Oncofusion

start / graph / mechanism-lineage-switch-kmt2a

Mechanismstrong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.Q4-cells

Lineage switch in KMT2A-rearranged leukemia — the same clone changes what kind of blood cell it is

also called lineage switch · lineage conversion · ALL-to-AML switch · phenotype switch · immunophenotypic switch · lineage plasticity in KMT2A-r leukemia

A KMT2A-rearranged leukemia can stop looking like a B-cell leukemia and start looking like a myeloid (AML) leukemia, while the DNA proves it is still the very same cancer clone. The cell has not been replaced; it has changed career, and that is how it walks away from treatments aimed at B cells.

The same thing, for a scientist

Lineage switch (LS) is the morphological and immunophenotypic conversion of an acute leukemia from one lineage to another with retention of the founding cytogenetic/molecular lesion, most classically KMT2A-rearranged (KMT2A-r) B-ALL relapsing as AML (usually with monocytic differentiation), and less often KMT2A-r AML converting to B-ALL. Clonal identity is established by an identical KMT2A fusion breakpoint, retained IGH/IGK (Ig) or TCR rearrangements, and shared SNVs. Mechanistically the switch is not a new mutation event: it is a large-scale reorganisation of chromatin accessibility and transcriptional programs, with recurrent involvement of chromatin modifiers (notably CHD4/NuRD), loss of B-lineage transcription-factor output (PAX5, EBF1) and gain of myeloid/CEBPA-driven programs. Paired sequencing shows that both routes exist — selection of a genetically divergent sister clone in roughly half of informative cases, and direct reprogramming of the bulk blast population in the other half — so 'subclone selection versus transdifferentiation' is a false dichotomy at the cohort level.

Go deeper
What is actually observed at the bedside2 deeper

The pre-immunotherapy baseline rate is low but real. In a single-centre paediatric series of 1,482 consecutive acute leukemias, nine patients (0.6%) converted lineage: seven from lymphoid (four pro-B, two pre-B, one common) to myelo-monocytic, and two from a bilineal myeloid-predominant leukemia to pro-B ALL. Eight of the nine were infants, 11q23/MLL (KMT2A) abnormalities were present in seven of nine, the switch was confirmed at a median of 15 days from the start of therapy (range 8 days to 6 months, with five before day 15), and all nine patients died despite chemotherapy being changed to match the new lineage 1.

That series also establishes the diagnostic standard the field still uses: a switch is only a switch if the clone is shown to be the same one. Rossi and colleagues did this with karyotype, RT-PCR for the fusion transcript, and Ig/TCR rearrangement studies in every patient 1. Modern series apply the same logic — in the international Project EVOLVE cohort, clonal relatedness was formally confirmed in 64 of 70 (91.4%) cases, with B-cell receptor gene rearrangements (PCR or clonoSEQ) available in 25 and retained in 20 of them 6.

The direction of travel, and the rarer reverse direction

B-ALL → AML dominates. Across 70 post-immunotherapy LS cases, 53 (75.7%) were B-ALL to AML and 17 (24.3%) were B-ALL to B/myeloid MPAL or undifferentiated leukemia 6. The myeloid product is characteristically monocytic or myelomonocytic — 21 of 33 (64%) in one multi-institutional pathology series 5.

The reverse happens too, and it matters because it shows the plasticity is bidirectional rather than a one-way collapse into myeloid identity. Two of nine paediatric switches in the Rossi series went myeloid → pro-B 1, and four of 33 cases in the Zhou series converted from AML (including one myeloid sarcoma) to B-ALL, two of them in infants with monocytic AML 5.

The spontaneous reversion case

The most informative single case in the literature is an infant with KMT2A-rearranged ALL in whom blinatumomab drove an ALL-to-AML switch that then spontaneously reverted — reported by Wölfl and colleagues as a leukemia 'that rapidly adapts to immunological stimuli, illustrating the high plasticity of this phenotype' 7.

This is not a single anecdote. Three of 33 patients in the Zhou multi-institutional series are recorded as having 'switched back' to the original lineage at last follow-up 5. Reversion is the strongest available argument that the switch is a reversible cell-state transition under environmental (immune) pressure, not an irreversible genetic commitment — because a subclone that has been positively selected does not un-select itself when the pressure is removed.

