Ordering the fusions developmentally — which stage of blood formation each one hijacks, and at what age the child presents
also called developmental ruler · stage-of-origin ordering · age-of-onset atlas
Each leukaemia fusion catches a blood cell at a particular moment in its development, and that moment predicts the age at which the child gets sick. Fusions that strike the earliest, most primitive cells cause disease in babies; fusions that strike more grown-up cells cause disease in older children and teenagers.
The same thing, for a scientist
The fusion atlas becomes far more interpretable when ordered by the developmental stage of the cell of origin rather than by chromosome or partner. The ordering runs from fetal-liver haematopoietic stem cells and lympho-myeloid primed progenitors (KMT2A-r infant ALL, CBFA2T3::GLIS2 infant AMKL, NUP98-r) through multipotent bone-marrow HSC (BCR::ABL1/CML, RUNX1::RUNX1T1, CBFB::MYH11, PML::RARA committed to the promyelocyte) to progressively more mature B-lineage progenitors (ETV6::RUNX1 and TCF3::HLF at pro-B/early-B, TCF3::PBX1 at pre-B, DUX4-r in adolescents and young adults). Age at presentation tracks this ordering closely, and so does latency: the earliest lesions have the shortest latency and the most primitive lesions the worst prognosis.
The ordering, in tiers2 deeper
Per-fusion values live in each fusion node under data.developmental_order_rank, data.stage_of_hematopoiesis and data.age_peak. The tiers are:
Tier 1 — fetal HSC / lympho-myeloid primed progenitor, presents under 1 year. fusion-kmt2a-aff1 and the other KMT2A-r infant cases; fusion-cbfa2t3-glis2 (infant acute megakaryoblastic leukaemia); fusion-nup98-kdm5a (also infant/young-child AMKL-like). Extremely short latency, a CD10-negative pro-B or megakaryoblastic phenotype with lineage promiscuity, and the worst outcomes in the whole atlas.
Tier 2 — multipotent bone-marrow HSC, any age. fusion-bcr-abl1 as p210/CML arises in a CD34+CD38- multipotent stem cell, which is why it is a myeloproliferative stem-cell disease that can undergo lymphoid blast crisis and why tyrosine kinase inhibitors suppress rather than eradicate the leukaemic stem cell. fusion-nup98-nsd1 sits here in myeloid progenitors.
Tier 3 — committed myeloid progenitors. fusion-runx1-runx1t1 and fusion-cbfb-myh11 (core-binding-factor AML, older children and adults); fusion-pml-rara, uniquely blocked at a single defined stage, the promyelocyte, which is exactly why one differentiating agent works.
Tier 4 — fetal/neonatal B-lineage progenitor with long latency. fusion-etv6-runx1: the first hit occurs in utero, the preleukaemic clone persists for years, and disease appears at age 2-5 after a postnatal second event. fusion-tcf3-hlf sits nearby in pro-B but with a stem-cell-skewed, hybrid, drug-resistant state.
Tier 5 — later B-lineage progenitors, older children and adolescents. fusion-tcf3-pbx1 at the pre-B stage (cytoplasmic mu positive, median age around 7 years); fusion-dux4-igh, which requires an active B-lineage IGH enhancer and therefore cannot arise before B-lineage commitment, presenting in adolescents and young adults.
Why the in-utero evidence is the anchor
Two classic studies fix the early end of the ruler. Clonotypic fusion sequences, including MLL-AF4, were amplified from the neonatal blood spots of children who later developed leukaemia, proving the rearrangement predates birth 1. Concordant leukaemia in monozygotic twins sharing a single MLL-AF4 genomic sequence establishes a single prenatal cell of origin with intraplacental spread 2. For ETV6::RUNX1 the same backtracking works but with a completely different timeline: the fusion is present at birth and the child presents at 2-5 years, and a discordant twin pair showed the first hit is insufficient without a postnatal promotional event 3.
That contrast — same in-utero origin, radically different latency — is the strongest single argument that the stage of the cell of origin, not merely the timing of the break, determines the disease.
Three things the ordering predicts, and one it does not
It predicts age almost perfectly, which is why the ruler is useful clinically.
It predicts latency: Tier 1 lesions transform in months, Tier 4 lesions take years, because a fetal stem cell already has the self-renewal machinery the fusion needs whereas a committed B progenitor must acquire it.
It predicts prognosis better than mechanism class does. Tier 1 (KMT2A-r infant ALL, CBFA2T3::GLIS2) has the worst outcomes; Tier 5 (DUX4-r) has among the best. That cuts across mechanism class — DUX4-r and KMT2A-r are both 'chromatin-adjacent' lesions with opposite outcomes — so any comparative strategy that sorts only by mechanism will mis-rank clinical urgency.
