Experiment: go back to the diagnostic sample and look for the myeloid clone before it existed
Take stored samples from the day of diagnosis, from children who later switched to a myeloid leukemia and from matched children who did not, and search very hard for a few myeloid-looking leukemia cells hiding among millions.
The same thing, for a scientist
A case-control archaeology on banked diagnostic material, designed around detection limit rather than around breadth. Cases are KMT2A-rearranged patients with subsequent confirmed lineage switch after CD19-directed therapy; controls are matched KMT2A-rearranged patients treated with the same modality who did not switch. The core requirement is an assay that reaches 10-3 to 10-4 for a transcriptional state, which standard discovery single-cell does not, so the design pairs enrichment with targeted profiling and includes an explicit sensitivity demonstration on spike-in mixtures before any patient sample is run.
Design
Step 0, sensitivity demonstration. Spike a known myeloid-primed KMT2A::AFF1 population into a B-lineage background at 10-2, 10-3, 10-4 and 10-5 and establish the assay's actual floor. Publish that floor. Without it, a negative result is uninterpretable, and Step 0 alone is the reason this experiment is worth designing carefully rather than running quickly.
Step 1, enrichment and profiling. Deplete the dominant CD19-positive population, or index-sort on candidate myeloid and progenitor markers (CD33, CD13, CD117, NG2, HLA-DR, IL7R), then run single-cell RNA-seq with paired surface protein and, where possible, single-cell ATAC. Per-cell fusion-transcript detection is essential — a myeloid cell without the fusion is normal myelopoiesis, not a precursor, and that distinction is the whole experiment.
Step 2, phylogeny. Where whole-genome sequencing of diagnostic and switch material is available, reconstruct the phylogeny and ask whether the switch clone branches before or after the start of immunotherapy, replicating in KMT2A-rearranged cases what has been shown in one KMT2A-wild-type case 2.
Step 3, covariates. Score CHD4 and chromatin-modifier status at diagnosis 3 and fit it alongside subpopulation frequency, so the analysis apportions between pre-existing substrate and induced transition rather than testing one against the other.
Feasibility: how many cases exist
Lineage switch is rare — 12 of 166 relapses among 420 CD19 CAR recipients, 7.2% of relapses and 2.9% of treated patients 1 — and banked diagnostic material with adequate cell numbers is rarer still. The Project EVOLVE consortium assembled 70 post-immunotherapy switch cases internationally 4, 45 of them KMT2A-rearranged, which is the realistic ceiling on case ascertainment and the obvious collaboration route.
Realistically this is a 15-to-25-case study with matched controls, powered to detect a large frequency difference and nothing subtler. That should be stated in the protocol, not discovered in the discussion.
The interpretive trap
Finding rare fusion-positive myeloid-primed cells at diagnosis in a switcher proves very little on its own, because they may be present in everyone. The comparison to matched non-switchers is the experiment; the case-only version is an anecdote with a sequencer attached. Controls must be matched on treatment modality, on age, on disease burden and on the depth of response achieved, because depth of response is itself hypothesised to drive the release.
| design | Case-control study on banked diagnostic samples: KMT2A-rearranged patients with confirmed post-CD19-therapy lineage switch versus matched KMT2A-rearranged non-switchers treated with the same modality, profiled by enrichment plus single-cell multi-omics with per-cell fusion detection, with a phylogenetic arm where paired whole-genome sequencing exists and chromatin-modifier status as a covariate. |
|---|---|
| readout | Frequency of fusion-positive myeloid-primed or lineage-ambiguous cells in the diagnostic sample; branch point of the switch clone relative to the start of immunotherapy; incremental predictive value over KMT2A genotype alone; independent contribution of CHD4 and chromatin-modifier status. |
| controls | Mandatory spike-in sensitivity demonstration establishing the assay floor before any patient sample is run; per-cell fusion detection to distinguish precursor cells from normal myelopoiesis; matched non-switcher controls matched on modality, age, disease burden and depth of response achieved. |
| cost tier | high |
| timeline | 18-24 months, dominated by sample ascertainment rather than by laboratory work |
| feasible n | realistically 15-25 cases with matched controls, powered only for a large frequency difference; this must be stated in the protocol rather than in the discussion |
| go no go | If the Step 0 spike-in demonstration cannot reach 10-3, redesign the assay before touching patient material; a negative result from an assay whose floor is unknown is worthless. |
| decision value | Delivers the first pre-treatment lineage-switch risk marker beyond genotype, or establishes that switch is reprogramming and that no diagnostic marker will exist — both of which change how KMT2A-rearranged children on CD19-directed therapy are monitored. |
tests hypotheses
[ "hyp-lineage-switch-is-competitive-release-of-preexisting-clone" ]
collaborators needed
[ "Project EVOLVE consortium for case ascertainment and banked diagnostic material", "a single-cell multi-omics group with rare-population enrichment experience" ]
- A positive result creates a risk score whose misuse — withholding CD19-directed therapy from KMT2A-rearranged infants — would cause more harm than the switch it predicts. That constraint belongs in the protocol and in the abstract, not in a limitations section.
- With 15-25 cases the study will be underpowered for anything but a large effect, and a null result will be ambiguous between absence of the population and absence of power. The spike-in floor is the only thing that makes a null interpretable at all.