The biology: chromatin, not a new mutation2 deeper

The definitive mechanistic study paired lymphoid presentations with their myeloid relapses in MLL/AF4 (KMT2A::AFF1) leukemia. Three findings anchor everything else 2:

  1. The myeloid relapses shared the oncogene fusion breakpoints with their matched lymphoid presentations — the fusion is retained, unchanged, at the same junction.
  2. The switched leukemias originated from a range of differentiation stages, from immature progenitors all the way through to committed B-cell precursors. There is no single 'switch-competent' cell.
  3. Switching is accompanied by substantial changes in chromatin accessibility and rewiring of transcriptional programs, including alternative splicing — i.e. the execution and maintenance of lymphoid differentiation is impaired rather than a myeloid driver being acquired.

Crucially, that study named a mechanism rather than describing a correlation: the myeloid phenotype was recurrently associated with altered expression, splicing or mutation of chromatin modifiers, including CHD4, the ATPase/helicase of the NuRD (nucleosome remodelling and deacetylation) complex, and perturbing CHD4 alone or with other mutated epigenetic modifiers induced myeloid gene expression in MLL/AF4+ cell models 2.

Transcription-factor rewiring: PAX5, EBF1, CEBPA

The B-lineage identity of a pro-B cell is actively maintained by a transcription-factor network; remove the maintenance and myeloid potential re-emerges. In murine pre-B ALL models under sustained CD19 CAR-T pressure, deletion of Pax5 or Ebf1 recapitulated the lineage reprogramming that occurred under CAR pressure, and whether the switch was partial or complete depended on the underlying genetic oncogenic driver 4. That is the cleanest causal link available between loss of B-lineage TF output and acquisition of a myeloid program in this setting.

On the myeloid side, single-cell work supplies the positive driver. Integrating scRNA-seq of 89 B-ALL samples with an atlas of normal human B-cell development showed that the subtypes prone to shift out of the B lineage — BCR::ABL1, KMT2A-R and DUX4-R — are enriched for multipotent progenitors, and that this developmental stage exhibits CEBPA activation and retains myeloid potential 3. A derived 'multipotency score' was most enriched in the plasticity-prone subtypes and was independently associated with inferior survival 3. This gives the clinical observation a developmental explanation: the switch-prone KMT2A-r cell was never a fully committed B cell to begin with.

What is not established: no study I could verify shows that a specific PAX5 or EBF1 lesion is acquired at the moment of clinical lineage switch in KMT2A-r patients. The PAX5/EBF1 evidence is causal in mouse models 4 and developmental-context evidence in human single-cell data 3; the human genetic proof is missing.

Fusion partner and breakpoint may set the plasticity ceiling

A human CRISPR/Cas9 model introduced MLL::AF4 or MLL::AF9 fusions with KMT2A breakpoints in intron 9 or intron 11 into cord-blood CD34+ cells from the same donor, removing genetic-background confounding. Transcriptomic profiling revealed breakpoint- and partner-specific expression patterns, and MLL(intron 9)::AF9 cells engrafted NSG mice and switched from a myeloid to a B-lymphoid identity in vivo 8. This is direct experimental evidence that lineage plasticity is intrinsic to the KMT2A-r cell rather than purely therapy-induced, and that the amount of plasticity is partner- and breakpoint-dependent.

Subclone selection or true transdifferentiation? Both, in about equal measure2 deeper

This is the question the field usually poses as either/or. The best available human data say it is neither exclusively.

In a 33-case multi-institutional series with clonal-relatedness required for inclusion, paired next-generation sequencing was available for the first and second leukemia in 10 patients. In five, there were shared alterations plus distinct additional mutations in each leukemia — divergent clonal evolution from a common leukemia-initiating cell that retained bipotential. In the other five, there was no clear branching: the full original mutation set was maintained, with or without additions, consistent with the bulk blast population undergoing direct reprogramming or dedifferentiation to a multipotent state followed by re-commitment 5.