It does not predict druggability. The two curable fusions sit in Tier 2 and Tier 3, the middle of the ruler, and their tractability came from retained protein modules, not from developmental stage. Stage tells you who gets sick and how badly; structure tells you whether a drug is possible.
The uncomfortable implication for the lab
If KMT2A::AFF1 leukaemia originates in a fetal-liver-like haematopoietic stem cell whose normal programme already includes high self-renewal and lineage plasticity, then part of the phenotype belongs to the cell rather than to the fusion. That predicts a ceiling on fusion-directed therapy: blocking the fusion program may release differentiation without removing the cell state that made the fusion sufficient in the first place, and it may explain why menin inhibitors give differentiation and remission but not eradication.
It also predicts that the right model system is a fetal or neonatal human progenitor, not an adult one — and a great deal of published KMT2A-r modelling uses adult murine bone marrow. That is a methodological warning the graph should carry prominently.
| ordering axis | developmental stage of the cell of origin, 1 = most primitive (fetal HSC) to 10 = most mature |
|---|---|
| what it does not predict | druggability — the two curable fusions sit in the middle tiers and their tractability came from retained protein modules, not from stage |
| methodological warning | Much KMT2A-r modelling uses adult murine bone marrow, whereas the disease originates in a fetal-like human progenitor; stage mismatch is a systematic source of irreproducibility. |
tiers — the numbers behind the chart
[
{
"tier": 1,
"stage": "fetal liver HSC / lympho-myeloid primed progenitor",
"age": "under 1 year",
"fusions": [
"fusion-kmt2a-aff1",
"fusion-cbfa2t3-glis2",
"fusion-nup98-kdm5a"
],
"latency": "months",
"prognosis_band": "worst"
},
{
"tier": 2,
"stage": "multipotent bone-marrow HSC (CD34+CD38-)",
"age": "any age; CML adult-predominant",
"fusions": [
"fusion-bcr-abl1",
"fusion-nup98-nsd1"
],
"latency": "variable",
"prognosis_band": "transformed by targeted therapy (BCR::ABL1) or very poor (NUP98::NSD1)"
},
{
"tier": 3,
"stage": "committed myeloid progenitor; promyelocyte for APL",
"age": "older children and adults",
"fusions": [
"fusion-runx1-runx1t1",
"fusion-cbfb-myh11",
"fusion-pml-rara"
],
"latency": "requires cooperating lesions",
"prognosis_band": "favourable to curable"
},
{
"tier": 4,
"stage": "fetal/neonatal pro-B or early-B progenitor with long postnatal latency",
"age": "2-5 years (ETV6::RUNX1); school-age and adolescent (TCF3::HLF)",
"fusions": [
"fusion-etv6-runx1",
"fusion-tcf3-hlf"
],
"latency": "years",
"prognosis_band": "excellent (ETV6::RUNX1) to dismal (TCF3::HLF)"
},
{
"tier": 5,
"stage": "pre-B and later B-lineage progenitors, requires active B-lineage enhancers",
"age": "median 7 years to adolescence/young adulthood",
"fusions": [
"fusion-tcf3-pbx1",
"fusion-dux4-igh"
],
"latency": "years",
"prognosis_band": "good to excellent"
}
]what the ordering predicts
[ "age at presentation", "latency from first hit to disease", "prognosis band, better than mechanism class does" ]
- If the cell state rather than the fusion supplies the self-renewal, fusion-directed therapy has a hard ceiling and residual disease will persist in a cell the drug never addressed.
- Fetal-stage origin means the relevant model system is a human fetal or neonatal progenitor, which is scarce, ethically constrained and rarely used, so the field may be systematically modelling the wrong cell.
- Ordering by stage highlights that some fusions need de-escalation rather than new drugs; a targeted-therapy programme that ignores this will expose good-prognosis children to unnecessary risk.
- If cell of origin is dose-dependent, then a stage ordering assembled from heterogeneous models is partly an ordering of experimental technique, and therapeutic inferences drawn from it (which differentiation state to target) inherit that artefact.
- Can the ordering be made quantitative by projecting each fusion subtype onto a single-cell reference atlas of human fetal and postnatal haematopoiesis, giving a continuous pseudotime coordinate instead of five tiers?
- How much of the KMT2A::AFF1 phenotype belongs to the fetal cell state rather than to the fusion, and would the same fusion in an adult HSC produce a milder disease?
- Does stage of origin, rather than fusion identity, determine menin-inhibitor sensitivity — and would that explain differences in response between infant ALL and older-child AML?