- Does banked diagnostic material with adequate viable cell numbers exist for the EVOLVE KMT2A-rearranged switch cases, and if not, can a prospective banking requirement be attached to the ongoing frontline blinatumomab trials at near-zero marginal cost?
- Should matched controls be drawn from patients who achieved equally deep responses, given that depth of response is itself the hypothesised trigger for release?
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Among 420 children and young adults treated with CD19 CAR T cells, 166 relapsed of which 12 (7.2% of relapses, 2.9% of treated patients) were lineage switch, and KMT2A rearrangement was the only pre-infusion factor associated with lineage-switch relapse.
Preinfusion factors impacting relapse immunophenotype following CD19 CAR T cells · Lamble AJ, Myers RM, Taraseviciute A, et al. · Blood Advances 2023What this evidence cannot show
Establishes the rarity that sets the feasible sample size, and establishes genotype as the comparator any new biomarker must beat.
Evidence that argues against this (1)
- Twelve events. KMT2A rearrangement being the only pre-infusion factor associated with lineage switch is a univariate finding on 12 outcomes, where any multivariable adjustment is impossible and the chance of a single factor reaching significance by itself is substantial. Statistics in Medicine 2006
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Whole-genome phylogenetic reconstruction in a post-immunotherapy lineage-switch case showed the AML derived from the initial ALL and carried distinct driver mutations detectable before the AML emerged, demonstrating that the phylogenetic arm of this design is technically achievable.
Clonal origin of KMT2A wild-type lineage-switch leukemia following CAR-T cell and blinatumomab therapy · Coorens THH, Behjati S, et al. · Nature Cancer 2023What this evidence cannot show
KMT2A wild-type, single case; this experiment is in part a replication of it in the genotype of interest.
Evidence that argues against this (1)
- The precedent this design is modelled on is a single case and, critically, a KMT2A *wild-type* one — the phylogeny that found pre-existing subclonal drivers was reconstructed in a child whose lineage switch occurred without a KMT2A rearrangement. The population this experiment targets is KMT2A-rearranged, where the published mechanistic account is different (chromatin-modifier-driven reprogramming with shared fusion breakpoints between lymphoid presentation and myeloid relapse). Generalising a design assumption from the wild-type case to the rearranged population is the specific step that needs justifying in the protocol. Tirtakusuma R et al. 2022
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Lineage switching in MLL/AF4 leukemia is recurrently associated with altered expression, splicing or mutation of chromatin modifiers notably CHD4, whose perturbation alone induces myeloid gene expression, making chromatin-modifier status a required covariate rather than an optional one.
Epigenetic regulator genes direct lineage switching in MLL/AF4 leukemia · Tirtakusuma R, Szoltysek K, Milne P, et al. · Blood 2022What this evidence cannot show
Supplies the competing mechanism the design must measure alongside, not against.
Evidence that argues against this (1)
- A negative archaeology result will be uninterpretable unless the reprogramming alternative is measured in the same samples. If switching is driven by acquired or altered chromatin-modifier function (CHD4 being the recurrent example, with CHD4 perturbation alone sufficient to induce myeloid gene expression), then failing to find a pre-existing myeloid-primed subclone is the expected result and says nothing about detection limits. The design should therefore pair deep genetic archaeology with a chromatin/ATAC or CHD4-status readout on the same diagnostic material, so that absence of a subclone and presence of a reprogramming lesion can be distinguished from assay failure. Tirtakusuma R et al. 2022
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Project EVOLVE assembled 70 post-immunotherapy lineage-switch cases internationally, 45 (64.3%) of them KMT2A-rearranged, with median time to switch of 1.5 months and 54.3% switching from documented MRD-negative complete remission.
Project EVOLVE: an international analysis of postimmunotherapy lineage switch, an emergent form of resistance · Project EVOLVE consortium · Blood 2025What this evidence cannot show
Whether diagnostic material with adequate cell numbers exists for those 45 cases is unknown and is the first question to ask the consortium.
Evidence that argues against this (1)
- Project EVOLVE assembles cases BECAUSE they switched, so 64.3% KMT2A-rearranged and 54.3% switching from MRD-negative complete remission are proportions conditional on switching and not rates. The median time of 1.5 months is also subject to ascertainment: early switches are the ones recognised as switches rather than as new leukaemias. Journal of Clinical Oncology 2025
- related to →Lineage switch as antigen escape from CD19-directed therapy in KMT2A-rearranged leukemia
a negative archaeology result is the expected outcome under the reprogramming account, so chromatin/CHD4 readouts must be run on the same diagnostic material for the result to be interpretable
case-control archaeology on banked diagnostic samples with a mandatory spike-in sensitivity floor
- depends on →Analysis: score fetal, myeloid-priming and differentiation-stage signatures across public KMT2A-r datasets
uses the existing myeloid-priming and differentiation-stage signatures to define the precursor state
tests whether the myeloid competence attributed to the arrested state is present as a pre-existing subclone before immunotherapy rather than acquired by reprogramming
CHD4 is one of the epigenetic-regulator lesions that experiment would be looking for at diagnosis
- tests →ZEB2 level sets how easily a KMT2A-rearranged blast can change lineage under CD19-directed pressure
paired diagnosis/switch samples would test whether the myeloid-primed subclone is ZEB2-low