Two cohort-level observations push in the same direction. First, the second leukemia carried a more complex karyotype than the first (three or more aberrations in 17/29 vs 10/30, P = 0.05) — some genetic evolution accompanies the switch 5. Second, and cutting the other way, LS in KMT2A-r cases occurs very early (median 1.5 months from the most proximal immunotherapy, and in Rossi's chemotherapy-era series a median of 15 days from starting therapy, five of nine before day 15) 16. Fifteen days is a short window in which to expand a rare pre-existing myeloid subclone to clinical dominance from a standing start, and it is the single strongest argument for reprogramming over pure selection.

KMT2A-r switches look different from non-KMT2A switches — which is itself an argument

In the Zhou series, 18 of 33 (54.5%) cases shared an 11q23/KMT2A fusion between the two leukemias, predominantly from t(4;11) (n = 11), then t(9;11) and t(11;19) (n = 2 each) 5. Splitting the cohort on KMT2A status produced two nearly non-overlapping clinical pictures: patients without KMT2A fusions were older (46.1 vs 25.2 years, P = 0.03), far less often had monocytic AML (5/15 vs 16/18, P < 0.001), had more complex karyotypes and more mutations, and 40% (6/15) had an antecedent chronic myeloid malignancy (MDS, PV, PMF, CML) preceding the first acute leukemia. No patient with an 11q23/KMT2A fusion had a preceding chronic myeloid neoplasm 5.

Read carefully, that is a mechanistic dissociation. In the non-KMT2A group the myeloid 'switch' is plausibly the unmasking of a pre-existing myeloid clone. In the KMT2A-r group there is no pre-existing myeloid clone to unmask, the AML is monocytic, and the switch is fast. The same paper notes the likely reason KMT2A-r needs less help: a founder effect, where the fusion alone suffices and additional leukemogenic events are not required 5.

The switched cell is not just phenotypically different — it is immunosuppressive

Multi-omic profiling of patient samples found that lineage-switched AML cells with KMT2A-r acquire monocytic myeloid-derived suppressor cell (M-MDSC)-like characteristics. Single-cell mass cytometry showed an increase in the M-MDSC-like population in LS AML relative to lineage-consistent KMT2A-r AML; single-cell transcriptomics identified distinct immunoregulatory gene expression in that population; and in vitro assays confirmed immunosuppressive capacity against T cells, analogous to MDSCs 9.

If this holds, lineage switch is not merely antigen loss. The switched cell actively suppresses the T-cell compartment that the immunotherapy depends on — which would explain both why switches emerge while CAR T cells are still present and why re-treating with another T-cell-engaging agent tends to fail.

Where this could be wrong

Three honest weaknesses.

(1) Lineage switch is not KMT2A-specific, and is not always bad. Among 726 consecutively diagnosed children with BCP-ALL, 8% experienced a switch to the monocytic lineage detectable by flow cytometry with loss of the B-cell immunophenotype including CD19 — and the genetic correlates were PAX5-P80R (5/5 cases), DUX4-r (30/41) and ZNF384-r (4/10), not KMT2A. Critically, despite substantial phenotypic change, the monocytic switch was not associated with a higher incidence of relapse or poorer prognosis in patients on standard ALL treatment 10. So 'immunophenotypic conversion' and 'lethal lineage switch' are not the same phenomenon, and a purely phenotype-based definition will over-call.

(2) Almost every human dataset is retrospective and assembled by case solicitation. Project EVOLVE explicitly states that the incidence of post-immunotherapy LS in KMT2A-r or infant ALL 'is not fully elucidated, and has ranged from 0% to 28.6% after CD19-directed CAR T cells across retrospective patient cohorts' 6. A 0–28.6% range is not an incidence estimate.

(3) The 'reversible cell state' model has been tested at scale in mice and cell lines, but reversion in humans rests on one published case plus three table entries 57. The human evidence for reversibility is thin, and I could not obtain the full text of the reversion case report to verify how the reversion was documented.