- For TCF3::HLF, why does a pro-B-stage lesion behave so much worse than the other pro-B-stage lesions, and is that a stem-cell-like state rather than a stage difference?
- Is the ordering a statement about the cell in which the translocation occurs, or about the cell in which the fusion can be tolerated and expanded? Chen 2008 separates those two and the entry does not.
- Does a haematopoietic stage ordering survive the niche result? CBFA2T3-GLIS2 transforms cord-blood HSPCs only with an endothelial coculture, and MLL-AF9 transforms GMPs or not depending on expression level. Both make the ordering a property of cell-plus-context rather than of developmental stage, and the experiment that separates them is a dose-controlled, niche-controlled transformation matrix across stages - which nobody has run.
- E1strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Clonotypic fusion-gene sequences, including MLL-AF4, can be amplified from neonatal blood spots of children who later developed leukaemia, showing the rearrangement arose before birth.
Backtracking leukemia to birth: identification of clonotypic gene fusion sequences in neonatal blood spots · Gale KB, Ford AM, Repp R, et al. · Proceedings of the National Academy of Sciences USA 1997What this evidence cannot show
Very small numbers, clone-specific PCR on archival Guthrie cards where negatives are uninformative, and a real contamination risk inherent to the method; the technique establishes prenatal origin but not the identity of the cell of origin.
Evidence that argues against this (1)
- Prenatal origin does not fix a cell of origin, and the ordering assumes it does. Under ENDOGENOUS Mll-AF9 regulation, LSK stem cells transform efficiently while granulocyte-macrophage progenitors are transformation-RESISTANT and produce no leukaemia; GMPs transform only at retroviral expression levels. The apparent stage of origin is a function of expression dose and model system, so an ordering built from published cell-of-origin assignments is ordering the experiments as much as the biology. Chen W et al. 2008
- E2strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Monozygotic twins with concordant infant leukaemia share a single identical MLL-AF4 genomic fusion sequence, indicating one prenatal cell of origin with intraplacental spread.
In utero rearrangements in the trithorax-related oncogene in infant leukaemias · Ford AM, et al. · Nature 1993What this evidence cannot show
A handful of infant cases including monozygotic twins; the concordance argument rests on n-of-few and on Southern-blot-era clonality evidence rather than on sequence-level breakpoint identity for every pair.
Evidence that argues against this (1)
- A shared identical genomic fusion sequence in concordant twins establishes a single prenatal clone and intraplacental spread; it does not establish that the fusion was the whole of the initiating event, which is the work this item does in a stage-ordering argument. Whole-genome, exome and transcriptome sequencing of 2,754 children with ALL found a median of four putative somatic driver alterations per sample, 376 putative driver genes varying in prevalence by subtype, and most samples carrying at least one rare gene alteration - including 70 putative drivers in categories a targeted exome panel does not call, such as ubiquitination, SUMOylation and non-coding transcripts. A twin pair is informative about timing and about clonal relationship, and is silent about how many other events were needed. Brady SW et al. 2022
- E3strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
ETV6::RUNX1 arises in utero but presents at 2-5 years, and twin studies show the prenatal fusion is insufficient for leukaemia without a postnatal promotional event — the same prenatal origin as KMT2A-r but a completely different latency.
Prenatal origin of acute lymphoblastic leukaemia in children · Wiemels JL, Cazzaniga G, Daniotti M, et al. · Lancet 1999What this evidence cannot show
Six informative cases of nine plus one twin pair; negative blood spots were uninformative rather than truly negative.
Evidence that argues against this (1)
- The 'prenatal fusion is insufficient without a postnatal event' inference is normally supported by the fusion being common in healthy newborns, and that support failed to replicate. Screening 1,417 cord blood samples isolated within 24 hours of birth by highly sensitive real-time RT-PCR, 14 were positive on first-run PCR at levels below 10^-5 and 9 survived specific hybridisation, but on repeat analysis of stored mRNA and of flow-sorted CD19+, CD8+ and double-negative fractions - a mean of 18 million cells sorted per sample - none was positive. The authors conclude the frequency of ETV6-RUNX1-positive cells is markedly lower than previous studies suggested. Backtracking the fusion from the blood spot of a child who did develop ALL remains sound; the population premise usually bolted onto it does not. Lausten-Thomsen U et al. 2010
- E4strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
CBFA2T3::GLIS2 defines an aggressive subtype of paediatric acute megakaryoblastic leukaemia occurring in very young children, anchoring the most primitive/earliest tier of the ordering on the myeloid side.
