A target map read off the diagnostic sample does not describe the same child's disease after a lineage switch
If doctors profile a child's leukaemia at diagnosis and keep that list of target genes for later, the list will be about half wrong once the disease changes type - no better than a list borrowed from a different child.
The same thing, for a scientist
Diagnostic chromatin or occupancy profiling is increasingly proposed as a way to individualise KMT2A-rearranged leukaemia therapy. Measured on the only chemotherapy-driven lineage switch in the public record with paired oncoprotein profiling at both timepoints, 45.4% of the diagnostic oncoprotein-target promoters are no longer called after the switch and 40.7% of the post-switch calls are new, and the resulting overlap sits at the 37th percentile of 27 unrelated KMT2A::AFF1 sample pairs. The risk is not that profiling is useless; it is that a diagnostic target map will be treated as durable when the evidence at n = 1 says it is not, and that the earlier reading of this same data - a briefed 92.6% retention - would have supported exactly that mistake.
The specific failure mode
- A child is profiled at diagnosis; a target list is stored.
- The disease switches lineage under therapy.
- A salvage decision - which cooperating pathway to co-target, which pharmacodynamic marker to follow - is made from the stored list.
- Roughly half the list no longer applies, and the half that does is the fusion's generic preference rather than anything specific to this child.
The inflated retention figure makes step 3 look safe. This corpus carried that figure in a dispatch brief, which is how a measurement error becomes a clinical premise.
Mitigation
- Re-profile at switch. The assay is automatable on primary material and the switch is the event that invalidates the old map, so re-profiling at that event is the direct mitigation.
- Prefer fusion-level targets over patient-level ones for salvage, since the shared component is fusion-level.
- Do not use a diagnostic target list as a pharmacodynamic reference after a switch; validate the marker in the new state.
- Report retention asymmetrically (fraction of old calls kept AND fraction of new calls that are new), never as a single symmetric overlap, because the symmetric number hides which direction the change ran.
The counter-consideration that limits this
The continuous promoter-signal profiles of the two states are highly correlated (Spearman ~0.85 on the KMT2A N channel, and 0.94 for H3K4me3, measured by an earlier beat on the same samples). So the underlying landscape is similar and it is the THRESHOLDED list that moves. A clinical readout built on continuous signal at a named locus may well carry over even though a binary target list does not - which would make this risk a property of how the profiling is reported rather than of profiling itself.
| kind | resistance |
|---|---|
| severity | moderate - affects target selection and pharmacodynamic interpretation at salvage, not first-line dosing |
| mitigation | Re-profile at lineage switch rather than relying on the diagnostic map; prefer fusion-level targets for salvage; report retention asymmetrically; validate any pharmacodynamic marker in the post-switch state. |
| measured retention of diagnostic calls | 0.546 |
| measured new calls after switch | 0.407 |
| percentile among unrelated pairs | 37 |
affects
[ "drug-revumenib", "drug-ziftomenib" ]
- Re-profiling at switch requires marrow at the moment the disease is changing, which is the point at which a child is sickest and sampling is hardest - so the mitigation is the least available exactly when it is most needed.
- Preferring fusion-level targets concentrates the whole cohort on one mechanism, which raises the stakes on a single resistance pathway: a MEN1 interface mutation then removes the option for every child at once.
- Does a continuous-signal readout at a named locus (for example MEF2C or PROM1) carry over between the two states where the thresholded list does not? That would separate 'profiling does not transfer' from 'binary lists do not transfer'.
- Is the same non-transfer seen for ENL, which the paper reports persisting at oncoprotein-target loci even where the oncoprotein is gone? If ENL's map transfers and the oncoprotein's does not, ENL is the better diagnostic channel.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
In the p279-to-p318 lineage switch, oncoprotein target sites differ substantially between the two states, with the paper identifying distinct B-ALL-specific and GMP-like target gene sets and reporting reduced oncoprotein levels after the switch.
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
One patient. The paper's set difference is on its own peak intervals; the 45.4% / 40.7% figures are this beat's re-implementation on refGene promoters with within-pair depth matching.
Evidence that argues against this (1)
- Because oncoprotein LEVEL falls at the switch, some of the lost calls are the weak tail of a ranked list rather than genuinely re-addressed sites; a dilution model would predict loss without redistribution and this data cannot separate the two. Janssens DH et al. 2024
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
The automatable assay that would be used for diagnostic profiling, and for re-profiling at switch, is established on primary KMT2A-rearranged leukaemia material.
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
Feasibility of the assay is not feasibility of obtaining marrow at the moment of a lineage switch, which is the real constraint on the mitigation.
Evidence that argues against this (1)
- If re-profiling at switch is not practically obtainable, the mitigation reduces to 'prefer fusion-level targets', and the risk is not mitigated so much as accepted. Janssens DH et al. 2024
the mechanism that makes the map non-durable
the overlap measurement that sizes the risk
the same event under a menin inhibitor rather than chemotherapy
a stored diagnostic target map would be used to interpret menin-inhibitor pharmacodynamics
the serial replication is what would establish or remove this risk