Facts and figures
mechanism classepigenetic/transcriptional cell-state transition under selective pressure
phenomenonlineage switch (LS)
canonical directionKMT2A-r B-ALL to AML, usually monocytic/myelomonocytic
reverse direction documentedtrue
reversion documentedtrue
frequencies
lineage conversion among …0.600lineage conversion among consecutive paediatric acute leukemias: 0.6monocytic switch detectab…8.000monocytic switch detectable by flow cytometry in unselected paediatric BCP-ALL (non-KMT2A drivers): 811q23/KMT2A fusion shared…54.511q23/KMT2A fusion shared between first and second leukemia among clonally confirmed lineage switches: 54.5monocytic/myelomonocytic …64.0monocytic/myelomonocytic differentiation of the switched AML: 64KMT2A rearrangement among…64.3KMT2A rearrangement among post-immunotherapy lineage switches from B-ALL: 64.3
clonality proof methods
[
  "identical KMT2A fusion breakpoint (RT-PCR / karyotype / FISH)",
  "retained IGH/IGK (Ig) and TCR rearrangements (PCR or clonoSEQ)",
  "shared somatic SNVs on paired NGS"
]
molecular effectors
{
  "chromatin": [
    "CHD4 (NuRD ATPase/helicase)",
    "EZH2 (in non-KMT2A lineage switch)"
  ],
  "b_lineage_tf_loss": [
    "PAX5",
    "EBF1"
  ],
  "myeloid_tf_gain": [
    "CEBPA"
  ],
  "other": [
    "genome-wide chromatin accessibility change",
    "alternative splicing rewiring"
  ]
}
frequencies — the numbers behind the chart
[
  {
    "metric": "lineage conversion among consecutive paediatric acute leukemias",
    "value": 0.6,
    "unit": "%",
    "n": 1482,
    "population": "paediatric acute leukemia, single centre, pre-immunotherapy era",
    "source_evidence": "E1"
  },
  {
    "metric": "11q23/KMT2A fusion shared between first and second leukemia among clonally confirmed lineage switches",
    "value": 54.5,
    "unit": "%",
    "n": 33,
    "population": "all ages, multi-institutional",
    "source_evidence": "E5"
  },
  {
    "metric": "KMT2A rearrangement among post-immunotherapy lineage switches from B-ALL",
    "value": 64.3,
    "unit": "%",
    "n": 70,
    "population": "international solicited cohort",
    "source_evidence": "E6"
  },
  {
    "metric": "monocytic/myelomonocytic differentiation of the switched AML",
    "value": 64,
    "unit": "%",
    "n": 33,
    "population": "clonally confirmed lineage switch",
    "source_evidence": "E5"
  },
  {
    "metric": "monocytic switch detectable by flow cytometry in unselected paediatric BCP-ALL (non-KMT2A drivers)",
    "value": 8,
    "unit": "%",
    "n": 726,
    "population": "consecutive paediatric BCP-ALL",
    "source_evidence": "E10"
  }
]
timing
[
  {
    "metric": "median time from start of therapy to confirmed switch (chemotherapy era)",
    "value": 15,
    "unit": "days",
    "range": "8 days to 6 months",
    "source_evidence": "E1"
  },
  {
    "metric": "median time from most proximal immunotherapy to switch",
    "value": 1.5,
    "unit": "months",
    "range": "0 to 36.5 months",
    "source_evidence": "E6"
  }
]
outcomes
[
  {
    "metric": "median overall survival after lineage switch",
    "value": 2.9,
    "unit": "months",
    "n": 33,
    "source_evidence": "E5"
  },
  {
    "metric": "median overall survival after post-immunotherapy lineage switch",
    "value": 4.8,
    "unit": "months",
    "n": 70,
    "source_evidence": "E6"
  }
]
origin model
{
  "divergent_clonal_evolution_cases": 5,
  "no_clonal_branching_cases": 5,
  "informative_paired_ngs_cases": 10,
  "interpretation": "both selection of a genetically divergent sister clone and direct reprogramming of the bulk blast population occur; neither explains all cases",
  "source_evidence": "E5"
}
How this could bite us back
  • Antigen escape without antigen loss: a CD19-directed agent can achieve a genuine MRD-negative remission of the B-ALL and still fail, because the same clone re-emerges as CD19-negative AML. Measuring CD19 on the relapse tells you nothing unless you also re-phenotype all lineages.
  • Lineage-restricted MRD assays go blind at exactly the wrong moment. B-cell-oriented flow cytometry underestimates MRD once a clone starts drifting out of the B lineage, which has been demonstrated directly in switch-prone subtypes; a clone can therefore look eradicated while it is converting.
  • The switched cell may actively disarm the therapy. If lineage-switched KMT2A-r AML acquires M-MDSC-like T-cell-suppressive function, then any subsequent T-cell-engaging salvage (second CAR, another bispecific) is being deployed into an immunosuppressed marrow of the leukemia's own making.
  • Multi-antigen targeting could select for a deeper de-differentiation rather than preventing escape. Targeting CD19 plus a myeloid antigen assumes the clone has only two states; the paired-NGS data and the multipotency-score work suggest the switch-competent cell sits at a multipotent progenitor stage from which several exits exist.
  • Differentiation-forcing agents and lineage plasticity interact in an unknown direction. Menin inhibitors act by relieving the differentiation block in the same cells whose problem is uncontrolled differentiation flexibility; this could either lock in a terminal fate (good) or accelerate exit from the B lineage (bad).
Still open
  1. Is lineage switch reversible in humans if antigen pressure is withdrawn? One published case reverted spontaneously and three patients in a 33-case series 'switched back', but nobody has prospectively stopped CD19-directed therapy at the moment of switch and serially profiled the clone. This is the single experiment that would decide between a reversible cell state and an irreversibly selected clone.
  2. Does a KMT2A-r patient's pre-treatment chromatin state predict who will switch? Tirtakusuma et al. showed the switched relapse has a different accessibility landscape, but no study has asked whether the diagnostic sample already contains an ATAC or methylation signature of switch-competence. A retrospective scATAC comparison of diagnostic marrows from KMT2A-r patients who later switched versus matched patients who relapsed in-lineage is cheap and has never been published.
  3. Is CHD4 loss necessary, or only sufficient, for lineage switch in KMT2A::AFF1? CHD4 perturbation induces myeloid gene expression in MLL/AF4+ cells, but the converse experiment — forcing CHD4/NuRD activity to block switching under CAR-T pressure in vivo — has not been reported.
  4. Does menin inhibition promote or prevent lineage switch? Revumenib works by forcing differentiation of KMT2A-r blasts, and differentiation syndrome is its signature toxicity. Whether pharmacologically pushing a plastic KMT2A-r cell toward a myeloid program makes subsequent CD19-directed therapy more or less likely to provoke a switch is completely unstudied and is directly relevant to the sequencing of menin inhibitors with immunotherapy.
  5. Why is the switched AML almost always monocytic (64% of cases)? Monocytic bias is one of the most reproducible features of KMT2A-r lineage switch and nobody has explained it. If the monocyte/macrophage program is the default attractor for a de-differentiating KMT2A-r cell, that program's dependencies are candidate targets for pre-empting the switch.
Every source behind this entry (10)
  1. E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In 1,482 consecutive paediatric acute leukemias, nine (0.6%) underwent lineage conversion — seven lymphoid to myelo-monocytic and two myeloid to pro-B — with eight of nine being infants, 11q23/MLL abnormalities in seven, confirmation at a median of 15 days from starting therapy, clonal stability proven by cytogenetics, RT-PCR and Ig/TCR rearrangement studies, and death in all nine.