An Inv(16)(p13.3q24.3)-encoded CBFA2T3-GLIS2 fusion protein defines an aggressive subtype of pediatric acute megakaryoblastic leukemia · Gruber TA, Larson Gedman A, Zhang J, et al. · Cancer Cell 2012What this evidence cannot show
A discovery cohort of modest size; the precise developmental stage of the cell of origin is inferred from immunophenotype and expression signature rather than demonstrated by transplantation of defined progenitors.
Evidence that argues against this (2)
- The stage anchor for this row depends on an intrinsic property of the target cell, and the transformation experiment says the decisive variable is the niche. Expression of CBFA2T3-GLIS2 transforms human cord-blood haematopoietic stem and progenitor cells only in an ENDOTHELIAL CELL COCULTURE system; in that context it reproduces the transcriptome, morphology and immunophenotype of the human disease and gives aggressive leukaemia in xenografts. If the same starting cell transforms or does not depending on whether an endothelial niche is present, then 'this fusion anchors the earliest tier' is a statement about the cells plus their environment, and the ordering is not a property of developmental stage alone. Le Q et al. 2022
- The fusion is also not sufficient on its own. In mouse models that recapitulate the phenotypic and transcriptional signature of human CBFA2T3-GLIS2 AMKL, an activating Ras mutation of the kind seen in human disease increases penetrance and shortens latency, and the dominant oncogenic properties map to GLIS2, which triggers AMKL in cooperation with oncogenic Ras. A tier anchored on a fusion that needs a cooperating RAS lesion to become leukaemia is anchored on a two-hit event, and the age at presentation reflects when the second hit lands as much as when the first cell existed. Neault M et al. 2023
- E5strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
DUX4-rearranged B-ALL depends on a relocated immunoglobulin heavy-chain enhancer and occurs in older children, adolescents and young adults, anchoring the latest tier of the ordering.
Identification of ETV6-RUNX1-like and DUX4-rearranged subtypes in paediatric B-cell precursor acute lymphoblastic leukaemia · Lilljebjorn H, Henningsson R, Hyrenius-Wittsten A, et al. · Nature Communications 2016What this evidence cannot show
A single-region consecutive paediatric cohort of 195 cases; the adolescent and young-adult skew of DUX4-r is better established in later larger cohorts than in this paper.
Evidence that argues against this (1)
- The DUX4-rearranged subtype is not one tier. Integrated whole-genome, exome and transcriptome analysis of 2,754 paediatric ALL cases found that the DUX4-rearranged subtype - like the KMT2A-rearranged subtype - separates into distinct CEBPA/FLT3-expressing or NFATC4-expressing subgroups with potential clinical implications. A subtype that splits on expression state into subgroups with different clinical behaviour cannot be used as a single age-anchored tier without saying which subgroup is being placed, and neither subgroup has been assigned a developmental stage. Brady SW et al. 2022
- E6strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Partner distribution in KMT2A-rearranged disease differs between infant lymphoblastic and infant myeloid leukaemia, showing that stage and lineage of origin, not the KMT2A break alone, shape the resulting disease.
The KMT2A recombinome of acute leukemias in 2023. · Meyer C, Larghero P, Almeida Lopes B, et al. · Leukemia 2023What this evidence cannot show
A referral-based diagnostic cohort, so partner frequencies reflect referral patterns rather than a true population distribution.
Evidence that argues against this (1)
- Mutation-pattern evidence is also a weak basis for stage inference here. Infant KMT2A-rearranged ALL has one of the lowest somatic mutation frequencies of any sequenced cancer (mean 1.3 non-silent mutations in the predominant clone), yet 47% of cases carry kinase-PI3K-RAS activating mutations that are often SUBCLONAL and frequently LOST AT RELAPSE, while KMT2A-rearranged leukaemia in older children carries a mean of 6.5 mutations and frequent epigenetic-regulator mutations. A cross-sectional ordering cannot distinguish developmental stage from clonal turnover. Andersson AK et al. 2015
- E7strong confidence 4 of 5: Replicated in peer-reviewed work by more than one group.paper
Under endogenous regulation, Mll-AF9 transforms LSK stem cells efficiently while GMPs are transformation-resistant and cause no leukaemia; GMP transformation occurs only at retroviral expression dosage.
Malignant transformation initiated by Mll-AF9: gene dosage and critical target cells. · Chen W, Kumar AR, Hudson WA, et al. · Cancer Cell 2008What this evidence cannot show
Abstract only. Mouse knock-in versus retroviral comparison for one fusion (Mll-AF9); does not settle the human cell of origin, and does not address KMT2A::AFF1. It does show that published cell-of-origin assignments are dose- and model-dependent, which is enough to weaken any ordering built from them.
