direction of this comparison is rule-dependent.
Powered for: a 0.15 difference in excess between same-patient and unrelated at the observed unrelated spread (p10-p90 of 1.19-1.62) needs roughly 8-10 anchored comparisons per arm, which 4 serial pairs supply.
Why this and not a re-analysis
Everything re-analysable in GSE252378 has now been re-analysed three times by three beats, and the remaining uncertainty is not analytical. One patient, confounded with chemotherapy, with a second pair whose post-switch sample has no measurable oncoprotein, cannot be made to answer this by better statistics. The honest next step is samples.
| design | Serial AutoCUT&RUN with paired KMT2A N-terminal and C-terminal antibodies in >=8 KMT2A-rearranged children, two timepoints each; >=4 same-lineage serial pairs (the missing control) and >=4 lineage-switching pairs including >=2 chemotherapy-only switches. One processing site, interleaved batches so batch is not collinear with patient. A fusion-negative calibrator (CD34+ or KMT2A-wild-type leukaemia) in every batch. Equal per-arm fragment totals by design. |
|---|---|
| readout | Excess over an activity-stratified permutation null of the between-timepoint promoter-set Jaccard, against the distribution over institution-matched, depth-matched, anchored unrelated pairs. Coverage printed above every rate. Both the absolute (1.75/1e-5) and the quantile calling rule reported. |
| controls | Fusion-negative calibrator per batch; same-epitope antibody contrasts (N1 vs N2, C1 vs C2) as the false-positive floor; IgG; a same-lineage serial arm as the patient-identity control. |
| cost tier | high |
| timeline | 18-30 months, gated on serial relapse marrow availability |
- If the same-lineage serial arm shows a patient's target set does predict itself, then diagnostic profiling becomes clinically actionable and the whole fusion-level reading collapses - so this experiment is designed to be able to refute the hypothesis that motivated it.
- Are serial diagnosis/relapse marrow pairs with adequate viable material available in any existing KMT2A-rearranged biobank, or does this require prospective collection? That single logistical answer decides whether the experiment is 18 months or 5 years.
- Can a same-lineage serial control be approximated by two independent aliquots of one diagnostic sample processed in different batches? That would bound the technical ceiling without needing a second timepoint.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
AutoCUT&RUN/CUT&Tag with paired KMT2A N- and C-terminal antibodies is an established, automatable assay on primary KMT2A-rearranged leukaemia material, which makes a multi-patient serial design feasible rather than aspirational.
Automated CUT&Tag profiling of chromatin heterogeneity in mixed-lineage leukemia · Janssens DH, Meers MP, Wu SJ, Babaeva E, Meshinchi S, Sarthy JF, Ahmad K, Henikoff S · Nature Genetics 53:1586-1596 2021What this evidence cannot show
That work established the platform on cell lines and a smaller set of patient samples; serial sampling of the same child at diagnosis and relapse adds a tissue-availability constraint the platform paper did not face.
Evidence that argues against this (1)
- Relapse marrow in infant KMT2A-rearranged ALL is often obtained after therapy has already begun, so 'diagnosis and relapse' pairs may be systematically on-treatment at the second timepoint - the same confound that makes p279/p318 hard to read. Janssens DH et al. 2024
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The existing public record contains three lineage-switching samples out of 36 profiled, of which the deposited N/C arm carries two paired sets and one of those two post-switch samples has undetectable oncoprotein - which is the binding constraint this experiment addresses.
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 abstract and the deposit; the beat did not establish whether the third lineage-switching sample's partner is deposited under an unrecognised id.
Evidence that argues against this (1)
- If a third paired set exists in the deposit under ids this beat did not pair, the design's urgency drops from n=1 to n=2 - still inadequate, but the same-lineage control arm remains missing either way. Janssens DH et al. 2024
supplies the same-lineage serial control arm the hypothesis needs
turns the post-hoc institution-matched result into a pre-registered one
the interleaved-batch requirement follows from the measured institution effect
the per-batch fusion-negative calibrator is the same missing control
the serial replication is what would establish or remove this risk