    Lineage switch in childhood acute leukemia: an unusual event with poor outcome · Rossi JG, Bernasconi AR, Alonso CN, Rubio PL, Gallego MS, et al. · American Journal of Hematology 2012
    PMID 22685031DOI
    What this evidence cannot show

    Single-institution retrospective series from Argentina; nine events is too few to model risk factors, and the 0.6% denominator is the whole acute-leukemia population rather than the KMT2A-r population, so it understates the per-patient risk in KMT2A-r. Pre-dates immunotherapy, so it cannot speak to antigen-directed pressure. 'All patients died' also reflects treatment options and supportive care in that era and setting — the same institution reported no benefit of Interfant protocols over BFM-based protocols in its infant population.

    Evidence that argues against this (1)
    • Phenotypic conversion out of the B lineage is far more common than 0.6% when looked for prospectively by flow cytometry, and in non-KMT2A genetic subtypes it is prognostically neutral — so the Rossi rate and the Rossi outcome are both specific to a severe, molecularly selected subset. Novakova M et al. 2021
  2. E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Myeloid relapses of MLL/AF4 leukemia share the oncogene fusion breakpoints with their matched lymphoid presentations, arise from differentiation stages ranging from immature progenitors to committed B-cell precursors, and are driven by substantial changes in chromatin accessibility and transcriptional rewiring including alternative splicing, with recurrent alteration of chromatin modifiers such as CHD4 whose perturbation alone induces myeloid gene expression in MLL/AF4+ cells.