Evidence that argues against this (1)
- This is a direct disagreement in the literature, not a nuance, and the entry cites only one side. MLL-AF9 introduced into committed granulocyte macrophage progenitors generated leukaemia stem cells that transferred leukaemia to secondary recipients when as few as FOUR cells were transferred, and those LSC carried the immunophenotype and global expression profile of normal GMPs with a limited stem-cell self-renewal programme re-activated on top. So a committed progenitor can be the cell of origin for MLL-AF9 leukaemia. The reconciliation offered in the entry - that GMP transformation happens only at retroviral dosage - is a reconciliation, not an observation, and it turns the stage-ordering claim into a claim about expression level rather than about developmental stage. Which of the two is true decides whether a stage ordering exists at all for this row. Krivtsov AV et al. 2006
- E8established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.paper
Infant KMT2A-rearranged ALL carries a mean of 1.3 non-silent somatic mutations in the predominant clone, with kinase-PI3K-RAS mutations in 47% of cases that are often subclonal and frequently lost at relapse, while KMT2A-rearranged leukaemia in older children carries a mean of 6.5 mutations and frequent epigenetic-regulator mutations.
The landscape of somatic mutations in infant MLL-rearranged acute lymphoblastic leukemias. · Andersson AK, Ma J, Wang J, et al. · Nature Genetics 2015What this evidence cannot show
Abstract only. 65 infants (47 KMT2A-r) plus 20 older children; the age comparison is cross-sectional, which is precisely the limitation this counter-evidence is raising about the ordering itself.
Evidence that argues against this (1)
- The 1.3 figure is a floor set by the assay and the clonal threshold, not a property of the disease. The largest integrated paediatric ALL study, using whole-genome as well as exome and transcriptome sequencing across 2,754 patients, reports a median of four putative somatic driver alterations per sample despite a generally low mutation burden, and identifies 70 putative driver genes in classes an exome-based non-silent-mutation count does not reach - ubiquitination and SUMOylation pathway genes and non-coding transcripts among them - with most samples carrying at least one rare alteration. 'Mean 1.3 non-silent somatic mutations in the predominant clone' and 'median 4 driver alterations per sample' are not in conflict; they are different measurements, and only the first is being used here to argue that KMT2A-r leukaemia needs almost nothing else. Brady SW et al. 2022
stage ordering of fusions across the hierarchy
KMT2A::AFF1 sits at the pro-B arrest position in that ordering
Counter-evidence on inferring stage from mutation pattern (Andersson 2015, PMID 25730765): infant KMT2A-r ALL carries a mean of 1.3 non-silent mutations, with RAS-pathway lesions in 47% that are often subclonal and lost at relapse, against 6.5 in older children. A cross-sectional ordering cannot separate developmental stage from clonal turnover.
- related to →ETV6::RUNX1 (TEL-AML1), t(12;21)(p13;q22)
Replication failure in the healthy-newborn premise (Lausten-Thomsen 2010, PMID 20713965): 1417 cord bloods, zero positives after repeat analysis and flow sorting of ~18 million cells per sample.
- related to →CBFA2T3::GLIS2, inv(16)(p13.3q24.3)
The niche, not the stage (Le 2022, PMID 36136600): CBFA2T3-GLIS2 transforms human cord-blood HSPCs only in endothelial coculture; and (Neault 2023, PMID 37716355) it needs a cooperating activating Ras lesion, with the oncogenic activity mapping to GLIS2.
- related to →KMT2A::MLLT3 (MLL-AF9), t(9;11)(p21.3;q23.3)
Direct disagreement on cell of origin (Krivtsov 2006, PMID 16862118): MLL-AF9 in committed GMPs generated LSC that transferred leukaemia from four cells, with a GMP expression profile plus a re-activated self-renewal programme.
- ← related toKMT2A::MLLT3 (MLL-AF9), t(9;11)(p21.3;q23.3)
Unresolved cell-of-origin disagreement (Chen 2008, PMID 18455126 vs Krivtsov 2006, PMID 16862118): under endogenous regulation GMPs are transformation-resistant and LSK cells transform, with GMP transformation appearing only at retroviral dosage.
- ← related toKMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
Prenatal origin is not single-hit sufficiency (Brady 2022, PMID 36050548): a median of four putative driver alterations per sample across 2754 paediatric ALL cases, with 70 putative drivers in classes an exome non-silent count does not reach.
- ← related toCBFA2T3::GLIS2, inv(16)(p13.3q24.3)
Falsification (Huang 2026, PMID 42660135): an adult patient with concurrent BCR::ABL1 and CBFA2T3::GLIS2 retires the 'never identified in adults' claim; this entry's paediatric-exclusivity statement is withdrawn.