    Epigenetic regulator genes direct lineage switching in MLL/AF4 leukemia · Tirtakusuma R, Szoltysek K, Milne P, Grinev VV, Ptasinska A, et al. · Blood 2022
    PMID 35839448DOI
    What this evidence cannot show

    The number of matched presentation/relapse pairs with full multi-omic data is small (lineage switch is rare), so the recurrence of CHD4 involvement is established on limited n. CHD4 perturbation inducing myeloid gene expression in cell models is a gain-of-phenotype experiment in vitro and does not by itself show that CHD4 loss is necessary for clinical switching. I read the abstract in full but not the figures, so the exact frequency of CHD4 alterations across the cohort should be read from the paper before being quoted as a number.

    Evidence that argues against this (1)
    • Shared fusion breakpoints establish common ancestry and not the direction of the transition: a myeloid relapse sharing the breakpoint is equally consistent with switching of the lymphoid clone and with outgrowth of a myeloid sibling present from the start. "Arising from differentiation stages ranging from immature progenitors to committed B-cell precursors" describes heterogeneity rather than a mechanism. Journal of Clinical Oncology 2025
  3. E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Integrating single-cell transcriptomes of 89 B-ALL samples with an atlas of normal human B-cell development shows that the subtypes prone to shift out of the B lineage (BCR::ABL1, KMT2A-R, DUX4-R) are enriched for multipotent progenitors, that this developmental stage exhibits CEBPA activation and retains myeloid potential, and that a derived multipotency score is independently associated with inferior survival.

    PMID 40579588DOI
    What this evidence cannot show

    Correlative mapping of leukemic cells onto a normal developmental reference; CEBPA 'activation' is inferred from expression/regulon analysis, not from a perturbation experiment in the patients' own cells. The multipotency score's association with survival is observational and not yet validated prospectively, and 89 samples spread over many B-ALL subtypes leaves few KMT2A-R cases per analysis.

    Evidence that argues against this (1)
    • Enrichment for multipotent progenitors in switch-prone subtypes is inferred by projecting single-cell transcriptomes onto a normal B-development atlas, and a leukaemia is not a point on a normal trajectory. Such projections assign blasts to whichever normal state they least mismatch, which will systematically place transcriptionally promiscuous blasts at immature positions. Oncofusion knowledge graph (internal analysis) 2026
  4. E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Sustained CD19 CAR-T pressure produces partial or complete lineage switch as a consistent resistance mechanism in murine pre-B ALL models depending on the underlying oncogenic driver, and deletion of Pax5 or Ebf1 recapitulates the lineage reprogramming seen under CAR pressure.

    CD19 CAR immune pressure induces B-precursor acute lymphoblastic leukaemia lineage switch exposing inherent leukaemic plasticity · Jacoby E, Nguyen SM, Fountaine TJ, Welp K, Gryder B, et al. · Nature Communications 2016
    PMID 27460500DOI
    What this evidence cannot show

    The causal Pax5/Ebf1 experiments are mouse genetics with engineered deletion, which models the endpoint rather than the mechanism of acquisition; no human KMT2A-r switch has been shown to acquire a PAX5 or EBF1 lesion at the time of switch. The murine ALL models are not KMT2A::AFF1 fusion models, so partner-specific biology is not captured.

    Evidence that argues against this (1)
    • Murine pre-B ALL under sustained CAR-T pressure, and the recapitulation by Pax5 or Ebf1 deletion is the informative part: it shows that removing a B-identity transcription factor produces the phenotype, which is a sufficiency result. Whether switching in patients proceeds through loss of those factors is a separate question. Journal of Clinical Oncology 2025
  5. E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In 33 clonally-confirmed lineage switches, 18 (54.5%) shared an 11q23/KMT2A fusion (t(4;11) in 11), the switched AML was monocytic/myelomonocytic in 21/33 (64%), paired NGS in 10 patients showed divergent clonal evolution in five and no clonal branching in the other five, the second leukemia had a more complex karyotype (three or more aberrations in 17/29 vs 10/30, P = 0.05), median survival after switch was 2.9 months, and patients without KMT2A fusions were older (46.1 vs 25.2 years, P = 0.03) with an antecedent chronic myeloid neoplasm in 6/15 (40%) whereas no KMT2A-fusion patient had one.

    Genetics and pathologic landscape of lineage switch of acute leukemia during therapy · Zhou T, Curry CV, Khanlari M, Shi M, Cui W, et al. · Blood Cancer Journal 2024
    PMID 38272888DOI
    What this evidence cannot show

    Retrospective multi-institutional case collection spanning 2003-2022 with a wide age range (0.1-83.2 years, median 34.6), so the adult and paediatric biology are pooled; only 10 of 33 patients had paired sequencing, so the 5-versus-5 split between divergent evolution and reprogramming rests on ten informative cases and targeted (not whole-genome) panels that could miss the subclonal variants needed to detect branching. Cases with MPAL as the first leukemia were excluded by design, which removes the most plastic presentations from the denominator.

    Evidence that argues against this (1)
    • Thirty-three clonally confirmed switches assembled retrospectively, and the informative finding is buried: paired sequencing in 10 patients showed DIVERGENT CLONAL EVOLUTION in five. That is half the sequenced cases behaving as separate clones rather than as one switching population, which the entry's framing does not accommodate. Journal of Clinical Oncology 2025
  6. E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    In the international Project EVOLVE analysis, 70 post-immunotherapy lineage switches from B-ALL comprised 53 (75.7%) to AML and 17 (24.3%) to B/myeloid MPAL or undifferentiated leukemia, KMT2A rearrangement was present in 45 (64.3%), clonal relatedness was confirmed in 64 (91.4%) with B-cell receptor gene rearrangements retained in 20 of 25 tested, switch emerged at a median of 1.5 months (range 0-36.5) with 81.4% within six months, and the reported incidence of post-immunotherapy lineage switch in KMT2A-r or infant ALL has ranged from 0% to 28.6% after CD19-directed CAR T cells across retrospective cohorts.

    Project EVOLVE: an international analysis of postimmunotherapy lineage switch, an emergent form of relapse in leukemia · Silbert SK, Rankin AW, Hoang CN, Semchenkova A, Myers RM, et al. · Blood 2025
    PMID 40193715DOI
    What this evidence cannot show

    A solicited international case series with no denominator: it can describe the phenotype, timing and outcome of lineage switch but cannot estimate its incidence, and the authors say so explicitly. Reporting is biased toward centres with sophisticated flow cytometry and molecular monitoring, and toward dramatic cases; 'lineage drift' cases show the boundary of the definition is soft.

    Evidence that argues against this (1)
    • Project EVOLVE assembles cases selected for having switched, so every proportion in it is conditional on switching and none of them is a rate. KMT2A rearrangement in 64.3% of switchers is compatible with a wide range of switch risks depending on how many KMT2A-rearranged patients received immunotherapy and did not switch. Journal of Clinical Oncology 2025
  7. E7moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    A blinatumomab-induced ALL-to-AML lineage switch in KMT2A-rearranged infant ALL spontaneously reverted, in a leukemia the authors describe as rapidly adapting to immunological stimuli and illustrating the high plasticity of this phenotype.

    PMID 29898879DOI
    What this evidence cannot show

    n = 1, and the full text is not open access — I verified the title, authors, journal, year, PMID and DOI and the published key-point sentence, but could not read how the reversion was documented (whether by flow cytometry alone or with molecular confirmation of clonal identity at both time points). Confidence is therefore capped at moderate and no numeric detail from this case is asserted anywhere in this node.

    Evidence that argues against this (1)
    • A single infant, and spontaneous reversion is where the competing explanation is hardest to exclude: two coexisting clones alternating as selection pressure is withdrawn produce the same observation as a reversible state transition. "Rapidly adapting to immunological stimuli" is the authors' interpretation. Journal of Clinical Oncology 2025
  8. E8moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    In a human CRISPR/Cas9 model introducing MLL::AF4 or MLL::AF9 with KMT2A breakpoints in intron 9 or 11 into cord-blood CD34+ cells from a single donor, transcriptomic profiling revealed breakpoint- and partner-specific expression patterns, and MLL(intron 9)::AF9 cells engrafted NSG mice and switched from a myeloid to a B-lymphoid identity in vivo.

    PMID 42001795DOI
    What this evidence cannot show

    An engineered model in cord-blood CD34+ cells rather than a patient-derived system; a single donor removes background confounding but also removes generalisability, and xenografts in NOD scid gamma mice have no adaptive immune system, so the in vivo switch here occurs without any immune selective pressure. The switch reported is myeloid-to-lymphoid, the opposite of the clinically dominant direction, so it demonstrates bidirectional plasticity rather than modelling the clinical event. Recent publication, limited independent replication.

    Evidence that argues against this (1)
    • A single cord-blood donor. Breakpoint- and partner-specific expression patterns from one genetic background cannot separate breakpoint effects from donor effects, and cord blood is neonatal while the disease is fetal in origin. Oncofusion knowledge graph (internal analysis) 2026
  9. E9moderate confidence 3 of 5: One strong study, or several smaller consistent ones.paper

    Multi-omic profiling of patient samples shows that lineage-switched AML cells with KMT2A rearrangement acquire monocytic myeloid-derived suppressor cell (M-MDSC)-like characteristics, with single-cell mass cytometry showing expansion of this population relative to lineage-consistent KMT2A-r AML, distinct immunoregulatory gene expression by single-cell transcriptomics, and confirmed immunosuppressive capacity against T cells in vitro.

    Multi-omics analysis identifies an M-MDSC-like immunosuppressive phenotype in lineage-switched AML with KMT2A rearrangement · Mikami T, Kato I, Nishimura A, Eguchi-Ishimae M, Kamitori T, et al. · Nature Communications 2025
    PMID 40858576DOI
    What this evidence cannot show

    Lineage-switched AML is rare, so the patient sample numbers behind the mass cytometry and single-cell transcriptomic comparisons are necessarily small and are not stated in the abstract; I did not read the methods, so I assert no n. The in vitro T-cell suppression assay establishes capacity, not that immunosuppression is what causes immunotherapy failure in vivo. Whether the M-MDSC-like state is a cause of switching or a consequence of adopting a monocytic program is not resolved.

    Evidence that argues against this (1)
    • M-MDSC-like characteristics are assigned by marker similarity in mass cytometry, and monocytic leukaemia cells share most of those markers with monocytic MDSCs by virtue of being monocytic. Calling the population MDSC-like imports an immunosuppressive function that the phenotype alone does not establish. Oncofusion knowledge graph (internal analysis) 2026
  10. E10strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper

    Among 726 consecutively diagnosed children with BCP-ALL, 8% experienced a switch to the monocytic lineage with loss of the B-cell immunophenotype including CD19; this correlated with PAX5-P80R (5/5), DUX4 rearrangement (30/41) and ZNF384 rearrangement (4/10) rather than KMT2A, caused underestimation of MRD by B-cell-oriented flow cytometry, and was not associated with higher relapse incidence or poorer prognosis on standard ALL treatment.

    DUX4r, ZNF384r and PAX5-P80R mutated B-cell precursor acute lymphoblastic leukemia frequently undergo monocytic switch · Novakova M, Zaliova M, Fiser K, Vakrmanova B, Slamova L, et al. · Haematologica 2021
    PMID 32646889DOI
    What this evidence cannot show

    This is monocytic switch detected by flow cytometry during early therapy, which is a milder and possibly distinct phenomenon from frank lineage switch to overt AML; the cohort is chemotherapy-treated, not immunotherapy-treated, so it does not test whether these subtypes would switch lethally under CD19-directed antigen pressure. It is included here precisely as the disconfirming case for a KMT2A-centric and uniformly fatal model of lineage switch.

    Evidence that argues against this (1)
    • This 8% is in BCP-ALL generally and the correlations the study reports are with PAX5-P80R, DUX4 and ZNF384 — explicitly NOT with KMT2A rearrangement. Citing it in a KMT2A lineage-switch entry imports a rate from a population in which the beat's lesion was not the driver. Oncofusion knowledge graph (internal analysis) 2026
Connected to